# SHAPING OF HUMAN IMMUNE SYSTEM AND METABOLIC PROCESSES BY VIRUSES AND MICROORGANISMS

EDITED BY : Marina I. Arleevskaya, Rustam Aminov, Wesley H. Brooks, Gayane Manukyan and Yves Renaudineau PUBLISHED IN : Frontiers in Microbiology and Frontiers in Immunology

#### Frontiers Copyright Statement

© Copyright 2007-2019 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.

The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.

Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.

Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.

As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.

All copyright, and all rights therein, are protected by national and international copyright laws.

The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use. ISSN 1664-8714 ISBN 978-2-88945-941-4 DOI 10.3389/978-2-88945-941-4

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# SHAPING OF HUMAN IMMUNE SYSTEM AND METABOLIC PROCESSES BY VIRUSES AND MICROORGANISMS

Topic Editors:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia Rustam Aminov, University of Aberdeen, United Kingdom Wesley H. Brooks, University of South Florida, United States Gayane Manukyan, Institute of Molecular Biology, Armenian National Academy of Sciences, Armenia Yves Renaudineau, INSERM U1227 Lymphocytes B et Autoimmunite (LBAI), France

Recent advances in the understanding of microbiota in health and diseases are presented in this special issue of *Frontiers in Immunology* and *Frontiers in Microbiology* as well as their impact on the immune system that can lead to the development of pathologies. Potential perspectives and biomarkers are also addressed.

We offer this Research Topic involving 64 articles and 501 authors to discuss recent advances regarding:

1. An overview of the human microbiota and its capacity to interact with the human immune system and metabolic processes,

2. New developments in understanding the immune system's strategies to respond to infections and escape strategies used by pathogens to counteract such responses,

3. The link between the microbiota and pathology in terms of autoimmunity, allergy, cancers and other diseases.

Citation: Arleevskaya, M. I., Aminov, R., Brooks, W. H., Manukyan, G., Renaudineau, Y., eds. (2019). Shaping of Human Immune System and Metabolic Processes by Viruses and Microorganisms. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-941-4

# Table of Contents

*10 Editorial: Shaping of Human Immune System and Metabolic Processes by Viruses and Microorganisms*

Marina I. Arleevskaya, Rustam Aminov, Wesley H. Brooks, Gayane Manukyan and Yves Renaudineau

#### SECTION

#### RELATIONSHIPS BETWEEN MICROBIOTA, VIRUSES AND THE HOST


Marco Toscano, Roberta De Grandi, Enzo Grossi and Lorenzo Drago

#### SECTION

#### OTHER MICROBIOTA


#### SECTION

#### INTERPLAY BETWEEN MICROBIOTA AND THE IMMUNE SYSTEM

*88 The Immune System Bridges the Gut Microbiota With Systemic Energy Homeostasis: Focus on TLRs, Mucosal Barrier, and SCFAs* Martina Spiljar, Doron Merkler and Mirko Trajkovski

*98 Fecal Microbiota Transplantation, Commensal* Escherichia coli *and*  Lactobacillus johnsonii *Strains Differentially Restore Intestinal and Systemic Adaptive Immune Cell Populations Following Broad-spectrum Antibiotic Treatment*

Ira Ekmekciu, Eliane von Klitzing, Christian Neumann, Petra Bacher, Alexander Scheffold, Stefan Bereswill and Markus M. Heimesaat


Angela M. Patterson, Imke E. Mulder, Anthony J. Travis, Annaig Lan, Nadine Cerf-Bensussan, Valerie Gaboriau-Routhiau, Karen Garden, Elizabeth Logan, Margaret I. Delday, Alistair G. P. Coutts, Edouard Monnais, Vanessa C. Ferraria, Ryo Inoue, George Grant and Rustam I. Aminov

*142 Free Fatty Acids Profiles are Related to Gut Microbiota Signatures and Short-Chain Fatty Acids*

Javier Rodríguez-Carrio, Nuria Salazar, Abelardo Margolles, Sonia González, Miguel Gueimonde, Clara G. de los Reyes-Gavilán and Ana Suárez


Wenkai Ren, Ranjith Rajendran, Yuanyuan Zhao, Bie Tan, Guoyao Wu, Fuller W. Bazer, Guoqiang Zhu, Yuanyi Peng, Xiaoshan Huang, Jinping Deng and Yulong Yin

*176 Variation of Carbohydrate-Active Enzyme Patterns in the Gut Microbiota of Italian Healthy Subjects and Type 2 Diabetes Patients*

Matteo Soverini, Silvia Turroni, Elena Biagi, Sara Quercia, Patrizia Brigidi, Marco Candela and Simone Rampelli

*184 Modulation of Immunological Pathways in Autistic and Neurotypical Lymphoblastoid Cell Lines by the Enteric Microbiome Metabolite Propionic Acid*

Richard E. Frye, Bistra Nankova, Sudeepa Bhattacharyya, Shannon Rose, Sirish C. Bennuri and Derrick F. MacFabe


Lorena Ruiz, Susana Delgado, Patricia Ruas-Madiedo, Borja Sánchez and Abelardo Margolles

*232 The Role of Lipoproteins in Mycoplasma-Mediated Immunomodulation* Alexei Christodoulides, Neha Gupta, Vahe Yacoubian, Neil Maithel, Jordan Parker and Theodoros Kelesidis

#### SECTION

#### IMMUNE SYSTEM: IMMUNE RESPONSE AND ALTERED IMMUNE RESPONSE DURING INFECTIONS


Nour Sherri, Noor Salloum, Carine Mouawad, Nathaline Haidar-Ahmad, Margret Shirinian and Elias A. Rahal

*267 Emergence of CD4+ and CD8+ Polyfunctional T Cell Responses Against Immunodominant Lytic and Latent EBV Antigens in Children With Primary EBV Infection*

Janice K. P. Lam, K. F. Hui, Raymond J. Ning, X. Q. Xu, K. H. Chan and Alan K. S. Chiang

*280 HHV-6A Infection of Endometrial Epithelial Cells Induces Increased Endometrial NK Cell-Mediated Cytotoxicity*

Elisabetta Caselli, Daria Bortolotti, Roberto Marci, Antonella Rotola, Valentina Gentili, Irene Soffritti, Maria D'Accolti, Giuseppe Lo Monte, Mariangela Sicolo, Isabel Barao, Dario Di Luca and Roberta Rizzo

*293 Age-Related Macular Degeneration: A Connection Between Human Herpes Virus-6A-Induced CD46 Downregulation and Complement Activation?*

Walter Fierz

#### SECTION

## ESCAPE STRATEGIES SUBSECTION

INNATE RESPONSE


Stéphane Rodriguez, Mikaël Roussel, Karin Tarte and Patricia Amé-Thomas

#### SECTION

#### ESCAPE STRATEGIES

#### SUBSECTION

miRNAs AND lncRNAs

*323 Herpesviral microRNAs in Cellular Metabolism and Immune Responses* Hyoji Kim, Hisashi Iizasa, Yuichi Kanehiro, Sintayehu Fekadu and Hironori Yoshiyama


Vandana Kaul, Kenneth I. Weinberg, Scott D. Boyd, Daniel Bernstein, Carlos O. Esquivel, Olivia M. Martinez and Sheri M. Krams

*352 Regulation of the Interferon Response by lncRNAs in HCV Infection* Saba Valadkhan and Puri Fortes

#### SECTION

## ESCAPE STRATEGIES SUBSECTION

#### OTHER ESCAPE STRATEGIES


Adriele Scopel-Guerra, Deiber Olivera-Severo, Fernanda Staniscuaski, Augusto F. Uberti, Natália Callai-Silva, Natália Jaeger, Bárbara N. Porto and Celia R. Carlini


Moffat Malisheni, Svetlana F. Khaiboullina, Albert A. Rizvanov, Noah Takah, Grant Murewanhema and Matthew Bates

#### SECTION

#### LINK WITH DISEASES SUBSECTION AUTOIMMUNITY

*419 Infectious Agents and Inflammation: The Role of Microbiota in Autoimmune Arthritis*

Andrea Picchianti-Diamanti, Maria M. Rosado and Raffaele D'Amelio

*428 Intestinal Microbiota Influences Non-intestinal Related Autoimmune Diseases*

Maria C. Opazo, Elizabeth M. Ortega-Rocha, Irenice Coronado-Arrázola, Laura C. Bonifaz, Helene Boudin, Michel Neunlist, Susan M. Bueno, Alexis M. Kalergis and Claudia A. Riedel

*448 Upregulation of Intestinal Barrier Function in Mice With DSS-Induced Colitis by a Defined Bacterial Consortium is Associated With Expansion of IL-17A Producing Gamma Delta T Cells*

Ming Li, Bing Wang, Xiaotong Sun, Yan Tang, Xiaoqing Wei, Biying Ge, Yawei Tang, Ying Deng, Chunyang He, Jieli Yuan and Xia Li

*462 Reduced Mass and Diversity of the Colonic Microbiome in Patients With Multiple Sclerosis and Their Improvement With Ketogenic Diet*

Alexander Swidsinski, Yvonne Dörffel, Vera Loening-Baucke, Christoph Gille, Önder Göktas, Anne Reißhauer, Jürgen Neuhaus, Karsten-Henrich Weylandt, Alexander Guschin and Markus Bock

*471 Associations Between Viral Infection History Symptoms, Granulocyte Reactive Oxygen Species Activity, and Active Rheumatoid Arthritis Disease in Untreated Women at Onset: Results From a Longitudinal Cohort Study of Tatarstan Women*

Marina I. Arleevskaya, Albina Z. Shafigullina, Yulia V. Filina, Julie Lemerle and Yves Renaudineau

*482 Association Between Systemic Lupus Erythematosus and Periodontitis: A Systematic Review and Meta-analysis*

Zoe Rutter-Locher, Toby O. Smith, Ian Giles and Nidhi Sofat

*490 Microbes and Viruses are Bugging the Gut in Celiac Disease. Are They Friends or Foes?*

Aaron Lerner, Marina Arleevskaya, Andreas Schmiedl and Torsten Matthias

#### SECTION

#### LINK WITH DISEASES SUBSECTION

ALLERGY AND CANCERS


Eduardo Mendes, Beatriz G. Acetturi, Andrew M. Thomas, Flaviano dos S. Martins, Amanda R. Crisma, Gilson Murata, Tárcio T. Braga, Niels O. S. Camâra, Adriana L. dos S. Franco, João C. Setubal, Willian R. Ribeiro, Claudete J. Valduga, Rui Curi, Emmanuel Dias-Neto, Wothan Tavares-de-Lima and Caroline M. Ferreira

*527 Multivariate Analysis as a Support for Diagnostic Flowcharts in Allergic Bronchopulmonary Aspergillosis: A Proof-of-Concept Study* Joana Vitte, Stéphane Ranque, Ania Carsin, Carine Gomez, Thomas Romain, Carole Cassagne, Marion Gouitaa, Mélisande Baravalle-Einaudi,

Nathalie Stremler-Le Bel, Martine Reynaud-Gaubert, Jean-Christophe Dubus, Jean-Louis Mège and Jean Gaudart


Melissa J. Blumenthal, Sylvia Ujma, Arieh A. Katz and Georgia Schäfer

### *555 Commentary: High Glucose Induces Reactivation of Latent Kaposi's Sarcoma-Associated Herpesvirus*

Fabrizio Angius, Maria A. Madeddu and Raffaello Pompei

*557 Pro-inflammatory State in Monoclonal Gammopathy of Undetermined Significance and in Multiple Myeloma is Characterized by Low Sialylation of Pathogen-Specific and Other Monoclonal Immunoglobulins* Adrien Bosseboeuf, Sophie Allain-Maillet, Nicolas Mennesson, Anne Tallet, Cédric Rossi, Laurent Garderet, Denis Caillot, Philippe Moreau, Eric Piver, François Girodon, Hélène Perreault, Sophie Brouard, Arnaud Nicot, Edith Bigot-Corbel, Sylvie Hermouet and Jean Harb

#### SECTION

## LINK WITH DISEASES SUBSECTION OTHER DISEASES

*574 Intestinal Microbiome Shifts, Dysbiosis, Inflammation, and Non-alcoholic Fatty Liver Disease*

Emma T. Saltzman, Talia Palacios, Michael Thomsen and Luis Vitetta


Lidia Sanchez-Alcoholado, Daniel Castellano-Castillo, Laura Jordán-Martínez, Isabel Moreno-Indias, Pilar Cardila-Cruz, Daniel Elena, Antonio J. Muñoz-Garcia, Maria I. Queipo-Ortuño and Manuel Jimenez-Navarro

*614 Microbiome-Derived Lipopolysaccharide Enriched in the Perinuclear Region of Alzheimer's Disease Brain*

Yuhai Zhao, Lin Cong, Vivian Jaber and Walter J. Lukiw

*620 Gut Dysbiosis and Adaptive Immune Response in Diet-induced Obesity vs. Systemic Inflammation*

Jana Pindjakova, Claudio Sartini, Oriana Lo Re, Francesca Rappa, Berengere Coupe, Benjamin Lelouvier, Valerio Pazienza and Manlio Vinciguerra

#### SECTION

#### NEW MODELS AND HYPOTHESIS


Anna Kłopot, Adriana Zakrzewska, Dorota Lecion, Joanna M. Majewska, Marek A. Harhala, Karolina Lahutta, Zuzanna Kaźmierczak, Łukasz Łaczmański, Marlena Kłak and Krystyna Dąbrowska

*671 Unveiling and Characterizing Early Bilateral Interactions Between Biofilm and the Mouse Innate Immune System*

Christiane Forestier, Elisabeth Billard, Geneviève Milon and Pascale Gueirard


Natalia N. Skvortsova and Levon G. Abrahamyan


Renée N. Douville and Avindra Nath

# Editorial: Shaping of Human Immune System and Metabolic Processes by Viruses and Microorganisms

Marina I. Arleevskaya<sup>1</sup> \*, Rustam Aminov <sup>2</sup> , Wesley H. Brooks <sup>3</sup> , Gayane Manukyan<sup>4</sup> and Yves Renaudineau1,5

<sup>1</sup> Central Research Laboratory, Kazan State Medical Academy, Kazan, Russia, <sup>2</sup> School of Medicine and Dentistry, University of Aberdeen, Aberdeen, United Kingdom, <sup>3</sup> Department of Chemistry, University of South Florida, Tampa, FL, United States, <sup>4</sup> Group of Molecular and Cellular Immunology, Institute of Molecular Biology, National Academy of Sciences, Yerevan, Armenia, <sup>5</sup> Laboratory of Immunology and Immunotherapy, INSERM U1227, University of Brest, Brest, France

Keywords: microbiota, immune system, autoimmunity, cancer, allergy, viruses

**Editorial on the Research Topic**

#### Edited by:

Juarez Antonio Simões Quaresma, Instituto Evandro Chagas, Brazil

#### Reviewed by:

Brian Gregory George Oliver, University of Technology Sydney, Australia

> \*Correspondence: Marina I. Arleevskaya marleev@mail.ru

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 24 February 2019 Accepted: 01 April 2019 Published: 17 April 2019

#### Citation:

Arleevskaya MI, Aminov R, Brooks WH, Manukyan G and Renaudineau Y (2019) Editorial: Shaping of Human Immune System and Metabolic Processes by Viruses and Microorganisms. Front. Microbiol. 10:816. doi: 10.3389/fmicb.2019.00816 **Shaping of Human Immune System and Metabolic Processes by Viruses and Microorganisms**

## INTRODUCTION

We are not alone in our bodies since we share them with a huge number of microorganisms. Such interactions represent a continuum, extending from mutualistic relationships, to commensal interactions and, at the end of the spectrum, development of human diseases (Lerner et al.). While substantial progress has been made in our understanding of the pathophysiology of infectious diseases, more subtle interactions exist with the microbiome and those interactions could promote or protect against the development of human disease according to the genetic/epigenetic susceptibility of the individual (**Figure 1**). By interfering with our bodies, modulating our immune system, controlling our metabolism, even our mood, microorganisms and viruses can actively contribute to the development of diseases that are major causes of mortality, and in particular by promoting cancers and autoimmune processes (**Figure 2**). At the same time, there is also the flip side of the coin as the indigenous microflora inhibits colonization by exogenous pathogens through both bacterial antagonism and host immune system stimulation. While and according to the hygiene hypothesis, a defective indigenous microflora and/or a lack of infectious agent exposure in childhood increase susceptibility to allergic diseases by suppressing the natural development of the immune system.

We offer this Research Topic involving 63 articles and 500 authors to discuss recent advances regarding:


## RELATIONSHIPS BETWEEN MICROBIOTA, VIRUSES AND THE HOST

#### Gut Microbiota

The gut microbiota results first from a tight equilibrium between the need for beneficial microbiota to ferment and digest incoming dietary material and the need to exclude competing pathogens. In addition, the gut microbiota depends on the individual's characteristics such as birth delivery mode, immune system activation state, host genetics and metabolism, and history of diseases; and to the influence of environmental factors, namely standards of hygiene, familial relationships, medication exposure including antibiotics, hospitalization after birth, lifestyle, cultural aspects, and contamination of food and water by fecal microbes; that all can alter transmission mechanisms.

What is the connection between the human microbiome and the soil microbial community? This obvious question is reviewed by Tasnim et al. Data analyses retrieved from published reports show that overlap between gut and soil and microbes is limited based on the observation that two phyla, Bacteriodetes and Firmicutes, dominate human fecal samples, whereas dominant phyla in soil samples are Proteobacteria and Verrucomicrobia. However, such a statement needs to be taken with caution since most soil and gut surveys were performed in North American cohorts, and it was not possible to determine whether this can be transposed to other geographical locations.

What is the impact of the human birth delivery mode on the gut microbiota? To answer this question, Stewart and coauthors have studied longitudinal development of the gut microbiome of preterm infants (24–31 weeks gestation) during the first 100 days of life following either cesarean or vaginal discharge (Stewart et al.). The main conclusion from their study is that there is no significant association of birth mode with the longitudinal alpha- or beta- diversity or composition of the microbiome. One possible explanation came from the fact that all infants were exposed to antibiotics at postpartum.

Toscano and colleagues have reviewed the importance of the human breast milk microbiota in the formation of the newborn's first gut microbiota during lactation which is a determining factor in the maturation of the immune system of newborns (Toscano et al.). Human breast milk microbiota is comprised of more than 200 different bacterial species and one key question is related to their origin as some microorganisms (Streptococcus spp. and Staphylococcus spp.) belong to the maternal skin or infant's oral cavity, suggesting a milk flow back into mammary ducts during lactation. Human breast milk contains also a great number of intestinal bacteria, which may spread from the maternal intestinal environment by a particular mechanism involving dendritic cells and neutrophils (Rodríguez and Wilson, 2014).

## Other Microbiota

Microbiota analysis is not restricted to the gut as Baker and coauthors summarized with regards to the current status of uterine microbiota (Baker et al.). The notion that a healthy uterine cavity is sterile is currently being revised since the uterine microbiota contains 100–10,000 times less bacteria than the vaginal microbiome. The most abundant uterine bacteria found consistently belong to the following phyla: Firmicutes, Bacteriodetes, Proteobacteria, and Actinobacteria. Within the Firmicutes, the genus Lactobacillus is the prominent one. Another emerging question regarding uterine microbiota in women is related to the transmission routes: (i) hematogenous spread through either an oral or gut route; (ii) ascension through the cervix; (iii) retrograde spread through fallopian tubes; and (iv) assisted reproductive technology-related procedures or insertion/removal of intrauterine devices. Such a list needs to be extended since Altamae suggests to add the seminal contribution in the uterine microbiota (Altmae).

What about human skin microbiota? Park and Lee discussed the protective role of the commensal skin and orogenital microbiota in protecting the host from chronic immunemediated inflammatory disease (Park and Lee). Indeed, commensal microbial species are effective for controlling (i) release of antimicrobial peptides by skin cells, (ii) the proinflammatory microbial sensor NOD2 (nucleotide-binding oligomerization domain-containing protein 2) pathway; and (iii) components of the complement system. In particular, Staphylococcus epidermidis as well as Corynebacterium pseudodiphtheriticum, Propionibacterium acnes, and Staphylococcus aureus impact in a "compartmentalized" manner IL-17A production and skin-resident Th17 cells (Pascal et al., 2018). Another example is Vitreoscilla filiformis, a Gramnegative bacterium, which induces dendritic cells to prime naive T cells to type 1 Treg cells after cutaneous exposure (Volz et al., 2014).

Regarding human oral microbiota, Vieira and colleagues discussed the impact of estrogen deficiency associated with menopause (Vieira et al.). Such an assertion is linked, on one hand, with the description of the estrogen receptor-beta in the oral mucosa and salivary glands (Välimaa et al., 2004), and, on the other hand, with the observation that menopause was associated in women with age-related hormonal changes in the exfoliated normal buccal mucosa. Specific bacterial species, such as Porphyromonas gingivalis and Tannerella forsythensis, were found to be important in the etiology of periodontitis in postmenopausal women. The presence or absence of estrogen is also suspected to alter the gut microbiota equilibrium and to promote intestinal permeability. Another unsuspected link between the gut microbiome and menopausal health is related to bone since, for instance, Lactobacillus reuteri treatment prevents post-antibiotic and post-menopausal bone loss by reducing the expression of two inflammatory cytokines (TNF-α and IL-1β), increasing osteoclastogenesis, and preventing barrier disruption (Britton et al., 2014; Schepper et al., 2019).

An investigation into the interplay between the oral cavity microbiome in humans with the host immune system has also been reported (Park and Lee; Vieira et al.). In particular it was demonstrated that P. gingivalis, a member of the phylum Bacteroidetes, manipulate the local host responses by preventing phagocytosis and rerouting the anti-microbial Toll-like receptor (TLR) pathway after acting on both TLR2 and C5a receptor (C5aR) (Maekawa et al., 2014). P. gingivalis inhibits the secretion of IL-8, lowering the number of neutrophils recruited to the site of inflammation. P. gingivalis is also known to maintain a hyper-inflammatory state by enhancing M1 macrophages and a TH1/TH17 immune response, which has been attributed in part to the ability to secrete the P.gingivalis-derived peptidyl arginine deiminase (PPAD) that opens up the possibility of converting arginine residues into citrulline residues. A loss of tolerance against citrullinated proteins can lead to the development of rheumatoid arthritis as reviewed by Sakkas and colleagues (Arleevskaya et al., 2016; Sakkas et al.).

As this editorial and the related special issue did not cover the entire microbiota spectrum, readers are invited to complete their overview with recent publications presenting the growing interest in the virome and parasitome in human conditions (Marzano et al., 2017; Mitchell and Glanville, 2018).

## Interplay Between Microbiota and the Immune System

Based on the observation that germ-free mice present an altered immunity with increased susceptibility to immunological diseases and metabolic alterations, Spiljar and coauthors reviewed the interconnection between gut microbiota, the immune system and systemic energy homeostasis (Spiljar et al.). The absence of gut microbiota leads (i) to the formation of isolated secondary lymphoid follicles and smaller Payer's patches; (ii) to decreased counts of immune cells; and (iii) to a reduced local production of immunoglobulin. Such an observation is linked to functional alterations in immune and intestinal epithelial cells, and in particular a down-regulation of the anti-microbial TLR pathway. At the same time germ-free mice show improved glucose and insulin tolerance, accompanied by reduced adiposity. The metabolic effects and the browning phenotype are mediated by the innate immune system and the shift from the pro-inflammatory M1 to the anti-inflammatory M2 macrophage polarization.

Another way to test the relationship of the microbiome on the immune system is to induce microbiota depletion by antibiotics and then explore the impact on the immune system after bacterial recolonization with E. coli, L. johnsonii or with fecal microbiota transplantation (FMT) as performed in mice by Ekmekciu et al. First, immune cell populations were decreased after antibiotic treatment but were completely or at least partially restored upon FMT or by recolonization with the respective bacterial species. In particular (i) L. johnsonii recolonization results in the highest CD4+ and CD8+ cell numbers in the small intestine and spleen; (ii) FMT most effectively increases the frequencies of Treg cells, dendritic cells and restores intestinal memory/effector T cell populations; (iii) E. coli recolonization increases the production of TH1/TH17 cytokines, particularly in the small intestine; and conversely (iv) L. johnsonii recolonization maintains colonic IL-10 production. In summary, FMT appears in this mouse model to be efficient in the restoration of antibiotic-induced collateral damage to the immune system. Similarly, the use of the environmental toxicant 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), an aryl hydrocarbon receptor (AhR), induces a shift in the gut microbiota by interacting with AhR ligand bacteria such as segmented filamentous bacteria, an immune activator, as presented by Stedtfeld et al.

Patterson and coauthors, using murine and in vitro models, have investigated the mechanisms of cross-talk between the host immune system and the anti-inflammatory human gut symbiont Roseburia hominis (Patterson et al.). In the bacterium, host gut colonization upregulated R. hominis genes involved in conjugation/mobilization, metabolism, motility, and chemotaxis. In the host cells, R. hominis colonization upregulated genes related to antimicrobial peptides, gut barrier function, TLR5-flagellin signaling, and Treg population expansion.

The Lactobacillus group is classically regarded as "beneficial" in humans, however this group is in a competitive relationship with other bacteria as reported by Rodríguez-Carrio and coauthors who have studied the impact of the main gut microbial populations (Akkermansia, Bacteroides sp, Bifidobacterium, Clostridium cluster XIVa, Lactobacillus sp, and Faecalibacterium) on fecal short-chain fatty acid (SCFA), serum free fatty acid (FFA) levels, and immune pro-inflammatory mediators (IFNγ, IL-6, and MCP-1) in healthy human adults (Rodriguez-Carrio et al.). From this study an opposing behavioral pattern between Akkermansia (negative relation) and Lactobacillus (positive relation) was demonstrated by the authors regarding their relation with FFA, SCFA, and pro-inflammatory IL-6 levels. The modulation of the immune response by lipid metabolites including SCFA is developed by Shibata et al., and a comprehensive analysis of SCFA in the gut microbiota based on the diet is presented by Soverini et al. Within SCFA, some have particular effects on the immune system such as propionic acid that modulates mitochondrial dysfunction with an impact on leukocyte antigen expression and immunoglobulin production as studied in silico by Frye et al.

Ilinskaya and coauthor's present arguments to consider that secreted compounds from two probiotics of the Bacilli class, genera Bacillus and Lactobacillus, can affect both the human microbiome composition and the immune system (Ilinskaya et al.). Indeed, these two probiotics are able to secrete compounds, referred to as the bacillary secretome, and which are effective in suppressing pathogenic bacteria and favor beneficial ones via competition for nutrients, especially for shared limited resource like iron, competitive attachment to the epithelium, formation of substrates for growth, production of waste products and antimicrobial compounds, and strengthening of the barrier function of the epithelium. Regarding their interaction with the immune system, the bacillary secretome controls key signaling pathways such as TLR, NF-κB, and MAPK pathways and, in turn, the emergence of pro-inflammatory cytokines is reduced for the benefit of anti-inflammatory cytokines.

Mu and coauthors reviewed the probiotic properties of Lactobacillus reuteri in humans (Mu et al.). First, L. reuteri produces antimicrobial molecules that are able to inhibit the colonization of pathogenic microbes and to remodel the commensal microbiota composition in the host. Second, L. reuteri can benefit the host immune system. For instance, some L. reuteri strains reduce the production of pro-inflammatory cytokines while promoting Treg cell development and function. Third, bearing the ability to strengthen the intestinal barrier, the colonization of L. reuteri decreases the microbial translocation from the gut lumen to the tissues.

Ruiz and coauthors discussed the immunomodulatory properties of Bifidobacteria and the mechanisms and molecular players underlying these processes (Ruiz et al.). Currently it is accepted that there is a critical role for bifidobacteria in the maturation of the host innate and acquired immune system from gestation to childhood. Contrarily, other bacteria such as Mycoplasma infections elicit inflammatory as well as immunomodulatory effects through surface lipoproteins as reviewed by Christodoulides et al.

## IMMUNE SYSTEM

## Immune Response and Altered Immune Response During Infections

Ghazarian and coauthors comprehensively characterized a population of mucosal-associated invariant T (MAIT) cells in humans, which are unconventional CD3+ CD161high T lymphocytes that recognize a bacterial vitamin B2 (riboflavin) precursor presented by the MHC-I related protein, MR1 (Ghazarian et al.; Shibata et al.). MAIT cells represent an abundant proportion of resident T cells in tissues (10–40%) and up to 10% of the circulating T cell pool. Upon TCRdependent recognition of microbial antigens, MAIT cells secrete inflammatory cytokines (IFNγ, TNF-α, IL-17, and sometimes IL-22) and perforin to induce cellular cytotoxicity against bacterially infected cells. Human MAIT cells can also be activated in vitro in a TCR-MR1 independent fashion during acute or chronic viral infections in response to cytokines such as IL-12, IL-18, IL-15, and/or interferon α/β.

Khaiboullina et al. have studied cytokine levels in patient sera samples collected during acute and late phases of hemorrhagic fever with renal syndrome (HFRS) and hantavirus pulmonary syndrome (HPS), two primary forms of Hantavirus infection (Khaiboullina et al.). The authors report a cytokine storm in both diseases and with particularly a robust TH1 and NK cell response in HPS.

The innate and adaptive immune responses to Epstein-Barr virus (EBV) are not fully understood and, due to the complexity of the mammalian immune system, Sherri and coauthors propose to use the insects Drosophila melanogaster in order to better characterize innate immune pathways that are activated in response to EBV DNA (Sherri et al.). Back to human, Lam and colleagues traced the dynamics of antigen-specific polyfunctional CD4 and CD8 T cell responses against lytic and latent antigens of EBV in children diagnosed to have primary symptomatic infectious mononucleosis and asymptomatic EBV infection (Lam et al.). Elevated viral loads, which declined steadily during a 12-month period from the time of diagnosis, is reported whilst a decrease in the magnitude of the CD8 cytotoxic T cell response with the greatest response being toward immunodominant epitopes in both lytic and latent proteins, is correlated to a steady decline in viral loads.

Starting from the observation that idiopathic infertile women present both a modified percentage of endometrial (e)NK cells and infection with the human herpes virus (HHV)-6A, Casseli and colleagues have explored the capacity of HHV-6 virus to induce cytotoxic eNK cells as such activation may have implications for embryo implantation and infertility (Caselli et al.). This hypothesis is supported by the observation that eNK cells infected with HHV-6 have an increased expression of chemokines (CCR2, CXCR3, and CX3CR1) and those endometrial epithelial cells up-regulated the corresponding ligands (MCP1, CXCL10, and CCL26). Another consequence of infection with HHV-6A is related to its impact on the complement pathway. Indeed, HHV-6A infects target cells by docking to the complement regulator CD46, which results in CD46 downregulation, a lack that can lead to hyperactivation of the complement pathway. This mechanism described in astrocytes may explain the connection reported between HHV-6A infection and age-related macular degeneration (AMD) (Fierz).

### Escape Strategies: Innate Response

Chen and coauthors discuss how the influenza A virus (IAV, Orthomyxoviridae family) has developed multiple strategies to escape from host immune surveillance for successful replication (Chen et al.). First, an exceptional variability of the most abundant surface glycoprotein of the virus, hemagglutinin (HA), delays the production of neutralizing antibodies by B cells. Second, neuraminidase (NA), the second most abundant surface glycoprotein, reduces infected cell recognition by NK cells through its capacity to cleave sialic acid (SA) moieties. Third, the nonstructural protein-1 (NS1) acts as an important interferon antagonist protein. Fourth, the viral M2 protein is able to block host autophagy by interfering with the TLR-interferon pathway. So, the host immune response to IAV infection comprises multiple intricate processes that coordinate together to play significant roles in the protection of the virus.

Rodriguez and coauthors have reviewed the impact of chronic viral infection on the control of the human T-cell dependent humoral immune response in secondary lymphoid organs (Rodriguez et al.). Such an impact varies according to the cellular partner implicated: B cells, follicular helper T cells (Tfh), and stromal cells with two main subsets: fibroblast reticular cells (FRC) involved in B and T cell recruitment and follicular dendritic cells (FDC) involved in the formation and maintenance of the germinal center. The first example is EBV that infects mature B cells and mimics BCR and CD40 signals in the absence of antigen by producing the latent membrane proteins LMP1/2. As a consequence EBV drives B cell proliferation in a germinal center independent way that leads to memory B cells infected by EBV that are not recognized by the immune system. The second example is HIV that infects CD4 T cells and proliferates predominantly in Tfh that become defective in driving an effective anti-HIV humoral response. The third example is related to Ebola, Lassa and Marburg Viruses plus the lymphocytic choriomeningitis virus that destroy FRC leading to lymph node disorganization and a defective humoral and cellular immune response. The fourth example is arboviruses that infect and kill FDC thus affecting the capacity of the germinal center to produce protective antibodies and inducing a transitory cellular immuno-depression.

## Escape Strategies: miRNAs and lncRNAs

Human herpes xeno-miRNAs regulate host cell proliferation, differentiation, apoptosis, and the cell cycle to establish latent viral infection or, contrarily, to produce viral reactivation progeny as discussed by Kim et al. When secreted, xenomiRNAs result in additional stimulation to the activation of the pro-inflammatory intracellular signaling and to the repression of various immune responses that can shift host cells to malignant transformation.

Rizzo and coauthors in an original research article have explored the impact of HHV-6 infection on the host miRNAs in NK cells (Rizzo et al.). To this end, a human NK cell line (NK-92) was infected with HHV-6A/B in vitro. HHV-6A/B induced significant modification in miRNA expression which are known for their role in inflammation, and NK cell development, maturation and effector functions (miR-146, miR-155, miR-181, miR-223). Similarly Kaul and coauthors have performed a comprehensive analysis of differentially expressed host peripheral blood miRNAs in the early stage of uncomplicated EBV infection in patients with acute infectious mononucleosis (Kaul et al.). They identified 215 differentially regulated miRNAs at the most acute stage of infection. Interferon signaling, T and B cell signaling and antigen presentation were the top pathways influenced by the miRNAs associated with this acute EBV infection. Altogether, this supports the idea that herpes viruses, HHV-6 and EBV, have developed an escape strategy by controlling host miRNA expression.

During the interferon response, several long non-coding RNAs (lncRNAs) are induced to control positively or negatively the expression of interferon-stimulated genes (ISGs) in order to stop the interferon response and to return to a normal state. This negative regulatory mechanism can be subverted for the virus's benefit by viruses such as hepatitis C virus (HCV) as reviewed by Valadkhan and Fortes.

## Other Escape Strategies

Zhang and co-authors reviewed another mechanism of interference used by members of the Flaviviridae family, HCV and dengue virus (DENV) (Zhang et al.). Since viruses lack the appropriate machinery to conduct their own lipid synthesis, they have evolved a means to interfere with the cellular homeostasis of lipid droplets in the endoplasmic reticulum. The authors also discuss the possibility of targeting the host lipid droplet metabolism as antiviral strategies. Besides allowing Helicobacter pylori survival within the gastric mucosa, H. pylori urease has an important pro-inflammatory effect on both neutrophils via ROS production and platelet aggregation as proposed by Scoppel-Guerra and colleagues (Scopel-Guerra et al.), as well as on angiogenesis as supported by Olivera-Severo et al.

Malisheni and colleagues have conducted meta-analysis to clarify clinical efficacy, safety, and immunogenicity of a Live Attenuated Tetravalent Dengue vaccine (CYD-TDV) in children (Malisheni et al.). The analysis demonstrated that CYD-TDV is rather effective against viral serotypes DENV4, DENV3, and DENV1, and sufficiently less effective against DENV2 serotype. The low vaccine efficacy could be due to the fact that the circulating DENV2 virus acquires mutations in the E protein hence becoming different from the one included in the CYD-TDV vaccine.

### LINK WITH DISEASES

#### Autoimmunity

Disbalanced gut microbiota is suspected of influencing a large panel of autoimmune diseases as reviewed by Opazo and Picchianti-Diamant (Picchianti-Diamanti et al.; Opazo et al.). In the case of inflammatory bowel disease (IBD), one way to improve the disease is to reduce gut permeability by restoring commensal gut microbiota. To this end, Li and coauthors have used the colitis-mice model to test the suitability of 10 strain bacterial consortium transplantation (BCT) (Li et al.). It was demonstrated that BCT resulted in the reduction of gut permeability and improvement of intestinal dysbiosis as well as in the elevation of IL-17A producing γδT cells in colonic lamina propria of mice. In addition, the expansion of γδT17 cells was related to Bifidobacterium sp and Bacillus sp in a TLR2 dependent pathway. Another way to control gut dysbiosis is to use a high-fat, adequate-protein, low-carbohydrate diet, namely a ketogenic diet, as reported in patients with multiple sclerosis by Swidsinski et al.

Arleevskaya and coauthors have reported an increased peak of infectious episodes at early stages of rheumatoid arthritis (eRA) development and in their first-degree relatives (FDR) when compared to healthy women without RA in their family history (Arleevskaya et al.; Arleevskaya et al., 2018). A history of herpes simplex virus (HSV) episodes and reactivation is associated with inflammation and disease activity (Arleevskaya et al.; Larionova et al., 2019), while an elevated incidence of anti-CCP2 autoantibody characterized eRA patients with a history of viral upper respiratory tract infection symptoms. Regarding the innate immune system, granulocyte reactive oxygen species (ROS) activity is altered in eRA patients, associated with viral symptoms including HSV exacerbation/recurrence, and such defects are positively correlated with disease activity. In addition to the well establish association between periodontitis due to P. gingivallis and RA, a meta-analysis conducted by Rutter-Locher and collaborators supports also an association with systemic lupus erythematosus, another systemic autoimmune disease (Rutter-Locher et al.).

#### Allergy and Cancers

As described for microbiota, commensal viruses control innate and adaptive immune defenses and an imbalance of the virome is thought to trigger the development of allergic diseases, like asthma which is an allergic airway inflammation. As a consequence, restoring the virome represents an attractive strategy to control allergies as proposed by Freer and colleagues based on the use of Anelloviridae, a recently discovered virus family and major component of the respiratory virome acquired early in life (Freer et al.). Another way is to control asthma by modulating the gut microbiota and this was effective when using Bifidobacterium longumin in ovariectomized allergic mice with albumin as reported by Mendes et al., This prophylactic effect of B. longumin involves fecal acetate production and, in turn, Treg induction. Type I hypersensitivity reactions can be seen in bronchial asthma and the diagnosis of allergic bronchopulmonary aspergillosis can be improved based on the use of anti-Aspergillus fumigatus (Af) IgE, IgG, and IgG4 biomarkers as proposed by Vitte et al.

The development of cancer can be promoted by infections through different mechanisms. First, an altered oral and gut microbiota can directly affect the inflammatory and carcinogenic response of the host as presented by Klimesova et al. Second and as presented by Blumenthal et al., a defective immune response to cancer can be promoted by the synergic effect of infectious agents like the human immunodeficiency virus (HIV) and an altered metabolic process such as type 2 diabetes. Hyperglycemia and type 2 diabetes contribution to lymphoma reactivation is also commented on by Angius et al. Third, an enhanced humoral response to infectious agents based on the characteristics of the monoclonal immunoglobulin associated with multiple myeloma development as proposed by Bosseboeuf et al. Monoclonal IgG purified from multiple myeloma patients are able to target pathogens known to cause latent infection; they are desialylated to better exert their anti-infectious effect; and they promote an inflammatory cytokine response including cytokines important for the anti-microbial response.

#### Other Diseases

Dysbalanced human gut microbiota is suspected of promoting inflammatory diseases and one proposed mechanisms is that the microbiota itself or its metabolism are able to influence gut permeability and, in turn, to promote endotoxin translocation into the blood circulation. The first example is provided by Saltzman and colleagues who discussed the composition of the gut microbiome and its biological consequences in regard to liver homeostasis and liver inflammatory diseases (Saltzman et al.), the impact of long-term intake of fructose is also discussed by Lambertz et al. The second example is proposed by Sanchez-Alcoholado and colleagues as the production of trimethylamine N-oxide (TMAO) from the gut microbiota promotes the development of cardiovascular diseases and impacts the production of the anti-inflammatory cytokine IL-10 and Treg development (Sanchez-Alcoholado et al.). The third example by Zhao and colleagues is related to the observation that a major source of pro-inflammatory signals in Alzheimer's disease (AD), at the brain neocortex and hippocampus, arises from pro-inflammatory neurotoxic bacterial lipopolysaccharide (LPS) secreted from Gram negative bacilli of the human gastrointestinal (GI) tract (Zhao et al.).

Host and microbiome share similar or identical nutritional substrates and generate common metabolic products, thus a metabolic cross-talk exists with profound implications regarding

(i) the pathogenesis of an infection as presented by Ren et al.; (ii) the immune system as discussed previously; and (iii) the risk to develop a metabolic syndrome according to Pindjakova and colleagues when using a non-inflammatory diet-mimicking healthy obesity or a pro-inflammatory and atherogenic paigen diet to fed mice (Pindjakova et al.).

#### NEW MODELS AND HYPOTHESIS

In order to better understand the cross-talk between microbiota and the immune system in complex systems, such as the gut or skin, new models are proposed based on the use of genetic and epigenetic in silico approaches as performed by Li et al., the use of bacteriophages to selectively target a species as claimed by Kłopot et al., or in vivo models based on the use of microinjections in the ear pinna dermis in transgenic mice with fluorescent immune cells as elegantly developed by Forestier et al.

Finally, in a review article, Lerner and colleagues address an important issue regarding fundamental genetic processes in the commensal gut microbiota, horizontal gene transfer (HGT), and how they may affect human health (Lerner et al.) or contribute to the emergence of hyper-virulent and drug resistant pathogens as reported by Khaertynov et al. Another important question is related to the putative impact of viruses on the nucleolus as such disruption can lead to generation of autoantigens as discussed by Brooks et al. (2010), Brooks and Renaudineau (2015), and Brooks. Finally, Douville and Nath in their review discussed a particular aspect of human immunodeficiency virus (HIV) associated neurological conditions, such as HIV encephalitis and HIV-associated neurocognitive disorders, mediated by

#### REFERENCES


neuronal human endogenous retrovirus-K (HERV-K) expression (Douville and Nath). This reinforces the contribution of HERV elements in human diseases (Renaudineau et al., 2005; Fali et al., 2014; Le Dantec et al., 2015).

#### CONCLUSIONS

In conclusion, articles in this Research Topic made a very significant contribution to our understanding of the role played by environmental factors, dysbiotic conditions, and infections in triggering diseases. Since this is a rapidly expanding area of research, many other factors contributing to the onset of these diseases are not covered here. We are confident, however, that further studies will expand the list as well as bring a better understanding of mechanisms involved in the onset of autoimmune and autoinflammatory diseases.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### FUNDING

This study was supported by research funding from the Russian Science Foundation (No. 17-15-01099).

#### ACKNOWLEDGMENTS

We are thankful to Simone Forest and Genevieve Michel for secretarial assistance.

multi-reactive anti-herpes IgM profile is associated with disease activity. Clin. Immunol. 200, 19–23. doi: 10.1016/j.clim.2019.01.004


loss by reducing intestinal dysbiosis and preventing barrier disruption. J. Bone Miner. Res. doi: 10.1002/jbmr.3635. [Epub ahead of print].


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Arleevskaya, Aminov, Brooks, Manukyan and Renaudineau. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Linking the Gut Microbial Ecosystem with the Environment: Does Gut Health Depend on Where We Live?

Nishat Tasnim, Nijiati Abulizi, Jason Pither, Miranda M. Hart\* † and Deanna L. Gibson\* †

Department of Biology, The Irving K. Barber School of Arts and Sciences, University of British Columbia, Kelowna, BC, Canada

Global comparisons reveal a decrease in gut microbiota diversity attributed to Western diets, lifestyle practices such as caesarian section, antibiotic use and formula-feeding of infants, and sanitation of the living environment. While gut microbial diversity is decreasing, the prevalence of chronic inflammatory diseases such as inflammatory bowel disease, diabetes, obesity, allergies and asthma is on the rise in Westernized societies. Since the immune system development is influenced by microbial components, early microbial colonization may be a key factor in determining disease susceptibility patterns later in life. Evidence indicates that the gut microbiota is vertically transmitted from the mother and this affects offspring immunity. However, the role of the external environment in gut microbiome and immune development is poorly understood. Studies show that growing up in microbe-rich environments, such as traditional farms, can have protective health effects on children. These health-effects may be ablated due to changes in the human lifestyle, diet, living environment and environmental biodiversity as a result of urbanization. Importantly, if early-life exposure to environmental microbes increases gut microbiota diversity by influencing patterns of gut microbial assembly, then soil biodiversity loss due to land-use changes such as urbanization could be a public health threat. Here, we summarize key questions in environmental health research and discuss some of the challenges that have hindered progress toward a better understanding of the role of the environment on gut microbiome development.

Keywords: gut microbiome, immunity, environment, human health, immune tolerance, microbial colonization, biodiversity, microbe-rich environments

## INTRODUCTION

Human health is closely linked to the diverse set of microorganisms in the intestine collectively known as the gut microbiota (Hooper and Gordon, 2001). This population of microorganisms and their genetic potential, or the gut microbiome, has been linked to human metabolism, intestinal homeostasis, immune development (Lynch and Pedersen, 2016), and brain processes and behavior (Mayer et al., 2015). A stable and diverse gut microbiota, optimal for maintaining health, produces metabolites that fuel physiological and metabolic processes. The gut microbiota also tunes local and systemic immune responses to confer protective immunity against pathogens while simultaneously maintaining immune tolerance toward commensals (Cerf-Bensussan and Gaboriau-Routhiau, 2010). Other functions of the gut microbiota include fermentation of indigestible dietary

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Michael Kogut, Agricultural Research Service (USDA), United States Maria Carmen Collado, Instituto de Agroquímica y Tecnología de Alimentos (CSIC), Spain

#### \*Correspondence:

Miranda M. Hart miranda.hart@ubc.ca Deanna L. Gibson deanna.gibson@ubc.ca

†Co-senior authors.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 31 July 2017 Accepted: 21 September 2017 Published: 06 October 2017

#### Citation:

Tasnim N, Abulizi N, Pither J, Hart MM and Gibson DL (2017) Linking the Gut Microbial Ecosystem with the Environment: Does Gut Health Depend on Where We Live? Front. Microbiol. 8:1935. doi: 10.3389/fmicb.2017.01935

components (Flint et al., 2012), breakdown of environmental pollutants and pharmaceuticals (Claus et al., 2017), and pathogen competitive exclusion (Kamada et al., 2013). Alterations to the gut microbiota, known as dysbiosis, can disrupt these essential health-promoting services and are associated with gastrointestinal, cardiovascular, autoimmune and metabolic diseases (Carding et al., 2015). Therefore, the gut microbiome is a microbial ecosystem that operates much like a microbial organ that functions to promote health and prevent disease.

We are only beginning to understand the ecological processes that lead to the growth and development of a stable and diverse gut microbiome that promotes host-health. The gut microbiota is a diverse ecosystem comprised of bacteria, archaea, fungi and viruses including a diverse bacteriophage community (Manrique et al., 2016). Bacteria dominate the microbiota in abundance and diversity, with commensal members from seven phyla (Firmicutes, Bacteriodetes, Actinobacteria, Fusobacteria, Proteobacteria, Verrucomicrobia, and Cyanobacteria), the majority of which are uncultivated and novel phylotypes (Eckburg et al., 2005). Members of the microbiota can be permanent "residents," transmitted through close contact between individuals, or transient "hitchhikers" from ingested food, water and various components of the environment (Ley et al., 2006; Harmsen and de Goffau, 2016). These transmission routes are important for establishing and maintaining microbial diversity in the gut (Browne et al., 2017). The mechanism of transmission can determine the pattern of colonization which shapes the gut microbial community of the host, but these patterns of transmission are poorly understood. Colonization that leads to the establishment of a stable and diverse adult gut microbiome lays the foundation for a homeostatic host-microbial relationship maintained by balanced immune responses. Colonizing gut microbes provide signals known as microbe-associated molecular patterns (MAMPs) that affect the maturation of the immune system and gut associated lymphoid tissue (GALT) (Wopereis et al., 2014). The development of the GALT is associated with bacterial activation of Toll-like receptors (TLRs) and downstream signaling pathways involved in maintaining host-microbial homeostasis, regulated through cytokines and chemokines (Hooper et al., 2015). Germ free animals have defects in the development of GALT, as well as cellular defects such as decrease in the number of lymphocytes, and molecular immune deficiencies such as reduced antibody production (Round and Mazmanian, 2009; Torrazza and Neu, 2011). Thus, colonization of the gut by microbes is not only important for the development of gut tissue, but also for the establishment of immune tolerance.

Gut bacterial community assembly begins pre-birth (Blaser and Dominguez-Bello, 2016), but rapid colonization takes place at birth and continues for the first 3 years of life (Lozupone et al., 2013). Two key factors that could influence the successful transmission of beneficial gut microbes to the infant are the mother and the external environment. Various studies that have sampled infant fecal microbiota have revealed that early gut microbial settlers that colonize the gut are derived from maternal vaginal, fecal, milk, mouth and skin microbiota during both gestation and birth through vertical transmission, and from the environment through horizontal transmission (Inoue and Ushida, 2003). Therefore, the infant gut microbiome is transmitted from a gut microbial species pool, comprised of gut symbionts from both the mother and the environment (**Figure 1**). The effect of the environment on the diversity and richness of the human gut bacterial species pool and gut microbiota transmission has yet to be explored. If the transmission of gut microbes is primarily parentchild, then environmental factors such as standards of hygiene, contamination of food and water by fecal microbes, delivery mode and hospitalization after birth can alter transmission mechanisms. On the other hand, if colonization patterns and gut microbiota diversity is linked to transmission of microbes from the external environment, then additional factors such as place of birth, geography, urban vs. rural living environment may also alter colonization of the gut microbiota affecting the human health. In this review, we discuss what is known about the role of environmental factors on the gut microbiota composition, diversity and assembly, identify major research challenges for research aiming to elucidate gut microbiota transmission patterns, and make suggestions for future studies that integrate the gut microbiome with environmental health research.

FIGURE 1 | Local microbial community assembly of the infant gut microbiota depends on dispersal from a bacterial source pool. This bacterial source pool is comprised of both maternal microbes, transmitted vertically, and environmental microbes, transmitted horizontally. The development of the local community is shaped primarily by host selection, based on interactions between host and bacterial cells.

## INFLUENCE OF ENVIRONMENT ON VARIATIONS IN GUT MICROBIOTA DIVERSITY

The composition and diversity of gut microbiota varies between individuals. Under germ free conditions, gut microbiota transplantation experiments between model organisms such as zebrafish and mice have shown that gut microbiota composition is host-specific (Rawls et al., 2006). In humans, many other factors contribute to variation, such as diet, host genetics and metabolism, familial relationships, culture (Dominguez-Bello and Blaser, 2011), and demographics (Lozupone et al., 2013). According to global surveys of fecal microbiota from healthy populations, variation between individuals in richness of gut microbiota is largely explained by age, ethnicity (Huttenhower et al., 2012), geography (Torrazza and Neu, 2011), medication exposure, blood parameters, bowel, diet, health, anthropometrics and lifestyle (Falony et al., 2016). Of particular interest is the observation that healthy adults from rural societies such as Papua New Guinea (Martínez et al., 2015), Amerindia and Malawi (Clemente et al., 2015), and hunter-gatherers from Tanzania and Amazon (Schnorr et al., 2014) have higher gut bacterial species richness compared to urban populations in Italy and US. Similarly, children (between ages 1 to 5) from rural communities have more diverse gut microbiotas compared to children from Western populations (De Filippo et al., 2010). These host-specific differences in gut microbiota may arise from distinct selective pressures within the host gut habitat including genetics and diet but also may be due, at least in part, to their unique environments.

## ROLE OF THE ENVIRONMENT ON GUT COMMUNITY ASSEMBLY AND IMMUNOREGULATION

The role of the environment in the assembly of the gut microbiota has yet to be elucidated, although there is good reason to believe they are linked. Urbanization leads to changes in living conditions such as increased sanitation and antibiotic use (Popkin, 1999), separation from the outdoors (Turner et al., 2004), and poor land management practices that may reduce soil microbial biodiversity (Wall et al., 2015). Accordingly, studies show that infants born via caesarian section have altered colonization patterns and lower total gut microbiota diversity (Biasucci et al., 2010), and individuals who grow up in city environments have a less diverse gut microbiome (Sjögren et al., 2009). Further, urbanites are more prone to inflammatory disorders like diabetes and multiple sclerosis (Kay, 2000) as well as allergic diseases such as asthma (Rook, 2012) during both infancy and adulthood (Garn and Renz, 2007). Although host genetics may in large part determine the composition of the adult gut microbiome, it has been shown that alien microbes from diverse habitats like soil can colonize the germ-free gut (Seedorf et al., 2014). Therefore, horizontal transmission of environmental microbes may be contributing commensal microbes to the gut ecosystem, altering patterns of colonization to increase variation in gut microbiota diversity.

Early-life exposure to microbe-rich environments may be beneficial for human health by increasing the gut bacterial species pool. The "microbial old friends" hypothesis, posits microbe-rich environments are a source of beneficial microbes that promote gut microbiota diversity (Zhou et al., 2015) reducing inflammatory disease risk (Rook et al., 2013). Indeed, growing up in microbe-rich environments, like traditional farms, result in healthier children (Mosca et al., 2016). Therefore, the prevalence of inflammatory disorders may be higher in modern cities because of reduced exposure to beneficial microbes from the environment, such as microbes from house dust or zoonotic microbes from animals. Indeed, exposure to household pets has been shown to alter the infant gut microbiota and reduce allergic disease (Tun et al., 2017). Reduced exposure to pathogenic microorganisms, largely as a result of modern hygienic practices, can also result in defective immunoregulation (Garn and Renz, 2007). The "hygiene hypothesis" makes the argument that infectious stressors are particularly important during early childhood (Wills-karp et al., 2001; Garn and Renz, 2007) and is supported by epidemiological studies showing rural children have reduced asthma (Ege et al., 2011), hay fever (Strachan, 1989) and ectopic eczema (Isolauri et al., 2000). Such allergic diseases are chronic inflammatory disorders caused by a decrease in immune tolerance (Garn and Renz, 2007). Decrease in tolerance is associated with a decrease in Treg cells expressing the transcription factor forkhead box P3 (FOXP3+ Treg cells) (Simon et al., 2015). FOXP3+ Treg cells produce anti-inflammatory cytokines such as interleukin 10 (IL-10) and transforming growth factor-β (TGF-β) which help to suppress exacerbating inflammatory responses and balance CD4+ helper T (Th) Th1 and Th2 cells. In allergic diseases, cytokine stimulation of naïve T cells from IL-4, IL-5 and IL-13 tilt the balance of Th cells toward the Th2 phenotype (Kay, 2000). In infants, there may be a normal Th2 bias observed in both mice (1–3 weeks old) and humans (0–2 years old) (Marchant and Goldman, 2005; Dowling and Levy, 2014). As the infant ages, the Th2 skew is balanced by Th1 responses and induced memory responses through mucosal-associated invariant T cells and interleukin-8 (CXCL8) secreting naïve T cells (Simon et al., 2015). In contrast, allergic infants have a persistent Th2 phenotype, resulting in long term Th2-skewed immunity (Barrios et al., 1996). Therefore, early life exposure to a broad range of immunoregulation-inducing commensal and pathogenic environmental microorganisms can provide a Th1 stimulus, conferring protection against immune disorders.

What is it about urban environments that reduces healthy gut microbiome functioning? Both "old friends" and the "hygiene hypothesis," are contingent on microbial biodiversity. Urban development leading to the loss of local habitats and biodiversity may be detrimental to human health by depleting or otherwise altering the reservoirs of environmental microbes including bacteria, fungi and viruses that may play a role in gut microbiotamediated immune health. The "biodiversity hypothesis" posits that clinical diseases, caused by poor microbiome, immune dysfunction and inflammation, are linked to biodiversity loss (Anderson et al., 2013). Biodiversity loss due to industrialization is associated with adverse health effects, including inflammatory diseases (Haahtela et al., 2013). Environmental biodiversity and immune function have been linked in epidemiological studies, which show individuals living in built environments have lower diversity of microbiota and higher allergic disposition (Wardle et al., 2004). The World Allergy Organization has proposed that loss of biodiversity is linked to loss of microbial diversity, resulting in microbial deprivation and ultimately, inflammatory disorders (Haahtela et al., 2013). This proposal extends the "old friends" and hygiene hypothesis to include environmental biodiversity as being important in the development of the immune system and gut microbiome (von Hertzen et al., 2011). A biodiverse environment that is microbe-rich may promote the development of healthy gut microbiota and lower disease risk.

Extending the "biodiversity hypothesis" to include soil biodiversity has the potential to provide more insight into the role of the environment and gut mediated immune health. Soils contain a dynamic reservoir of biodiversity (Torsvik and Øvreås, 2002) and this diversity is essential for maintaining biogeochemical processes and ecosystem functioning (Wardle et al., 2004). In this way, soil biodiversity provides benefits to human health indirectly through suppression of soil-borne pathogens, provision of clean air, water and food, and exposure to immunoregulation-inducing soil microorganisms (Wall et al., 2015). Although unknown, we ask if there is a direct link between soil microbial diversity and human health? Certainly, soil microbial diversity varies in taxonomic composition between biomes (Fierer et al., 2012b), physical and chemical gradients (Fierer et al., 2012a; Lauber et al., 2013), and anthropogenic activity (Ramirez et al., 2010). Whether it is species richness that is important, or the composition of key taxa has not been determined. There is some indirect evidence that soil biodiversity and human microbiota are interrelated (Hanski et al., 2012), to provide "natural immunity" (von Hertzen et al., 2011). Further, exposure to soil microbes has been experimentally shown to increase gut microbiota diversity (Zhou et al., 2015). There is also some evidence to suggest that exposure to possible soil pathogens could contribute to immune tolerance (Wall et al., 2015). However, little is known about the impact of soil exposure on gut microbiota transmission and colonization patterns in humans.

#### EXPLORATION OF THE CONNECTIONS BETWEEN SOIL MICROBIAL COMMUNITIES AND GUT MICROBIAL COMMUNITIES

For soil biodiversity to be relevant to human health requires microbes from local soil to be transmitted horizontally to humans and then established in the gut. If so, then people exposed to similar soil microbial communities should have more similar gut communities. We analyzed soil and gut studies from publicly available datasets on Qiita (http://qiita.microbio.me), to investigate the link between soil and gut microbial diversity, in terms of richness, diversity and species identity. We combined OTU (operational taxonomic unit) tables from 7 gut studies (n= 2,497 human fecal samples) and 4 soil studies (n = 1,123 soil samples) that used 16 s amplicon sequencing to study bacterial communities (**Figures 2A,B**). The human fecal samples were collected from 14 countries, although the vast majority of samples were from USA (n = 1,062), Malawi (n= 1,042) and Venezuela (n = 99). These samples were collected from a range of ages (0– 77 years). A small proportion of adult humans were diagnosed with obesity, atherosclerosis (n = 52). Soil studies were from 17 different locations and most samples were from North America (n = 1062). Soil samples ranged from wetland to garden soil from tundra to tropical biomes.

## US Studies Dominate

Although our dataset was diverse, we lacked sufficient data to explore global variation in soil-gut microbiota. At the time of analysis, there were 244 studies on Qiita of which we picked four large-scale soil and seven gut studies to pool into a combined dataset (**Figure 2A**). Most studies on soil and human bacterial communities were located in the US. Future efforts should survey populations from different countries and physiographic regions to provide global geographical gut microbial datasets.

## There Is Little Overlap between Soil and Gut Microbes at Lower Taxonomic Levels

We performed downstream analysis to compare the relative proportion of bacterial phyla in human gut and soil samples (**Figure 2B**). We visualized the OTUs in human gut and soil samples using taxa summary plots. Samples were grouped and averaged by sample type (gut or soil) and taxonomic composition was summarized on multiple taxonomic levels (e.g., phylum, order, etc.) (Navas-Molina et al., 2013). We found that human fecal samples were dominated by Bacteriodetes and Firmicutes phyla, whereas soil samples were dominated by Proteobacteria and Verrucomicrobia. These differences in taxonomic composition between soil and gut samples were also consistent at lower taxonomic levels (see **Figure 2B** table).

## Study Effects Account for Variation

Differences in DNA extraction protocol, primer selection, sequencing platform and sequence analysis pipelines introduce bias to datasets known as study effects. To evaluate the influence of study effects, we pooled all human gut studies from a single investigator (Rob Knight, University of California), and excluded all studies outside the US resulting in four studies. We tested for study effects (n = 935) by considering research group (**Figure 2C**) and primer subfragment (**Figure 2C**). We found that primer target region or research group contributed to strong study-based clustering, similar to clustering patterns found in other meta-analyses of the human microbiota (Lozupone et al., 2013). Our results indicate that soil and gut bacterial communities have few overlapping taxa, but because most gut and soil studies survey North American cohorts, we were not able to determine whether local soil microbial communities influence the composition of gut microbial communities of individuals from different geographical locations.

samples were taken from a variety of habitats, including wetlands, garden soils, tundra, and tropical biomes. (B) Relative proportion of bacterial phyla in human feces (n = 2,497) and soil (n = 1,123) samples combined from seven gut and four soil studies in QIITA show little overlap of bacterial taxa. (C) Non-metric multidimensional scaling (NMDS) ordination plot of Bray-Curtis community dissimilarities on OTUs from 16 s gene sequences from four US gut studies conducted by the same principal investigator (Rob Knight) (2D stress value = 0.16). Samples show clustering according to study center (n = 935) (plot on left) as well as primer choice (plot on right) where V2 subfragment (n = 35) or V4 subfragment (n = 900) is targeted. Symbols represent individual fecal samples.

## CHALLENGES FOR STUDYING ENVIRONMENT-GUT MICROBIOTA INTERACTIONS IN HEALTH AND DISEASE

Global surveys on the relationships between environment, gut microbiota and inflammation are yet to be explored, such as how traditional diets consumed in a region may contribute to the gut microbial community or how local soil influences diversity of the gut microbiota of the population through horizontal transmission. The mechanism of horizontal transmission of environmental microbes, whether inhalation, ingestion or cutaneous, also remains to be elucidated. The rate of urbanization and soil degradation may be related to changes in the composition of the gut microbiota, such as an increase abundance of bacterial indicators of dysbiosis such as Proteobacteria (Shin et al., 2015). We cannot begin to understand the link between soil microbial diversity and gut microbial assembly until studies adopt standardized collection, extraction and sample preparation procedures with complete and transparent metadata reporting and appropriate analysis platforms. Some additional challenges for study design and analysis of environment-gut studies are outlined below.

## Challenge 1: Quantifying Microbe-Richness and Diversity of the Environment

To link gut microbiota to environmental microbial diversity, it will be important for future studies to develop standardized methods that reliably reflect microbial biodiversity in the environment. In addition to microbial diversity of the direct environment (home, air, soil, water, etc.), biodiversity of the surrounding environment should be estimated by recording information about the landscape, including land use type and predominant vegetation structure, abiotic factors such as climatic factors and information about the biodiversity of resident communities (i.e., plants, animals, etc.; as described by Hanski et al., 2012). Given the logistical challenges associated with such efforts, we recommend choosing sampling locations strategically in relation to desired environmental (Metzger et al., 2013) and other characteristics. This approach will help elucidate associations between microbial exposure, environmental biodiversity, and gut microbiota assembly.

Together, these parameters will help elucidate direct effects of microbial exposure and environmental biodiversity on gut microbiota assembly.

### Challenge 2: Collection, Storage and Analysis of Host-Microbiome-Environment Interactions

Developing analysis tools and platforms that are able to store and analyze large datasets will be critical to link gut microbiota assembly to external factors. Currently, limitations in sample collection, processing and storage (Gorzelak et al., 2015), as well as systems of reporting, study design, sample size, variation in demographics and statistical approaches prevent cross study comparisons (Hunter, 2005). The two publiclyavailable platforms for microbiome-environment studies are Qiita and SourceTracker (Knights et al., 2011), yet these have

## REFERENCES


had little uptake by the community as a whole. To fully understand demographic factors in gut microbial assembly, this will need to be a globally coordinated effort. The utility of the NIH Human Microbiome Project (http://www.hmpdacc. org/) could be enhanced by including protocols and repositories for environmental biodiversity (microbial and otherwise). Applications such as SourceTracker could then be easily used to investigate source-sink dynamics of the microbiota, to investigate microbiome-exposure interactions on the ecology of the microbiome. Once the challenge of data collection, handling and analysis are met, microbiome changes can be used as biomarkers to indicate individual health and disease outcome (Segata et al., 2011).

## CONCLUSION

The study of environmental influences on gut microbiota structure and function is especially pertinent because the human living environment is becoming rapidly urbanized. Such drastic changes to the human environment may interrupt the healthy development of the microbiota and increase risk of inflammatory diseases. Moving forward, we must incorporate gut microbiota surveys into a broader framework of environmental exposure, for a thorough understanding of how ecosystem processes contribute to gut microbiota development, and affect the quality of human health.

### AUTHOR CONTRIBUTIONS

NT: data collection and analysis; statistical analysis; intellectual design; writing and editing of the manuscript. NA: data collection and analysis; statistical analysis. JP: data analysis; statistical analysis oversight; editing of the manuscript. MH: data analysis; intellectual design; writing and editing of the manuscript; funding support. DG: intellectual design; writing and editing of the manuscript; funding support.

## FUNDING

JP and MH are funded through grants from the Natural Sciences and Engineering Research Council. DG is funded through grants from the Natural Sciences and Engineering Research Council and Crohn's and Colitis Canada.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Tasnim, Abulizi, Pither, Hart and Gibson. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Cesarean or Vaginal Birth Does Not Impact the Longitudinal Development of the Gut Microbiome in a Cohort of Exclusively Preterm Infants

Christopher J. Stewart<sup>1</sup> \*, Nicholas D. Embleton<sup>2</sup> , Elizabeth Clements<sup>3</sup> , Pamela N. Luna<sup>4</sup> , Daniel P. Smith<sup>1</sup> , Tatiana Y. Fofanova<sup>1</sup> , Andrew Nelson<sup>5</sup> , Gillian Taylor<sup>3</sup> , Caroline H. Orr<sup>3</sup> , Joseph F. Petrosino<sup>1</sup> , Janet E. Berrington<sup>2</sup> and Stephen P. Cummings<sup>3</sup>

<sup>1</sup> Alkek Center for Metagenomics and Microbiome Research, Department of Molecular Virology and Microbiology, Baylor College of Medicine, Houston, TX, United States, <sup>2</sup> Newcastle Neonatal Service, Royal Victoria Infirmary, Newcastle upon Tyne, United Kingdom, <sup>3</sup> School of Science and Engineering, Teesside University, Middlesbrough, United Kingdom, <sup>4</sup> Department of Statistics, Rice University, Houston, TX, United States, <sup>5</sup> Faculty of Health and Life Sciences, Northumbria University, Newcastle upon Tyne, United Kingdom

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Larry J. Dishaw, University of South Florida St. Petersburg, United States Misty Good, Washington University in St. Louis, United States

\*Correspondence:

Christopher J. Stewart christopher.stewart@bcm.edu

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 25 April 2017 Accepted: 22 May 2017 Published: 06 June 2017

#### Citation:

Stewart CJ, Embleton ND, Clements E, Luna PN, Smith DP, Fofanova TY, Nelson A, Taylor G, Orr CH, Petrosino JF, Berrington JE and Cummings SP (2017) Cesarean or Vaginal Birth Does Not Impact the Longitudinal Development of the Gut Microbiome in a Cohort of Exclusively Preterm Infants. Front. Microbiol. 8:1008. doi: 10.3389/fmicb.2017.01008 The short and long-term impact of birth mode on the developing gut microbiome in neonates has potential implications for the health of infants. In term infants, the microbiome immediately following birth across multiple body sites corresponds to birth mode, with increased Bacteroides in vaginally delivered infants. We aimed to determine the impact of birth mode of the preterm gut microbiome over the first 100 days of life and following neonatal intensive care unit (NICU) discharge. In total, 867 stool samples from 46 preterm infants (21 cesarean and 25 vaginal), median gestational age 27 weeks, were sequenced (V4 region 16S rRNA gene, Illumina MiSeq). Of these, 776 samples passed quality filtering and were included in the analysis. The overall longitudinal alpha-diversity and within infant beta-diversity was comparable between cesarean and vaginally delivered infants. Vaginally delivered infants kept significantly more OTUs from 2 months of life and following NICU discharge, but OTUs lost, gained, and regained were not different based on birth mode. Furthermore, the temporal progression of dominant genera was comparable between birth modes and no significant difference was found for any genera following adjustment for covariates. Lastly, preterm gut community types (PGCTs) showed some moderate differences in very early life, but progressed toward a comparable pattern by week 5. No PGCT was significantly associated with cesarean or vaginal birth. Unlike term infants, birth mode was not significantly associated with changes in microbial diversity, composition, specific taxa, or overall microbial development in preterm infants. This may result from the dominating effects of NICU exposures including the universal use of antibiotics immediately following birth and/or the lack of Bacteroides colonizing preterm infants.

#### Keywords: birth mode, cesarean, vaginal, gut microbiome, preterm infants, 16S rRNA sequencing

**Abbreviations:** LOS, late onset sepsis; NEC, necrotising enterocolitis; NICU, neonatal intensive care unit; OTU, operational taxonomic unit; PCoA, principal coordinates analysis; PGCT, preterm gut community type.

## INTRODUCTION

fmicb-08-01008 June 1, 2017 Time: 17:26 # 2

Immediately following birth, a neonate encounters large numbers of viable microbes. Despite emerging evidence suggesting the potential for prenatal exposure to microorganisms during the fetal stages (Jiménez et al., 2008; Aagaard et al., 2014; Collado et al., 2016), the main colonisation event occurs at birth, where for term infants birth mode shapes what microbes are passed from the mother to the offspring (Aagaard et al., 2016). Within the first 24 h of life, the microbiome of multiple distinct sites across the neonate reflect the route of delivery, with vaginally delivered neonates harboring vaginally derived organisms (typically Lactobacillus) and cesarean delivered neonates harboring skin-like microbes (typically increased Staphylococcus) (Dominguez-Bello et al., 2010). In the subsequent days and weeks following birth, the microbes colonizing different body sites begin to show more distinction (Chu et al., 2017). In infants delivered at term, the impact of birth mode has been studied in several longitudinal studies, most showing infants delivered vaginally have increased Bacteroides throughout the 1st year of life (Azad et al., 2013; Jakobsson et al., 2014; Bäckhed et al., 2015; Bokulich et al., 2016; Yassour et al., 2016). Contrary to these reports, no difference in the overall microbial community or in specific taxa between vaginal and cesarean infants at 6 weeks of life has also been reported (Chu et al., 2017). Differences between cohorts and methods (e.g., sequencing depth) may account for these discrepancies.

In preterm infants, studies directly exploring associations between birth mode and the temporal microbiome are lacking, although evidence suggests other factors such as feeding practices, postnatal age, and diseases like sepsis or necrotizing enterocolitis (NEC) likely have a bigger association with the microbiome (Stewart et al., 2013a, 2015b, 2016; La Rosa et al., 2014; Cong et al., 2016; Hill et al., 2017). A recent meta-analysis in preterm infants found reduced Bacteroides and increased Firmicutes in cesarean infants (Pammi et al., 2017). Infants delivered by cesarean have increased risks of later life obesity (Yuan et al., 2016), allergy (Roduit et al., 2009), and asthma (Thavagnanam et al., 2008; Sevelsted et al., 2014). Notably, while the microbiome has been suggested to be involved in the pathobiology of these diseases, direct causality has not been demonstrated.

In the current study we combined our existing publically available datasets (Abdulkadir et al., 2016; Stewart et al., 2016) and our previously unpublished data to directly explore how the birth mode impacts the temporal development of preterm infants while on the NICU and following discharge.

#### MATERIALS AND METHODS

#### Ethics Statement

Ethical approval was obtained from the County Durham and Tees Valley Research Ethics Committee. Parental written informed consent was given.

## Participants and Study Design

The study design, setting, participants, and methods of data collection have been reported previously (Stewart et al., 2012, 2013b; Abdulkadir et al., 2016). Briefly, all infants were cared for in a single NICU with standardized feeding, antibiotic and antifungal guidelines. Due to a change in NICU practice in 2013 infants born after this routinely received the probiotic Infloran <sup>R</sup> (Bifidobacterium bifidum-ATCC15696 and Lactobacillus acidophilus-NCIMB701748) soon after initial introduction of feeds, where half an Infloran capsule was given twice daily (125 mg b.d. at 10 9 organisms per dose). Infants contributing a minimum of seven samples in their first 100 days were included in the study.

#### 16S rRNA Gene Bacterial Profiling

Nucleic acid extraction was carried out on 100 mg of stool using the PowerLyzerTM PowerSoil <sup>R</sup> DNA Isolation Kit (MoBio, CA, United States) in accordance with the manufacturer's instructions. The V4 region of the 16S rRNA gene was amplified by PCR using barcoded Illumina adapter-containing primers 515F and 806R (Caporaso et al., 2012) and sequenced on the MiSeq platform (Illumina; San Diego, CA, United States) by NU-OMICS using the 2 × 250 bp paired-end protocol yielding pair-end reads that overlap almost completely. Sequencing read pairs were demultiplexed based on the unique molecular barcodes, and reads were merged using USEARCH v7.0.1090 (Edgar, 2010). Merging allowed zero mismatches and a minimum overlap of 50 bases, and merged reads were trimmed at the first base with a Q ≤ 5. In addition, a quality filter was applied to the resulting merged reads and those containing above 0.05% expected errors were discarded. Sequences were stepwise clustered into OTUs at a similarity cutoff value of 97% using the UPARSE algorithm (Edgar, 2013). Chimeras were removed using USEARCH v7.0.1090. OTUs were determined by mapping the centroids to the SILVA database (Quast et al., 2013) containing only the 16S rDNA V4 region to determine taxonomies. A custom script constructed a rarefied OTU table (rarefaction was performed at only one sequence depth) from the output files generated in the previous two steps for downstream analyses. We utilized multiple quality control measures, including the use of non-template controls during microbial DNA extraction and 16S rRNA gene amplification. Resulting OTU tables were rarified to 4300 reads per sample.

#### Bioinformatic and Statistical Analysis

Data analysis was conducted in R version 3.3 using ggplot2 (R Core Team, 2014). Alpha diversity analyses, specifically observed OTUs and Shannon diversity, are presented between infants. Weighted and unweighted UniFrac distances (Lozupone et al., 2011) were performed within infants based on consecutive samples. The number of OTUs kept (retained from one sample to the next), OTUs lost (present in the previous but not current sample), OTUs regained (any OTU detected in the current sample and previously detected within the infant, but not the preceding sample), and new OTUs gained (OTU detected in an

infant for the first time) was performed within infants based on consecutive samples.

Inferred metabolic capacity of the bacterial community was determined by Tax4Fun (Asshauer et al., 2015). FishTaco was then performed at the pathway level using genomic content inference to determine which species attenuated and drove the significantly altered functions (Manor and Borenstein, 2017).

Preterm gut community types were determined based on a publically available script for linear mixed-effects modeling, medoid-based clustering, and Markov chain modeling (DiGiulio et al., 2015). Weighted UniFrac (Lozupone et al., 2011) was used to calculate the distance between all samples and this was denoised by extraction of the most significant PCoA eigenvectors before applying the PAM algorithm. Gap statistic was used to determine the appropriate number of clusters based on the section of the plot where the curve markedly flattens (i.e., the elbow phenomenon).

Cross-sectional analyses were performed at discrete time points (1, 3, 5, and 8 weeks of age) to overcome bias of repeated measures in longitudinal analyses. At a given time point, samples within ±10 week days included, where the sample nearest to the specific week were chosen, giving preference to samples collected prior to the time point. Significance of categorical variables were determined using the non-parametric Mann– Whitney test for two category comparisons or the Kruskal– Wallis test when comparing three or more categories (Kruskal and Wallis, 1952). All P-values were adjusted for multiple comparisons using the false discovery rate (FDR) algorithm (Benjamini and Hochberg, 1995). Linear regression models adjusted for age (day of life), sex, birth weight, gestational age, diagnosis of NEC and/or LOS, receipt of expressed breast milk, and antibiotics (< or >10 days of antibiotics while on the NICU).

#### RESULTS

#### Study Population

In total, 63,592,993 sequencing reads from 867 samples (46 patients) mapped to the database, with 776 (760 NICU and 16 post discharge) samples remaining in the analysis following rarefication to 4300 reads. Of these 46 infants, 25 infants were delivered vaginally and 21 infants were delivery by cesarean section (**Table 1**). All infants received at least 48 h of antibiotics immediately following birth and the total number of days of antibiotic treatment while on the NICU was comparable between birth modes. Due to a change in unit practice, probiotics were administered in 4/21 of the cesarean infants born after 2013 and no vaginal infants received probiotics (P = 0.07). Use of probiotics was associated with increased relative abundance in only the genera contained within the probiotic (Infloran; Bifidobacterium and Lactobacillus), but not in other taxa (Supplementary Figure 1). Nonetheless, receipt of probiotics (along with other important covariates – see Materials and Methods) were also adjusted for in all significance testing of taxa.

TABLE 1 | Characteristics of 46 preterm infants born by either vaginal or cesarean delivery.


<sup>∗</sup>Samples remaining after QC and rarefaction.

NICU, neonatal intensive care unit; NEC, necrotising enterocolitis; LOS, late onset sepsis.

### Longitudinal Alpha and Beta Diversity Was Comparable between Cesarean and Vaginally Delivered Infants

The number of observed OTUs of samples decreased initially following birth, then increased from 11 OTUs on day 12 to 17 OTUs on day 100 of life. While vaginal infants had slightly more observed OTUs, there was no significant difference between cesarean and vaginally delivered infants (Mann–Whitney crosssectional comparison at each week P-value was 0.43 or higher) (**Figure 1A**). The Shannon diversity increased from around 0.75 in the days following birth to 1.25 at NICU discharge, with comparable development between cesarean and vaginally delivered infants (Mann–Whitney cross-sectional comparison at each week P-value was 0.49 or higher) (**Figure 1B**). The observed OTUs and Shannon diversity continued to increase following discharge but no significant difference between cesarean and vaginally delivered infants occurred in post discharge samples (observed OTUs P = 0.212; Shannon P = 0.428) (Supplementary Figures 2A,B). Birth mode was also comparable between weighted and unweighted UniFrac distance between consecutive samples during NICU sampling (**Figures 1C,D**) and post discharge (weighted UniFrac P = 1; unweighted UniFrac P = 0.875) (Supplementary Figures 2C,D). Comparing the crosssectional weighted UniFrac distance between birth modes at weeks 1, 3, 5, 8, and post discharge showed no significant difference at any time point (Supplementary Figure 3).

### Vaginally Delivered Infants Have Increased OTU Stability, But Comparable OTU Acquisition While on the NICU and Following Discharge

The individual OTUs were tracked through time in consecutive samples, showing that vaginally delivered infants kept significantly more OTUs from month 2 of life (P < 0.001)

than those delivered by cesarean (**Figure 2A**). The number of OTUs kept in month 1 of life was comparable between birth modes (P = 0.947). The microbiome stabilized rapidly from birth to week 4 of life, where the number of 'OTUs lost' and 'new OTUs gained' declined, but the 'OTUs regained' (previously present but not in the preceding sample) increased (**Figure 2**). However, there was no difference in birth mode between OTUs lost, regained, or newly gained during the first 100 days while on the NICU. Following discharge, vaginally delivered infants continued to have significantly (P = 0.021) increased kept OTUs compared to cesarean, relative to the last NICU sample collected (Supplementary Figure 2E). As with the NICU samples, no significant difference between OTUs lost, regained, or newly gained was found for post discharge samples, relative to the last NICU sample (Supplementary Figures 2F–H).

## Bacterial Genera Were Comparable between Cesarean and Vaginally Delivered Infants

Klebsiella (28% overall relative abundance in NICU samples), Escherichia (22% overall relative abundance), Enterococcus (15%), Staphylococcus (14%), and Bifidobacterium (5%) dominated NICU samples, accounting for 84% of the total relative abundance in the first 100 days of life (**Figure 3**). Klebsiella and Enterococcus remained relatively consistent during the NICU period, Staphylococcus and Escherichia declined from birth, and Bifidobacterium gradually increased through the NICU period. To determine significant differences in the relative abundance of genera through the NICU, while accounting for repeated measures, the first sample from each infant in weeks 1, 3, 5, 8, and 10+ were included. This cross-sectional comparison showed no significant difference between cesarean and vaginal infants in any genera at any time point (Supplementary Table 1). Bacteroides was the 8th most abundant genera from all NICU samples (Supplementary Figure 4), but was 3rd most abundant in the post discharge samples (Supplementary Figure 5). Despite associations in term infants, the relative abundance of Bacteroides was comparable between cesarean and vaginal infants during NICU and following discharge in this preterm population. Comparing delivery mode in the post discharge period also showed the relative abundance of predominant genera was comparable (Supplementary Figure 5).

## Preterm Gut Community Type Development Is Comparable between Different Birth Modes

Using Partitioning Around Medoids (PAM) clustering based on weighted UniFrac, the early preterm microbiome was defined by six distinct clusters, termed PGCTs (**Figure 4A**). With the exception of PGCTs 1 and 4, all PGCTs had significantly different

FIGURE 2 | Longitudinal OTU tracker analysis of the gut microbiome in preterm infants by birth mode. Post discharge samples omitted from the analysis. Shaded bars represent the 95% confidence interval. (A) OTUs kept. (B) OTUs lost. (C) OTUs regained (previously detected in the infant). (D) New OTUs gained (not previously detected in the infant).

Shannon diversity, with PGCTs 2 and 4 showing higher Shannon diversity and PGCT 6 (Staphylococcus dominant) showing the lowest Shannon diversity (**Figure 4B**). Birth mode showed moderate differences in the PGCTs detected over the initial weeks of life (**Figures 4C,D**). Most notable was the increase of PGCT 3 in cesarean delivery and PGCT 4 in vaginal delivery at week 1 of life. PGCT 4 continued to be increased in vaginal infants at week 3 of life, but by week 5 of life and thereafter the PGCTs detected in cesarean and vaginal infants were comparable. Despite these trends in PGCT development, the trajectory of

gut microbiome development was highly variable within and between infants and no significant difference (P = 0.125) in PGCTs was found between cesarean and vaginal infants (Supplementary Figure 6).

## Inferred Bacterial Metabolic Potential Were Comparable between Cesarean and Vaginally Delivered Infants

Tax4Fun (Asshauer et al., 2015) was employed to infer the metabolic potential of the microbiome between birth modes. FishTaco (Manor and Borenstein, 2017) was performed to determine significance of inferred function at each time point (weeks 1, 3, 5, 8, and post discharge), finding no pathway was significantly altered between cesarean and vaginally delivered infants at any time point.

## DISCUSSION

The role of birth mode on the initial acquisition and subsequent development of the infant microbiome remains an important research question. We investigated the longitudinal development of the microbiome during the first 100 days of life and following discharge. In a cohort of exclusively preterm infants (24–31 weeks gestation), sampled during NICU and post discharge, birth mode was not significantly associated with the alpha- or beta- diversity (both within and between patients). Furthermore, birth mode had no significant association with the relative abundance of any bacterial genera or PGCTs. A novel analysis temporally tracking OTUs within infants in consecutive samples showed vaginally delivered infants retain more OTUs from month 2 of life and post discharge, suggesting increased microbiome stability associated with vaginal delivery.

In a large meta-analysis of preterm gut microbiome, infants delivered by cesarean section had increased Firmicutes and reduced Bacteroides, but the overall microbiome profiles were comparable (Pammi et al., 2017). In a recent study of term and preterm infants, birth mode was associated with altered stool microbiome at weeks 1, 4, 8, and 24 for term infants only (Hill et al., 2017). While only four preterm infants delivered vaginally were included in the study by Hill et al. (2017), microbiome profiles of NICU and post discharge samples appeared comparable between the preterm infants born by cesarean or vaginal delivery. In a previous longitudinal study of 58 preterm infants, postconceptional age was associated with the gut microbiome, with delivery mode reported to have minimal influences (La Rosa et al., 2014). Furthermore, in a study of 29 preterm infants, gender and feeding were more associated with the gut microbiome development compared to other demographics, including birth mode (Cong et al., 2016). While no clear associations in preterm infants have been reported, birth mode has been associated with significantly altered gut microbiome in term infants in the 1st year of life (Azad et al., 2013; Jakobsson et al., 2014; Bäckhed et al., 2015; Bokulich et al., 2016; Yassour et al., 2016). The discrepancy between term and preterm infants may result from the greater use of antibiotics in preterm populations (Berrington et al., 2012), or other NICU practices that have a dominant effect on the microbiome. Specifically, the frequency with which preterm infants are considered to be at risk of early onset infection, and thereby determining the use of antibiotics for the first 48 h of life, may supersede perinatal maternal influences such as birth mode or reasons for preterm labor and/or expedited delivery. Alternatively, the difference may relate to the differential abundance of Bacteroides between term and preterm infants. Bacteroides is the primary genera significantly altered between cesarean and vaginal term infants over the 1st year of life (Azad et al., 2013; Jakobsson et al., 2014; Bäckhed et al., 2015; Bokulich et al., 2016; Yassour et al., 2016). However, Bacteroides was not dominant in the current study, nor previous studies of preterm populations during NICU (Taft et al., 2014; Stewart et al., 2015b; Ward et al., 2016; Warner et al., 2016; Hill et al., 2017). Following discharge, Bacteroides represented the third most abundant genera and was higher in vaginally delivered infants. However, low numbers of post discharge samples preclude robust analysis of significance.

Previous studies collecting samples immediately at birth have shown vaginally delivered term born neonates have increased Lactobacillus and cesarean delivered neonates have increased Staphylococcus, Propionibacterium, and Streptococcus across multiple body sites (Dominguez-Bello et al., 2010; Chu et al., 2017). The earliest sample collected in the current study was on day 2 of life, at which stage the infant had received antibiotics (most commonly penicillin and gentamicin). This may account for why no specific bacterial genera or PGCT was significantly associated with the birth mode. It is also noteworthy that samples collected within the initial hours of life generally reflect (viable and non-viable) microbes transmitted from the mother, and not true colonization per se (Aagaard et al., 2016). Thus, it is also possible that no difference was reported in the current cohort because the non-viable and non-colonizing organisms are no longer detected by day 2 of life. Furthermore, preterm infants have a greater exposure time to microbes in the NICU environment (e.g., surfaces, bedside equipment, and staff), which are also responsible for shaping the preterm microbiome (Brooks et al., 2014; Hartz et al., 2015), and might add to the discrepancies between the effects of birth mode on the developing microbiome between preterm and term infants.

In the current study a novel OTU tracker was applied to determine the OTUs kept, lost, gained, or regained through consecutive samples. This was the only analysis that found a significant association, showing that vaginal infants had an increased number of kept OTUs in consecutive NICU samples from month 2 of life, which remained in samples collected post discharge. This is potentially reflective of increased gut microbiome stability in vaginal infants, compared to cesarean (Bäckhed et al., 2015). No difference in kept OTUs was found in the 1st month of life, reflecting the dynamic nature of the microbiome in both cesarean and vaginal infants during this phase (Koenig et al., 2010; Stewart et al., 2016). Notably, the number of new OTUs gained

while on the NICU was typically low (usually a single OTU per consecutive sample), but regained OTUs increased through NICU stay. This is likely a consequence of NICU practices, with limited environmental microbial exposure (e.g., use of sterile incubators, use of hand sanitizer, minimal skin-to-skin contact) restricting the introduction of new OTUs.

The current study has several limitations. First, a larger cohort comprised of preterm infants from multiple NICUs may reveal novel associations not seen in this single unit study. Second, collection of post discharge samples was challenging, with additional costs for sample collection and variable response rates to sampling requests, resulting in inclusion of post discharge samples from only 35% (16/46) of infants. Further study of post discharge samples is needed to determine the longterm associations of birth mode and specifically if Bacteroides establishes in higher relative abundance in vaginal infants (Azad et al., 2013; Jakobsson et al., 2014; Bäckhed et al., 2015; Bokulich et al., 2016; Yassour et al., 2016). Third, the universal use of antibiotics in this cohort prevents direct analysis on the role of antibiotics in superseding birth mode effects. Either much larger cohorts, or additional experimentation utilising animal models will be necessary to determine if antibiotics alone account for the discrepancies between preterm and term infants. Finally, the current analysis was performed only on 16S rRNA gene sequencing data, but differences in microbiome and host function may occur between vaginal and cesarean infants, requiring the use of proteomics and metabolomics to explore these elements further (Stewart et al., 2015a).

In summary, in a single NICU preterm population exposed to antibiotics postpartum, birth mode was not significantly associated with the diversity or composition of the microbiome. Vaginal infants tended to have greater stability following month 2 of life and post discharge. These findings highlight key differences between preterm and term infants. The long-term effects on host immune development from the transfer of microbes during delivery and the subsequent risk of obesity, autoimmunity, and allergy warrant further investigation.

#### AUTHOR CONTRIBUTIONS

CS, NE, GT, CO, JB, and SC were responsible for the study concept and design. CS and AN were responsible for extraction

and sequencing of samples. CS, EC, PL, DS, TF, and JP were involved in the data processing and analysis. CS drafted the manuscript and all authors contributed to critical revisions and approved the final manuscript.

## FUNDING

This work was supported by funding from Tiny Lives charity (Newcastle upon Tyne, United Kingdom), Newcastle upon Tyne Hospitals NHS Charity and the Newcastle Healthcare Charity. The content is solely the responsibility of the authors.

### REFERENCES


## ACKNOWLEDGMENTS

We gratefully acknowledge the help and support of parents who have helped with this and our other studies. We also acknowledge Jonathan Geselle for assistance in processing raw data and biom file formation.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmicb. 2017.01008/full#supplementary-material



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Stewart, Embleton, Clements, Luna, Smith, Fofanova, Nelson, Taylor, Orr, Petrosino, Berrington and Cummings. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Role of the Human Breast Milk-Associated Microbiota on the Newborns' Immune System: A Mini Review

#### Marco Toscano<sup>1</sup> , Roberta De Grandi<sup>1</sup> , Enzo Grossi<sup>2</sup> and Lorenzo Drago1,3 \*

<sup>1</sup> Laboratory of Clinical Microbiology, Department of Biomedical Sciences for Health, University of Milan, Milan, Italy, <sup>2</sup> Villa Santa Maria Institute, Como, Italy, <sup>3</sup> Clinical-Chemistry and Microbiology Lab, IRCCS Galeazzi Orthopedic Institute, University of Milan, Milan, Italy

The human milk is fundamental for a correct development of newborns, as it is a source not only of vitamins and nutrients, but also of commensal bacteria. The microbiota associated to the human breast milk contributes to create the "initial" intestinal microbiota of infants, having also a pivotal role in modulating and influencing the newborns' immune system. Indeed, the transient gut microbiota is responsible for the initial change from an intrauterine Th2 prevailing response to a Th1/Th2 balanced one. Bacteria located in both colostrum and mature milk can stimulate the antiinflammatory response, by stimulating the production of specific cytokines, reducing the risk of developing a broad range of inflammatory diseases and preventing the expression of immune-mediated pathologies, such as asthma and atopic dermatitis. The aim of the present Mini Review is to elucidate the specific immunologic role of the human milk-associated microbiota and its impact on the newborn's health and life, highlighting the importance to properly study the biological interactions in a bacterial population and between the microbiota and the host. The Auto Contractive Map, for instance, is a promising analytical methodology based on artificial neural network that can elucidate the specific role of bacteria contained in the breast milk in modulating the infants' immunological response.

#### Edited by:

Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia

#### Reviewed by:

Eugene Michael Dempsey, University College Cork, Ireland Ryo Inoue, Kyoto Prefectural University, Japan

> \*Correspondence: Lorenzo Drago lorenzo.drago@unimi.it

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 21 July 2017 Accepted: 13 October 2017 Published: 25 October 2017

#### Citation:

Toscano M, De Grandi R, Grossi E and Drago L (2017) Role of the Human Breast Milk-Associated Microbiota on the Newborns' Immune System: A Mini Review. Front. Microbiol. 8:2100. doi: 10.3389/fmicb.2017.02100 Keywords: human milk, milk microbiota, colostrum, immunomodulation, newborn's immune system, AutoCM

## INTRODUCTION

The human milk is a rich and complete nourishment that is essential for the correct development of the infant's organism (Ballard and Morrow, 2013).

The first milk produced by mothers after the delivery is called colostrum and it is biochemically and functionally different from the mature milk (Castellote et al., 2011). Colostrum, indeed, contains high concentration of lactoferrin, Immunoglobulin A (IgA), leukocytes and specific developmental factors, and a low amount of lactose, potassium and calcium, underlying its immunological functions rather than nutritional (Kulski and Hartmann, 1981; Pang and Hartmann, 2007).

From 5 days to 2 weeks postpartum, there is the production of transitional milk which shares some characteristics of colostrum, although its main function is to support newborns at nutritional level (Henderson et al., 2008; Nommsen-Rivers et al., 2012).

Finally, 2 weeks after the delivery the milk can be considered as mature and its composition tends to be stable over the time, even if slight variations can occur during lactation (Ballard and Morrow, 2013). The main components of human milk are: (i) macronutrients, such as protein, fat and lactose, which concentration depends on the stage of lactation and maternal characteristics; (ii) micronutrients, including vitamins A, B1, B2, B12 and D that vary in human milk in relation to maternal diet and body stores; (iii) growth factor, which are strongly active on the endocrine system, nervous system, vasculature and intestinal tract; (iv) immunological factors, which are essential to defend the newborn from inflammation and infection, and for this reason, the early milk is rich in immune components that can support infants in the first delicate stages of their life; (v) the microbiota, which comprises more than 200 different bacterial species with a pivotal role in the formation of the newborn's first gut microbiota (Drago et al., 2017).

The aim of the present Mini Review is to highlight the specific and fundamental role of human milk-associated bacteria in modulating and influencing the newborns' immune system during their life.

#### THE MILK MICROBIOTA AND THE NEWBORN'S IMMUNE SYSTEM

The specific mechanisms that lead to the formation of the human milk microbiota are still unknown; however, there are different hypothesis that can explain the origin of milkassociated bacteria. Indeed, some microorganisms belonging to the maternal skin or infant's oral cavity may become an integral component of the milk microbiota by means of a milk flow back into mammary ducts during lactation (Rodríguez, 2014). This mechanism may justify the presence of cutaneous and oral bacteria that are recovered in the milk microbiota, such as Streptococcus spp. and Staphylococcus spp. (Gao et al., 2007; Grice et al., 2009). Interestingly, human milk contained also a great number of intestinal bacteria, which may spread from the maternal intestinal environment by a mechanism involving dendritic cells (DCs) and CD18<sup>+</sup> cells (Rodríguez, 2014); these cellular types would be able to capture intestinal microorganisms from the gut lumen and transfer them to lactating mammary glands by means of translocation, which results to be increased during late pregnancy and lactation (Rodríguez, 2014). Consequently, the milk microbiota can shape the initial intestinal microbiome of newborns, together with the maternal intestinal and vaginal microorganisms that are ingested by the neonate during the passage through the birth canal (Houghteling and Walker, 2015). Human milk can stimulate the proliferation of numerous Bifidobacterium and Lactobacillus strains, the main probiotic microorganisms present in the gut, creating an acidic environmental rich in short chain fatty acids (SCFAs) with a protective and nutritive role at intestinal level (Bode, 2012; Walker and Iyengar, 2015). The constant intake, during lactation, of bacteria contained in the human milk leads to the formation of a transient intestinal microbiota that deeply impacts on the newborn's development, acting mainly on the maturation of his immune system (Houghteling and Walker, 2015). Indeed, several studies underscored the strict link between the gut microbiota signals, the mucosal host defense and the maturation of immune system, both at intestinal and systemic level (Smith et al., 2007; Sekirov et al., 2008; Walker and Iyengar, 2015). It has been showed that an altered colonization of newborns' gut may lead to a persistent intestinal dysbiosis and, consequently, to immune-mediated and metabolic diseases during infancy and childhood (Gareau et al., 2010; Johnson and Versalovic, 2012). Moreover, breast-fed newborns have shown to possess a more stable intestinal bacterial population and a well-balanced mucosal immune response if compared to the formula-fed ones (Gronlund et al., 2000; Bezirtzoglou et al., 2011); indeed, a healthy intestinal microbiota can induce specific T cell responses and modulate substrates oxidation, decreasing the impact of autoimmune and allergic diseases not only during childhood but also in adulthood (Guaraldi and Salvatori, 2012; Palma et al., 2012). Finally, breastfeeding has been observed to have a protective role against respiratory and gastrointestinal infections between the ages of 7 and 12 months, leading to a general improvement of symptoms associated to gastrointestinal infections (Duijts et al., 2010).

Intestinal bacteria can also stimulate lymphoid elements and positively influence the maturation of both innate and adaptive immune system, as clearly demonstrated by studying germ-free animals (Cash and Hooper, 2005). It has been shown that in germ-free mice the villus capillaries develop poorly during weaning and remained in this condition also during adulthood, suggesting that the intestinal microbiota is fundamental for intestinal blood vessel to be completely developed (Martin et al., 2010). More interestingly, intestinal bacteria can promote B cell development in Peyer's Patches and increase the production of mucosal IgA, the main antibody class in secretions that acts as first line of defense (Martin et al., 2010).

Moreover, bacterial surface-expressed or secreted ligands can interact with specific receptors on mucosal immune system and enterocytes leading to a self-limited inflammatory response for preventing pathogen mucosal penetration (Round and Mazmanian, 2009; Walker and Iyengar, 2015).

As underlined by Latuga et al. (2014), all newborns have an immature immune system and the cord blood rich in anti-inflammatory T regulatory cells; furthermore, infants have a high T helper 2 (Th2) that promotes humoral immunity with the production of IL4, IL6 and IL21, thus promoting an increased B cell response and, potentially, a higher allergic sensitization (Latuga et al., 2014). The pivotal role of milk-associated microbiota in influencing the neonates' immune system is over-emphasized by the cytotoxic function promoted by microbial ligands in breast milk (Donnet-Hughes, 2008). Indeed, in vitro stimulation of DCs with lipopolysaccharide can lead to T-cells differentiation, supporting the hypothesis that mature milk may implement the maturation of cytotoxic Th1 cells and improve their activity against infections (M'Rabet et al., 2008). Probably, commensal colonic

TABLE 1 | Immunomodulatory activities of the human milk microbiota.


bacteria may stimulate the release of specific cytokines that create a balanced microenvironment suitable for naive Th0 cells to ripen toward Th1 cell type (Walker and Iyengar, 2015).

Bacteroides is a bacterial genus that is very abundant in human colostrum and it may have a main role in the early stages of newborns' gut colonization, as reported by Mazmanian and Kasper (2006). In particular, the polysaccharide A located on the surface of Bacteroides fragilis can interact with Toll receptor 2 on intestinal DCs to stimulate cytokine production which, in turn, favor the proliferation of FOXP3 T cells in the lamina propria. FOXP3 belongs to the forhead transcription factor family bindweed in the expansion of regulatory T cells, thus having a suppressive role in the host's immune system (Kim, 2009). Therefore, it is clear that a correct stimulation of neonates' intestinal environment is fundamental for the physiological development of mucosal immune system and tolerance; the latter, in particular, is extremely important to avoid developing allergy or autoimmune diseases (Walker and Iyengar, 2015). Indeed, germ-free animals cannot develop tolerance due to the lack of intestinal bacteria and only the adequate colonization of newborns' gut can lead to a complete tolerance generation (Karlsson et al., 1999; Olszak et al., 2012).

Consequently, breastfeeding is essential for oral tolerance in newborns, as it is extremely important for the establishment of local and systemic immune tolerance to antigens ingested during lactation (Verhasselt, 2010). Infants are daily exposed to specific antigens, part of which belong directly to the human milk microbiota, and that can translocate across the intestinal barrier, being involved in the presentation by antigen-presenting cells to T lymphocytes. Furthermore, bacteria located in the human milk are fundamental to correctly stimulate the Peyer's patches, increasing the number of IgA-producing plasma cells in the intestinal environment of newborns (Gross, 2007). Consequently, IgA can trap food antigens favoring their

FIGURE 1 | Microbiota network of Italian mature milk. The figure shows the distribution of bacterial population contained in the Italian human mature milk. The network was obtained applying the AutoCM. The main hubs of the bacterial network are underlined with a blue line; red circle shows the central node of the network (Drago et al., 2017). As authors of the previously published manuscript, we maintain the ownership of copyright as stated in the license-to-publish we signed for the Nature Publishing Group.

elimination by specific enzymes, avoid the adherence of viruses and microorganisms to intestinal mucosa also counteracting the proliferation of pathogens and exert a direct immunomodulatory activity (Verhasselt, 2010).

Finally, the oral tolerance seems to be actively involved in the prevention of allergic diseases onset in babies, avoiding also the impact of respiratory and gastrointestinal infections during the early stages of their life (Lack, 2008).

Interestingly, some strains contained in the breast milk can stimulate immune responses in both animal and human models and their activity seems to have a moderate suppleness based on the gut environment (Fernández et al., 2013). Some Lactobacillus strains can enhance the production of Th1 cytokines as well as TNF-alpha even without inflammatory stimuli and activate NK cells, CD4+ and CD8+ T cells and regulatory T cells.

**Table 1** summarizes the main immunomodulatory activities of human milk microbiota.

### MICROBIAL NETWORK: NEW INSIGHTS INTO BACTERIAL ECOLOGY

The study of microbial interactions within a bacterial population is of extreme importance to clearly understand the specific role of microbiome. Indeed, microorganisms compete for nutrients, exchange genetic material and metabolites, being responsible of influencing the microbiota composition and the host's health (Layeghifard et al., 2017). Due to its dynamic nature and high heterogeneity, the microbiota can be considered a complex and variable ecosystem not often well understandable. For this reason, in the last years a novel approach has been developed to study the microbiota, by using graphtheoretical, systems-oriented method able to facilitate the understanding of evolutionary and complex ecological processes (Layeghifard et al., 2017). Bacterial network is becoming essential to study microbial relationships and clarify the impact of various interactions on the host by identifying the main "hubs" that may represent the most influential member in a bacterial community (Layeghifard et al., 2017). Moreover, a central node is thought to have more links with other hubs, having a pivotal role in the stability of the whole microbial network.

Numerous methods to identify network hubs exist, but in the last years there was a great interest in using and apply the Auto-Contractive map (AutoCM). AutoCM system is a fourthgeneration unsupervised artificial neural network (ANN) that can outperform numerous unsupervised algorithms (Buscema and Sacco, 2016). The system uses the minimum spanning tree (MST) theory to underline the natural connections between variables (Buscema and Grossi, 2008; Buscema et al., 2008). A MST is a spanning tree of a connected, undirected graph that links all vertices together with the minimal total pondering for its edges. The concept is that as all biological and natural systems are inclined to a minimal energetic condition, the graph delineates the essential biological information of this process. The final goal of this approach is finding out all specific correlations between variables, creating a semantic map of connections in which non-linear associations are maintained. This approach allows to both highlight the relevant nodes of the system and show the network of the main relations among variables. Nodes, also defined as "hubs," are defined as variables with the highest number of relations in the map (Buscema and Grossi, 2008; Buscema et al., 2008; Buscema and Sacco, 2016).

AutoCM has already been used in the microbiological field to study the bacterial network of human colostrum and mature milk in Italian and Burundian populations, highlighting the main microbial hubs that represent the biological leading structure of the whole network (Drago et al., 2017). This mathematical approach could partially clarify the complexity of human-bacteria mutualism, and above all the role that some specific microorganisms have in the human milk.

**Figure 1** shows a graphical representation of the human mature milk microbiota network obtained using AutoCM. The main bacterial hubs (blue lines) represent those microorganisms that may have a pivotal and leader role in the whole microbiota (Drago et al., 2017).

In the future, the AutoCM may be useful in understanding which microbe is directly involved in the stimulation and manipulation of newborns' immune system, maybe defining a "immunomodulatory bacterial pattern" with a pivotal role in the development of infant's immunity.

#### CONCLUSION

The human milk microbiota seems to have a central role in stimulate newborns' immune system, contributing to create the first transient intestinal microbiota with strong immunomodulatory activities. However, further studies are needed to highlight the direct and strong connection between the human milk microbiota and the stimulation of newborns' immune system, as to date there are no clear and specific evidence about this association.

Application of network biology will significantly improve our knowledge on bacterial interactions among the milk microbiota, with important applications for eventual targeted modification of bacterial composition, aimed to enhance the abundance of those microorganisms that may be essential not only for the modulation of the infants' immune system, but also for improving the whole host's health.

## AUTHOR CONTRIBUTIONS

LD provided the general concept. LD and EG drafted part of the manuscript. MT and RDG wrote the manuscript. All authors revised and approved the manuscript.

#### REFERENCES

fmicb-08-02100 October 23, 2017 Time: 15:55 # 5


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Toscano, De Grandi, Grossi and Drago. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Uterine Microbiota: Residents, Tourists, or invaders?

#### *James M. Baker1,2, Dana M. Chase3 and Melissa M. Herbst-Kralovetz1,4\**

*1Department of Basic Medical Sciences, College of Medicine-Phoenix, University of Arizona, Phoenix, AZ, United States, 2Department of Biology and Biochemistry, University of Bath, Bath, United Kingdom, 3Arizona Oncology (US Oncology Network), University of Arizona College of Medicine, Creighton University School of Medicine at St. Joseph's Hospital, Phoenix, AZ, United States, 4Department of Obstetrics and Gynecology, College of Medicine-Phoenix, University of Arizona, Phoenix, AZ, United States*

Uterine microbiota have been reported under various conditions and populations; however, it is uncertain the level to which these bacteria are residents that maintain homeostasis, tourists that are readily eliminated or invaders that contribute to human disease. This review provides a historical timeline and summarizes the current status of this topic with the aim of promoting research priorities and discussion on this controversial topic. Discrepancies exist in current reports of uterine microbiota and are critically reviewed and examined. Established and putative routes of bacterial seeding of the human uterus and interactions with distal mucosal sites are discussed. Based upon the current literature, we highlight the need for additional robust clinical and translational studies in this area. In addition, we discuss the necessity for investigating host–microbiota interactions and the physiologic and functional impact of these microbiota on the local endometrial microenvironment as these mechanisms may influence poor reproductive, obstetric, and gynecologic health outcomes and sequelae.

Keywords: endometrium, microbiome, host–microbe interactions, gynecologic and reproductive health, inflammation, infertility, endometrial cancer, pathophysiology

#### BACKGROUND

For almost a century and based on the work of Henry Tissier in 1900, consensus was that a healthy uterine cavity is sterile (**Figure 1**) (1). This sterility was hypothesized to be maintained by the cervical plug, which was compared with the "Colossus of Rhodes" in providing an impermeable barrier to bacterial ascension from the vagina (2). This assumption was challenged by multiple reports in the mid to late 1980s, using culture-dependent methods, of uterine-dwelling bacteria even in healthy asymptomatic women (**Figure 1**) (3–6). Furthermore, the cervical mucus plug has been shown to not be entirely impermeable to bacterial ascension from vaginal bacteria (7, 8). It was also shown that in a non-pregnant state, particles can translocate from the vagina to the uterus through the cervical canal within minutes during the follicular and luteal phases of the cycle (9). The naturally occurring uterine peristaltic pump aids in sperm transport from the cervical canal to the uterus, and these peristaltic contractions have been shown to move macrospheres from the canal into the uterus and other areas of the upper female reproductive tract (FRT), and therefore may play a role in seeding the uterus with bacteria (8). The follicular phase of the menstrual cycle has been shown to be associated with an increased frequency of uterine contractions (10). Uterine conditions may also promote bacterial seeding of the uterus through hyper- and dysregulation of uterine contractions (10). In addition, it was argued that the position of the uterus in such close proximity to a consistently colonized site such as the vagina would make some movement of bacteria to the

#### *Edited by:*

*Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia*

#### *Reviewed by:*

*Elisabeth Margaretha Bik, uBiome, United States Luke McNally, University of Edinburgh, United Kingdom*

#### *\*Correspondence:*

*Melissa M. Herbst-Kralovetz mherbst1@email.arizona.edu*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 04 October 2017 Accepted: 24 January 2018 Published: 02 March 2018*

#### *Citation:*

*Baker JM, Chase DM and Herbst-Kralovetz MM (2018) Uterine Microbiota: Residents, Tourists, or Invaders? Front. Immunol. 9:208. doi: 10.3389/fimmu.2018.00208*

**40**

uterine cavity inevitable. The presence of a uterine microbiome has been reported in animal models, most notably in cows, where the impact of the uterine microbiome on fertility is a key contemporary research question (11–15). Specific bacterial species have shown a tendency for colonizing the uterus, such as *Fusobacterium*, which has been found in both mice and cow uteri (16). Colonization with this particular bacterium in mice has been shown to be transmitted through the hematogenous route (**Figure 2**) (17). Hematogenous (through the bloodstream) spread of bacteria through either oral (18) or the gut route (19) allows bacteria from mucosal sites such as the oral cavity and the gastrointestinal tract to colonize distal mucosal sites and occurs during epithelial barrier breach (e.g., gingivitis and leaky gut) (17, 20–23). However, other sources of uterine microbiota seeding may include inadvertent bacterial transmission of vaginal bacteria into the uterus through assisted reproductive technology (ART)-related procedures or during placement of intrauterine contraceptive devices (**Figure 2**) (24, 25).

Later the assumption became that any detection of bacteria in the uterus was the result of ascension from the lactobacilli (LB) rich vaginal microbiome. The presence of bacteria in the uterus has been associated as causative agents in adverse conditions such as recurrent abortion and preterm labor (23, 26). An enormous hurdle to twentieth century research in this area was the necessity to culture bacterial specimens for analysis, which severely limited quantification and resolution of bacterial communities, leading to low and inconsistent bacterial yields (**Figure 1**). This is particularly important when one considers that the uterus is a low abundance site (27, 28). Estimations of uterine bacteria load are estimated between 100 and 10,000 times less bacteria than the vaginal microbiome (27, 28).

The advent of next-generation sequencing (NGS) technologies in 2007 has enabled a far more global assessment of bacterial composition of the uterus than could be measured solely with culturedependent methods. Furthermore, culture studies focused on the ability to culture a finite variety live bacteria, whereas sequencing technologies enabled the identification of the full range of uterine bacteria (26). Indeed, it is now appreciated that only approximately 1% of bacteria are culturable (29–31). NGS has enabled species-level quantification utilizing the variable (V) regions of the 16S rRNA gene, potentially allowing for the determination of the full scope of uterine microbiome signatures in both healthy and diseased hosts.

The purpose of this review is to provide a comprehensive summary of the uterine microbiome literature to date, focusing on detailing studies of "healthy" bacterial residents to bacterial

tourists or invaders that are present in particular diseased states. This review will consider the limitations of NGS at a low abundance site such as the uterus and discuss key methodical considerations pertaining to sampling the uterus and unique challenges to collecting these patient specimens. In addition, we identify gaps in the area of uterine microbiota research that will provide exciting research opportunities for future studies. The ascertainment of uterine microbiota signatures in various states of health and disease could potentially lead to effective clinical interventions for a variety of conditions and have a positive impact on obstetric and gynecologic health.

#### RESIDENTS, TOURISTS, OR INVADERS: DEFINING UTERINE MICROBIOTA

The current NGS literature on the uterine microbiome has provided provocative glimpses into the putative role of the uterine microbiota in multiple disease states and the potential impact on women's health. It is important, however, to take into consideration the limitations of the studies to date. These considerations include reagent controls, subject cohort size and patient demographics, and sample collection, sequencing methods, and downstream analyses.

The possibility of sample contamination is a significant hurdle to ascertaining whether uterine bacteria are residents, tourists, or invaders due to the low abundance of bacteria in the uterus (32). Contamination may contribute to a larger percentage of microbiota enriched and reported at low abundance sites. The placenta and lung are also low abundance sites that have divergent reports in the literature with some groups detecting a distinctive placental microbiome (33), while others find it to be indistinguishable from negative controls (34). These studies illustrate the difficulty in distinguishing low abundance sites from false positives and the requirement for larger studies and need for robust controls.

Contamination controls, when reported, did not fully describe how these control samples and data were accounted for during downstream analysis of samples. Walther-António et al. reported contamination in 9 of their 14 negative controls, yet it was not described how these controls were accounted for in the sample analysis (35). This is fairly typical of the current literature, whereby a detailed description of control samples are not clearly described. However, Chen et al. included more rigorous controls and reported inclusion of both extraction controls and PCR controls (27). This group also included additional controls when culturing peritoneal fluid, using diluent controls and swabs taken from sterilized skin of the patient as well as swabs of the doctor gloves (27). Most uterine sampling is performed transcervically, which makes it difficult to avoid cross-contamination with the cervical microbiota (see **Table 1**). In addition, uterine manipulators and cervical dilators may further contribute to cross-contamination from the cervix if used during hysterectomy procedures; however, studies rarely report if these instruments were used. In most studies in this review (**Table 1**), precautions were taken to limit contamination through methods such as vaginal disinfection, which coupled with careful sampling, reduces contamination. Nevertheless, rigorous application of contamination controls and detailed descriptions of clinical procedures are an important aspect of future research (32). Future studies may consider including negative controls that can measure reagent contamination that can be subtracted from the experimental samples (36).

Subject cohort size is another key limitation of the field, due in part to the difficulty of enrolling patients and the technical challenges in obtaining uterine samples. Six of the nine studies focused on in this review had a cohort of ≤35. This small sample size severely reduces the statistical power of the studies (**Table 1**). A related issue is the lack of ethnic diversity across these smallscale studies. It is already well documented that bacterial vaginosis (BV) differs significantly in Caucasian women compared with other racial and ethnic groups (37). It is therefore likely that the same would be true of the uterine microbiota as suggested by a recent abstract (38). If the uterine microbiome varies with ethnicity this may have potential impact on risk factors associated with gynecologic and reproductive sequelae, but should also account for differences in socioeconomic factors and environment

through the endocervical canal.

Table 1 | Patient demographics, study design, and profile results of current literature describing next-generation sequencing studies of uterine microbiota in chronological order.


(*Continued*)

Uterine Microbiota Drive Obstetric/Gynecologic Sequelae


(*Continued*)

Uterine Microbiota Drive Obstetric/Gynecologic Sequelae


(*Continued*)

Uterine Microbiota Drive Obstetric/Gynecologic Sequelae

Baker et al.


*IVF, in vitro fertilization; EP, endometrial polyps; CE, chronic endometritis; MFI, male factor infertility; NR, not reported; GnRHa, gonadotropin-releasing hormone agonist; LD, Lactobacillus dominant; NLD, non-Lactobacillus dominant. aNo report of any prevailing medical conditions which may modulate microbiota.*

(e.g., diet) (39, 40). Clearly, larger and more inclusive studies are needed.

One seemingly unavoidable limitation of this line of research is the lack of healthy controls which results from the fact that healthy excised uteri are rarely obtained. All hysterectomies were carried out due to an underlying benign or non-neoplastic condition or a symptomatic condition such as fibroids. However, the issue of appropriate controls also extends to studies assessing the microbiota of *in vitro* fertilization (IVF) patients; even though there may not be frank disease as such, these women can still not be considered healthy controls due to infertility. IVF studies where the inclusion criterion is restricted to male factor infertility provide a better control population (**Table 1**). Other factors that affect "healthy" controls include antibiotic usage and collection of a detailed medical history and use of clear exclusion criteria. For example, women with an intrauterine device (IUD) should be excluded due to their potential impact on uterine colonization, unless this is related to the question being addressed. Mitchell et al. were the only group that excluded IUD users despite IUDs being known to harbor bacteria and aid in uterine colonization (25, 28).

The specific 16S rRNA gene V region primers used by studies in this area (shown in **Table 1**) are a potential cause of incongruence as certain 16S rRNA gene V regions have been shown to over- or underrepresent certain taxa (41, 42). In addition to the choice of 16S rRNA gene V region primers, DNA extraction methods and operational taxonomic unit classification have also been identified as potential sources of variation in microbiome studies (43). Adoption of standardized methodology in these areas would greatly facilitate comparisons across studies.

While NGS provides a useful tool in bacterial quantification, it only quantifies bacterial the 16S rRNA gene, it does not represent viability. As pointed out in the recent review by Perez-Muñoz et al., this is a significant limitation in the field (32). While bacteria have been cultured from the uterus in numerous studies since the 1950s and in the recent report by Chen et al. there is still a question as to whether these bacteria quantified by NGS represent viable bacteria.

The ability for germ-free mice to be generated provides some evidence against a resident uterine microbiome as the process involves the removal of the pregnant uterus from conventional mice, placing in a germicidal bath and then transferring them to a germ-free mother. However, low abundance uterine microbiota may be removed as a result of the germicidal bath. While not the focus of this review, the bacterial seeding of the uterus has important ramifications related to the highly debated topic of maternal–fetal transfer of microbiota and postnatal health (32, 44). The presence or absence of a placental microbiome remains a controversial topic as it relates to maternal–fetal transfer of the microbiome and is beyond the scope of this review (21, 45). Currently, the data available suggest that maternal gut microbiota impacts fetal health outcomes (46). Whether this is through interaction of bacteria and the placenta/amniotic fluid/ meconium directly, or whether the interaction is through microbial products or metabolites, remains to be fully elucidated and may not be mutually exclusive.

While there are certainly limitations in the studies to date, the current literature demonstrates significant changes in microbiota compositions related to various disease states, rates of IVF success, and risk for endometrial cancer. These studies have provided a starting point for future studies in uterine microbiome research and to expand our fundamental understanding of this emerging aspect of human health.

## RESIDENTS: UTERINE MICROBIOTA IN "HEALTHY" ASYMPTOMATIC WOMEN

This review mainly focuses on the negative consequences of the presence of uterine microbiota due to inherent difficulties in sampling the uterus in healthy women; however, clearly the maintenance of homeostasis is also important if a "normal" resident microbiome in the uterus is defined. The uterine microbiota reported in healthy subjects, as defined by NGS, varies greatly throughout the nine reports that exist to date (**Table 1**). With little consistency extending all the way up to the phyla level, it is currently difficult to define a consensus "healthy" or "core" uterine microbiota. However, certain generalizations can be made from the existing data. The most abundant bacteria consistently belong to the following phyla: Firmicutes, Bacteriodetes, Proteobacteria, and Actinobacteria (27, 35, 47–53).

Within the Firmicutes, the genus *Lactobacillus* is a very prominent component in the majority of the uterine microbiome studies and is a consistent finding among reports to date (**Table 1**) (28, 47–49, 51, 52). Again, however, comparison of the relative abundance (or even absence) of *Lactobacillus* between sequencing reports highlights the inconsistency among reports and warrants further investigation. For example, Fang et al. reported higher levels of *Lactobacillus* in diseased groups of women with endometrial polyps (EP) or in women with EP and chronic endometritis (EP + CE) (38.64 and 33.21%, respectively) compared with healthy controls (6.17%) (49). By contrast, Moreno et al. reported that high levels of *Lactobacillus* (>90% as defined by the group) are significantly associated with increased reproductive success in women undergoing IVF, although whether only certain (undefined) *Lactobacillus* species may be capable of conferring this benefit is not clear from this study (47).

Notably, *Lactobacillus* dominance is generally considered to be a predictor of vaginal health (54, 55). However, increased level of *Lactobacillus* may act as a risk factor or marker for EP + CE through a breach in the cervical barrier that subsequently allows for ascension of *Lactobacillus* from the vagina to the uterus. The increased reproductive success in women with high levels of *Lactobacillus* may simply reflect the composition of the vaginal microbiome at time of IVF ET (56). It is also important to consider that assessing the uterine microbiota by catheter tip analysis may not be true representation of the bacteria in the uterus. The sampling and surface area that the catheter tip assesses is dramatically smaller compared with swabbing the uterus. This likely decreases the quantity of bacteria obtained and therefore increases the impact of contaminants either from the cervicovaginal environment or the reagents. However, the uterine microenvironment is unique from other mucosal sites in that it serves as the starting point for embryo implantation and placentation and is tightly regulated by female sex hormones. Therefore, unlike the vagina, *Lactobacillus* may not be a predictor of uterine health and could have pathophysiological consequences following ascension from the vagina (**Figure 3**).

The source of *Lactobacillus* in the uterus is easily explained by the abundance of this bacterial genus in the nearby vagina (although again the *Lactobacillus* species could differ). It is also possible that *Lactobacillus* reported in many uterine reports is a result of contamination from the vagina. The presence of other taxa that, in some cases, constitute a significant portion of the uterine microbiome (**Table 1**) may result from other routes of seeding outlined in **Figure 2** (57).

Franasiak et al. investigated the uterine microbiome at the time of IVF and embryo transfer and found that *Flavobacterium* comprised one of the two most abundant taxa of the uterine microbiome (48). *Flavobacterium* was not found in any of the other nine NGS sequencing papers to date. This is particularly surprising due to the prevalence of *Flavobacterium* reported by the study in both ongoing and non-ongoing pregnancy. In addition, the study was unique from other studies in the sequencing methodology employed (The Ion 16S Metagenomics Kit rather than Illumina sequencing or pyrosequencing). Furthermore, *Flavobacterium* has been shown to be a common contaminant in reagents by Salter et al. and Laurence et al., specifically in ultrapure water (35, 47–52, 58, 59). However, it is worth noting that Franasiak et al. included a positive (*Escherichia coli*) and negative control (reagent only control), unlike many of the studies covered in this review (48).

Additional work was carried out by Tao et al. to assess the limits of accurate detection by NGS on single species and polymicrobial cultures (52). It was shown that bacteria culture lysates above 60 cells had accurate taxonomic identification (52). The authors were confident that this method was sufficient in detecting microbiota at this low level. Furthermore, it was shown that none of the taxa present in the negative control were one of the four bacterial strains used to assess the limits of detection using their method (52).

In addition to variability in methodology, patient populations and controls, the FRT is modulated by circulating sex hormones leading to physiological changes that influence microbiota compositions and *vice versa* (60). It is not clear if the uterine

of autophagy. (B) Downregulation of cell–cell junction expression is a key method of epithelial barrier breach and allows for the movement of bacteria in between epithelial cells. Similarly, the degradation of the extracellular matrix by matrix metalloproteinases also disrupts epithelial barrier integrity. (C) Microbial-secreted metabolites such as short-chain fatty acids (SCFAs) can encourage the growth of specific species and suppress growth of other bacteria. Reactive oxygen species (ROS) and changes in the pH of the uterine microenvironment may also drive disease. (D) Inflammation triggered by TLR activation and subsequent proinflammatory pathways can recruit immune cells and lead to the secretion of antimicrobial peptides (AMPs), which leads to the depletion of bacterial abundance. TLR-mediated signaling can also regulate mucin synthesis of both membrane-associated and secreted mucins that may impact colonization.

microbiome changes over time or during the menstrual cycle. However, Moreno et al. evaluated IVF catheter tips to assess the uterine microbiome across two different time points (47). One sample was taken at the prereceptive phase and another at the receptive phase of the same menstrual cycle to assess a putative shift in microbiome composition in IVF patients (47). In this single study, the uterine microbiome was similar at these hormonal stages in 9 out of 13 patients sampled, which is similar to the vaginal microbiome in its stability through hormonal stages (47, 61). However, the fact that this study was conducted over a short period of time with a small sample size suggests that these results should be viewed with some caution (47). Given the significant impact of menopause on the vaginal microbiome reported in many other studies it would be important to determine the impact of hormonal fluctuations and therapies on the uterine microbiota throughout a woman's lifespan, as well as, in the setting of gynecologic cancers (60, 62, 63).

### TOURISTS/INVADERS: UTERINE MICROBIOTA LINK TO POOR REPRODUCTIVE OUTCOMES AND ENDOMETRIOSIS

A healthy endometrium is the foundation for successful implantation and intrauterine infection has been deemed the cause of many reproductive complications (64). The endocervical barrier, as a means of preventing ascension of bacteria from the vagina, can be breached. Kunz et al. performed a study demonstrating that radioactively labeled macrospheres reached the uterine cavity within minutes of being administered at the external cervical os and documented the mechanism of the uterine peristaltic pump that actively moves vaginal content to the uterus (8). Zervomanolakis et al. extended these findings by demonstrating that particles could ascend through the cervix within minutes during the follicular and luteal phases of the cycle (9), clearly establishing the plausibility of bacterial ascension as a route of seeding of the uterine microbiome (**Figure 2**) (9). In addition, ART procedures may seed the uterine microbiota and drive adverse reproductive and gynecologic outcomes through modulating the local microenvironment (24). A reduction in clinical pregnancy rates has been shown when bacteria were cultured from the IVF catheter tip during ART procedures (65). Alternatively, it may be the ascension of key bacterial species/ taxa that may lead to increased susceptibility to reproductive complications rather than simply bacterial seeding of any taxa. Notably, Swidsinski et al. demonstrated that half of the women presenting with BV had a polymicrobial biofilm adhered to the endometrium (66).

The presence of bacteria in the uterus has been associated with poor reproductive outcomes and endometriosis; however, a cause and effect relationship has not been clearly established. In addition, the association between endometriosis and poor fertility has been well documented (67). Uterine microbiota composition has been shown to be significantly different in women with endometriosis (27). Furthermore, Cicinelli et al. have reported that endometriosis patients treated with antibiotics before implantation had significantly better reproductive outcomes compared with those not treated with antibiotics (31), suggesting that the negative impact of endometriosis on reproductive outcomes may be in part attributable to the presence of uterine bacteria. It would not be surprising that the anatomical and physiological changes elicited by endometriosis would result in a significantly different uterine microbiome composition due to the proximity of endometriotic lesions to the uterus. Endometriosis patients have been shown to exhibit a uterine bacterial composition with low levels of *Lactobacillaceae* species and enrichment of *Streptococcaceae, Staphylococcaceae*, and *Enterobacteriaceae* species relative to healthy controls (53). Conversely, changes in the microbiome may potentially trigger endometriosis through modification of the microenvironment. As highlighted in the review by Sirota et al., inflammation in the uterus due to the presence of bacteria may influence the balance of cytokines needed for successful blastocyst development and implantation (68). Taking this concept a step further, an inflammatory cytokine signature of endometriosis may have a significant impact on the microenvironment and reproductive outcomes (69). Correlations exist between various pro-inflammatory cytokines such as IL-6 (70) and anti-inflammatory cytokines and adverse reproductive conditions such as polycystic ovary syndrome, tubal factor infertility, or infertility of unknown origin (71).

Despite these links between endometriosis and reproductive outcomes, however, there is only one report that demonstrates a statistical difference in microbiome profiles, rather than just presence of bacteria, between successful vs. unsuccessful reproductive outcomes (47) (**Table 1**). These investigators enrolled a cohort of 35 subjects undergoing IVF. Endometrial samples were collected before blastocyst implantation to assess the uterine microbiota profiles, which were classified as either *Lactobacillus* dominant (LD), defined as consisting of >90% *Lactobacillus* spp*.*, or non-*Lactobacillus* dominant (NLD), consisting of <90% *Lactobacillus* spp. These investigators found a significant difference in the reproductive outcomes between these two groups. Women with an LD uterine microbiome had markedly higher rates of implantation [60.7 vs. 23.1% (*P* = 0.02)], pregnancy [70.6 vs. 33.3% (*P* = 0.03)], ongoing pregnancy [58.8 vs. 13.3% (*P* = 0.02)], and live births [58.8 vs. 6.7% (*P* = 0.002)] compared with those with an NLD uterine microbiome composition (47). Interestingly, germ-free mice have been shown to have reduced reproductive success after embryo transfer compared with conventionally housed mice, suggesting a role for the presence uterine microbiota in pregnancy (72). As mentioned earlier, the association between *non-Lactobacillus* species and adverse reproductive outcome has been demonstrated in the vaginal microbiome (56). Consequently, the association between the uterine microbiome obtained from the IVF catheter tip and reproductive success following IVF may simply be a reflection of the vaginal microbial community (e.g., LD) through cross-contamination and its association with reproductive success.

By contrast, Franasiak et al. found no significant difference in uterine microbiota between groups with non-ongoing vs. ongoing pregnancy (48). Similar to the previous study by Moreno et al., they found *Lactobacillus* to be one of the most abundant taxa, but they also reported that *Lactobacillus* was not significantly different between non-ongoing and ongoing pregnancy groups. As mentioned earlier, they identified *Flavobacterium*, as one of the most abundant taxa, which is inconsistent with the current NGS literature. This as-yet-unresolved discrepancy between the Moreno and Franasiak studies highlights the need for additional studies. Data analysis may also play a role in the disparity between these two studies. For example, if Franasiak et al. had used the same 90% cutoff as Moreno et al. to determine LD, they may have reached significance. Larger sample size and standardized procedures to avoid vaginal cross-contamination as discussed herein will be important aspects of future studies aimed at determining the role of vaginal and uterine microbiota and reproductive success.

#### INVADERS: UTERINE MICROBIOTA IN CANCER AND DISEASE

Success in identifying unique species in the uterine microbiome that are associated with a particular disease state could potentially be used as microbial biomarkers for prevention, screening, diagnosis, or even treatment to improve health and reproductive success. Besides reproductive outcomes and endometriosis, uterine colonization with BV-associated bacteria has been hypothesized to promote carcinogenesis through microbiotamediated pathophysiologic changes in the microenvironment (62, 73). Indeed, the microbiome is suspected of playing a general role in carcinogenesis through stimulating host secreted proinflammatory cytokines or growth factors as a result of dysbiosis (74). For example, pelvic inflammatory disease has been shown to increase risk of developing endometrial cancer by 1.89-fold in a nationwide population-based retrospective cohort study (75).

A recent study by Walther-António et al. compared the microbiome at various sites in the FRT in patients with endometrial cancer, endometrial hyperplasia (as a cancer precursor group) and those with benign uterine conditions. These investigators collected uterine, Fallopian, ovarian, and peritoneal samples posthysterectomy and preoperative vaginal, cervical, urine, and stool samples from patients in these three study groups and reported the FRT sites and stool sampled in the cancer and hyperplasia patients. Using the microbiome results combined with patient demographic data, it was possible to statistically associate the presence of *Atopobium vaginae* and a *Porphyromonas* sp. in the FRT as being associated with cancer (35). Future studies should be geared at better understanding the functional impact of these bacterial species on hallmarks of cancer illustrated in **Figure 3**.

Microbiota can drive cancer through numerous mechanisms including preventing apoptosis, stimulating proliferation, and driving genomic instability that are hallmarks of cancer highlighted in **Figure 3** (76). The relationship between these taxa and disease may not be limited to resulting inflammation and secretion of cytokines by the host cells, but may also be influenced by the hormonal status of the host. In particular, sex hormones such as estrogens, which are key drivers in certain cancers, have been implicated in carcinogenesis, raising the question of whether estrogens might influence the microbiomes of the uterus similar to the vagina. Use of the gonadotropin-releasing hormone agonist is associated with a shift in composition of the uterine microbiome, demonstrating the uterine microbiome may be hormonally regulated (53). Gut microbiota have been shown to facilitate the reuptake of estrogen to contribute to the progression of estrogen-driven cancer (77). In support of this notion, it has been reported that both gut microbiota composition and systemic estrogen levels are significantly different in patients with breast cancer compared with healthy patients (78–82). In addition, levels of free estrogens have been shown to be modulated by gut bacteria, through the "estrobolome," *via* secretion of β-glucuronidase, which deconjugates estrogens into their active metabolites (79, 83). However, studies directed at investigation of the relationship between hormonal status and uterine microbiome composition are still lacking.

In addition to women with endometrial cancers, differences between microbiome profiles in "healthy" (albeit underlying conditions, see **Table 1**) women vs. those with EP and chronic endometritis (EP + CE) have also been reported (see **Figure 3**). As previously mentioned, Fang et al. divided women into healthy, EP only and EP + CE groups (see also **Table 1**) and analyzed samples of both vaginal and uterine microbiota (49). These investigators reported that, compared with samples from healthy subjects, EP or EP + CE samples contained higher proportions of Firmicutes at the phylum level and *Lactobacillus*, *Gardnerella*, *Bifidobacterium*, *Streptococcus*, and *Alteromonas* at the genus level, and confirmed that these differences were statistically significant using AMOVA and ANOSIM analyses (49). The finding that *Lactobacillus* was over three times more abundant in the uterine microbiome of both diseased groups EP and EP + CE, compared with healthy controls may suggest ascension of vaginal bacteria (49). We hypothesize that the cervical barrier may have been disrupted in these disease states, which allowed the ascension of the dominant vaginal bacteria, *Lactobacillus*, into the upper FRT. The question of whether bacterial ascension is causative of EP or whether EP results in the increased cervical permeability and ascension could potentially be addressed by analysis of samples collected longitudinally to determine the timing of *Lactobacillus* expansion in the uterus. This strategy could also address whether there may be a positive feedback loop whereby increased cervical permeability leads to increased colonization of vaginal bacteria in the uterus. Another next step would be to determine the functional impact of specific organisms or groups of organisms on the host epithelium using robust human model systems (54, 55, 84).

High vaginal pH (an indicator of vaginal dysbiosis) was also significantly associated with endometrial cancer in the Walther-António study (35). However, vaginal pH, as a single variable, was not significantly different in the benign group compared with the hyperplasia group. A limitation of this study is that the overall microbiota community structure was not fully reported; thus, any firm relationships between microbiome composition, pH, and cancer remain unclear and require further investigation. Indeed, it would be worthwhile to assess *Lactobacillus* spp. in patients to determine whether low *Lactobacillus* abundance correlates with high pH as previously reported (85), or whether it is some other factor within the tumor microenvironment that increases vaginal pH. Interestingly, increased vaginal pH has also been shown to be associated with endometriosis and GnRHa therapy (86), which may suggest a relationship between the vaginal microenvironment on proliferative uterine diseases driven by hormones or *vice versa*.

#### MUCOSAL AXES AND THE ENDOMETRIAL MICROENVIRONMENT

This review mainly focuses on the FRT microbiota; however, it is important to consider that the uterine microbiome may be impacted by, or exert impact on, other distal mucosal sites, which extend beyond the spatial relationship between the vagina and uterus (**Box 1**).

For example, women with endometriosis show significantly lower levels of *Lactobacillaceae* in the uterine microbiome when undergoing treatment with GnRHa, compared with GnRHauntreated women (53). Women with endometriosis have also been shown to exhibit increased levels of the pro-inflammatory cytokine IL-6 in follicular fluid, with implications for reproductive function (70). It may be this altered inflammatory profile that drives the uterine microbiota composition seen in women with endometriosis or *vice versa* (53). Production of pre-IL-1β in patients with endometriosis has also been found to induce inflammation in the peritoneum (87). Another study that provides evidence of endometriosis impacting inflammation-linked sequelae is a nationwide Danish study, which assessed 37,661 women hospitalized with endometriosis. The results showed that women, after developing endometriosis, were significantly more likely to develop inflammatory bowel disease, ulcerative colitis or Crohn's disease, compared with controls (88). Even 20 years after initial hospitalization with endometriosis, these patients had an increased risk of developing ulcerative colitis and Crohn's disease, underscoring the importance of a better understanding of these complex relationships between mucosal sites (88).

Beyond an inflammatory milieu, we further hypothesize that, similar to immune mediators that communicate through the common mucosal immune system, the mucosal sites of the body interact through exchange of bacteria, metabolites or immune signaling between sites and that dysbiosis at one site could impact the mucosal immune environment at another site

Box 1 | Future areas of study and clinical implications of uterine microbiome research.


(**Figure 2**). Evidence for this concept includes the following studies. First, using culture-dependent methods, it has been shown that Rhesus monkeys with endometriosis exhibited significantly different proportions of *Lactobacillus* spp. and aerobic and facultative anaerobic Gram-negative bacteria in the intestinal microbiota compared with healthy controls (89). A second study supporting the transfer of microbiota from one mucosal site to another is provided by Fardini et al., who reported transmission of bacteria from the oral microbiome to the placenta in mice (18) (presumably *via* the maternal circulation) as a potential cause of intrauterine infection. Further evidence of translocation of viable bacteria is provided in the review by Potgieter et al. that postulates the translocation of dormant but viable bacteria through the blood (90). These authors further determined that injection of saliva and gingival plaque samples into the tail veins of pregnant mice, to mimic the bacteremia of an oral infection, resulted in species-specific colonization of the placenta by such species as *Neisseria flavescens* or *Neisseria subflava* normally found in oral flora (18). The association between subgingival plaque bacteria and placental bacteria has been demonstrated through comparing hypertensive to normotensive individuals (91). Periodontal pathogens are more abundant in subgingival plaques and the placenta in hypertensive women (91). It has also been shown that the detection of *Gardnerella* or *Ureaplasma* in the vaginal microbiome is associated with preterm labor, which is often associated with intrauterine-infection-driven preterm birth following ascension of these vaginal microbiota (45). However, it was questioned as to whether the preterm labor was induced by microbial risk factors of intrauterine infection since none of the women in the study had documented intra-amniotic infection, therefore suggesting an inflammation-related preterm birth (92). As the authors point out, however, ascending infection and documented intra-amniotic infection is not the only possible mechanism microbiota-related risk of preterm birth and this may also be related to inflammatory factors (93). Bacterial translocation to the uterus through the vasculature has also been demonstrated in *Fusobacterium nucleatum* (**Figure 2**) (17). However, critically, *F*. *nucleatum* did not persist in the liver and spleen, declining in abundance as time progressed (17). This contrasts with the placenta in which bacterial load increased with time (17). Specificity of *F*. *nucleatum* to the uterine cavity is evidence of the plausibility of the transport of bacteria from the blood to the uterus (17). Furthermore, this specificity also suggests that the uterine microenvironment provides a uniquely favorable niche for certain bacterial taxa. However, hematogenous spread of bacteria has strong critics whom refute the plausibility of the spread of bacteria through the body (32).

In addition to the microbiome and immune environment, metabolites produced by microbiota can also interact with host cells to have a positive or negative impact on the host (**Figure 3**). An example of a mutually beneficial relationship is the production of vitamins and SCFAs that are produced by the gut microbiome and can act not only as nutrients for cells but may also elicit beneficial epigenetic changes in the host as well as (in the case of propionate and acetate) serving as important satiety signal (94). Many other examples exist and a full description would be beyond the scope of this review (95). Conversely, metabolites produced by an unfavorable microbiome (or diet) may have negative impact on the host (**Figure 3**).

## FURTHER AREAS OF STUDY

Microbiota interactions with the host endometrial microenvironment will be an important area of research as we continue to elucidate potential mechanisms that drive disease in the uterus as well as reproductive outcome (69). The uterine microbiota composition may have unique consequences for the endometrial microenvironment due to the site-specific differences in anatomical and physiologic features throughout the FRT (2). We and others have shown the site-specific differences in host responsiveness to microbial products and bacteria throughout the FRT epithelia (2, 55, 84, 96, 97). Various models can effectively recapitulate the complex microenvironment of the FRT and have shown utility to understand host–microbiota interactions (98). For example, Laniewski et al. established and characterized a novel 3-D endometrial epithelial model to better understand host–microbiota interactions at this site (55). Assessing the impact of multiple bacterial species using synthetic combinations or patient-derived samples, to mimic the complex uterine microbiota, may help elucidate the host immune mechanisms in response to microbiota at this site (**Box 1**).

Similarities may exist between the host response mechanisms of the uterine microenvironment and other sites in the FRT (**Box 1**). For example, vaginal LB lower the pH of the vaginal microenvironment, which inhibits the colonization of dysbiotic species (85, 99). However, it is unclear how the pH of the intrauterine environment is altered by the presence of microbiota. While one study investigating the role of endometrial pH on reproductive outcome did not demonstrate a significant association with *Lactobacillus* abundance and low pH, this could be due to the relative levels of LB required to lower the pH of the endometrium or other biochemical mechanisms.

The physiological pH is an understudied aspect of the uterine microenvironment, which is likely to be influenced by the presence of and composition of microbiota. The influence of the vaginal microbiome on vaginal pH is well documented and profound (85). The uterine pH may have a similarly important association with particular microbiome compositions. It is even plausible that the vaginal pH may impact uterine pH through direct or indirect mechanisms. In the vaginal microbiome, LB play a crucial role in the modulation of pH through their production of lactic acid (85). In lieu of the consistent finding of LB in the uterine microbiome, important questions are raised as to whether uterine pH is altered by LB presence. However, based on decades of research it is unlikely that LB are found at high enough levels to maintain an acidic environment in the uterus; however, a lower physiological pH at this site may result in damage. The limited data currently available concerning uterine pH suggest that it resides at ~pH 7 (47, 100). There is a discrepancy between studies with it being reported that the uterine pH never exceeds 7.2 (100); however, Moreno et al. reported a range of 6.6–8.51 uterine pH across different patients. Further research is needed to define "normal" uterine pH as the implications of this finding may extend to fertility as well as reproductive and gynecologic sequelae. For example, the neonatal Fc receptor (FcRn), which plays a key role in trafficking immunoglobulins across mucosal tissues, including the uterus has been shown to be pH sensitive. At pH of 6–6.5 the receptor is functional; however at pH of 7, it is non-functional and inhibits transport of IgG, which has significant implications for sexually transmitted infections such as *Chlamydia trachomatis* as it has been shown that IgG translocation *via* FcRn significantly reduces infection (101). Moreno et al. found no association with pH and uterine LB dominance or reproductive success (47). However, additional research is needed in this area to better understand the physiological impact of the presence of uterine bacteria on uterine pH and the local microenvironment (**Figure 3**).

Future studies should aim at studying the functional relevance of the presence of microbiota in the uterine cavity in terms of pathophysiological mechanisms that contribute to disease pathogenesis (mechanisms outlined in **Figure 3**). Longitudinal studies assessing the stability of uterine microbiota could help discern whether bacteria colonize transiently (tourists/invaders) or whether there is a stable population (residents). Ascertaining the viability of bacteria will also aid in distinguishing whether microbiota are truly residents of the uterus that contribute to homeostasis or represent microbial DNA left behind from previous transient bacterial tourists or invaders.

## CONCLUSION

Based on the current literature evaluated in this review, the evidence for a "core" or bacterial resident population in the uterus is lacking and therefore the presence of uterine microbiota are likely reflective of bacterial tourists or invaders rather than a resident population that contributes to health and homeostasis. Uterine microbiota and specific bacterial species may be linked to critical health issues such as endometriosis, endometrial cancer and rates of IVF success. Public health programs will benefit from expanded studies of host–microbiota and host–metabolome interactions within the FRT (summarized in **Figure 3**). For optimal success, future studies require well-designed and larger patient cohorts to elucidate interactions between the uterine microbiota and host in the context of women's health. Specific species or microbiota compositions may provide indicators or predictors of disease, participate as mere passengers or act as microbial drivers of disease. As evidence for interactions between the microbiome at mucosal sites increases, other diseases of dysbiosis may drive poor reproductive and gynecologic health outcomes by impacting the uterine microbiota. Studies are needed to further investigate if a "core" or resident uterine microbiota exists and the contributions to health and homeostasis. Furthermore, additional research is warranted to elucidate the functional impact of uterine microbiota or specific bacterial species that may participate as tourists or microbial invaders of this mucosal site and the impact these microbes have on the physiology of the local endometrial microenvironment.

## AUTHOR CONTRIBUTIONS

MH-K designed the scope and organization of the review and supervised writing. JB and MH-K conducted literature reviews, figure, and table construction and contributed to the writing of the manuscript. DC and MH-K critically edited and reviewed the complete manuscript, tables and figures. All authors approved the final manuscript for submission.

### ACKNOWLEDGMENTS

The authors would like to thank the members of the Herbst-Kralovetz research team and Dr. Charles Armitage for their thoughtful discussion on this topic and Dr. Kerr Whitfield for

#### REFERENCES


critical review of the manuscript. The authors would also like to thank Arizona Health Sciences library for providing the program "Visio" for timeline construction.

## FUNDING

This work was supported by the Valley Research Partnership Grant #VRP26 (DC and MH-K) and the Mary Kay Foundation Translational Research Grant #017-48 (MH-K).

implication of oral bacteria in preterm birth. *Infect Immun* (2004) 72(4):2272– 9. doi:10.1128/IAI.72.4.2272-2279.2004


salpingo-oopherectomy. *Fertil Steril* (2017) 107(3):813.e–20.e. doi:10.1016/j. fertnstert.2016.11.028


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2018 Baker, Chase and Herbst-Kralovetz. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Commentary: Uterine Microbiota: Residents, Tourists, or Invaders?

#### *Signe Altmäe1,2\**

*1Department of Biochemistry and Molecular Biology, Faculty of Sciences, University of Granada, Granada, Spain, 2Competence Centre on Health Technologies, Tartu, Estonia*

Keywords: endometrium, menstrual cycle, microbiome, microbiota, seminovaginal microbiota, uterus

#### **A Commentary on**

#### **Uterine Microbiota: Residents, Tourists, or Invaders?**

*by Baker JM, Chase DM, Herbst-Kralovetz MM. Front Immunol (2018) 9:208. doi: 10.3389/fimmu. 2018.00208*

The recently published review by Baker et al. summarizes the current status of uterine microbiota with the aim to promote research priorities and discussion on this novel research field (1). The authors are to be congratulated on this much anticipated review as microbiota in the uterus is one increasing research area, though poorly investigated microbial niche relative to other organs. However, emerging evidence is beginning to indicate that the uterine microbiota has important implications for female (reproductive) health and disease, and it is becoming evident that the concept of sterile uterus is outworn, although the true core uterine microbiota still needs to be assessed.

#### *Edited by:*

*Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia*

#### *Reviewed by:*

*Douglas Mark Ruden, Wayne State University, United States*

#### *\*Correspondence:*

*Signe Altmäe signealtmae@ugr.es*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 21 May 2018 Accepted: 30 July 2018 Published: 24 August 2018*

#### *Citation:*

*Altmäe S (2018) Commentary: Uterine Microbiota: Residents, Tourists, or Invaders? Front. Immunol. 9:1874. doi: 10.3389/fimmu.2018.01874*

In their comprehensive review, Baker et al. present established and putative bacterial transmission routes between uterine microbiota and distal sites, where they highlight (a) hematogenous spread of bacteria through either oral or gut route, (b) ascension of bacteria through the cervix, and (c) other routes such as retrograde spread through fallopian tubes, assisted reproductive technology-related procedures or insertion/removal of intrauterine devices together with its potential aid in ascension through the "tails" of the device (1).

There is, however, another important bacterial transmission route that has high potential to influence uterine microbiota that the authors have missed to present—the seminal microbiota. Even before the era of 16S RNA analysis, it was postulated that "it is difficult to envision that a mucosa continuously exposed to microorganisms present in the lower genital tract and that is regularly invaded by sperm that can carry microorganisms into the endometrial cavity may be free of bacteria" (2). Indeed, a term "complementary seminovaginal microbiota" has been recently proposed (3). Recent studies are demonstrating that bacteria are shared among partners and that partners influence the species composition of each other's reproductive tract microbiota (3–6), with sexual debut and activity having significant impact (4, 5). In line with sexual activities and hematogenous spread of bacteria emanating from the gut and oral microbiota, oral and anal sex can influence the microorganismal continuum, as is known with different diseases caused by sexually transmissible pathogens (e.g., oral lesions, proctitis, proctocolitis, and enteritis) (7, 8). Interestingly, the placental microbiome resembles that of the oral cavity more than that of the gut or even vagina (9). In short, semen serves as a perfect medium for the transmission of microorganisms (being slightly basic and enriched with carbohydrates it creates an ideal habitat for microorganisms), which should be considered as one important route of microorganismal tourism or invasion, with potential to become residents in the uterus.

Furthermore, Baker et al. mention briefly in their review that bacterial seeding of the uterus has important ramifications on maternal–fetal transfer of microbiota and postnatal health (1). Also here the paternal contribution should be highlighted, as it is clear that male contribution to offspring is more than just the haploid genome complement in sperm. It has been recently proposed that fathers may transmit information *via* microbiota to their partners and progeny (10). Novel studies are providing knowledge of possible mechanisms of microbiota's role on offspring, where the influence on methylome and transcriptome changes, and on microglia has been shown (11, 12).

There is, however, one aspect that needs to be clarified, as Baker et al. conclude in their review that it is not clear if the uterine microbiota changes during the menstrual cycle (1). The authors mention that the only study assessing uterine microbiota across two different time points of the menstrual cycle has been Moreno et al. (13). In that study, the uterine microbiome was similar at the two hormonal stages, but as the authors adequately conclude that these results should be viewed with some caution (13). Given the fact that hormonal changes influence vaginal microbiota (14, 15), that microbiota is influenced by hormones (16), and that the use of gonadotrophin-releasing hormone agonist resulted in a shift of uterine microbiome composition (17), one would expect that also uterine microbiota is influenced by sex hormones during natural menstrual cycle. Indeed, what Baker et al. have missed to present in their review, is the study results by Chen et al. (18), where microbiota continuum along the female reproductive tract on 95 women in the proliferative and secretory phases were studied.

TABLE 1 | Enrichment analysis of microbial KEGG pathways in the proliferative and the secretory phases in the endometrium from 80 reproductive-aged women (18) (adapted with permission from Nature Publishing Group).

Top microbial KEGG pathways enriched in the endometrium


*This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94942, USA.*

#### REFERENCES


Operational taxonomy units that led to optimal classification between the two phases in the uterus included *Sphingobium* sp., *Propionibacterium acnes*, and *Carnobacterium* sp. Interestingly, *P. acnes*, that has previously been identified in the placenta and follicular fluid, was more abundant in the secretory phase uterus (18). Enrichment analysis identified different pathways associated with increased bacterial proliferation (pyrimidine and purine metabolism, and aminoacyl-tRNA biosynthesis) in the proliferative phase compared to the secretory phase (**Table 1**).

In addition, there are two new studies published that support the concept of microbial changes throughout the menstrual cycle, where bacterial continuum between proliferative and secretory phases differed in (1) endometria from dysmenorrhea and menorrhagia patients (19) and (2) in the fallopian tubes (20). Clearly, more studies are required for identifying the "baseline" microbial continuum in the uterus, nevertheless the first studies are showing the uterine microbiota differences along the menstrual cycle. Furthermore, the importance of the microbiota in regulation of rhythmic biological changes has recently been proposed (21), and that fluctuating microbial community structures might direct hormonal changes (22), it is tempting to hypothesize that similar dynamics might be involved in the female menstrual cycle (23).

As it stands, the assessment of uterine microbiota suffers from many limitations, there is a need for functional studies as well as for well designed and larger sample cohorts to unravel the role of microorganisms (and not only bacteria but also viruses, fungi, microscopic eukaryotes, and archaea) in uterine health and pathologies. Nevertheless, the novel studies are indicating that microbiota is another piece in the complex mechanism contributing to the cogwheels of hormones and physiological adaptations that are required for successful embryo implantation and pregnancy.

#### AUTHOR CONTRIBUTIONS

SA conceived and wrote the manuscript.

#### FUNDING

SA is funded by grants RYC-2016-21199 and ENDORE SAF2017-87526 from the Spanish Ministry of Economy, Industry and Competitiveness (MINECO), and European Regional Development Fund (FEDER).


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be constructed as a potential conflict of interest.

*Copyright © 2018 Altmäe. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The Role of Skin and Orogenital Microbiota in Protective immunity and Chronic immune-Mediated inflammatory Disease

*Young Joon Park1 and Heung Kyu Lee1,2\**

*1Graduate School of Medical Science and Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea, 2KAIST Institute for Health Science and Technology, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea*

The skin and orogenital mucosae, which constitute complex protective barriers against infection and injuries, are not only the first to come into contact with pathogens but are also colonized by a set of microorganisms that are essential to maintain a healthy physiological environment. Using 16S ribosomal RNA metagenomic sequencing, scientists recognized that the microorganism colonization has greater diversity and variability than previously assumed. These microorganisms, such as commensal bacteria, affect the host's immune response against pathogens and modulate chronic inflammatory responses. Previously, a single pathogen was thought to cause a single disease, but current evidence suggests that dysbiosis of the tissue microbiota may underlie the disease status. Dysbiosis results in aberrant immune responses at the surface and furthermore, affects the systemic immune response. Hence, understanding the initial interaction between the barrier surface immune system and local microorganisms is important for understanding the overall systemic effects of the immune response. In this review, we describe current evidence for the basis of the interactions between pathogens, microbiota, and immune cells on surface barriers and offer explanations for how these interactions may lead to chronic inflammatory disorders.

Keywords: immune response, microbiota, skin mucosa, orogenital mucosa, inflammation

## INTRODUCTION

The skin and mucosae constitute complex protective barriers against infection and injuries. Oral and genital mucosae differ from those of lung or gut as they are in constant contact with extrinsic stimulation, such as food, medication, and physical trauma. The skin and orogenital mucosae are similar with respect to function, as these barriers are not only the body's first line of defense against pathogens but also hosts for a substantial number of commensals, including bacteria, fungi, and viruses. These organisms influence the immune response at the barrier sites and can lead to aberrant responses and chronic inflammation. Additionally, dysbiosis caused by specific bacteria species in the skin and orogenital mucosae is associated with tissue-specific chronic immune-mediated disease. The specific species responsible for this disease include *Staphylococcus aureus* on skin, *Porphyromonas gingivalis* on oral mucosa, and *Gardnerella vaginalis* on vaginal mucosa. The association between certain systemic diseases and changes in the microbiota of these barrier surfaces has just recently been reported, due to the fact that the microbiota in these areas is given less

#### *Edited by:*

*Yves Renaudineau, University of Western Brittany, France*

#### *Reviewed by:*

*Mario M. D'Elios, University of Florence, Italy Sandip Chakraborty, College of Veterinary Sciences and Animal Husbandry, India*

*\*Correspondence:*

*Heung Kyu Lee heungkyu.lee@kaist.ac.kr*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 08 September 2017 Accepted: 19 December 2017 Published: 10 January 2018*

#### *Citation:*

*Park YJ and Lee HK (2018) The Role of Skin and Orogenital Microbiota in Protective Immunity and Chronic Immune-Mediated Inflammatory Disease. Front. Immunol. 8:1955. doi: 10.3389/fimmu.2017.01955*

attention compared to the microorganisms residing in the gut. Understanding the interactions between the local commensals and the immune system at these barrier sites is important, however, as the initial interaction occurs at these sites and can lead to widespread disease. Thus, in this review, we describe the major immune mediators of interactions between the surface barrier tissue and local microorganisms with respect to the representative disease.

#### SKIN MICROBIOTA

Human skin, which comprises the body's largest organ, is home to many commensals. Across the 1.8 m2 of the skin surface, one million bacteria reside on each square centimeter, yielding a total of more than 1010 bacterial cells on the human body (1). While the gut supports 1014 bacterial cells per healthy individual, the skin contains the second highest number and diversity of bacterial cells with as many as 40 different species per individual (2, 3). The difference between skin microbiota and that of the large intestine stems largely from the direct contact of the skin with the surrounding environment. The combination of the multiple niches with various pH and temperatures created by these interactions with the environment and the diversity of the epidermal compositions and dermal appendages of the skin leads to the differences and diversity of the resident bacteria. The initial microbial skin colonization depends on the delivery mode. Vaginal delivery results in babies with microbiota similar to that of mother's vagina, whereas babies born *via* cesarean section acquire microbiota related to skin. During puberty, the skin microbiota goes through a major transition with domination of lipophilic bacteria, such as *Propionibacterium* and *Corynebacterium.*

#### The Skin Immune System

The immune system of the skin is composed of complex network of keratinocytes and immune cells, including the skin-specific antigen-presenting cells, the Langerhans cells. As the first line of defense against infection, the innate and adaptive immune systems are delicately controlled, and even a slight interference to the network can initiate an inflammatory response. The immunemicrobial interaction is therefore quite important, despite the fact that the skin microbiota are not required for immune system organization (4). When in symbiosis, which is defined as a persistent balanced state between the skin and skin-resident microorganisms, a specific member of the microbiota acts to preserve barrier functions, but when the barrier is breeched by intrinsic or extrinsic factors, the very same member can initiate an immune response.

#### Skin Microbiota and Innate Immunity

Epidermal keratinocytes release antimicrobial peptides (AMPs), such as cathelicidins and β-defensins, which comprise the majority of these AMPs (**Figure 1A**). The AMPs, which are also produced by sebocytes, provide microbicidal activity against pathogens and may also trigger the inflammatory response. Some of these AMPs are actually controlled by the microbiota. For example, *Propionibacterium* species induce the expression of AMPs in human sebocytes (5).

In addition to microbiota regulation of AMPs, adipose tissue that interfaces with the skin may also contribute to the innate responses. In invasive *S. aureus* infection, local preadipocytes rapidly proliferate, leading to an expansion of the dermal fat layer, and simultaneously, cathelicidin is markedly increased in the adipose tissue and the infected skin (6). This response by the local adipose tissue provides both physical and immunologic antimicrobial defense.

Another response of the skin cells to bacterial pathogens is *via* pattern recognition receptors (PRRs). Nucleotide-binding oligomerization domain containing 2 (NOD2), an intracellular PRR, recognizes bacterial peptidoglycans of both Gram-positive and Gram-negative bacteria. Experimental manipulation of NOD2 led to bacterial dysbiosis with local changes in AMPs (7). Hence, although the specific mechanism has yet to be elucidated, NOD2-mediated immune surveillance appears to function in the immune response in association with the cutaneous microbiota.

Skin-resident microbes are also capable of regulating components of the complement system. Mice deficient in complement develop altered skin microbial communities (8). As inhibition of complement component C5a receptor in germ-free mice leads to decreased levels of AMPs and proinflammatory factors, the skin microbiome closely interacts with the complement system.

A recent study using a Leishmaniasis skin infection model demonstrated that human skin microbiota is altered following infection with *Leishmania* and that this dysbiosis is transmissible and capable of exacerbating skin inflammation *via* recruitment of neutrophils and production of interleukin (IL)-1β [**Figure 1B** (9)]. The skin microbiota also controls expression of IL-1α independently (4). The IL-1 cytokine is essential for the initiation and amplification of immune responses, and thus, the acute immune response is considered to be under the influence of host skin commensal interactions.

#### Skin Microbiota and Adaptive Immunity

Skin microbiota is capable of promoting both the innate and the adaptive immune responses to limit pathogen invasion and maintain homeostasis. Mice without adaptive immunity fail to control their skin microbiota, allowing pathogenic microbial invasion to occur (10). The adaptive response, however, does not occur alone and is an extension of the innate response. For example, increased IL-1 production is followed by production of IL-17 and interferon-γ (IFN-γ) from dermal T cells. Reciprocally, T helper 17 (Th17) cells along with IL-17A-producing γδ T cells are reduced in mice lacking IL-1R1 (4). Introduction of *Staphylococcus epidermidis* to germ-free mice restores the production of IL-17A, indicating that *S. epidermidis* as part of the skin commensal microbiome potently induces Th17 cells as well as other T cells that express IL-17A. Other human skin commensals, such *as Corynebacterium pseudodiphtheriticum*, *Propionibacterium acnes*, *and Staphylococcus aureus*, also increase the number of skin IL-17A<sup>+</sup> cells and IFN-γ+ cells but do not induce as prominent a response as the introduction of *S. epidermidis*. In addition, *S. epidermis* is the only commensal species that increased the frequency of CD8<sup>+</sup> T cells in the skin. These CD8<sup>+</sup> T cells are capable of producing either IFN-γ or IL-17A and homed only to the epidermis. These commensal-specific

Figure 1 | Skin microbiota and immunity. (A) The microbiome is more diverse in healthy skin. *Staphylococcus epidermidis*, *Acinetobacter* spp., and Gram-positive anaerobe cocci (GPAC) exhibit protective features against atopic disease. Keratinocytes and sebocytes release antimicrobial peptides (AMPs), and associations with skin commensals, such as *Propionibacterium* spp., have been demonstrated. Pattern recognition receptors (PRRs), such as nucleotide-binding oligomerization domain containing 2, recognize bacterial peptidoglycans to increase AMP production. Skin commensals also control the expression of interleukin (IL)-1, and increased IL-1 production is followed by IL-17A and subsequent interferon-γ production by dermal T helper 17 and γδ T cells. Regulatory T (Treg) cells reside primarily around fair follicles and interact with commensals within a specific time window to achieve immune tolerance. In addition to classical Foxp3+ Treg cells, Foxp3− Treg cells also interact with the bacterium *Vitreoscilla filiformis* to induce Foxp3− Treg cell differentiation. (B) The dysbiosis induced by Leishmaniasis infection is not only transmissible but also exacerbates skin inflammation *via* neutrophil recruitment and production of IL-1β. (C) In atopic dermatitis, *Staphylococcus aureus* proliferates and microbial diversity decreases concomitantly. Additionally, along with epithelial barrier disruption, proinflammatory cytokines are produced. Activation of T cells to Th2 cells occur *via* two mechanisms: degranulation of mast cells from δ-toxin and downregulation of IP-10 and other Th1 cell-recruiting chemokines.

IL-17A-producing T cells successfully enhance barrier immunity and limit pathogen invasion (11). Skin-resident Th17 cells that are affected by the skin microbiota are also independent of those from the gut microbiota, suggesting that Th17 cells of the barrier sites are regulated in a "compartmentalized" manner by local commensals (4).

In the skin of both mice and humans, Foxp3<sup>+</sup> regulatory T (Treg) cells are present in the dermis, especially surrounding the hair follicles, where skin-resident microorganisms also reside (12). Occupation of these hair follicles by commensals may be coupled with a regulatory response by Treg cells to limit abnormal inflammatory response against them (13). The time window during which this commensal-specific tolerance is achieved may be as early as the neonatal period. Precolonization of neonatal mice with *S. epidermidis* suppresses skin inflammation upon *S. epidermidis* challenge. A wave of Foxp3<sup>+</sup> Treg cell infiltration in the skin occurs in the second week in neonate mice, and this infiltration is accompanied by higher levels of CTLA-4 and ICOS, which are both critical mediators of immune tolerance (14). The skin of germ-free mice, however, has a high frequency of Foxp3<sup>+</sup> Treg cells compared with the skin of specific pathogen-free (SPF) mice (4). Thus, the underlying mechanism of Tregs in controlling host-microorganism dialog has yet to be fully elucidated. Foxp3<sup>+</sup> Treg cells are known for their function in promoting microbial persistence (15), but Foxp3<sup>−</sup> Treg cells may also be involved in establishing immune tolerance of certain microbiota. *Vitreoscilla filiformis*, a Gram-negative bacterium, induces dendritic cells to prime naïve T cells to type 1 Treg cells without Foxp3 expression. Furthermore, epicutaneous application of *V. filiformis* lysate induces IL-10high T cells and inhibits T-cell proliferation in NC/ Nga mice that exhibit atopic dermatitis (AD)-like inflammation (16). Nonetheless, as Treg cells are not the sole inducers of immune tolerance, much remains to be revealed about the basis of tolerance of constitutive commensals.

## ATOPIC DERMATITIS

Atopic dermatitis is a prototype immune-mediated inflammatory skin disorder with prominent association with commensals. AD is considered to be an initiating stage of abnormal systemic Th2 response that progresses to allergic rhinitis and asthma. Some *Staphylococcus* spp., and in particular, *S. aureus*, have been found to be associated with AD and AD flares. Colonization of *S. aureus* on barrier-disrupted murine skin increases expression of IL-1β, IL-6, and TNF-α, demonstrating a pivotal role of *S. aureus* in promoting skin inflammation [**Figure 1C** (17)]. *S. aureus* colonization results in a Th2 immune response instead of a Th1 response despite the presence of its superantigen, which elicits a predominant Th1 cytokine profile, has been explained by two studies. First, δ-toxin released by *S. aureus* induces the degranulation of dermal mast cells and in turn, promotes the Th2 response (18). Second, components of the *S. aureus* cell wall downregulate IP-10 independent of IL-10; trigger activation of MAPK, p38, and ERK; and inhibit STAT1 signaling in monocytes, contributing to the abrogation of Th1 cell-recruiting chemokines (19). In contrast, *S. epidermidis* colonization provides innate immune signals to establish a functional threshold for adaptive immunity to facilitate pathogen control (20). As expected, overabundant colonization of *S. aureus* correlates with worsening AD in both mice and humans, but whether the loss of microbiome diversity leads to skin flares or skin flares result in reduced microbiotic diversity has not been illuminated (21). Dysbiosis with abundant *Staphylococcus* spp. and *Corynebacterium* spp., however, was observed in a genetically engineered murine model of AD. Also, administration of antibiotics to mice deficient in epidermal ADAM17 (Adam17<sup>Δ</sup>Sox9) prevented eczematous lesion development, prevented an increase in skin-infiltrating T-cell numbers, and paradoxically, increased diversity of the skin microbiome. The study clearly showed that dysbiosis and overabundance of *S. aureus* contribute to AD-like lesions and are likely responsible for acute atopic flares in humans (22). The most recent reports also reveal a role for *S. aureus* in acute AD and new-onset pediatric AD, both of which show Th17 immune polarization. Epicutaneous *S. aureus* exposure induces skin inflammation by triggering IL-36R/MyD88 signaling and consequent IL-17 production from T cells (23, 24). The findings provide an explanation of increased IL-36α/γ transcripts and the increased number of Th17 cells in human AD skin (25, 26). On the other hand, skin commensals, including *S. epidermidis,* were found to inhibit *S. aureus via* production of AMPs that selectively kill *S. aureus*, and reintroduction of these commensals to human subjects decreased colonization by *S. aureus*, confirming the role of *S. epidermis* to prevent dysbiosis and initiation of inflammation (27). In addition to *S. epidermidis*, *Acinetobacter* species and Gram-positive anaerobe cocci (GPAC) may also play a protective role. *Acinetobacter* species were found to not only be more abundant in healthy subjects compared to atopic subjects but also to be associated with expression of anti-inflammatory molecules by peripheral blood mononuclear cells. *Acinetobacter*

species induced strong Th1 and anti-inflammatory responses by immune cells and skin cells, suppressing allergic sensitization and lung inflammation in the skin (28). The abundance of GPAC was found to be low in human skin with a filaggrin deficiency, an important factor involved in AD pathogenesis. The cocci generated an inflammatory response distinct from that of *S. aureus in vitro* (29). These studies suggest that the loss of microbiome diversity is linked with a higher allergic response and subsequent inflammation.

## ORAL MICROBIOTA

The oral microbiota is also an important part of the human microbiota. A minimum of 700 different species are present in the human oral cavity, inhabiting diverse locations such as the tongue, hard tissues, and dentures (**Figure 2A**) (30). Similar to the skin, the microbiota of the oral cavity differs substantially based on the ecological niche (31). One unique characteristic of the oral microbiota is that due to saliva, which washes the mucosa and soft tissue surfaces, only a non-pathogenic monolayer exists on these surfaces. On the contrary, biofilm develops on hard tissue surfaces. Biofilm is an aggregate of extremely heterogeneous bacteria and contains sufficient nutrients to sustain the metabolic needs of the microbiota (32). Dental plaque, which is a typical form of biofilm, is known for its potential to induce inflammation. Under certain circumstances, for example, a change in pH, oxygen tension, or host immune status, the resident microbiota of the biofilm can transform to a pathogenic population (33). Pathogenic colonization with or without mucosal epithelial disruption impairs host immune responses and may result in disease, such as gingivitis or periodontitis.

The majority of bacteria in the oral cavity belong to five phyla: *Actinobacteria*, *Bacteroidetes*, *Firmicutes*, *Fusobacterium*, and *Proteobacteria*. As the microbial communities from different oral sites are distinct, the salivary microbiome is generally used as a representative microbiome (34). In a Japanese study on the salivary microbiome of 2,343 adults aged 40 years or older, the population could be divided in to two groups based on the composition of the microbiome. The group with more abundant *Prevotella histicola*, *Prevotella melaninogenica*, *Veillonella parvula*, *Veillonella atypica*, *Streptococcus salivarius*, and *Streptococcus parasanguinis* was associated with poorer oral hygiene and poorer general conditions, suggesting that the oral microbiota reflects both the local and the systemic immune status (35). Despite the characterization of the overall composition of the oral microbiome (36), studies on the immunologic role of each organism in the oral microbiome, notably bacteria without virulence factors, are remarkably scarce. The diversity of the oral microbiome based on spatial niches and constant temporal change due to its nature as an open system exposed to exogenous microbes may underlie the rarity of such studies, although a few studies have been completed. One example of such a study showed that bacteriocin produced by *S. salivarius* inhibits the growth of Gram-negative species associated with periodontitis and halitosis *in vitro* (37), in agreement with evidence that *S. salivarius* as a probiotic is beneficial to halitosis *in vivo* (38, 39).

### PERIODONTAL MICROBIOME, PROTECTIVE IMMUNITY, AND PERIODONTITIS

The gingival sulcus and periodontal pocket are the most-studied niches of microbial colonization in the oral mucosa. The crevice between the hard surface of the teeth and the gingiva harbors microbial communities that interact with the mucosal epithelial cells. Phylum *Proteobacteria*, particularly the *gammaproteobacteriae* of genus *Acinetobacter, Haemophilus,* and *Moraxella*, were most prevalent in healthy gingival sulci less than 4 mm deep. Periodontal pockets are formed when the attachment between the gingivae and teeth is lost. The space then can become colonized with anaerobic bacteria. The microbiota highly associated with pockets greater than 4 mm in depth include *Spirochetes* genus *Treponema*; *Synergistetes* genus *Sinergistes*; *Bacteroidetes,* such as genera *Porphyromonas*, *Prevotella,* and *Tannerell*a; and *Fusobacteria* genera *Fusobacterium* and *Leptotrichia* (40, 41). Development of culture-independent techniques has expanded the range of associated species that were previously uncultivable or underappreciated. These species include *Filifactor alocis*, *Peptostreptococcus stomatis*, *Prevotella denticola*, *Porpyromonas*  *endodontalis*, *Anaeroglobus geminatus*, and *Eubacterium saphenum* (40, 42).

Toll-like receptors (TLRs) and other PRRs act as immunologic sensors of the biofilm. The epithelial cells activated by microbeassociated molecular patterns (MAMPs) of the biofilm increase its proliferative rate, expression of adhesion molecules, IL-1β production, and production of AMPs, such as calprotectin (S100A8/ A9) and defensins (27). Calprotectin from gingival epithelial cells (GECs) functions both extracellularly and intracellularly *via* incorporation into neutrophil extracellular traps and aiding innate intracellular immunity, respectively, providing resistance against pathogenic bacteria (31, 43).

In periodontal disease (also termed periodontitis), a chronic inflammatory disease that results in the destruction of bony apparatus, the components of the pattern recognition system, for example TLR2, initiate the sustained inflammation and ultimately induce the induction of bone loss in mice (44). Recently, the NOD1 cytoplasmic PRR was found to be responsible for the majority of periodontal bone destruction. Interestingly, monocolonization of germ-free animals with a single specific bacterium NI1060, which exhibits a greater than 60% homology with the coding sequences of *Aggregatibacter actinomycetemcomitans*, was sufficient to promote periodontal bone destruction in a NOD1-dependent manner (45). These findings demonstrate that a single bacterial component from a normal commensal microbiome can cause destructive disease through activation of a PRR in oral mucosa. Calprotectin is also abnormally activated in periodontitis and has been implicated in the pathogenesis of this disease (46).

T helper 17 cells have been implicated in mediating protective immunity as well as pathogenic inflammatory response at multiple barrier sites, including the oral cavity, skin, and gut. Gingival Th17 cells develop independently of commensal microbe colonization. As the frequencies of gingival Th17 cells are unchanged in germ-free mice compared to SPF mice, the oral microbiome does not appear to be a primary driver of gingival Th17 development, and this finding is in contrast to the situation in the skin and gastrointestinal tract. IL-6 expression in response to mechanical damage of epithelial cells instead is the major promoting factor for accumulation of Th17 cells and subsequent elevation of epithelial antimicrobial peptides and neutrophil chemo-attractants in the oral cavity (47). These data should be carefully interpreted, because it is evident that dysbiotic microbial communities aggravate periodontitis possibly in relation to accumulated Th17 cells (48–50). Mechanical damage may be critical for Th17 cell proliferation but also serves as an amplifier of the oral inflammatory response.

### *Porphryomonas gingivalis*

*Porphyromonas gingivalis*, a member of the phylum *Bacteroidetes*, is a keystone oral pathogen that evades the immune system and alters local host responses. While this Gram-negative anaerobic rod-shaped bacterium mainly resides in biofilms and behave as a commensal, it also invades host cells, such as GECs in certain situations. This bacterium occupies its own replicative niche within autophagosomes (51). This microorganism also underlies the "dysbiosis hypothesis," which surmises that *P. gingivalis*, even at low colonization levels, triggers changes in the amount and composition of the oral commensal microbiota, leading to inflammation and ultimately periodontal bone loss. Interestingly, the commensal microbiota and complement are both required for this *P. gingivalis*-induced bone loss, as *P. gingivalis*, although necessary, is not sufficient to induce inflammation (52). Thus, *P. gingivalis* exerts its pathogenic effects *via* its ability to induce dysbiotic microbial communities (**Figure 2B**). Manipulation of the host immune system (called immune subversion) by *P. gingivalis*, therefore, is essential and is therefore discussed in detail in this review.

#### *P. gingivalis* and Innate Immunity

As described above, PRRs, such as TLR2 and NOD1, are mediators of the innate immune response. *P. gingivalis* coactivates TLR2 and C5a receptor (C5aR) in neutrophils, resulting in crosstalk that leads to ubiquitination and proteasomal degradation of MyD88 and inhibits the host-protective antimicrobial response. Moreover, this C5aR-TLR2 crosstalk activates PI3K, which prevents phagocytosis through inhibition of RhoA activation and actin polymerization and stimulates an inflammatory response (53). Upon initiation of inflammation, GECs secrete the chemoattractant IL-8 to recruit neutrophils. *P. gingivalis* inhibits the secretion of IL-8 from GECs, lowering the number of neutrophils recruited to the site of inflammation. Another way that *P. gingivalis* inhibits the polymorphonuclear leukocyte function is through the binding of whole cells and lipopolysaccharides of the bacteria to adhesion molecules, interrupting leukocyte diapedesis. Other mechanisms to manipulate neutrophil activity include bacterial binding to fMLF receptor and PPAD-citrullinated C5a, resulting in inhibition of chemotaxis, and regulation of triggering receptor expressed on myeloid cells 1 (TREM-1) *via* both gingipain-dependent and -independent mechanisms, rendering the inflammatory response appropriate for the bacteria (54).

Gingipains are cysteine proteases that act as critical enzymes in periodontitis pathogenesis. In addition to cleavage of TREM-1 by Arg-gingipain, gingipain is responsible for proteasomal degradation of *P. gingivalis* and increased inflammation, as this enzyme cleaves C5 to C5a, leading to My88D degradation and PI3K activation (53). Gingipain is also associated with increased IL-33 production by GECs *via* PAR-2-p38/NF-κB signaling (55). IL-33 is an alarmin released in response to tissue damage, and increased IL-33 downregulates AMPs, such as LL-37 (56). Moreover, IL-33 induces receptor activator of NF-κB ligand, a crucial osteoclastogenic factor (57).

In macrophages, C5aR-TLR2 crosstalk by *P. gingivalis* suppresses production of nitric oxide and IL-12 simultaneously, inhibiting *P. gingivalis* killing but preserving the ability of these cells to elicit an inflammatory response (58, 59). *P. gingivalis* binds to complement receptor 3 (CR3) on macrophages to downregulate IL-12 selectively. This binding further allows *P. gingivalis* to persist intracellularly, as CR3 is not coupled to microbicidal mechanisms directly (60).

#### *P. gingivalis* and Adaptive Immunity

Macrophages engage microbes not only to eliminate them but also to work as antigen-presenting cells. Currently, the M1 subpopulation of macrophages is considered to be responsible for inflammatory function, and *P. gingivalis* clearly triggers M1-type-associated inflammatory pathways (61, 62). A recent study, however, indicated a novel M1 macrophage appears in response to *P. gingivalis* with the addition of IFN-γ, which significantly upregulated proinflammatory cytokines, such as IL-1β, IL-6, and TNF-α, and lowered production of chemokines related to T-cell recruitment (IL-12, IL-23, and CXCL10). These effects are in stark contrast to those induced by *A. actinomycetemcomitans*, another representative oral pathogen. Thus, *P. gingivalis* may have a unique capacity to alter the programmed course of the hyperinflammatory and T-cell immunomodulatory M1 macrophages (63).

*Porphyromonas gingivalis* appeared to drive dendritic cells, however, toward a Th2 antigen-presenting cell phenotype, at least *in vitro* (64). A Th2 cytokine-mediated inflammatory response was observed in association with *P. gingivalis*, and IL-33 was again identified recently using *P. gingivalis* gingipain-null mutant KDP cells. Of the various virulence factors, gingipain was entirely responsible for the IL-33 increase that was observed in human GECs (55). As evaluation of the protein levels in cells activated by *P. gingivalis* revealed no obvious Th1- or Th2-skewed profiles (65) and IL-1β and IL-6 were produced by CD4<sup>+</sup> T cells (66), the specific response of Th cells induced by this bacterium remains to be fully delineated. Notably, *P. gingivalis* was posited to stimulate myeloid antigen-presenting cells to drive Th17 polarization, and inactive gingipains selectively generated a Th17 phenotype in an IL-6-dependent manner. Inhibition of IL-6 signaling in dendritic cells led to a significant depletion of the Th17 population without similar effects on other T-cell subsets. These studies unveiled the possibility that Th17 cells are involved in the pathogenesis of periodontitis and confirmed that IL-6 signaling is an attractive target for treatment of this disease (49).

Despite the paucity of *in vivo* evidence of *P. gingivalis* manipulation of the adaptive immune response, this organism should still be considered a keystone species and successful pathogen with the ability to modulate adaptive immunity, as *P. gingivalis* disables the overall host response while simultaneously enhancing the inflammatory response and the pathogenicity of a polymicrobial community.

#### VAGINAL MICROBIOTA

The human vaginal mucosa is home to abundant microflora. The environment is exposed to unique foreign substances, such as spermatozoa and sexually transmitted pathogens. The environment is also very versatile, constantly changing due to menstrual cycle changes of the female body. In addition, the composition of the microbial community is largely influenced by events such as pregnancy and menopause. The commensals of the vagina change throughout a woman's life and function as a main component of the vaginal mucosal defense against pathogens.

Out of the ~200 bacterial species that reside in the vagina, *Lactobacillus* species predominantly colonize the vaginal tract (**Figure 3A**). Currently, more than 120 *Lactobacillus* species have been described and comprise more than 70% of resident bacteria in women. *Lactobacilli* are believed to contribute to the immunity in a healthy vagina. *Lactobacilli* produce significant amounts of lactic acid, ensuring the environment maintains a relatively low pH. Furthermore, these bacteria produce H2O2, which is presumed to contribute to protection of the vagina from pathogens, as H2O2-producing vaginal *Lactobacillus* spp. are more protective against bacterial vaginosis (BV) than those that do not produce H2O2 (67, 68).

The vaginal mucosa of a single woman is usually dominated by one or two species of *Lactobacillus*, and attempts to classify women by lactobacill*i* and other microbiota species have shown that ethnicity affects the microbiome. Caucasian women are dominated by *L. iners*, black and Hispanic women are dominated by *L. genseni*, and Asian women are dominated by *L. crispatus*. Similarly, reproductive-aged women have been classified to five community state types (CSTs). These types include women who harbored *L. crispatus* preferentially (CST I), women who were dominated by *L. gasseri* (CST II), and women who were dominated by *L. iners* (CST III). Furthermore, CST IV was initially defined as a complex mixture of bacteria, including *Prevotella*, without any Lactobacillus species dominance. Additionally, the presence of *L. crispatus* and/or *L. iners* without dominance was assigned as CST IV-A, and no or minimal lactobacilli was assigned as CST IV-B. The clinical significance of these groupings has yet to be fully explored, but once established, the vaginal microbiota is stable, despite frequent interruptions, such as sexual behavior (69, 70).

The vaginal microbiota is also influenced by gene polymorphisms, and the innate immune response to the microbiota in the female genital tract has been found to be associated with these genetic variants. Single-nucleotide polymorphisms that disrupt immune recognition are associated with increased susceptibility to disruption of vaginal microbiota and vaginal infections (71, 72). Especially, polymorphisms in IL-4, IL-10, TLR4, and TNF-α genes have been shown to induce aberrant responses to BV-associated bacteria and preterm delivery (72, 73). Of note, periodontal disease and BV are both influenced by gene polymorphisms and are both associated with preterm birth (74).

#### Vaginal Microbiota and Protective Immunity

Pattern recognition receptors, including TLRs, dectin-1 receptors, and NOD receptors, work as inspectors for MAMP on both commensal and pathogenic microbes (75, 76). Upon binding, PRRs initiate cytokine/chemokine signaling cascades for host defense. The cytokine/chemokines, including IL-1β, IL-6, and IL-8, recruit and activate various immune cells, such as macrophages, natural killer cells, and T cells. *In vitro* colonization of vaginal epithelial cell multilayers by *Lactobacilli* demonstrated that these bacteria do not elicit an inflammatory response in contrast to *S. aureus*. Rather, isolates of *Lactobaclli,* especially *L. jensenii,* tempered the induction of cytokines following TLR2/6 and 3 agonist treatment of the epithelial cells (77). Within the CST IV classification, *Prevotella amnii*, *Mobiluncus mulieris*, *Sneathia amnii*, and *Sneathia sanguinegens* were found to induce higher levels of inflammatory cytokines and chemokines, such as IL-1α/β and IL-8, relative to microbiota dominated by *L. crispatus* or *L. iners*. The study also showed significant increases of IL-1α/β and TNF-α during transition of communities from CST I to CST IV longitudinally, supporting the notion that the innate immune response is affected by vaginal bacterial community states (78).

Furthermore, although the number of vaginal antigenpresenting cells (VAPCs) does not vary with vaginal microbial populations, the CST IV VAPCs were more activated and mature and induced a marked response to LPS as well to IFN-γ and IL-1β. The study suggests that VAPCs are stimulated by contact with the resident flora to upregulate inflammatory cytokine/chemokine expression, resulting in a significant increase in vaginal CCR5<sup>+</sup> CD4<sup>+</sup> T cells (78). The increase of the accessible CD4 T-cell is important, as the vaginal mucosa is a restricted space for lymphocyte migration (79).

Soluble factors such as mannose-binding lectin (MBL), vaginal AMPs, and immunoglobulins (Igs) also contribute to vaginal defense (80). MBL binding to mannose, N-acetylglucosamine and fucose carbohydrate moieties present on microbial cell surfaces leads to cell lysis or recognition by immune cells (81). Women with MBL deficiency due to genetic polymorphism are more susceptible to recurrent vulvovaginal candidiasis (82). Defensins are a class of AMPs that act against various pathogens, including bacteria and viruses. The concentration of defensin

lectins (MBLs), and immunoglobulins (Igs), contribute to the homeostatic immunity of the vaginal surface. In addition, the surveillance of commensals and pathogenic microbes is achieved by pattern recognition receptors (PRRs). (B) In cases of disrupted vaginal microbiota, such as bacterial vaginosis, community state type IV type microorganisms dominate to initiate an inflammatory response. Short-chain fatty acids produced by these microorganisms are likely to induce the production of proinflammatory cytokines. IL-33 has recently been identified as the key cytokine in association with antiviral immunity modulation by the vaginal microbiome. IL-33 is also responsible for the Th2-type immune response elicited by proteases that are secreted by pathogenic microbes.

three is associated with dysbiosis and BV during pregnancy (83). Human β-defensin-2 expression was associated with colonization by *L. iners*, *Atopovium vaginae,* and *Prevotella bivia* in *ex vivo* organotypic models of the vaginal epithelium and with *L. jensenii* but not *G. vaginalis in vitro* (71, 84). Natural Igs, such as IgG, present in vaginal mucus prevent viral infections, such as HSV, by forming multiple low-affinity bonds between the virus and mucus gel. A sufficient number of low-affinity bonds ensures that viruses are effectively trapped in mucus, thereby reducing the flux of infectious virions (85).

*Lactobacilli* dominance of vaginal microbiota has been shown to protect the host from other opportunistic microbial infections, including bacteria (e.g., *Neisseria gonorrhoeae* and *Chlamydia trachomatis*), viruses (e.g., HIV and HPV), and fungi [e.g., *Candida albicans* (86–90)]. The lactic acid and antimicrobial compounds produced by *Lactobacilli* are hypothesized to underlie such protection, and many studies are now focused on discovering the specific mechanisms (90, 91). Notably, of the *Lactobacillus species*, *L. crispatus* is most prominently involved in inhibition of the growth of other microbes and in suppression of cytokine production, whereas *L. iners*, despite its affiliation with *Lactobacilli*, shows conflicting results.

## BACTERIAL VAGINOSIS

A disrupted vaginal microbiota may be related to various diseases but is directly connected with two pathogenic states, BV and aerobic vaginitis (AV) (**Figure 3B**). Both of these disease entities originate from the CST IV category. AV is mainly differentiated from BV by the presence of prominent inflammatory response associated with aerobes, such as group B *Streptococcus*, *S. aureus*, *E. coli*, and *Enterococcus* (80, 92). Despite prominent clinical symptoms and a higher risk of preterm labor and preterm birth than BV (93), the identity of AV as dysbiosis-induced inflammation is still controversial, as some believe that AV is primarily an immunologic disorder with secondary dysbiosis or a dermatological disease (94). Thus, we will focus our discussion on BV in this review.

Bacterial vaginosis is defined as a replacement of lactobacilli with characteristic groups of bacteria and subsequent change in vaginal fluid. Clinically, Amsel et al. proposed the first diagnostic criteria for BV: gray-white milky homogeneous discharge, vaginal fluid pH > 4.5, clue cells upon microscopy, and release of fishy odor on 10% potassium hydroxide solution (95). An alternative for diagnosis is to use a vaginal smear for Gram-staining of *Lactobaccillus*, *Gardnerella,* and other Gram-variable rods to yield a Nugent criteria (96). BV may or may not elicit overt inflammatory responses, and investigation of inflammatory cytokines in BV has led to inconsistent results (97).

Bacterial vaginosis has long been known to be associated with *G. vaginalis*; however, following the development of nonculture-based methods, it became evident that *G. vaginalis* is often present in the absence of BV, suggesting that colonization by this bacterium is not a precondition of BV. Furthermore, BV is not caused by transferring *G. vaginalis* to a healthy woman, but transfer of discharge from a woman with BV is enough to elicit BV. Hence, it is not a single type of bacteria, even at high numbers, that is important but a critical mixture of bacteria with potential pathogenic properties and a lack of favorable lactobacilli that is essential for the development of BV (98, 99).

#### BV-Altered Immune Response

Short-chain fatty acids (SCFAs) produced by commensal microbiota play an anti-inflammatory role in the gut (100); however, high concentrations (20 mM) of some SCFAs (acetate and butyrate) that are prevalent in BV induce PBMC production of the proinflammatory cytokines IL-8, TNF-α, and IL-1β at neutral pH, which resembles the pH of vagina. Lower levels of these SCFAs also significantly enhance TLR2 ligand- and TLR7 ligand-induced production of IL-8 and TNF-α in a time- and dose-dependent manner (101). The discordance between the effect of SCFAs in the gut and the effect of SCFAs in the vagina may be ascribed to inherent differences in these organs, including cell-type, SCFA concentration, and pH (101, 102). *In vitro* studies using vaginal epithelial cells and BV-associated bacteria have shown a similar proinflammatory response to that elicited by BV-associated SCFAs (71, 77, 103). In addition, a number of studies have reported higher levels of proinflammatory cytokines, such as IL-1β and IL-8, in women with BV than in controls with normal Nugent scores (104, 105). Taken together, these studies indicate that these metabolites may contribute to BV and the sub-clinical inflammation present in women with BV.

With respect to altered T-cell responses, CD4<sup>+</sup>, CD4<sup>+</sup> CCR5<sup>+</sup>, and CD4<sup>+</sup> CD69<sup>+</sup> T cells were found to be decreased in the cervixes of Kenyan sex workers following BV treatment (106). Gamma delta T cells in the endocervix were also decreased in women with BV (107). On the contrary, qPCR-confirmed presence of both *L. crispatus* and *L. jensenii* was associated with lower numbers of cervical CD3<sup>+</sup> HLADR<sup>+</sup> and CD3<sup>+</sup> CD4<sup>+</sup> CCR5<sup>+</sup> cells in healthy Belgian women (108). These results suggest a connection between BV and T cells, although the specific details of this relationship warrant further research.

Another important consequence of BV-altered immunity is the increased risk of viral infection, including sexually transmitted viruses, such as HIV and HSV. Commensal *Lactobacilli* inhibit HIV-1 replication in human tissue *ex vivo* by medium acidification, lactic acid production, and direct viricidal effects (89). The proinflammatory cytokines and chemokines that are increased by BV-associated bacteria *in vitro* and are associated with BV *in vivo* enhance the risk of HIV transmission by directly stimulating HIV replication in latent viral reservoirs and by facilitating the trafficking and activation of CD4<sup>+</sup> host cells, which are normally sparse in the cervicovaginal mucosa (88). A recent study highlighted the importance of IL-33 in antiviral immunity of vaginal mucosa. The study demonstrated that the innate immune responses, including type I IFN and proinflammatory cytokine production at infection sites as well as induction of virus-specific CD4 and CD8 T-cell responses in draining lymph nodes, were not impaired when vaginal dysbiosis was induced using oral antibiotics. IL-33 alone was able to suppress local antiviral immunity by blocking the migration of effector T cells to the vaginal tissue, thereby inhibiting IFN-γ production (109). The study solidified the relationship between dysbiosis and HSV infection and revealed the mechanisms through which dysbiosis modulates antiviral immunity. The researchers further investigated the T-cell response from bacteria- or protozoansecreted proteases. Challenge with the prototype protease papain on vaginal mucosa induced Th2 immunity that was dependent on IL-33. Furthermore, dendritic cells that express interferon regulatory factor 4 were responsible for the induction of Th2 differentiation (110).

## CONCLUDING REMARKS

Skin, oral, and genital mucosa function similarly as the initial barriers of host defense from pathogens; however, the interactions of these mucosae with microbes and the microbial rendering of these environments are unique, which can also be described as "compartmentalized" (4). Still, researchers should not overlook the possibility of connections between the microbiota of different tissues. Recently, ectopic colonization of oral bacteria in the intestine was found to induce gut inflammation (111). Diseases, such as Behçet's disease, which affects orogenitalia, gut, and skin, exhibit distinct salivary microbiota as well as gut microbiota (100, 112). Thus, despite the absence of continuity, it remains feasible that the microbiota of one tissue affects another tissue.

The interaction between the major pathogenic microbes and subsequent chronic inflammatory immune response sheds light on the diverse mechanisms of the host response against dysbiosis and chronic systemic disorders. In addition, these microbes with pathogenic potential can become more than commensals, and thus, illuminating the key metabolic alterations responsible for their expansion will increase our understanding of tissue-specific pathology.

*In vivo* studies describing the host adaptive immunity and its association with compartmentalized microbiota outside of the gut remain scarce, likely due to the difficulty of such research. Skin and orogenital mucosa are profound reservoirs of T cells, especially tissue-resident memory T (TRM) cells (12, 113–115), which are cells recently identified as lymphocytes that are distinct from recirculating central and effector memory T cells (116). Thus, studies that illustrate the relationship between the microbiota and TRM cells are very likely to increase our understanding of the continuous interaction between our body and microbes. Moreover, novel strategies targeting the host manipulation of microbes with immune cells can also be developed *via*

#### REFERENCES


information we learn from further studies on microbiota and the human immune system.

#### AUTHOR CONTRIBUTIONS

YJ performed literature research and wrote the review. HK conceived the idea for the review, provided insightful discussion when necessary and edited the review.

#### ACKNOWLEDGMENTS

The authors would like to thank the members of the Laboratory of Host Defenses for helpful advice and discussions.

#### FUNDING

This study was supported by the National Research Foundation of Korea (NRF-2016R1A2B2015028 and NRF-2015M3D6A1065121) and KAIST Future Systems Healthcare Project funded by the Ministry of Science and ICT of Korea. Young Joon Park is a recipient of Global PhD fellowship supported by the National Research Foundation of Korea (NRF-2017H1A2A1045742).


IL-36-mediated T cell responses. *Cell Host Microbe* (2017) 22(5):653–66.e5. doi:10.1016/j.chom.2017.10.006


lectin, and a mannose-binding lectin gene polymorphism in Latvian women. *Clin Infect Dis* (2003) 37(5):733–7. doi:10.1086/377234


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2018 Park and Lee. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Influence of Oral and Gut Microbiota in the Health of Menopausal Women

Angélica T. Vieira<sup>1</sup> , Paula M. Castelo2,3, Daniel A. Ribeiro3,4 and Caroline M. Ferreira2,3 \*

<sup>1</sup> Department of Biochemistry and Immunology, Institute of Biological Sciences, Federal University of Minas Gerais, Belo Horizonte, Brazil, <sup>2</sup> Department of Pharmaceutics Sciences, Institute of Environmental, Chemical and Pharmaceutical Sciences, Universidade Federal de São Paulo, Diadema, Brazil, <sup>3</sup> Pathology Graduate Program, Universidade Federal de São Paulo, São Paulo, Brazil, <sup>4</sup> Department of Biosciences, Universidade Federal de São Paulo, Santos, Brazil

Sex differences in gut microbiota are acknowledged, and evidence suggests that gut microbiota may have a role in higher incidence and/or severity of autoimmune diseases in females. Additionally, it has been suggested that oral, vaginal, and gut microbiota composition can be regulated by estrogen levels. The association of vaginal microbiota with vulvovaginal atrophy at menopause is well described in the literature. However, the relevance of oral and gut microbiota modulation in the immune system during estrogen deficiency and its effect on inflammatory diseases is not well explored. Estrogen deficiency is a condition that occurs in menopausal women, and it can last approximately 30 years of a woman's life. The purpose of this mini- review is to highlight the importance of alterations in the oral and gut microbiota during estrogen deficiency and their effect on oral and inflammatory diseases that are associated with menopause. Considering that hormone replacement therapy is not always recommended or sufficient to prevent or treat menopause-related disease, we will also discuss the use of probiotics and prebiotics as an option for the prevention or treatment of these diseases.

Keywords: saliva, oral health, mouth diseases, gut microbiota, estrogen, menopause

## INTRODUCTION

We harbor trillions of microorganisms that associate with specific tissues and are termed microbiota. This rich community of microorganisms, mostly bacteria, has co-evolved in a symbiotic relationship with humans in such a way that it is now essential for several physiological functions and controls many aspects of host physiology (Backhed et al., 2005; Backhed, 2012; Grover and Kashyap, 2014).

One of the factors that plays a pivotal role in microbiota modulation, although broadly understudied in current research, is the change in female sexual hormones throughout life. Two phases occur in a woman's life that are characterized by several physiological, metabolic and immunological changes: menarche, or the first menstruation of a woman, which occurs during adolescence between 10 and 15 years of age (Hoffmann et al., 2004), and menopause, which occurs between age 45 and 55 and includes the cessation of menstrual periods and loss of the reproductive function of the ovaries (Brotman et al., 2014). In fact, estrogen and the microbiota of a woman's body tend to be investigated more extensively during the woman's reproductive years than during menopause or the phase of estrogen decline. One exception is the vaginal microbiota,

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Sergueï O. Fetissov, University of Rouen, France Sunil Joshi, Old Dominion University, United States

#### \*Correspondence:

Caroline M. Ferreira caroline.nu.ferreira@gmail.com

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 28 July 2017 Accepted: 14 September 2017 Published: 28 September 2017

#### Citation:

Vieira AT, Castelo PM, Ribeiro DA and Ferreira CM (2017) Influence of Oral and Gut Microbiota in the Health of Menopausal Women. Front. Microbiol. 8:1884. doi: 10.3389/fmicb.2017.01884

**72**

which has been widely investigated during menopause. Here, we consider menopause or the menopausal phase, including perimenopause (before menopause), menopause and postmenopause (after menopause).

Considering that menopause can last for approximately 30 years of a woman's life (Brotman et al., 2014), the purpose of this mini-review is to highlight the importance of alterations in the oral and gut microbiota during estrogen deficiency and determine their relevance in oral infections and inflammatory diseases that are associated with menopause.

### THE INTERACTION BETWEEN ORAL MICROBIOTA AND FEMALE SEX HORMONES

The oral cavity (mouth) is composed of several distinct microbial habitats, including the lips, the teeth, the gingival sulcus, the tongue, the cheeks, the palate and the tonsils, which are colonized by hundreds of different bacterial, viral, and fungal species (Dewhirst et al., 2010; Yost et al., 2015). The microbial communities associated with these structures are in symbiosis with the host (Sanz et al., 2017). However, in the presence of stressors that can perturb this homeostasis, several oral infectious diseases may appear, including dental caries and periodontitis (Almstahl et al., 2010; Dewhirst et al., 2010). Many of these disease are recognized to be caused by the consortia of organisms in a biofilm rather than a single pathogen (Jenkinson and Lamont, 2005). In addition, poor oral health and oral diseases may be associated with many systemic diseases (Seymour et al., 2007), such as cardiovascular diseases (Joshipura et al., 1996; Montebugnoli et al., 2004; Belenguer et al., 2006), stroke (Joshipura et al., 2003), preterm birth (Offenbacher et al., 1998), diabetes (Genco et al., 2005), and pneumonia (Awano et al., 2008).

In healthy individuals, the microorganisms found in the mouth with the largest representation include Streptococcus, Actinomyces, Veillonella, Fusobacterium, Porphyromonas, Prevotella, Treponema, Neisseria, Haemophilus, Eubacteria, Lactobacterium, Capnocytophaga, Eikenella, Leptotrichia, Peptostreptococcus, Staphylococcus, and Propionibacterium (Jenkinson and Lamont, 2005; Liu et al., 2012). The behavior of these organisms can be very dynamic and adapt to a wide range of environments and interactions with other microbial species while aggregated in biofilms over the oral surfaces.

Estrogen receptor-beta has been detected in the oral mucosa and salivary glands (Valimaa et al., 2004), and some evidence shows age-related hormonal changes in the exfoliated normal buccal mucosa of women (Donald et al., 2013). Moreover, the vaginal and buccal epithelia share some microscopic similarities. As observed by Thompson et al. (2001), the patterns of surface keratinization and the distribution and appearance of the lipid lamellae in the intercellular spaces were similar between vaginal and buccal epithelial samples of postmenopausal women. Therefore, given that many menopausal women also suffer from oral discomforts in addition to climacteric symptoms (Meurman et al., 2009), an understanding of the impact of female sex hormones on the characteristics of the oral microbiota may be clinically relevant, especially during menopause. Some of the main complaints from women in menopause include dry mouth and tooth loss, and the existing data have focused on the salivary microbial composition and the microbiota characteristics of the gingival sulcus. Therefore, this review will explore the main findings of the relationship between the oral microbiota and menopause in saliva and periodontal support.

#### Saliva

Saliva plays an important role in the maintenance of oral health integrity and the protection against dental caries and other oral diseases (Marsh et al., 2016; Wang et al., 2016). The salivary microbiota is highly diverse and complex (Curtis et al., 2011).

Estrogen and menopause-related hormonal imbalances are believed to affect oral health (Cao et al., 2007). According to the literature (Meurman et al., 2009), together with climacteric complaints, various oral discomforts are reported in menopausal women. The main peri- and postmenopausal symptoms include xerostomia (subjective oral dryness) and/or hyposalivation (Mahesh et al., 2014), which may increase the occurrence of mucosal and dental diseases, such as candidiasis. Few studies have investigated the effects of hormone replacement therapy in such patients (Mahesh et al., 2014; Lago et al., 2015), although the existing results show an improvement in symptoms following such treatment (Mahesh et al., 2014; Lago et al., 2015).

The quantitative and qualitative changes in saliva may alter the regular homeostasis of oral health, subsequently leading to specific changes in the salivary bacterial composition (Nasidze et al., 2009, 2011; Belstrøm et al., 2014). However, recent findings have shown that patients with severe hyposalivation do not differ in their bacterial profiles compared with those with normal salivary flow rates (Belstrom et al., 2016), although the corresponding study did not focus on the evaluation of such differences between menopausal and non-menopausal women.

Because the salivary composition may be influenced by the presence of oral diseases, prescribed medications and general health (Belstrom et al., 2016), researchers must pay attention to the sample size and control for confounding factors when revising the existing literature to confirm the external validity of any quantitative and qualitative changes in saliva related to menopause.

#### Periodontal Support

The periodontium is the specialized tissue that both surrounds and supports the teeth. Periodontal disease, which includes gingivitis and periodontitis, is highly prevalent in adults, and disease severity increases with age. This inflammatory disease develops over time with the accumulation of biofilm (dental plaque), bacterial dysbiosis, the formation of periodontal pockets, gum recession, and tissue destruction (including alveolar bone loss), which can ultimately lead to tooth loss (Michaud et al., 2017).

Fluctuating female sexual hormone levels in menopausal women may represent key factors that respond to changes detected in the oral cavity (Dutt et al., 2013). Menopause is accompanied by decreased bone density, which may have

implications for oral health such as the risk of enhanced progression of periodontal infections and tooth loss (Hernandez-Vigueras et al., 2016). According to the literature, sex-related hormonal changes may cause the gums to become more susceptible to plaque and create a much higher risk for gingivitis and advanced periodontitis (Suresh and Radfar, 2004).

Periodontitis is a chronic inflammatory process that occurs in response to an increase in Gram-negative bacteria in the biofilm (Ruby and Barbeau, 2002), affecting the tissues that surround and support the teeth. Specific bacterial species, such as Porphyromonas gingivalis and Tannerella forsythensis, were found to be important in the etiology of periodontitis in postmenopausal women (Brennan et al., 2007). In addition, changes in periodontal status were found to be associated with variations in sex hormone levels (Mascarenhas et al., 2003), and the occurrence of periodontitis was reported to be greater in postmenopausal women who did not receive hormone replacement than in premenopausal women (Haas et al., 2009). Therefore, from a clinical point of view, the roles of sex hormones and hormone therapy in the prevalence of subgingival bacterial infection in peri- and postmenopausal women are of great interest.

In a cohort study that included 106 women aged 50–58 years, hormone replacement therapy led to a decreased number of positive samples showing the periodontal pathogens P. gingivalis, Prevotella intermedia, and T. forsythia from the subgingival plaque (Tarkkila et al., 2010). Consistent with this result, a previous study found improved periodontal probing depths and tooth mobility in 190 randomized women who received hormone therapy for 1 year (López-Marcos et al., 2005). Conversely, Pilgram et al. (2002) investigated 135 women in a randomized, controlled trial who received estrogen replacement for 3 years and did not find any changes in clinical parameters such as the attachment of teeth or the bone mineral density of the lumbar spine. In mice, estrogen seems to modulate IL-1 production and participate in the resistance of females to disseminating dentoalveolar infections, leading to the enhanced localization of these infections (Youssef and Stashenko, 2017), which draws attention to the potential role of sex-related hormones in the modulation of oral mucosal infections.

Non-conventional treatment approaches for oral infections, with a particular emphasis on dental biofilm-related diseases, have gained attention in recent years. The use of probiotics and prebiotics to improve gastrointestinal health has now led to an interest in using these treatments to control oral diseases (Allaker and Ian Douglas, 2015). However, few studies have focused on recovery of the oral equilibrium by promoting beneficial microbiota. Despite differences in the composition of the gut and oral microbiota, the community types observed in the gut are predictive of the community observed in the mouth and vice versa (Ding and Schloss, 2014). Among other host factors, the oral microbiota serves as an inoculum for the intestine, and the microorganisms that find adequate conditions in the mouth give rise to distinct types of communities in the intestine. Interestingly, oral inoculation with P. gingivalis in experimental models leads to a change in the intestinal microbiota, which is a possible mechanism for the establishment of diseases associated with periodontitis, such as cardiovascular diseases (Arimatsu et al., 2014). In this sense, understanding the role of health-associated microorganisms may have utility in the application of these approaches for the prevention and treatment of disease (Sanz et al., 2017).

## THE INTERACTION BETWEEN GUT MICROBIOTA AND FEMALE SEX HORMONES

As mentioned earlier, female sex hormones levels influence the composition of the microbiota in many sites of the body, especially the gut. Due to intimate contact with the larger gut immune system, the gut microbiota has been shown to influence many diseases outside of this organ (**Figure 1**). Accordingly, imbalance of the gut microbiota, called dysbiosis, has been extensively related to metabolic and immunological diseases. Interestingly, the presence or absence of estrogen may be able to alter the gut microbiota equilibrium and corresponding disease pathways. Some autoimmune diseases affect more often women than men, including systemic lupus erythematosus (Jiang et al., 2005), Sjogren's syndrome (Patel and Shahane, 2014) and rheumatoid arthritis (Oliver and Silman, 2009). Gender differences have also been reported for the outcome of microbial infections (Fischer et al., 2015). Interestingly, the onset of autoimmune diseases, asthma (Akinbami et al., 2016) and other diseases occurs after menarche or during the reproductive period of women. Experimental findings in mice have shown that the interactions among the microbiota, female sexual hormones, and immunity are associated with the development of autoimmune diseases (Yurkovetskiy et al., 2013, 2015), including type 1 diabetes (Markle et al., 2013) and rheumatoid arthritis (Wu et al., 2010). The non-obese diabetic (NOD) mouse exhibits spontaneous, immune-mediated pancreatic beta cell destruction causing type 1 diabetes (T1D) with a complex genetic and environmental etiology. The NOD T1D incidence shows a strong 2:1 female to male sex bias (Markle et al., 2013). Interestingly, germ-free NOD female mice lack this gender bias for diabetes. Additionally, after castration, males exhibit a similar microbiota composition and T1D incidence to females (Markle et al., 2013). In general, this study shows that the microbiome is a causal factor and not simply a consequence of autoimmune disease.

### The Relevance of the Gut Microbiota in the Health of Menopausal Women

When the interaction between the gut microbiota and estrogen is altered due to a lack of estrogen, this relationship is restructured according to the new circumstances. However, host functional alterations, such as metabolic and immunological changes, also occur.

Obesity affects 65% of postmenopausal women and is associated with the onset of metabolic dysfunction (Leeners et al., 2017). Multiple studies have suggested that postmenopausal women exhibit increased total fat mass and abdominal fat and

decreased lean body mass compared with those of premenopausal women, regardless of aging (Aloia et al., 1995; Schreiner et al., 1996; Cordina-Duverger et al., 2016). The accumulation of abdominal fat in postmenopausal women appears to be a critical factor in the development of insulin resistance and type 2 diabetes (Lobo et al., 2014), and the relationship between the gut microbiota and a lack of estrogen is likely responsible for weight gain and lipid deposition during menopause (**Figure 1**). The gut microbiota can metabolize estrogen-like compounds such as isoflavonoids, which are found in soy foods, and promote the growth of some specific bacteria (Frankenfeld et al., 2014; Chen and Madak-Erdogan, 2016; Miller et al., 2017). Indeed, the administration of soy isoflavones to postmenopausal women was shown to increase the concentration of Bifidobacterium and suppress Clostridiaceae, which are known to be involved in inflammatory diseases (Frankenfeld et al., 2014; Nakatsu et al., 2014). This suppression of Clostridiaceae, a family of Clostridia associated with obesity (**Figure 1**), likely explains why diets containing phytoestrogens have been shown to improve weight gain in menopausal women.

Few studies have investigated whether prebiotics and probiotics can improve insulin sensitivity in postmenopausal women or body fat in mice. The intake of flaxseed mucilage, a prebiotic, is known to improve insulin sensitivity and alter the gut microbiota in obese postmenopausal women (Brahe et al., 2015). Thus far, the implications of the gut microbiota with low levels or the absence of estrogen hormone in the metabolism of women have not been sufficiently studied and require further clarification.

Another link between the gut microbiome and menopausal health is related to bone. Interesting, the gut microbiota has also been found to influence bone homeostasis. Approximately one in two women over age 50 will break a bone because of osteoporosis. A study that involved twenty postmenopausal women with a mean age of 65 years showed that the group that consumed Lactobacillus helveticus-fermented milk had increased serum calcium levels and reduced bone reabsorption compared with those of the control milk consumption group (Narva et al., 2004). Experimental studies have also demonstrated similar results. For instance, L. reuteri treatment significantly protected ovariectomized mice from bone loss and increases in bone marrow CD4+ T-lymphocytes, which promote osteoclastogenesis (Britton et al., 2014). Another study that investigated probiotic treatment for cortical bone loss found reduced expression of two inflammatory cytokines, TNF-α and IL-1β, and increased expression of osteoprotegerin, a potent inhibitor of osteoclastogenesis, in the cortical bone of ovariectomized mice (Ohlsson et al., 2014). Additionally, sex steroid deprivation has been reported to promote intestinal permeability (**Figure 1**), and the oral administration of L. rhamnosus GG (LGG) or VSL#3 (a combination of tree probiotics) to estrogen-deficient mice significantly reinforced intestinal barrier integrity and completely protected the mice against sex steroid depletion-induced bone loss (Li et al., 2016). Importantly, to confirm the role of the gut microbiota in bone health, another experiment also showed that germ-free mice are protected against the bone loss induced by the absence of sex steroids (Li et al., 2016).

We must mention that the gut microbiota may influence the risk for breast cancer through effects on endogenous estrogens produced by adipose tissue in postmenopausal women (**Figure 1**) (Key et al., 2003). A cross-sectional study on 60 healthy postmenopausal women found that women with a more diverse gut microbiome and an abundance of four Clostridia taxa exhibited an elevated urinary ratio of hydroxylated estrogen metabolites to parent estrogens (Fuhrman et al., 2014), which is related to the etiology of breast cancer (Flores et al., 2012; Kwa et al., 2016). However, another study compared 48 postmenopausal breast cancer patients and 48 control patients and observed that postmenopausal women with breast cancer exhibited an altered composition and estrogen-independent low diversity of their microbiota (Goedert et al., 2015). These different findings on gut microbiota diversity and breast cancer could be explained by the fact that disease outcome or disease stage can also affect the microbiota. In this scenario, the consumption of the soy isoflavone daidzein, which is metabolized by some bacteria of the microbiota to generate equol and O-desmethylangolensin (ODMA), could represent a therapeutic strategy for breast cancer prevention. Some, but not all, studies have shown a lower risk of breast cancer associated with equol production (Hullar et al., 2014). However, only approximately 30–50% of the population can metabolize daidzein (Frankenfeld et al., 2004; Uehara, 2013; Nakatsu et al., 2014) to equol, likely due to the host microbiota. Therefore, an investigation into the administration of the soy isoflavone daidzein together with probiotic bacteria to produce equol is warranted and could offer benefits in the prevention of breast cancer in menopausal women.

### CONCLUDING REMARKS

fmicb-08-01884 September 27, 2017 Time: 16:37 # 5

Many chronic diseases can emerge after estrogen levels decline, which will affect a considerable part of a woman's life. Understanding the role of the microbiota in women's health at the menopausal phase could help to improve strategies for microbiota modulation and prevent dysfunction. The oral and gut microbiotas have been extensively studied in women of reproductive age, while the menopausal period has been somewhat overlooked. The use of hormone replacement is not indicated for all menopausal women, and considering that probiotics and prebiotics can affect the dysfunction of bone, adipose tissue, oral and other tissues, such treatments may constitute an important therapeutic strategy. Pro- and prebiotics can also be used in conjunction

#### REFERENCES


with menopause hormone therapy and may attenuate the side effects that can arise from hormone replacement. In conclusion, the scientific findings published to date do not definitively demonstrate how non-vaginal microbiota sites influence the health of menopausal women. Thus, many questions remain unanswered and warrant further investigation to improve the quality of life of menopausal women.

#### AUTHOR CONTRIBUTIONS

AV, PC, DR, and CF drafted and revised the manuscript.

#### FUNDING

This work was supported by grants awarded by the São Paulo Research Foundation (FAPESP), project number 2012/50410-8 to CF and the National Council for Scientific and Technological Development (CNpq), project number 486037/2012-6.

trial in obese postmenopausal women. Br. J. Nutr. 114, 406–417. doi: 10.1017/ S0007114515001786



spine and hip bone mineral density: data from healthy postmenopausal women. J. Periodontol. 73, 298–301. doi: 10.1902/jop.2002.73.3.298


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Vieira, Castelo, Ribeiro and Ferreira. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Infectious Basis of ACPA-Positive Rheumatoid Arthritis

Lazaros I. Sakkas <sup>1</sup> \*, Dimitrios Daoussis <sup>2</sup> , Stamatis-Nick Liossis <sup>2</sup> and Dimitrios P. Bogdanos <sup>1</sup>

*<sup>1</sup> Department of Rheumatology and Clinical Immunology, Faculty of Medicine, School of Health Sciences, University of Thessaly, Larissa, Greece, <sup>2</sup> Division of Rheumatology, Department of Internal Medicine, Faculty of Medicine, University of Patras, Patras, Greece*

Rheumatoid arthritis (RA) is associated with HLA-DRB1 shared epitope (HLA-DRB1SE) and anti-citrullinated protein autoantibodies (ACPAs). ACPAs precedes the onset of clinical and subclinical RA. There are strong data for three infectious agents as autoimmunity triggers in RA, namely *Porphyromonas gingivalis* and *Aggregatibacter actinomycetemcomitans* causes of periodontal disease (PD), and Epstein-Barr virus (EBV). *P. gingivalis* expresses arginine gingipains, that cleave proteins at the arginine residues, and peptidyl arginine deiminase (PPAD), which citrullinates arginine residues of proteins, thus forming neoantigens that lead to ACPA production. Peripheral blood plasmablasts from ACPA+RA patients produce ACPAs the majority of which react against *P. gingivalis*. *A. actinocycetemcomitans* produces leukotoxin A, a toxin that forms pores in the neutrophil membranes and leads to citrullination and release of citrullinated autoantigens in the gums. EBV can infect B cells and epithelial cells and resides as latent infection in resting B cells. Abs against citrullinated peptides derived from EBV nuclear antigen appear years before RA and cross-react with human citrullinated fibrin. Citrullinated proteins are potential arthritogenic autoantigens in RA. The conversion of arginine to citrulline increases the peptide binding affinity to HLA-DRB1SE. Also, citrullinated fibrinogen induces arthritis in HLA-DRB1∗0401 transgenic mice, and transfer of their splenic T cells causes arthritis to recipient mice.

#### Edited by:

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### Reviewed by:

*Jan Potempa, University of Louisville, United States Angelo A. Manfredi, Vita-Salute San Raffaele University, Italy*

#### \*Correspondence:

*Lazaros I. Sakkas lsakkas@med.uth.gr*

#### Specialty section:

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology*

Received: *03 July 2017* Accepted: *11 September 2017* Published: *27 September 2017*

#### Citation:

*Sakkas LI, Daoussis D, Liossis S-N and Bogdanos DP (2017) The Infectious Basis of ACPA-Positive Rheumatoid Arthritis. Front. Microbiol. 8:1853. doi: 10.3389/fmicb.2017.01853* Keywords: anti-citrullinated protein antibodies, arthritis, Ebstein-Barr virus, HLA-DRB1 shared epitope, Porphyromonas gingivalis

## INTRODUCTION

Rheumatoid arthritis (RA) is a systemic inflammatory disease mainly manifested with peripheral polyarthritis. The aetiopathogenesis of the disease is incompletely understood. Risk factors for RA include HLA-DR genes, periodontal disease (PD), and smoking (Bartold et al., 2005; Scher et al., 2012; Mikuls et al., 2014; Kharlamova et al., 2016). The early HLA-DR4 association of RA classified RA by many investigators as an immune-mediated disease and suggested that T cells recognized an antigen presented on HLA-DR4 molecules. The discovery of HLA-DRB1 shared epitope (SE, HLA-DRB1SE), a hypervariable DRβ chain sequence shared by all alleles associated with RA, reinforced this concept (Gregersen et al., 1987; Wordsworth et al., 1989). The discovery of autoantibodies against citrullinated antigens (ACPAs) greatly advanced our understanding of the pathogenetic mechanisms in this disease. ACPAs appear years before clinical onset of RA (Nielen et al., 2004), predict subsequent development of the disease, occur in 50–67% of RA patients, are associated with severe disease, and are highly specific for the disease (van Gaalen et al., 2004; van der Helm-van Mil et al., 2005; Alexiou et al., 2007; Barouta et al., 2017; Hensvold et al., 2017).

Citrullination is a post-translational modification of proteins in which arginine residues are converted to citrulline by the action of enzyme peptidylarginine deiminase (PAD). There are five PAD isoforms (PAD1-4, PAD6), and PAD2 and PAD4 have been implicated in RA. The production of ACPAs means break of tolerance. Tolerance is no immune response to unmodified self. Many proteins are extensively post-translationally modified that including citrullination. In this context, citrullination is a physiological process in many tissues and only in specific circumstances this leads to immune response. Thus, citrullination could create particular neoantigens that would activate T cells, which in turn will provide antigenspecific help to B cells to produce ACPA. Indeed, citrullination increases the affinity of citrullinated antigen to HLA-DRB1SE allele (Hill et al., 2003; Scally et al., 2013). ACPAs in RA recognize many citrullinated autoantigens (**Table 1**) and are associated with HLA-DRB1SE (Snir et al., 2009), and HLA-DRB1SE appears to be a risk factor for ACPA production in RA rather than an independent risk factor for RA development (van der Helm-van Mil et al., 2006). These findings and the fact that ACPAs are of IgG and IgA class suggest that T cells provide help to B cells for the subsequent ACPA production.

Although smoking is a risk factor for RA (van der Helm-van Mil et al., 2007; Lundberg et al., 2013; Hensvold et al., 2015), and increases citrullination in bronchial tissues (Makrygiannakis et al., 2008), other environmental factors, in addition to smoking, appear to play a predominant role in the development of ACPA+RA (Lee et al., 2007; Hensvold et al., 2015) and infections are likely candidates (Bogdanos and Sakkas, 2017).

### INFECTIONS AS GENERATORS OF ACPA IN RA

ACPAs precede the subclinical joint inflammation in pre-RA patients (van de Sande et al., 2011) and can be detected in joints and epithelial sites, such as periodontium in PD (Nesse et al., 2012) and bronchial tissues in early RA (Reynisdottir et al., 2014). Identical citrullinated peptides were found in pulmonary bronchial tissue and synovial membrane and were found to be targets of ACPAs in RA thus providing a link between lungs and joints in ACPA+RA (Ytterberg et al., 2015). A monoclonal ACPA derived from RA patients cross-reacted with many viral, bacterial fungal and plant proteins (Tsuda et al., 2015). The most widely studied infection has been with P. gingivalis and Ebstein-Barr virus. The mechanisms by which these infectious agents could trigger RA are illustrated in **Figure 1**.

## Porphyromonas Gingivalis

Chronic PD is very common affecting nearly 30% of adult population (Brown and Loe, 1993) and is caused by various microbes including Porphyromonas gingivalis (P. gingivalis).

P. gingivalis infection, detected by abs against P. gingivalis components, have been associated with ACPA in HLA-DRB1SE+RA patients. Anti-P. gingivalis abs, detected as abs against RgpB, potent virulent factors of P. gingivalis (Haffajee and Socransky, 1994; Kadowaki et al., 1998), showed stronger association with ACPA+RA (Kharlamova et al., 2016). Furthermore, there was additive interaction between these two factors. Anti-RgpB abs also showed more than additive interaction with HLA-DRB1SE in ACPA+RA (Kharlamova et al., 2016). Using anti-P. gingivalis lipopolysaccharide abs, one study reported association of anti-P. gingivalis abs with ACPA in HLA-DRB1SE+ RA patients and their relatives (Hitchon et al., 2010) whereas another study did not find an association with RA or ACPA status (Seror et al., 2015).

P. gingivalis has two unique enzymes, peptidylarginine deiminase (PPAD) and arginine ginpains (Rgps) which are expressed on the bacterial outer membrane and can also be secreted (Potempa et al., 1995; McGraw et al., 1999). Rgps are proteases that cleave proteins at arginine residues, and PPAD citrullinates both bacterial and human proteins (Wegner et al., 2010). P. gingivalis PAD citrullinates carboxyterminal arginine of human proteins following proteolytic cleavage by P. gingivalis arginine-gingipains (Wegner et al., 2010). Crystal structure of PPAD and the use of synthetic peptides also revealed that PPAD exhibits a definitive specificity for C-terminal arginine residue created by Rgps, whereas PAD2 and PAD4 preferentially citrullinate internal arginine residues (Goulas et al., 2015; Montgomery et al., 2016). Thus P. gingivalis creates neoantigens, not formed by PAD2 and PAD4 and this may explain its pathogenic potential.

It is reasonable to assume that neoantigens, created by Rpgs in conjunction with PPAD in the periodontium of PD, can lead to loss of tolerance and ACPA production. In PD, increased concentrations of anti-CCP and anti-α-enolase autoAbs are detected (Lappin et al., 2013). A peptide 1 of human citrullinated α-enolase (CEP1), an immunodominant epitope, shares 92% homology with P. gingivalis α-enolase and cross-reacts with it (Lundberg et al., 2008). This links periodontitis with RA and suggests that periodontal infection can be the inciting agent that breaks immune tolerance in ACPA+RA, although other studies did not find association of PD with RA (Arkema et al., 2010; Eriksson et al., 2016). Using a single-cell ab cloning method, Li et al showed that peripheral blood plasmablasts in ACPA+RA patients produce ACPAs the majority of which cross-react with outer membrane antigens and/or citrullinated a-enolase from P. gingivalis (Li et al., 2016).

In addition, P. gingivalis can induce neutrophil extracellullar trap (NET) formation (Delbosc et al., 2011), another source of citrullinated autoantigens. NETs are externalized chromatin fibers containing DNA and histones, and decorated with cytoplasmic granular peptides, such as myeloperoxidase, proteinase 3, neutrophil elastase, cathepsin G, LL37, and others, in a process of programmed neutrophil death called NETosis (Yang et al., 2016). PAD4-induced citrullination is an important step in NETosis during which citrullinated histones, vimentin, α-enolase and others are externalized


TABLE 1 | Examples of citrullinated peptides which are targeted by immune responses against self and non-self immune responses in patients with rheumatoid arthritis.

and recognized by ACPAs (Li et al., 2010; Pratesi et al., 2014). PAD4 is also essential for the antibacterial neutrophil immunity (Li et al., 2010). NETosis is enhanced in RA peripheral blood and synovial fluid neutrophils (Khandpur et al., 2013). A positive feedback loop between NETosis and ACPA has been proposed. ACPAs induce NETosis and NETosis provides citullinated autoantigens for ACPA production, as NET components are recognized by RA autoantibodies (Khandpur et al., 2013). Also, B cells from synovial ectopic lymphoid structures (ELS), recognize citrullinated histones of NETs. For instance, monoclonal abs generated from synovial ELS single B cell cloning from patients with ACPA+RA, recognized citrullinated histones from NETs (Corsiero et al., 2016). Neutrophils provide citrullinated autoantigens to the attention of the immune system also via immune-mediated membranolytic pathways (Romero et al., 2013), and this has led to introduction of another pathogen of PD as candidate trigger of autoimmunity in RA, namely Aggregatibacter actinomycetemcomitans.

#### Aggregatibacter Actinomycetemcomitans

#### Aggregatibacter actinomycetemcomitans

(A.actinomycetemcomitans) a periodontal pathogen associated with aggressive PD (Haubek and Johansson, 2014) can cause citullination of a broad range of proteins by a completely different mechanism. A.actinomycetemcomitans produces leukotoxin A (LtxA), which forms pores on the cell membrane of neutrophils at the crevicular fluid of PD. This leads to neutrophil PAD activation and citrullination of a broad range of proteins, which are released from neutrophils (Konig et al., 2016a). In addition, 47% of RA patients show evidence of A.actinimycetemcomitans infection that is associated with ACPA presence. More interestingly, in patients with RA HLA-DRB1SE is associated with ACPA only in patients exposed to A.actinomycetemcomitans (Konig et al., 2016a). Thus A.actinomycetemcomitans is identified as a strong bacterial candidate triggering autoimmunity in RA.

#### Epstein-Barr Virus

Epstein-Barr virus (EBV) is a herpes virus infecting most of the adult population. EBV can infect B cells and epithelial cells and cause primary infection usually asymptomatic in childhood and then a life-long latent infection in resting memory B cells (Kalla and Hammerschmidt, 2012). High titers of anti-EBV abs are detected in RA patients (Alspaugh et al., 1981), and the EBV DNA load was found to be increased 7–10 fold in PBMCs from RA patients compared to healthy EBV carriers (Balandraud et al., 2003; Lunemann et al., 2008). Furthermore, substantial expansions of CD8+Tcells specific for EBV antigens were detected in PB (Lunemann et al., 2008) and expansions of CD8+T cells reactive against key transactivators of EBV lytic infection were also detected in RA joints (Scotet et al., 1996). Latent membrane protein (LMP) 2A through its immunoreceptor tyrosine activation motif (ITAM) phosphorylates (activates) downstream proteins of B cell receptor thus positively regulating B cell survival and activation (Swanson-Mungerson and Longnecker, 2007).

Abs against citrullinated peptide corresponding to EBV nuclear antigen (EBNA)1 (viral citrullinated peptide 1, VCP1) were detected in RA. Furthermore, affinity-purified anti-VCP1 abs reacted with citrullinated fibrinogen (Pratesi et al., 2006). More importantly, Abs against citrullinated peptides derived from EBNA2 (VCP2) along with abs against histone-4-derived citrullinated peptides appear years before the onset of clinical RA and predict subsequent development of RA (Johansson et al., 2016). These abs were associated with HLA-DRB1SE (Johansson et al., 2016). Abs against VCP1 and VCP2 cross-react with human citrullinated peptides (Pratesi et al., 2011). In particular, competition assays showed that abs to citrullinated peptide EBNA (VCP1) strongly crossreacted with the citrullinated peptide β60-74 which bears the immunodominant epitope of citrullinated fibrin in RA (Cornillet et al., 2014, 2015). EBV latent transcripts and EBV latent and lytic proteins were detected in germinal center-like ectopic lymphoid structures (ELS) of RA synovial membrane (Croia et al.,

2013). Also ACPA producing plasma cells (anti-citrullinated fibrinogen abs) at the periphery of ELS were infected with EBV (expressed lytic proteins). Furthermore, ELS-containing RA synovia transplanted onto severe combined immunodeficiency (SCID) mice produced abs against citrullinated EBV proteins (VCP1 and VCP2) (Croia et al., 2013). These findings provide strong circumstantial evidence that EBV may initiate an immune response which subsequently may be re-directed against self antigens by way of cross-reactivity and epitope spreading.

## CITRULLINATED PROTEINS AS ARTHRITOGENIC AUTOANTIGENS

ACPAs are associated with severe disease and are strong predictors of joint erosions in RA (Alexiou et al., 2007; Jilani and Mackworth-Young, 2015). Although association does not prove causation, several lines of evidence suggest that citrullinated proteins are likely to be arthritogenic autoantigens in RA (Sakkas et al., 2014). This means that citrullinated peptides are recognized by and activate T cells, which in turn (a) produce pro-inflammatory mediators and talk to other cells causing joint damage, and (b) provide help to B cells for ACPA production, which by themselves may be pathogenetic.

As already mentioned, the conversion of arginine to citrulline increases the affinity of citrullinated antigen binding to HLA-DRB1SE alleles (Hill et al., 2003; Scally et al., 2013). This has been confirmed in a study by Scally et al who showed that citrulline but no arginine is accommodated within the electropositive P4 pocket of RA-susceptible HLA-DRB1<sup>∗</sup> 0401/04 alleles (Scally et al., 2013). Furthermore, using HLA-DR4 tetramers, the authors found that citrullinated vimentin- and citrullinated aggrecanspecific T cells were present in peripheral blood of RA patients and their numbers were correlated with disease activity (Scally et al., 2013). Also, a T cell line recognizing citrullinated fibrinogen, abundant in RA joints, induced proinflammatory cytokines, and transfer of this T cell line to mice with CIA exacerbated arthritis (Cordova et al., 2013). Immunization with human fibrinogen (containing citrullinated peptides) in complete Freud adjuvant enhanced arthritis and T cells from these mice were fibrinogen-reactive and produced high levels of IL-6, IFNγ and IL-17 (Ho et al., 2010). Furthermore, adoptive transfer of plasma or T cells caused arthritis in naïve mice (Ho et al., 2010). Citrullinated fibrinogen but not unmodified fibrinogen induced arthritis in HLA-DRB1<sup>∗</sup> 04:01-IE transgenic mice but not in wild-type C57BL/6 mice (Hill et al., 2008). Furthermore, transfer of splenocytes from these transgenic arthritic mice caused arthritis to recipient mice, indicating that activated citrullinated fibrinogen-specific T cells are crucial for arthritis development (Yue et al., 2010). Also, a pan-PAD inhibitor (Clamidine) decreased the clinical severity of collagen-induced arthritis (CIA) and joint and serum protein citrullination (Willis et al., 2011). These studies show that citrullinated peptides in conjuction with HLA-DRB1SE activate T cells which become arthritogenic.

Infection with P. ginvivalis further contributes to joint inflammation that is dependent on citrullination. For instance, infection with P. gingivalis caused exacerbation of collageninduced arthritis that was dependent on P. gingivalis PAD (PPAD) (Maresz et al., 2013). High levels of citrullinated proteins at the site of infection with P. gingivalis were detected as well as ACPAs (Maresz et al., 2013). Also, CIA was much less severe in the presence of PAD-deficient P. gingivalis (Gully et al., 2014). On the other hand, P. gingivalis components may cause arthritis through molecular mimicry. For instance, immunization of HLA-DR4-IE-transgenic mice with P. gingivalis α-enolase either citrullinated or noncitrullinated caused arthritis and abs reactive with human α-enolase (Kinloch et al., 2011). As already mentioned, P. gingivalis α-enolase shares sequence similarity with human α-enolase.

ACPAs contribute to joint inflammation and damage since they induce secretion of inflammatory cytokines and differentiation of osteoclasts. ACPA-containing immune complexes induced TNFα secretion by peripheral blood-derived macrophages (Clavel et al., 2008; Laurent et al., 2011; Sokolove et al., 2011), via toll-like receptor 4 (TLR4) and FcγR (Sokolove et al., 2011), whereas citrullinated fibrinogen stimulated TNFα production through TLR4 (England et al., 2017). Also, citrullinated histones increase macrophage TNFα production via TLR4, and immune complexes containing citrullinated histones activate macrophage production of TNFα via TLR4 and FcγR and neutrophils. Over 90% of RA patients have abs against neutrophil-derived citrullinated histones (citrullinated H2B).

P. gingivalis may also contribute to arthritis through inflammatory cytokine release. For instance, periodontal disease induced by P. gingivalis (Marchesan et al., 2013) or Prevotella nigrescens (de Aquino et al., 2014) exacerbated collagen-induced arthritis and promoted Th17 responses.

The effect of citrullinated proteins and ACPAs in wild-type animals (not transgenic animals) in the induction or exacerbation of arthritis is not certain. For instance, immunization of mice with citrullinated histone did not cause arthritis but exacerbated collagen-induced arthritis (Sohn et al., 2015). Similarly, some studies show exacerbating effect of ACPAs on collagen-induced arthritis and others suppressing effects (Kuhn et al., 2006; Shoda et al., 2011; Cantaert et al., 2013). For instance, it has been reported that co-administration of anti-citrullinated fibrinogen abs with anticollagen II abs enhanced CIA (Kuhn et al., 2006).

NETs, apart from providing targets for ACPAs, contribute to inflammatory process in RA. The nuclear and cytoplasmic molecules in NETs have antimicrobial properties, and stimulate adaptive and innate immune responses. LL37 NETs increase fibroblast-like synoviocyte IL-6 and IL-8 production (Khandpur et al., 2013). Also, LL37 can form complexes with DNA and RNA and stimulate innate TLRs (Lande et al., 2007; Ganguly et al., 2009).

ACPAs also contribute to joint erosions. ACPAs and monoclonal ACPAs derived from RA synovial fluid (SF) single B cells enhanced differentiation of osteoclasts through PAD-dependent IL-8 production. Furthermore, transfer of monoclonal ACPAs into mice induced IL-8-mediated bone loss (Krishnamurthy et al., 2016). Also, affinity-purified abs against mutated citrullinated vimentin (MCV) bind to osteoclast surface and induce osteoclastogenesis, whereas adoptive transfer of anti-MCV abs into mice causes osteopenia (Harre et al., 2012).

## RELEVANCE TO TREATMENT

Although there may be few disagreements (Konig et al., 2016b), the research data outlined above imply that citrullination, in conjunction with genetic factors, such as HLA-DRB1SE, protein tyrosine phosphatase nonreceptor type 22 (PTPN22) risk allele (Joshua et al., 2016) encoding an R620W amino acid change that allows survival of autoreactive B cells (Menard et al., 2011), is the key element in breaking tolerance. Thus citrullinated peptides may offer new therapeutic strategies for RA. For instance, CTLA-4Ig blocked the development of arthritis induced by citrullinated fibrinogen in HLA-DRB1<sup>∗</sup> 0401 transgenic mice (Yue et al., 2010). This concept is re-enforced by the study of Gertel et al. who used citrullinate multiepitope peptide derived from prevalent citrullinated autoantigens in RA to reduce disease severity in adjuvant-induced arthritis in rats (Gertel et al., 2015). Also, citrullinated peptide autologous dendritic cells immunotherapy administered to once reduced effector T cells and increased regulatory T cells at 1 month (Benham et al., 2015).

Another strategy could be inhibition of TLR4. TLR4 is an innate immunity receptor for various ligands, including immune complexes containing ACPAs, mainly against citrullinated fibrinogen. Inhibition of TLR4 has been shown to decrease inflammatory arthritis in mouse models (Abdollahi-Roodsaz et al., 2007; Pierer et al., 2011). More importantly, the presence of ACPAs against citrullinated peptides from α-chains and βchains of fibrinogen and histone 2A in RA patients predicts the anti-inflammatory response of TLR4 inhibition by a therapeutic ab (NI-0101) in an ex vivo model of RA (Hatterer et al., 2016). Therefore, it is likely that these new therapeutic strategies will be fruitful in human RA in the near future.

#### AUTHOR CONTRIBUTIONS

LS, DD, SL, and DB substantially contributed on drafting the work and revising the manuscript, and approved the final version to be published. LS, DD, SL, and DB agreed to be accountable for all aspects of the work in ensuring that questions related to the

#### REFERENCES


accuracy or integrity of any part of the work are appropriately investigated and resolved. LS had the original idea of drafting the manuscript and overall supervision of manuscript's shaping.

### FUNDING

This work was supported by the Research Committee of the University of Thessaly (Grant No. 4052).


rheumatoid arthritis. Arthritis Rheum. 64, 3083–3094. doi: 10.1002/art. 34539


rheumatoid arthritis. Arthritis Rheum. 62, 2662–2672. doi: 10.1002/art. 27552


joints of patients with rheumatoid arthritis: identification and validation. Ann. Rheum. Dis. 74, 1772–1777. doi: 10.1136/annrheumdis-2013-204912

Yue, D., Brintnell, W., Mannik, L. A., Christie, D. A., Haeryfar, S. M., Madrenas, J., et al. (2010). CTLA-4Ig blocks the development and progression of citrullinated fibrinogen-induced arthritis in DR4-transgenic mice. Arthritis Rheum. 62, 2941–2952. doi: 10.1002/art.27597

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Sakkas, Daoussis, Liossis and Bogdanos. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Immune System Bridges the Gut Microbiota with Systemic Energy Homeostasis: Focus on TLRs, Mucosal Barrier, and SCFAs

*Martina Spiljar1,2, Doron Merkler <sup>3</sup> and Mirko Trajkovski1,2,4\**

*<sup>1</sup> Faculty of Medicine, Department of Cell Physiology and Metabolism, Centre Médical Universitaire, University of Geneva, Geneva, Switzerland, 2Diabetes Center, Faculty of Medicine, Centre Médical Universitaire, University of Geneva, Geneva, Switzerland, 3 Faculty of Medicine, Department of Pathology and Immunology, Centre Médical Universitaire, University of Geneva, Geneva, Switzerland, 4 Institute of Genetics and Genomics in Geneva (iGE3), University of Geneva, Geneva, Switzerland*

The gut microbiota is essential for the development and regulation of the immune system

and the metabolism of the host. Germ-free animals have altered immunity with increased susceptibility to immunologic diseases and show metabolic alterations. Here, we focus on two of the major immune-mediated microbiota-influenced components that signal far beyond their local environment. First, the activation or suppression of the toll-like receptors (TLRs) by microbial signals can dictate the tone of the immune response, and they are implicated in regulation of the energy homeostasis. Second, we discuss the intestinal mucosal surface is an immunologic component that protects the host from pathogenic invasion, is tightly regulated with regard to its permeability and can influence the systemic energy balance. The short chain fatty acids are a group of molecules that can both modulate the intestinal barrier and escape the gut to influence systemic health. As modulators of the immune response, the microbiota-derived signals influence functions of distant organs and can change susceptibility to metabolic diseases.

Keywords: gut microbiota, immune system, toll-like receptors, short chain fatty acids, mucosal barrier, metabolism, dysbiosis

### MICROBIOTA SHAPES THE IMMUNE SYSTEM AND THE HOST METABOLISM

The human microbiota comprises an enormous amount and variety of microorganisms. Among archea, eukarya, and viruses, bacteria are the most abundant inhabitants of the human host. The human gastrointestinal tract is one of the world's most densely packed microbe habitats (1, 2). By coevolution with humans, a symbiotic relation evolved with the host providing a living environment and nutrients in exchange for support in nutrient degradation and protection from pathogenic microorganisms. The development of 16S rRNA sequencing allowed estimates of the abundance of microbiota in greater detail, revealing the gram-negative Bacteroidetes and grampositive Firmicutes as the most abundant phyla in the human gut. Changes in the Firmicutes to Bacteroidetes ratio, as inducible among others by high-fat diet (3), can impact metabolism (4–7), and lead to symptoms of type 1 and 2 diabetes, colitis, and obesity (8). Interestingly, transplantation of microbiota from obese human or murine donors to germ-free (GF) mice is sufficient to induce insulin resistance and increased adiposity compared to the lean microbiota transplanted controls

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Angela M. Zivkovic, University of California, Davis, United States Francois-Pierre Martin, Nestle Institute of Health Sciences, Switzerland*

#### *\*Correspondence:*

*Mirko Trajkovski mirko.trajkovski@unige.ch*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 25 July 2017 Accepted: 03 October 2017 Published: 30 October 2017*

#### *Citation:*

*Spiljar M, Merkler D and Trajkovski M (2017) The Immune System Bridges the Gut Microbiota with Systemic Energy Homeostasis: Focus on TLRs, Mucosal Barrier, and SCFAs. Front. Immunol. 8:1353. doi: 10.3389/fimmu.2017.01353*

(9–12). The Firmicutes and the Bacteroidetes, together with the other inhabitants of the gut, can also modulate immune function (13). The importance of the microbiota for the formation of a fully functional immune system first became evident by studying the GF animals, which are bred and housed in an environment devoid of microorganisms. The immune tissues or local immune cell subsets that are in direct contact with microbiota, as in the gut, are different in GF animals. The gut of the GF mice shows fewer and smaller Payer's patches (14), an altered mucus layer (15), not fully developed gut-associated lymphoid tissues, and no formation of isolated lymphoid follicles, which help inducing local immune responses (16). Counts of the immune cells residing within the gut are decreased, including the intraepithelial CD8<sup>+</sup> T cells, the lamina propria CD4<sup>+</sup> T cells, and their subsets type 17 T helper cells (Th17) and regulatory T cells (Tregs), as well as immunoglobulin G counts (17, 18). The absence of gut microbiota leads to functional alterations in immune and intestinal epithelial cells, which express less microbe sensing toll-like receptors (TLR) (19) and major histocompatibility complex II molecules for antigen presentation (20).

Certain aspects of an underdeveloped, less responsive gut immune system seem to have beneficial effects in obesity. GF, or microbiota depleted mice using antibiotics show improved glucose and insulin tolerance, accompanied by reduced adiposity (9), increased browning of their white fat depots (21), and are protected from diet-induced obesity (DIO) (22–24). The fat browning promotes energy dissipation in the form of heat, enabled by the uncoupling of oxidative phosphorylation from ATP biosynthesis (25). In part, the metabolic effects and the browning phenotype are mediated by the innate immune system and the increased M1 to M2 macrophage polarization that potentiates the fat browning either after cold (26, 27), or after microbiota depletion (21). The GF metabolic phenotype is partially reversible once mice are transplanted with microbiota from donor mice. These observations from GF animals, or from the mice transplanted with microbiota, illustrate that the microbiota is essential in the progression of metabolic imbalances and in the development of a functional immune system. This review will summarize how microbial molecules and changes in the microbiota composition are sensed by the immune system and how microbial products and consequently the immune response can spread into the circulation to cause systemic and metabolic consequences, starting in the gut and moving toward the periphery. We highlight TLR signaling, gut barrier modulation, and short chain fatty acid (SCFA)-mediated interactions as important mechanisms in this process.

### TLRs SENSE MICROBIAL PRODUCTS IN THE GUT AND PERIPHERY TO INFLUENCE SYSTEMIC IMMUNITY AND METABOLISM

One of the immune system's tasks is to decide whether a microbe is a harmless commensal or an invading and potentially pathogenic one. As the first line of defense, the innate immune system will recognize both kinds of microbes as they include certain non-self patterns. This is particularly important for ensuring correct intestinal homeostasis (28). One class of molecules that recognizes such patterns is the pattern recognition receptor family that includes the TLRs. There are 10 TLRs in humans and 12 in mice found on the cell or endosomal membranes of different cell types including macrophages, dendritic cells (DCs), and nonimmune cells such as epithelial cells, hepatocytes, or adipocytes. Activation of such receptors initiates downstream signaling cascades that often result in induction of cytokine expression. These cytokines can further influence other immune cells and accordingly dictate the tone of an immune response.

Toll-like receptor 2 signals as a heterodimer with TLR1 or TLR6 and recognizes a wide variety of signals on fungi and bacteria. One ligand of TLR2 is polysaccharide A (PSA) from *Bacteroides fragilis* that induces anti-inflammatory responses, activates plasmacytoid DCs, interleukin 10 (IL-10) production of CD4<sup>+</sup> T cells, promotes clonal expansion and induction of Treg cells, and suppresses Th17 production in the gut (**Figure 1**) (29–31). These aspects potentially contribute to the amelioration of inflammation in animal models such as experimental autoimmune encephalomyelitis and colitis after PSA administration (29, 32, 33) or after colonization of GF mice with Bacteroidetes (34). Anti-inflammatory responses upon TLR2 activation were suggested as a way to recognize commensals as non-pathogenic. This suggests that specific microbial-derived mechanisms actively promote immunologic tolerance to symbiotic bacteria. In contrast, TLR2 signaling can also result in proinflammatory responses, as for instance upon detection of *Lactobacillus plantarum* teichoic acid d-alanylation (35). Although the highest encounter opportunity between microbes and immune cells is the gut, microbial products are also detected by TLRs at peripheral sites, influencing systemic and metabolic effects. *Tlr2* expression is increased in visceral adipose tissue in mice fed a high fat diet (HFD) compared to a normal chow diet, resulting in tumor necrosis factor alpha (TNF-α) expression, thereby supporting a low grade inflammation in tissues (**Table 1**) that is typical of obesity (36). *Tlr2* inhibition or ablation results in improved insulin sensitivity and decreased inflammation and adiposity in a DIO mouse model (37–39). These findings support the role of TLR2 as a proinflammatory mediator of metabolic symptoms.

Toll-like receptor 4 mainly recognizes the Gram-negative bacterial cell wall component lipopolysaccharide (LPS). LPS levels in blood are increased in obesity or after high caloric diet (HCD) feeding (56) and are associated with increased Firmicutes to Bacteroidetes ratios. TLR4 signals through the MyD88-dependent pathway, or by TIR-domain-containing adapter-inducing interferon-β regulation of type 1 interferons *via* interferon regulatory factor 3 (IRF3). *Tlr4* expression is increased in adipose tissues, peripheral blood or muscle of obese or type 2 diabetes patients (57, 58) and in adipose tissues of obese *db/db* mice (59) and correlates with insulin resistance. LPS induces adipose tissue inflammation through TLR4 (**Figure 1**), and increases *Ccl2* expression on adipocytes. This chemokine contributes to a microbiota-induced macrophage accumulation and WAT inflammation in lard-fed mice (40). Loss of function mutation in *Tlr4*, or TLR4 deletion reduces macrophage infiltration into the adipose tissue, promotes their

epithelial barrier is composed of an outer mucus layer that is habitat for microbes and an inner, impenetrable layer, in addition to tight-junction connected epithelial cells. Under normal physiologic conditions and when sufficient anti-inflammatory microbial products (green) are present in the gut, the gut lining is well protected, with plasma B cell secreted Immunoglobulin A (IgA), regulatory T cells (Tregs), and eosinophils. The polysaccharide A (PSA) of *Bacteroides fragilis* is a TLR2 ligand that promotes secretion of the anti-inflammatory cytokine interleukin 10 (IL-10) from the plasmacytoid dendritic cells (pDCs)-activated CD4+ T cells, or from the expanded Tregs cells. Under microbial dysbiosis, proinflammatory microbes (red) predominate in the gut and can be sensed, e.g., by TLR4 on CD103+ dendritic cells (DCs) or macrophages. Such microbes can degrade and invade the second colonic mucus layer and escape the gut when the epithelial cell lining and tight junctions are disrupted. Subsequently the inflammatory microbial products, like the LPS of gram-negative bacteria, can reach distant organs through circulation. LPS and fatty acids are elevated under high fat diet (HFD) and activate TLR4 signaling. Organ-specific effects after HFD include upregulation of TLR4 and macrophage attractant CCL2 in the adipose tissue, which leads to inflammation and supports obesity.

anti-inflammatory M2 polarization (41), decreases tissue and circulating inflammatory marker levels, and diminishes inflammation in the streptozotocin-induced mouse model for type 1 diabetes (42). *Irf3* knockdown improves insulin sensitivity, mediates an anti-inflammatory phenotype, and also increases white fat browning (60). Intestinal epithelial cell-specific *MyD88* deletion decreases fat mass accumulation, body weight gain and glucose intolerance in diet-induced obese mice (61). Apart from LPS, fatty acids can also increase TLR4 signaling in several cell types including adipocytes and macrophages (62) and promote visceral obesity and insulin resistance (63). Interestingly, adipocyte-specific *Tlr4* KO mice demonstrate two distinct effects of TLR4 on the adipose tissue. Specifically, these mice show increased whole body and muscle insulin resistance after HFD feeding, but also improved insulin sensitivity after an acute lipid challenge during a hyperinsulinemic euglycemic clamp (64). Adipocyte-specific *Tlr4* deletion also modifies *Tlr4* expression in other tissues, as it decreases *Tlr4* expression in peritoneal macrophages and liver. Alteration of TLR levels in the liver has important systemic metabolic effects. Hepatocytespecific *Tlr4* deletion improves glucose tolerance, insulin sensitivity, and hepatic steatosis in HFD fed obese mice (65). In addition to the local signaling in the gut or the peripheral tissues, the TLR-mediated effects also reach the brain. TLR4 inhibition decreases inflammation and leptin resistance in the hypothalamus (43). Deletion of the TLR downstream signaling molecule *MyD88* in the central nervous system prevents obesity and leptin resistance (66). Furthermore, in older or HCD-fed



*TLR and SCFA signaling occurs in response to microbiota changes in the intestine and can reach distant organs, influencing the inflammatory and metabolic state to initiate systemic effects.*

*DIO, diet-induced obesity; HFD, high-fat diet; TLR, toll-like receptor; Tregs, regulatory T cells; T1D, type 1 diabetes.*

mice, *Tlr4* expression was increased in the pro-opiomelanocortin neurons from the arcuate nucleus of the hypothalamus, a central metabolism regulating brain area (67).

Toll-like receptor 5 recognizes bacterial flagellin, a component of the bacterial locomotion system. This signaling pathway causes an anti-inflammatory response by inducing interleukin 1 receptor antagonist secretion and diminishing IL-1β and inflammasome activity (68). *Tlr5* KO mice were first reported to be prone to develop metabolic syndrome, including insulin resistance and increased adiposity, which correlated with changes in their microbiota composition (69). This phenotype was also confirmed in an epithelial cell-specific *Tlr5* KO (70) and was associated with low-grade inflammation. Interestingly, these initial findings could not be reproduced in the same *Tlr5* KO mouse line by a different research group (71, 72), nor in a new *Tlr5* KO model (73). While a clear explanation for this discrepancy is missing, a possible reason could lay in the differences in the microbiota composition between different mouse facilities, and/ or may depend on whether initially pro- or anti-inflammatory molecules are predominant in the local gut environment. Specific *Tlr5* deletion in hepatocytes confers predisposition to diet-induced liver pathology. This is accompanied by elevated expression of proinflammatory cytokines and dependent on the Nod-like-receptor C4 inflammasome and rescued by microbiota depletion (44). Similarly, hepatocyte-specific deletion of *MyD88* leads to inflammation, glucose intolerance, and hepatic insulin resistance, accompanied by alterations in the microbiota composition (74). These observations suggest that hepatocyte TLR5 plays a role in protecting the liver and in preventing diet-induced liver disease.

## DYSREGULATION OF GUT MUCOSAL SURFACES CAN CONTRIBUTE TO SYSTEMIC MICROBIAL EFFECTS

Changes in the gut microbiota can modulate systemic microbederived metabolite levels by affecting their biosynthesis, or by changing the intestinal permeability and gut barrier. The mucosal surfaces are another first line defense component of the innate immune system, at which most of the microbe–host interactions take place. In the gastrointestinal tract, the mucosal surface consists of an epithelial cell layer, in which the mucus-secreting goblet cells are responsible for forming a second protective barrier, the mucus layer. Due to its physiologic function in the gut for food absorption, the mucosal surface is naturally a thin and permeable layer. The protection against microbial invaders under natural physiologic conditions is ensured by several mechanisms, including tight junctions between the epithelial cells, which are constantly renewed from stem cells allowing immediate repair in case of cell loss or damage. Mucus is composed of charged glycoproteins called mucins and its density and stickiness traps microbes and their products, preventing contact to the epithelial cells. Within the large intestinal mucus, there are two layers, a loose outer layer that is a natural habitat for microbes and a dense inner layer, impermeable for bacteria (15).

Dysregulation of the mucus structure by changes in its thickness or penetrability can lead to inflammation. This is evident in the *mucin 2* (*Muc2*) KO mice where microbiota is in direct contact with the epithelial cells (75, 76). TLRs were shown to mediate MUC2 secretion (77) and were implied in epithelial homeostasis (28, 78). *MyD88* deletion in intestinal epithelial cells leads to defects in mucosal barrier functions with reduced *Muc2* mRNA expression and increases epithelial cell penetration by bacteria (79, 80). GF mice have easily penetrable mucus in the large intestine, whereas conventionally raised mice show a thick, impenetrable mucus layer (15, 81). The level of penetration and degradation depends largely on the gut microbiota composition (82). Although most bacteria metabolize non-digested food polysaccharides (83, 84), many bacteria can use mucin glycans as an energy source (85–87). This substantial role of the microbiota also indicates that an imbalance in the bacterial species composition can influence the protective capacity of the mucus barrier. This kind of dysbiosis can result from a variety of different factors, including a decreased bacterial diversity, lack of secretory antibodies that line the mucus layer, and lack of Tregs (88) or eosinophils (89, 90). Once the mucus barrier is invaded and microbes are in closer contact to epithelial cells, mucus production and stem cell number is increased in the intact epithelium, and this is triggered by MyD88 signaling (91). This illustrates that the epithelium counteracts the potential threat of microbes invading the mucus layer. However, similar to the mucus layer, the epithelial cell lining can also become permeable. Damage to the epithelial layer can originate from altered tight junction composition or epithelial cell stress (92). Autophagy, a process in which a phagosome is formed in the cytoplasm to engulf various contents, is also important for maintaining an intact epithelial barrier and in the regulation of mucus secretion (93, 94). Disruption of autophagy induced by coding polymorphism (Thr300Ala) in the autophagy-related 16-like 1 leads to decreased antibacterial autophagy, induces epithelial cell stress, and allows systemic penetration of bacteria and inflammation (95).

Gut permeability is increased with HFD, in obesity and in diabetes (96). HFD-induced nonalcoholic fatty liver disease (NAFLD) at thermoneutral housing conditions (30°C) is associated with an altered microbiome, increased intestinal permeability and induction of proinflammatory responses that are active in this disease in humans. Depletion of hematopoietic *Tlr4* or gram-negative microbiota leads to altered immune responsiveness and protects from NAFLD at thermoneutrality (97). With dietary fiber deficiency, a subset of mucindegrading bacteria increases, which express mucin-degrading carbohydrate-active enzymes (CAZymes) that enable access to host-secreted mucus glycoproteins as a nutrient (98), resulting in the degradation of the colonic mucus barrier. This increases LPS levels systemically. Glucagon like peptide (GLP)-1 and 2 levels, which have beneficial metabolic effects (99) and could affect gastric barrier function (100), can be regulated by LPS. In addition, obese mice have impaired IL-22 induction from innate lymphoid and CD4<sup>+</sup> T cells under immune challenge, and IL-22 reverses the HFD-induced epithelial cell stress (92). IL-22 depletion results in defects in mucosal defense and metabolic symptoms, many of which can also be reversed by exogenous IL-22 (101).

A group of molecules that can both modulate the intestinal barrier and escape the gut to influence systemic health are the SCFAs. SCFAs acetate, butyrate, and propionate are produced by bacterial anaerobic metabolism of indigestible dietary components, including fiber. They signal *via* G-proteincoupled-receptors such as GPR41, GPR43, and GPR109a (102, 103) and are important regulators of gut homeostasis and epithelial barrier maintenance. Acetate enhances protection against infection (104), while butyrate is an energy source for colonocytes (105–107) and can regulate stem cell proliferation and anti-inflammatory macrophage polarization (54, 108) with its histone deacetylase inhibiting function. Butyrate also increases colonic mucus secretion, both directly by promoting *Muc2* and glycoltransferase expression (55, 109), and indirectly by promoting autophagy (93, 94, 110, 111). Further protection against inflammation through SCFAs is accomplished by their regulation of Tregs (112). The elevated circulating SCFA levels in mice fed a high-fiber diet result in allergy protection by impaired Th2 differentiation and increased phagocytosis by DCs, while low-fiber diet promotes allergic inflammation (113). In type 1 diabetes, the SCFA amount correlates with improvement of symptoms *via* limitation of autoreactive T cells, induction of Tregs, and enhanced gut barrier (50). The beneficial SCFA effects (52, 114) can act through local induction of gut-derived hormone secretion, including GLP-1 and peptide YY (45–47), and through the circulation (115). SCFA-mediated beneficial metabolic effects on health can be mediated by induction of intestinal gluconeogenesis (48). Specifically, supplementation with acetate in drinking water or through nanoparticles, butyrate gavage alone or combined with propionate, or adding propionate, butyrate, and acetate to the diet, improve host physiology and glucose metabolism (49, 51, 52, 116) (**Table 1**), which in the case of propionate seems to be mediated by vagus-nerve stimulation by the peripheral nervous system (48). Conversely, the microbiota-mediated increase in acetate turnover that occurs during exposure to a HFD diet might mediate a feedback loop between the gut microbiota and the parasympathetic nervous system, promoting hyperphagia owing to increased ghrelin secretion, and increased energy storage (53). The site of stimulation seem to be important for the outcome of SCFA-mediated effects (117, 118) and points to the need for further exploration of the general role of SCFAs in regulating obesity.

## CONCLUSION

The encounter of microbial molecules by TLRs initially provokes a local immune response in the gut. Such local reactions can spread by escape of the microbial products from the gut, which can then reach distant organs. In addition to immune cells, the adipose tissue, liver, and brain are important TLR-expressing targets and are all potent metabolic regulators with systemic effects. The escape of microbial factors from the gut can result from perturbations in the gut microbiota and its interaction with immune system components that weakens the mucosal and epithelial lining of the gut. When the integrity of the mucus layer and intestinal epithelium is impaired, harmless commensals can become a threat, crossing the epithelium to invade the bloodstream, causing systemic infection or inflammation, which favors development of immune-mediated and metabolic diseases. Consequently, peripheral anti- or proinflammatory responses take place and modulate host metabolism, emphasizing to which substantial extent microbiota can influence systemic health.

In addition to the above discussed mechanisms, bile acids, which can be modified by microbiota or result from microbiota changes, could modulate host metabolism (119, 120) and also have immunomodulatory effects (121). Further gut microbiota derived metabolites, including tryptophan, phenylalanine, tyrosine, and polyamines, can regulate the immune system and the host metabolism with potential effects on health. For instance, the molecular signature of such interactions is described for the regulation of mucosal integrity and inflammatory signaling (122–124) by bacterial indoles binding to the pregnane X receptor (125) *via* pathways involving TLR4 (126) and NFkB. Indoles (127–132) and dietary ligands such as flavonoids (133) can be sensed by aryl hydrocarbon receptor nuclear translocator 2, which modulates epithelial barrier integrity (134–139) and provokes immune changes (140–144). Further immunologic components that can bridge microbiota with host metabolism include the nucleotide-binding oligomerization domain-like receptors (NOD-like receptors) and innate lymphoid cells that have been reviewed elsewhere (145). While linking host metabolism with microbiota alterations is a challenging task, the integrative metabolomics–metagenomics approaches are a promising way to provide a better understanding of these interactions. For example, using such approaches Pedersen and colleagues identified *Prevotella copri* and *Bacteroides vulgatus* as the main species driving the association between biosynthesis of branched-chain amino acids and insulin resistance, where *P. copri* increased their circulating levels, and induced insulin resistance and glucose intolerance (146).

Reprogramming of microbiota in patients can be an accessible and promising anti-obesity treatment. Thus, understanding the complex and reciprocal interplay between the microbiota and the immune system, and how this relationship can modulate metabolic parameters could result in advances toward the treatment of metabolic diseases.

### AUTHOR CONTRIBUTIONS

The authors reviewed literature and conceived the manuscript. MS wrote the manuscript and prepared the figure. DM provided

### REFERENCES


insightful comments and corrections. MT wrote paragraphs, corrected the manuscript, and provided guidance.

#### ACKNOWLEDGMENTS

We wish to thank our colleagues from the Faculty of Medicine, University of Geneva for discussions. We apologize that due to space limitations, we could not cite all relevant literature.

### FUNDING

This work was supported by the CONFIRM grant of the Hôpitaux Universitaires de Genève (HUG) Foundation (no. RC2-09) and the Swiss Multiple Sclerosis Society Grant to DM and MT; the European Research Council under the European Union's Seventh Framework Programme (FP/2007-2013)/ERC Grant Agreement no. 336607 (ERC-2013-StG-336607) to MT; Swiss National Science Foundation (SNSF) grant to DM (310030\_173010); and SNSF Professorship grants (PP00P3\_144886 and PP00P3\_172906) to MT.

infectious diseases. *Clin Transl Immunology* (2017) 6:e125. doi:10.1038/ cti.2016.91


high-fat diet-induced liver disease. *Cell Mol Gastroenterol Hepatol* (2016) 2:584–604. doi:10.1016/j.jcmgh.2016.04.007


layers requires long-term microbial colonization. *Cell Host Microbe* (2015) 18:582–92. doi:10.1016/j.chom.2015.10.007


obesity upon cold exposure. *Cell Metab* (2016) 23:1216–23. doi:10.1016/j. cmet.2016.05.001


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Spiljar, Merkler and Trajkovski. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

,

\*

# Fecal Microbiota Transplantation, Commensal Escherichia coli and Lactobacillus johnsonii Strains Differentially Restore Intestinal and Systemic Adaptive Immune Cell Populations Following Broad-spectrum Antibiotic Treatment

Ira Ekmekciu<sup>1</sup> Alexander Scheffold2,3, Stefan Bereswill<sup>1</sup> and Markus M. Heimesaat<sup>1</sup>

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Rajaraman D. Eri, University of Tasmania, Australia Lisa Rizzetto, Fondazione Edmund Mach, Italy Francisco José Pérez-Cano, University of Barcelona, Spain

> \*Correspondence: Markus M. Heimesaat

markus.heimesaat@charite.de

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 29 August 2017 Accepted: 23 November 2017 Published: 11 December 2017

#### Citation:

Ekmekciu I, von Klitzing E, Neumann C, Bacher P, Scheffold A, Bereswill S and Heimesaat MM (2017) Fecal Microbiota Transplantation, Commensal Escherichia coli and Lactobacillus johnsonii Strains Differentially Restore Intestinal and Systemic Adaptive Immune Cell Populations Following Broad-spectrum Antibiotic Treatment. Front. Microbiol. 8:2430. doi: 10.3389/fmicb.2017.02430 1 Intestinal Microbiology Research Group, Institute of Microbiology, Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany, <sup>2</sup> Department of Cellular Immunology, Clinic for Rheumatology and Clinical Immunology, Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany, <sup>3</sup> German Rheumatism Research Center, Leibniz Association, Berlin, Germany

, Christian Neumann2,3, Petra Bacher<sup>2</sup>

, Eliane von Klitzing<sup>1</sup>

The essential role of the intestinal microbiota in the well-functioning of host immunity necessitates the investigation of species-specific impacts on this interplay. Aim of this study was to examine the ability of defined Gram-positive and Gram-negative intestinal commensal bacterial species, namely Escherichia coli and Lactobacillus johnsonii, respectively, to restore immune functions in mice that were immunosuppressed by antibiotics-induced microbiota depletion. Conventional mice were subjected to broadspectrum antibiotic treatment for 8 weeks and perorally reassociated with E. coli, L. johnsonii or with a complex murine microbiota by fecal microbiota transplantation (FMT). Analyses at days (d) 7 and 28 revealed that immune cell populations in the small and large intestines, mesenteric lymph nodes and spleens of mice were decreased after antibiotic treatment but were completely or at least partially restored upon FMT or by recolonization with the respective bacterial species. Remarkably, L. johnsonii recolonization resulted in the highest CD4+ and CD8+ cell numbers in the small intestine and spleen, whereas neither of the commensal species could stably restore those cell populations in the colon until d28. Meanwhile less efficient than FMT, both species increased the frequencies of regulatory T cells and activated dendritic cells and completely restored intestinal memory/effector T cell populations at d28. Furthermore, recolonization with either single species maintained pro- and antiinflammatory immune functions in parallel. However, FMT could most effectively recover the decreased frequencies of cytokine producing CD4+ lymphocytes in mucosal and systemic compartments. E. coli recolonization increased the production of cytokines such as TNF, IFN-γ, IL-17, and IL-22, particularly in the small intestine. Conversely, only L. johnsonii recolonization maintained colonic IL-10 production. In summary,

**98**

FMT appears to be most efficient in the restoration of antibiotics-induced collateral damages to the immune system. However, defined intestinal commensals such as E. coli and L. johnsonii have the potential to restore individual functions of intestinal and systemic immunity. In conclusion, our data provide novel insights into the distinct role of individual commensal bacteria in maintaining immune functions during/following dysbiosis induced by antibiotic therapy thereby shaping host immunity and might thus open novel therapeutical avenues in conditions of perturbed microbiota composition.

Keywords: commensal intestinal microbiota, Escherichia coli, Lactobacillus johnsonii, fecal microbiota transplantation, secondary abiotic (gnotobiotic) mice, intestinal mucosal and peripheral and central immunity, immune-modulating effects, immune restoration

#### INTRODUCTION

Approximately 10<sup>13</sup> microorganisms, collectively known as microbiota, reside in the human gastrointestinal tract (Sender et al., 2016) and have increasingly received well-deserved attention regarding their deep impact on the physiology and well-being of the mammalian host. The microbiota composition is characterized by vast inter-individual variations and is furthermore influenced by numerous factors, including genetics (Org et al., 2015), mode of delivery (Biasucci et al., 2010), age (Palmer et al., 2007), diet (Ericsson and Franklin, 2015), hospitalization (Penders et al., 2006), exposure to pathogens and/or antibiotics (Buffie et al., 2012; Carding et al., 2015). It has been shown that bacteria belonging to the phyla Firmicutes, Bacteroidetes, Proteobacteria, Actinobacteria, and Fusobacteria comprise the vast majority of the mammalian intestinal microbiota (Eckburg et al., 2005; Palmer et al., 2007; Hill et al., 2012) and both bacterial density and diversity increase from the proximal to the distal gut (Sommer and Backhed, 2013). Firmicutes are Gram-positive bacteria, which include the large class of Clostridia and the lactic acid bacteria (Zhang et al., 2015), while Escherichia coli (Gram-negative, rodshaped bacteria belonging to the family Enterobacteriaceae and phylum Proteobacteria) represent the predominant facultative anaerobe member of the mammalian gastrointestinal tract (Finegold et al., 1983; Tenaillon et al., 2010). The small intestinal microbiota is dominated by the families Lactobacillaceae and Enterobacteriaceae, whereas species from the families Bacteroidaceae, Prevotellaceae, Rikenellaceae, Lachnospiraceae, and Ruminococcaceae can be detected in the colon (Donaldson et al., 2016). While the microbiota is of paramount importance for numerous metabolic processes, including vitamin synthesis (LeBlanc et al., 2013) and digestion of dietary compounds (Backhed et al., 2005), compelling evidence points toward its impact on the maturation, development and function of the innate and adaptive immune system of the host (Macpherson and Harris, 2004; Sommer and Backhed, 2013).

Lactic acid bacteria exert important functions in the modulation of immune responses and have successfully been used as probiotics in inflammatory conditions of mice and men. For instance, Lactobacillus rhamnosus GG has been shown to exert preventive and therapeutic effects in atopic eczema and dermatitis (Isolauri et al., 2000; Kalliomaki et al., 2001, 2003; Viljanen et al., 2005). Moreover, treating IL-10 deficient mice with L. plantarum attenuated the severity of colonic inflammation by reducing mucosal IL-12p40 and IFN-γ levels (Schultz et al., 2002). In our own previous work, we could further demonstrate that a single commensal L. johnsonii strain was able to attenuate both intestinal mucosal and systemic proinflammatory immune responses upon murine infection with the enteropathogen Campylobacter jejuni (Bereswill et al., 2017). Kwon et al. (2010) have proposed that the observed beneficial anti-inflammatory properties may be elicited due to induction of regulatory DCs and regulatory T cells (Treg).

Gram-negative commensals, such as members of the family Enterobacteriaceae, tend to be neglected, but may, nevertheless, be potent immune-modulators (Zeuthen et al., 2006). This is best corroborated by the probiotic strain E. coli Nissle 1917, which has conclusively been shown efficient in the treatment of ulcerative colitis and is considered as an effective alternative to the standard maintenance therapy, i.e., mesalazine (Kruis et al., 2004; Henker et al., 2008; Schultz, 2008).

Importantly, Hessle et al. (2000, 2005) and Skovbjerg et al. (2010) have shown the differing impact of Gram-positive and Gram-negative bacteria on the modulation of cytokine production patterns, given that in vitro co-culturing of human PBMC with Gram-positive bacteria leads to elevated levels of IL-12, IL-1β, IFN-γ, and TNF, while Gram-negative bacteria rather induce IL-6, IL-8, and IL-10. In contrast, from monocytes derived human DC have been shown to produce comparable levels of IL-12 and TNF in response to commensal Gram-positive and Gram-negative bacteria (Karlsson et al., 2004; Zeuthen et al., 2006).

Murine studies have further underlined the impact of individual bacterial species on distinct immune cell populations. For instance, segmented filamentous bacteria (SFB) have been identified to induce the expansion of IL-17 producing Th17 cells (Ivanov et al., 2008), while Clostridium species of clusters IV and XIVa promoted accumulation of Treg in the colonic lamina propria (LP) of mice (Atarashi et al., 2011). The differentiation of CD4+ T cells into Treg locally in the LP as well as in the

**Abbreviations:** ABx, secondary abiotic; BSA, bovine serum albumin; CFU, colony forming units; DC, dendritic cells; FMT, fecal microbiota transplantation; IFN-γ, interferon-γ; IL, interleukin; LPL, lamina propria lymphocytes; LPS, lipopolysaccharide; MLN, mesenteric lymph nodes; PBS, phosphate buffered saline; PBMC, peripheral blood mononuclear cells; PP, Peyer's patches; PSA, polysaccharide A; SFB, segmented filamentous bacteria; SPF, special pathogen free; Th, T helper cell; TLR, toll-like receptor; TNF, tumor necrosis factor; Treg, regulatory T cells.

circulation is also supported by TLR-2 mediated sensing of the Gram-negative bacterium Bacteroides fragilis (Round et al., 2011). Moreover, treatment of IL-10−/<sup>−</sup> mice with L. plantarum has been shown to attenuate the severity of colonic inflammation by reducing mucosal IL-12p40 and IFN-γ levels (Schultz et al., 2002). However, elucidating the differential impact of Grampositive and Gram-negative commensals on the immune cell homeostasis in in vivo models, and identifying the species specific mechanisms of immunomodulation remains difficult and challenging, yet of utmost interest.

In our previous works, we have shown that normally developed mice rendered void of intestinal microbiota through a quintuple antibiotic therapy (i.e., secondary abiotic mice, ABx mice), display numerous changes of intestinal mucosal and systemic immune cell subsets, which can, however, almost completely be restored through reintroduction of intestinal antigens via fecal microbiota transplantation (FMT) (Ekmekciu et al., 2017a,b). Furthermore we were able to demonstrate that recolonization of ABx mice with VSL#3, a probiotic mixture consisting of eight bacterial species (namely Streptococcus thermophilus, Bifidobacterium breve, B. longum, B. infantis, Lactobacillus acidophilus, L. plantarum, L. paracasei, and L. delbrueckii subsp. bulgaricus), elicits IL-10 production by lymphocytes in the small and large intestinal LP, MLN and spleen, but does not influence the production of pro-inflammatory cytokines (Ekmekciu et al., 2017a,b). The impact of probiotics on cytokine production has been extensively investigated, and it has been proposed, that some strains might promote mainly IL-12 production, thus inducing Th1 type immune responses, which are known to protect from pathogens. Other strains, however, more effectively impacted the production of the anti-inflammatory IL-10, thereby limiting excessive immune responses as observed in autoimmune and allergic diseases, for instance (Fujiwara et al., 2004; Foligne et al., 2007; Shida et al., 2011).

Niess et al. (2008) have previously further addressed the role of commensal organisms in the recruitment of pro-inflammatory cytokine producing CD4+ cells in the colonic lamina propria and proposed that these cell types might contribute to the immunopathogenesis of colitis. However, a fine-tuned balance between pro- and anti-inflammatory immune responses at mucosal and systemic sites is of utmost importance for the vertebrate host. For instance, a reduction of the IL-17 producing Th17 cell compartment playing a pivotal role in protection against bacterial and fungal pathogens (Aujla et al., 2007) may explain the increased susceptibility of microbiota depleted mice to pathogens (Croswell et al., 2009; Stiemsma et al., 2014).

In the present study, we investigated the role of two important representatives of Gram-negative (namely E. coli) and Gram-positive (namely L. johnsonii) commensals, both derived from the gut microbiota of healthy mice, in restoring numbers of distinct immune cell subsets following antibiotic treatment as compared to FMT in C57Bl/6j ABx mice. To address this, we analyzed the immune responses exerted by lymphocytes within the small and large intestinal LP, MLN and spleen of microbiota-depleted mice upon reassociation with E. coli or L. johnsonii and upon FMT at day (d) 7 and d28 post recolonization. We furthermore surveyed the cytokine production patterns assessing TNF, IFN-γ, IL-17, IL-22, and IL-10 secretion by CD4+ lymphocytes in respective compartments.

#### MATERIALS AND METHODS

#### Ethics Statement

All animal experiments were conducted according to the European Guideline for animal welfare (2010/63/EU) with approval from the commission for animal experiments headed by the "Landesamt für Gesundheit und Soziales" (LaGeSo, Berlin, Germany, registration numbers G0097/12 and G0184/12). Animal welfare was examined twice daily by assessment of clinical conditions including weight loss.

### Generation of Secondary Abiotic (Gnotobiotic) Mice

All animals were bred, raised and housed in the facilities of the "Forschungseinrichtungen für Experimentelle Medizin" (FEM, Charité – University Medicine Berlin, Germany) under specific pathogen-free (SPF) conditions. Secondary abiotic mice were generated through quintuple antibiotic treatment for 8 weeks via the drinking water as previously described (Heimesaat et al., 2006; Bereswill et al., 2011; Fiebiger et al., 2016; Ekmekciu et al., 2017a,b).

#### Bacterial Recolonization

Three days prior bacterial recolonization experiments, the antibiotic cocktail was withdrawn and replaced by sterile drinking water (ad libitum). Successful depletion of the intestinal microbiota was confirmed and FMT performed as described previously (Ekmekciu et al., 2017a,b). Furthermore, secondary abiotic mice were recolonized with 10<sup>9</sup> CFU of either E. coli or L. johnsonii in 0.3 ml PBS (Gibco Life Technologies, Paisley, United Kingdom) by gavage on day 0. The applied E. coli strain constitutes a commensal isolate derived from a naive conventionally colonized C57BL/6j wildtype mouse and did not express known virulence factors such as stx 1 and 2, hlyA, cspA, catA, katA, and astA as confirmed in a reference laboratory (Heimesaat et al., 2007; Haag et al., 2012). The L. johnsonii strain had initially been isolated from the feces of a healthy female 3 months old C57BL/6j wildtype mouse as previously described (Bereswill et al., 2017). For FMT fresh murine fecal samples were collected from 10 age and sex matched SPF control mice, pooled, dissolved in 10 ml sterile PBS and the supernatant perorally applied by gavage (in 0.3 ml PBS) in order to reconstitute secondary abiotic (i.e., gnotobiotic) mice with a complex intestinal microbiota as shown previously (Heimesaat et al., 2006; Ekmekciu et al., 2017a,b).

#### Sampling Procedures

Mice were sacrificed by isoflurane treatment (Abbott, Greifswald, Germany) at d7 or d28 following bacterial

recolonization. Luminal large intestinal samples as well as ex vivo biopsies from spleen, MLN, ileum, and colon were taken under sterile conditions. Intestinal samples were collected in parallel for microbiological and immunological analyses.

## Quantitative Analysis of Bacterial Colonization

Total intestinal loads of E. coli and L. johnsonii were quantitated in fecal and colonic samples over time post recolonization or upon necropsy. Respective samples were dissolved in PBS and serial dilutions streaked onto appropriate solid culture media: E. coli was detected on Columbia agar supplemented with 5% sheep blood and MacConkey agar (Oxoid, Wesel, Germany) following aerobic incubation at 37◦C for 48 h. L. johnsonii loads were determined on Columbia agar supplemented with 5% sheep blood, Columbia-CNA agar supplemented with colistin and nalidixic acid and MRS agar (all from Oxoid) in parallel and incubated under microaerobic and obligate anaerobic conditions (in jars using CampyGen and AnaeroGen gas packs, respectively; both from Oxoid) at 37◦C for at least 2 days. Bacterial species were identified according to their typical morphological appearances and biochemical properties. The detection limit of viable bacteria was ≈100 CFU/g.

### Lymphocyte Isolation from Spleen and Mesenteric Lymph Nodes

Single cell suspensions were generated from spleens and MLN, and erythrocytes were removed from splenic samples by 1.66% ammonium chloride. All samples were resuspended in volumes of 5 ml (spleen) and 2 ml (MLN) PBS/0.5% BSA (Sigma–Aldrich, St. Louis, MO, United States) and subjected to further processing (Cording et al., 2013).

#### Lamina Propria Lymphocyte Isolation

Lamina propria lymphocyte isolation followed a standard protocol with minor modifications as described previously (Sheridan and Lefrancois, 2012; Ekmekciu et al., 2017b). Briefly, the intestines were cut into 0.5 cm pieces and incubated twice in 25 ml 1 mM dithioerythritol (DTE; Carl Roth) in Hanks balanced salt solution (HBSS; Gibco) for 20 min at 37◦C at 220 rpm. Afterward the intestines were introduced to 1.3 mM ethylenediaminetetraacetic acid (EDTA; life technologies, Eugene, OR, United States) solution in HBSS and were shaken again twice herein for 30 min at 37◦C at 220 rpm. After each incubation, the epithelial cell layer containing the intraepithelial lymphocytes was removed by intensive vortexing and passing through a 70 µm cell strainer, and new solution (DTE or EDTA) was added. After the second EDTA incubation the cells were washed with RPMI 1640 (Gibco) containing 5% fetal calf serum (FCS; Biochrom, Berlin, Germany) and were subsequently placed in 35 ml digestion solution containing 0.5 mg/ml collagenase A (Roche, Mannheim, Germany), 0.5 mg/ml DNAse I (Roche), 10% FCS, 1 mM of each CaCl<sup>2</sup> and MgCl<sup>2</sup> (both Carl Roth) in RPMI 1640 (Gibco). Digestion was performed through incubation for 45 min at 37◦C and 220 rpm. After incubation the digested tissues were washed in RPMI supplemented with 5% FCS and centrifuged for 6 min at 4◦C and 350 × g. The pellets were resuspended in 5 ml 44% Percoll (GE Healthcare, Uppsala, Sweden) and overlaid on 5 ml 67% Percoll in a 15 ml Falcon tube. Percoll gradient separation was performed by centrifugation at 600 × g for 20 min at room temperature. LPL were collected from the interphase, washed once and suspended in 1 ml of PBS/0.5% BSA.

## Surface and Intracellular Stainings and Flow Cytometry

Surface staining was performed using a mix of the following antibodies: FITC-anti-CD4 (Clone RM4-5; 1:200), PerCPanti-CD8 (Clone 53-6.7; 1:100), PacBlue-anti-B220 (Clone RA3-6B2, 1:200), APC-Cy7-anti-CD25 (Clone PC61, 1:200), PE-anti-CD44 (Clone IM7, 1:200), APC-anti-CD86 (Clone B7-2, 1:200) (all from BD Biosciences, San Jose, CA, United States).

For intracellular staining cells from spleen, MLN and intestinal LP were restimulated for 5 h with 10 ng/ml phorbol myristate acetate (PMA) and 1 µg/ml ionomycin, in a tissue culture incubator at 37◦C (both Sigma–Aldrich). The 10 µg/ml brefeldin A (Sigma–Aldrich) were added to the cell suspensions after 1 h of polyclonal restimulation. Then cells were treated with LIVE/DEAD Fixable Aqua Dead Cell Stain kit (life technologies) and hereafter fixed with 2% paraformaldehyde (PFA; Sigma–Aldrich) for 20 min at room temperature. Cells were stained in 0.5% saponin (Sigma–Aldrich) using a mix of the following antibodies: PacBlue-Anti-CD4 (Clone RM4-5; 1:400), PE-Cy7-anti-IFN-γ (Clone XMG 1.2; 1:400), APC-Cy7-anti-TNF-α (Clone MP6-XT22; 1:400) (all three from BD Biosciences), FITCanti-IL17A (Clone TC11-18H10.1; 1:200, BioLegend, San Diego, CA, United States), PE-anti-IL10 (Clone JESS-16E3; 1:100), APC-anti-IL22 (Clone IL22JOP; 1:100) (both from eBioscience). After gating for lymphocytes and excluding doublets only living cells were included in further analyses. CD4 and CD8 cells were gated on living cells, whereas CD86+ (activated dendritic) cells were identified in the CD4-CD8 compartment. All data were acquired on a MACSQuant analyzer (Miltenyi Biotec, Bergisch Gladbach, Germany) and were analyzed with FlowJo Software v10.1 (Tree star, Ashland, OR, United States).

## Statistical Analysis

Medians, means, standard deviations (SD) and significance levels were determined using Mann–Whitney U test or oneway analysis of variance (ANOVA) with Tukey's post hoc test for multiple comparisons (GraphPad Prism Software v6, La Jolla, CA, United States) as indicated. Two-sided probability (p) values ≤ 0.05 were considered significant. Experiments were reproduced twice and pooled data are shown (n = 8–15 per group).

FIGURE 1 | Kinetics of intestinal bacterial colonization densities following bacterial recolonization of secondary abiotic mice. Secondary abiotic mice were generated by broad-spectrum antibiotic treatment and perorally recolonized with (A) E. coli (open squares) or (B) L. johnsonii (open circles) on day (d) 0 as described in "Materials and Methods." Bacterial colonization densities were assessed in fecal samples (colony forming units per gram, CFU/g) over time upon recolonization as indicated by culture. Medians (black bars) are indicated. Data were pooled from three independent experiments.

#### RESULTS

### Kinetics of Intestinal Colonization Densities Following Recolonization of Secondary Abiotic Mice with E. coli or L. johnsonii

In the present study, we investigated the immunomodulatory properties of representative Gram-negative or Gram-positive intestinal bacterial species (namely, E. coli and L. johnsonii, respectively) in comparison to a complex gut microbiota in mice that had been treated with broad-spectrum antibiotic compounds. To address this, conventionally reared mice were subjected to a quintuple antibiotic cocktail rendering them secondary abiotic (Ekmekciu et al., 2017a,b), and after a sufficient wash-out period ABx mice were perorally recolonized with respective bacterial species or the complex intestinal microbiota by FMT. Importantly, the intestinal microbiota composition of ABx mice that had been subjected to FMT was comparable to naive conventionally colonized counterparts as confirmed previously (Ekmekciu et al., 2017a,b). As early as 3 days upon initial recolonization we assessed intestinal colonization efficiencies of respective species over time (**Figure 1**). Cultural analyses of fecal samples revealed that E. coli (**Figure 1A**) and L. johnsonii (**Figure 1B**) stably colonized the murine intestinal tract, both with high median loads of >10<sup>8</sup> CFU per gram feces until necropsy at d28. Of note, none of the interventions carried out in this study (i.e., antibiotic treatment, recolonization with E. coli/L. johnsonii, FMT) led to any adverse clinical effects in mice such as wasting, diarrhea, occurrence of blood in feces or microscopic signs of intestinal inflammation including epithelial apoptosis (data not shown).

#### T Cell Populations in Murine Intestinal and Systemic Compartments Following FMT or Recolonization of Secondary Abiotic Mice with E. coli or L. johnsonii

To examine the impact of E. coli and L. johnsonii versus complex gut microbiota on distinct immune cell populations after antibiotic microbiota depletion, we isolated lymphocytes from small and large intestinal LP, MLN and spleen and analyzed defined immune cell populations by flow cytometry on d7 and d28 post recolonization. Firstly, we analyzed the CD4+ (**Figures 2**, **3**) and CD8+ lymphocytes (**Supplementary Figures S1**, **S2**) and assessed both the relative abundances (i.e., percentages) and the absolute cell numbers of respective immune cell populations. After the antibiotic treatment the abundance of CD4+ T helper lymphocytes declined in both the small and large intestines, but could be fully restored upon FMT (p < 0.05–0.001; **Figures 2A,B**). Mice harboring E. coli or L. johnsonii displayed slightly higher percentages of CD4+ cells in their small intestine without reaching significant levels (n.s. vs ABx; **Figure 2A**). In the colon, however, E. coli recolonization resulted in higher frequencies of CD4+ cells at d7, but decreased again until d28 (**Figure 2A**). Interestingly, in the MLN the percentage of the CD4+ lymphocytes was at its lowest at d7 post FMT (p < 0.05 vs. N; **Figure 2C**), but reached naive levels thereafter. In the spleens of L. johnsonii colonized mice we detected higher frequencies of CD4+ lymphocytes than in their with E. coli colonized counterparts at d28 (p < 0.05–0.01; **Figure 2D**). Furthermore, L. johnsonii recolonization led to the highest absolute CD4+ cell numbers in the murine small intestine (p < 0.05–0.001; **Figure 3A**). At d7 following either recolonization higher numbers of CD4+ lymphocytes could be observed in the colonic LP as compared to ABx mice (p < 0.05–0.01; **Figure 3B**), but declined until d28. Notably, at d28 after L. johnsonii colonization mice also displayed the highest numbers of CD4+ cells in the

splenic (and hence systemic) compartment (p < 0.05–0.001; **Figure 3D**).

Interestingly, the microbiota depletion-induced decreases of the CD8+ cell abundances within the small and large intestinal LP were accompanied by increases of these cells in the spleen (p < 0.01–0.001 ABx vs. naive; **Supplementary Figures S1A,B,D**). Single strain mono-colonization could restore this cell population in the small intestine rather late, i.e., until d28 post recolonization with either E. coli or L. johnsonii (p < 0.05–0.001; **Supplementary Figure S1**). Remarkably, neither E. coli nor L. johnsonii colonization impacted the decline of colonic CD8+ cell numbers, which could only be restored upon FMT at d28 (**Supplementary Figure S1B**). At d7 post FMT, mice displayed the lowest frequencies of CD8+ cells in their MLN, which, however, increased until d28 (**Supplementary Figure S1C**). Moreover, at d7 following either recolonization, mice exerted lower percentages of splenic CD8+ cells than their ABx counterparts (p < 0.05–0.001; **Supplementary Figure S1D**). This effect, however, could be preserved over time in E. coli recolonized mice only. Remarkably, mice harboring single commensal species in their gut displayed higher CD8+ lymphocyte numbers in the small intestinal LP at d28 as compared to mice that had been subjected to FMT (p < 0.05–0.001; **Supplementary Figure S2A**). Until d7, however, only L. johnsonii but not E. coli could sufficiently restore the antibiotics-induced decreased colonic CD8+ lymphocytes (p < 0.01 vs. ABx; **Supplementary Figure S2B**). In line with the splenic CD4+ lymphocytes, the highest splenic CD8+ lymphocyte numbers could be detected in L. johnsonii colonized mice at d28 post recolonization (p < 0.001; **Supplementary Figure S2D**).

#### Activated T Cells (Including Treg), Memory/Effector T Cells and Activated Dendritic Cells in Murine Intestinal and Systemic Compartments Following FMT or Recolonization of Secondary Abiotic Mice with E. coli or L. johnsonii

We further expanded our investigations by analyzing the activation status of defined immune cell populations. Therefore we stained for surface markers CD25 (**Figure 4**), CD44 (**Figure 5** and **Supplementary Figure S3**) and CD86 (**Supplementary Figure S4**) characteristic for activated T cells (including Treg), memory/effector T cells and activated DCs, respectively (Sprent and Surh, 2002; Wallet et al., 2005;

Ostman et al., 2006). Broad-spectrum antibiotic treatment led to a significant reduction of the CD4+CD25+ abundances in all analyzed immunological compartments (**Figures 4A–D**). At d7 post FMT, mice displayed the highest percentages of CD4+CD25+ cells, which were even higher than in naive untreated SPF controls. However, both E. coli and L. johnsonii induced a significant increase of CD25 expression in lymphocytes derived from the small intestine, colon and spleen as early as d7 post recolonization (p < 0.05–0.001 vs. ABx; **Figures 4A,B,D**), whereas notably, L. johnsonii exerted a more pronounced effect than E. coli in the colon at d7 (p < 0.05–0.001 vs. Ec; **Figure 4B**). Moreover, only L. johnsonii was able to restore CD4+CD25+ cells in the MLN (p < 0.001 vs. ABx; **Figure 4C**).

Similarly, microbiota-depleted mice exhibited an overall reduction of the CD4+CD44+ memory/effector cell subset (p < 0.001 vs. N; **Figures 5A–D**). In the small and large intestine E. coli, L. johnsonii and FMT were equally capable of restoring this cell population as early as d7 post recolonization (p < 0.01–0.001 vs. ABx; **Figures 5A,B**). This also held true for the MLN, albeit in this compartment L. johnsonii could restore the abundances of CD4+CD44+ cells only rather late in the course, i.e., until d28 post recolonization. In contrast, E. coli, but not L. johnsonii colonization resulted in restored frequencies of splenic memory CD4+ cells at d28 post recolonization (p < 0.001 vs. ABx; **Figure 5D**).

Moreover, reduced abundances of CD8+CD44+ cells could be observed in the small intestine, colon, MLN and spleen of with antibiotics treated mice (p < 0.001 vs. N; **Supplementary Figures S3A–D**). Already at d7 following recolonization (regardless of the regimen) frequencies of CD8+ memory cells were elevated in the small intestine and spleen, whereas, in contrast to FMT, either single species colonization could normalize colonic CD8+ memory cells only until d28. In the MLN, E. coli was able to restore the CD8+ memory/effector cells more effectively than L. johnsonii, considering that abundances of CD8+CD44+ cells were comparable in E. coli mono-associated and naive mice (**Supplementary Figure S3C**).

The expression of the surface molecule CD86, a co-stimulatory protein marking activated DC, showed a strong intestinal

FIGURE 4 | Activated T cells (including Treg) in intestinal and systemic compartments of secondary abiotic and recolonized mice. Secondary abiotic mice were generated by broad-spectrum antibiotic treatment and perorally recolonized by gavage. Subsequently, lymphocytes from small intestinal and colonic lamina propria, MLN and spleen were isolated, and analyzed by flow cytometry as described in "Materials and Methods." The frequencies of activated T cells (including Treg, CD4+CD25+, gated on CD4+ cells) in the (A) small intestine, (B) colon, (C) MLN and (D) spleen of naive conventional mice (N), secondary abiotic mice (ABx) and mice re-associated with either E. coli (Ec), L. johnsonii (Lj) or complex intestinal microbiota by FMT on d7 and d28 post-recolonization are depicted. Columns represent means +SD. Significance levels (p-values) determined with one-way ANOVA test followed by Tukey post-correction test for multiple comparisons are indicated. Significant differences as compared to secondary abiotic mice are indicated by asterisks (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data were pooled from three independent experiments.

microbiota dependence, as indicated by a strong reduction of CD86+ cells in mucosal and systemic compartments of secondary abiotic mice (p < 0.001 vs. N; **Supplementary Figures S4A–D**). Small intestinal CD86+ cells in mice harboring single bacterial species were more abundant than in ABx mice, however, lower than in their naive or with FMT treated counterparts (**Supplementary Figure S4A**). Moreover, only FMT, but neither E. coli nor L. johnsonii, could recover this antibiotics-induced reduction of CD86+ cell frequencies in the MLN until d28 (**Supplementary Figure S4C**). Splenic CD86+ cells displayed the highest abundance at d7 post FMT, but could also be restored to basal naive levels by either E. coli or L. johnsonii recolonization alone (p < 0.001 vs. ABx; **Supplementary Figure S4C**).

Taken together, our data indicate that depending on the respective immunological compartment and immune cell subset, single bacterial commensal species may be as effective as complex microbiota in reversing antibiotics induced collateral damages on intestinal mucosal and systemic immunity.

#### Production of Pro- and Anti-inflammatory Cytokines by CD4+ Cells in Murine Intestinal and Systemic Compartments Following FMT or Recolonization of Secondary Abiotic Mice with E. coli or L. johnsonii

We further addressed the impact of E. coli and L. johnsonii recolonization on the cytokine production pattern of CD4+ lymphocytes in intestinal mucosal, peripheral and systemic immunological sites and therefore determined the frequencies of TNF (**Figure 6**), IFN-γ (**Figure 7**), IL-17 (**Figure 8**), IL-22 (**Figure 9**), and IL-10 (**Figure 10**) producing CD4+ lymphocytes in the small and large intestines, MLN and spleens of recolonized mice.

The antibiotics-induced decreases of TNF producing CD4+ cells in the small and large intestines were accompanied by increased abundances in the MLN and spleen (p < 0.01–0.001 ABx vs. N; **Figures 6A–D**). Small intestinal TNF production was at it highest at d7 following E. coli recolonization,

but further declined thereafter, while it remained unaffected by L. johnsonii during the entire observation period. In addition, either recolonization regimen resulted in higher abundances of TNF producing CD4+ cells in the colonic LP at d7, whereas none of them could sustain this effect until d28. Of note, microbial reassociation of ABx mice dampened TNF production in the MLN (p < 0.05–0.01 vs. ABx; **Figure 6C**), except for FMT at d28. E. coli and complex microbiota could additionally reduce the TNF production in the spleen at d7 (p < 0.01–0.001 vs. ABx, n.s. vs. N; **Figure 6D**).

Furthermore, a significant reduction of IFN-γ producing CD4+ cells could be detected in the small and large intestines of secondary abiotic mice (p < 0.001 vs. N; **Figures 7A,B**). E. coli at d7 and d28, as well as L. johnsonii and FMT at d7 could completely restore the IFN-γ production in the small intestine. Colonic IFN-γ expressing CD4+ cells also increased upon either recolonization regimen at d7 (p < 0.001; **Figure 7B**). In contrast, IFN-γ production remained largely unaffected by antibiotic treatment and subsequent bacterial reassociation in the MLN and spleen of mice (**Figures 7C,D**).

Furthermore, IL-17 expressing CD4+ cells were strongly diminished in all analyzed immunological compartments upon microbial depletion, but could be fully restored by FMT (p < 0.05–0.001; **Figures 8A–D**). Mono-association with E. coli resulted in elevated IL-17 production in the small intestine and colon at both d7 and d28, as well as in the MLN at d7 and spleen at d28 (**Figures 8A–D**). Mice harboring L. johnsonii alone displayed higher abundances of CD4+IL17+ cells than ABx mice in the colon and spleen at d7 post recolonization. In addition, decreased percentages of IL-22 producing CD4+ cells following broad-spectrum antibiotic treatment were detected in either compartment (p < 0.01–0.001; **Figures 9A–D**). Small intestinal IL-22 expression was most distinctly induced by E. coli as early as d7 and by FMT at d7 only (**Figure 9A**). Mono-associated mice further displayed slightly higher abundances of colonic IL-22 producing CD4+ cells than their ABx counterparts, which were, however, only significantly higher in the case of E. coli at d7 (p < 0.05 vs. ABx; **Figure 9B**). Either recolonization regimen could restore the CD4+IL22+ immune cell subset in the MLN at d7, but only E. coli was able to sustain this effect until d28. Moreover, recolonized mice showed similar frequencies of IL-22 producing CD4+ lymphocytes in the systemic compartment as their naive, untreated counterparts, regardless of the intervention or time point (p < 0.05–0.001; **Figures 9C,D**).

Overall, the IL-10 expressing CD4+ lymphocytes were down-regulated in the absence of the intestinal microbiota

(p < 0.05–0.001; **Figures 10A–D**). Both E. coli and L. johnsonii mono-colonization, however, could sufficiently restore the CD4+IL10+ cells in the small intestines, whereas only L. johnsonii could achieve this effect in the colon (**Figures 10A,B**). Increased abundances of IL-10 producing CD4+ cells were also detected upon E. coli recolonization at d7 in the MLN and d28 in both MLN and spleen. The IL-10 production in the MLN and spleen could also be restored at d7 following L. johnsonii recolonization, but this effect could not be sustained until d28 post recolonization.

#### DISCUSSION

In the present study, we aimed at comparing the shortand long-term effects of a single representative Gram-negative and Gram-positive intestinal commensal species, namely of E. coli and L. johnsonii, respectively, versus the complex gut microbiota in shaping host immunity following antibioticsinduced microbiota depletion. These species were chosen for the following reasons. Firstly, both commensals are common inhabitants of the intestinal ecosystem, and although lactobacilli are quantitatively more abundant than E. coli (Castillo et al., 2006), the latter represents the predominant facultative anaerobe Gram-negative strain within the mammalian gastrointestinal tract (Finegold et al., 1983; Tenaillon et al., 2010). Furthermore, in our previous works we could confirm differential immunomodulatory properties of either strain under defined immunopathological conditions. For instance, E. coli, but not L. johnsonii, aggravated Th1-driven proinflammatory immune responses in murine ileitis (Heimesaat et al., 2006, 2007). Furthermore, L. johnsonii was able to attenuate intestinal and systemic pro-inflammatory and to augment anti-inflammatory immune responses upon C. jejuni infection of secondary abiotic mice (Bereswill et al., 2017), and has further been shown to be effective against enteric including infectious morbidities (Lievin-Le Moal and Servin, 2014), hence resulting in its commercial probiotic application (e.g., Nestlé LC1).

Here, we unraveled immunomodulatory properties of respective commensals at two different time points following recolonization of antibiotics-treated mice, but in the absence of immunopathological conditions. Performing kinetic analysis in this context is important, given that intestinal responses are

highly dynamic over time after the transfer of microbiota into germ-free animals (El Aidy et al., 2012; Tomas et al., 2015).

Cultural analysis of fecal samples revealed that either commensal strain was able to stably colonize the murine intestinal tract at high intestinal loads throughout the observation period. Our previous work revealed that upon cessation of antibiotic therapy neither regrowth of intestinal bacterial commensals nor changes in the immune cell populations could be observed, further supporting that the observed immune responses of the mucosal and systemic sites can be exclusively attributed to defined bacterial commensal recolonization (Ekmekciu et al., 2017a,b).

Overall, single-strain recolonization appeared to be, to some extent, less effective in restoring immune cell populations after broad-spectrum antibiotic treatment than the complex microbiota upon FMT, given that mono-colonized mice displayed a lack of full recovery of affected cell populations. Following mono-colonization of mice both common and strict strain-specific immune responses could be observed over time. For instance, E. coli and L. johnsonii had a minor impact on the relative abundances of intestinal CD4+ and CD8+ cells when compared to antibiotics-treated mice, while FMT resulted in cell percentages comparable to naive conventionally colonized counterparts. Furthermore, E. coli, but not L. johnsonii recolonization could increase colonic CD4+ abundances at d7, but not later on. Whereas small intestinal CD8+ cells could be reestablished only rather late upon mono-colonization, colonic CD8+ cells were virtually unaffected. In the systemic compartment, however, the observed increases in percentages of splenic CD8+ cells in microbiota-depleted mice could be reversed by both E. coli and L. johnsonii until d7, and remained lower in E. coli recolonized mice later on. These results underline the importance of examining immunological sequelae of gut microbiota-interventive strategies, not only at mucosal surfaces, but also on the systemic level of the immune system, given that the presence/absence and distinct composition of the gut microbiota might also affect splenic lymphocytes. This rationale is supported by the former findings that the Bacteroides fragilis driven differentiation of Treg was not limited to the LP, but could also be detected in the circulation (Round et al., 2011).

Interestingly, recolonization with L. johnsonii resulted in highest CD4+ lymphocyte numbers in the small intestinal LP, supporting former evidence from a clinical investigation, where the application of another Lactobacillus strain, namely L. reuteri,

led to a significant increase in ileal CD4+ cells (Valeur et al., 2004). In our study, either commensal strain was able to promote CD25 expression on CD4+ lymphocytes in the small and large intestines as well as in the spleen, but only L. johnsonii was able to achieve the same effect in MLN. Of note, at d7 following recolonization, L. johnsonii was a stronger inducer of the colonic activated T cells (including Treg) than E. coli. The dependence of Treg on intestinal bacterial antigens has been demonstrated before. Atarashi et al. (2011) showed that treatment of mice with vancomycin, an antibiotic compound directed mainly against Gram-positive bacteria, resulted in reduced numbers of colonic Treg (Zarkan et al., 2016). A study conducted with germ-free rats revealed that colonic CD25+ cells were higher in animals co-colonized with E. coli and L. plantarum 299v, than in those harboring E. coli alone (Herías et al., 1999). This indicates that increased microbiota diversity and synergistic effects between species may elicit a stronger immunomodulatory effect on activated T cells and the Treg population, which is supported by our findings that FMT induced the most prominent expression of CD25 in all analyzed immunological compartments.

Unlike FMT, single-strain recolonization resulted in an only partial restoration of the intestinal activated DC (CD86+) cells, which remained, however, lower than in untreated SPF mice. Former evidence suggests, that the activation, maturation and cytokine production of DC upon bacterial stimulation is highly dose dependent (Evrard et al., 2011). Our data indicate that the activation of DC may furthermore be dependent on the diversity of presented microbial antigens.

In vitro studies applying PBMC revealed that Gram-positive and Gram-negative bacteria differentially induced cytokine expression patterns (Hessle et al., 2000, 2005; Skovbjerg et al., 2010). Therefore we analyzed the production of pro- and anti-inflammatory cytokines, including TNF, IFN-γ, IL-17, IL-22, and IL-10 in mice following recolonization with either E. coli or L. johnsonii. Intestinal E. coli colonization had a more prominent effect on TNF production by lymphocytes than L. johnsonii. At d7 following E. coli application, TNF levels measured in intestinal, peripheral and systemic immune compartments were comparable to those in naive mice. However, this effect could

be preserved until d28 in MLN only. While excessive TNF production has conclusively been linked to chronic inflammation in numerous organs including the gastrointestinal tract (Kollias et al., 1999; Popa et al., 2007), it is important to recognize the protective roles of TNF against epithelial injury and disease susceptibility. In a murine ileitis model increased TNF levels resulted in improved epithelial barrier functions and prevention from disease onset (Pagnini et al., 2009). Moreover, TNF deficient mice were more prone to acute colitis, thus leading to the conclusion that TNF might have protective functions in normal gut homeostasis and intestinal epithelial integrity (Naito et al., 2003). Rakoff-Nahoum et al. (2004) suggested that the production of intestinal epithelial tissue-protective factors, including TNF, is regulated via TLR-mediated recognition of commensal bacteria, while we here provide further evidence that the adaptive immune system may also contribute to this effect.

Furthermore, both E. coli and L. johnsonii were able to promote the small intestinal and colonic production of IFNγ by CD4+ cells as early as d7 post recolonization, whereas this cytokine remained highly expressed until d28 only in the small intestines of E. coli recolonized mice. The abundance of IFN-γ expressing CD4+ lymphocytes in the MLN and spleen remained unaffected by any of the interventions. These data seem to contradict in vitro findings, which suggest that Grampositive bacteria tend to be stronger inducers of IFN-γ and TNF than Gram-negative species (Hessle et al., 2000, 2005; Skovbjerg et al., 2010). Apart from the general difficulty of directly applying in vitro results to vertebrates exhibiting a complex intestinal microbiota and distinct intra-luminal milieu interacting with the fine-tuned immune system in health and disease, it is also highly likely that human PBMC and murine CD4+ cells differently respond to bacterial stimuli, given former evidence regarding the differences in cytokine patterns between human monocytes and human DC upon bacterial stimulation (Karlsson et al., 2004).

Moreover, E. coli recolonization seemed to favor a Th17 immune cell differentiation, particularly in the small intestine, given that the production of the two key cytokines of this cell population, i.e., IL-17 and IL-22, was significantly higher in CD4+ cells isolated from with E. coli recolonized mice as compared to mice harboring L. johnsonii only. Th17 polarization

is determined by the co-stimulation and cytokine provision from the DC (Perona-Wright et al., 2009), whereby IL-23 plays a particularly important role in the sustenance of Th17 cell responses in vivo (Mangan et al., 2006; McGeachy et al., 2009). It is tempting to speculate that commensal E. coli more strongly induce IL-23 production of DC, thus leading to a Th17 cell differentiation. Interestingly, a study conducted with human PBMC revealed that myeloid DC do indeed produce higher levels of IL-23 when stimulated with E. coli (Manuzak et al., 2012). Furthermore, E. coli heat-labile enterotoxin (synergizing with LPS) has been shown to induce IL-1β and IL-23 secretion by DC, which in turn promoted IL-17 and IFN-γ production by CD4+ T cells (Brereton et al., 2011).

Lactobacillus johnsonii recolonized mice, however, displayed higher abundances of CD4+IL-17+ cells in the colon and spleen and of CD4+IL22+ cells in the small intestinal LP and MLN as compared to their secondary abiotic counterparts at d7 post recolonization. Nonetheless, all the aforementioned immune cell populations declined back to the levels observed in secondary abiotic mice until 4 weeks after L. johnsonii recolonization. It has been known for a while now, that specific bacterial species such as SFB are required for the differentiation of small intestinal Th17 cells in germ-free mice (Ivanov et al., 2008; Gaboriau-Routhiau et al., 2009). A more recent investigation revealed that extracellular pathogens, such as Citrobacter rodentium and E. coli O157:H7 can also induce Th17 differentiation following adhesion to intestinal epithelial cells (Atarashi et al., 2015). Importantly, a common pilus adherence factor, which mediates adhesion to epithelial cells, has been reported for both pathogenic and commensal E. coli species (Rendón et al., 2007), indicating that the observed effects on the Th17 cells may be mediated through adherence of this commensal E. coli.

Furthermore, both E. coli and L. johnsonii could restore the antibiotics-induced reduction of IL-10 producing CD4+ cells in the small intestines, but only the latter could achieve the same effect in the colon, indicating commensal species-specific anti-inflammatory properties. Our results are well in line with a comprehensive analysis of numerous Lactobacillus species suggesting that species derived from healthy mice can be largely classified as potentially antiinflammatory (Gorska et al., 2016). Notably, these data support the already proposed concept and evidence, that probiotic bacteria including L. johnsonii exert health-promoting effects through induction of regulatory/anti-inflammatory host immune responses (Di Giacinto et al., 2005; Kwon et al., 2010). However, it is highly likely that L. johnsonii and other probiotic species exert additional beneficial functions through

IL-10 independent mechanisms, given their capability to attenuate the severity of colonic inflammation in IL-10 deficient mice.

The reintroduction of complex microbiota into the host via FMT is a well-known therapy dating back to the Chinese Dong-jin dynasty in the fourth century (Zhang et al., 2012) and has undergone a renaissance recently as a therapeutic option for the treatment of recurrent and refractory Clostridium difficile toxin induced acute necrotizing pseudo-membranous enterocolitis (van Nood et al., 2009; Rohlke et al., 2010; Brandt et al., 2012; Fischer et al., 2016; Scaldaferri et al., 2016).

In summary, the restoration of immune cell populations following broad-spectrum antibiotic treatment by commensal bacterial mono-colonization appears to be less effective than by complex intestinal microbiota association upon FMT, but depends both on the respective immune cell subtypes and analyzed compartments and does hence not follow a clear cut pattern. Nevertheless, mono-colonization of secondary abiotic mice with defined Gram-negative or Gram-positive species has a deep impact on shaping the host immune system and drives both pro- as well as anti-inflammatory immune responses on the mucosal, peripheral and systemic level in parallel. However, E. coli appears to favor pro-inflammatory immune responses, while L. johnsonii has a stronger impact on anti-inflammatory immune cell subsets.

### CONCLUSION

Future studies should further unravel the fine-tuned interplay between the well-orchestrated concert within the complex intestinal microbiota intestinal on one side and the different members of the immune system on the other in health and disease subsequently providing potential novel treatment options for immunopathological intestinal and extra-intestinal morbidities.

## AUTHOR CONTRIBUTIONS

IE: Performed experiments, analyzed data, and wrote paper. EvK: Performed experiments. CN: Suggested critical parameters in design of experiments, supplied antibodies. PB: Suggested critical parameters in design of experiments, supplied antibodies. AS: Provided advice in design and performance of experiments. SB: Provided advice in design and performance of experiments, co-edited paper. MH: Designed and performed experiments, analyzed data, and co-wrote paper.

#### FUNDING

This work was supported by grants from the German Research Foundation (DFG) to SB (SFB633, TP A7), MH (SFB633, TP B6), AS (SFB633, TP A1), IE and EvK (SFB633; Immuco) and by the German Federal Ministries of Education and Research (BMBF) to SB and MH (PAC-Campy 01KI1725D).

### ACKNOWLEDGMENTS

The authors thank Michaela Wattrodt, Ursula Rüschendorf, Alexandra Bittroff-Leben, Ulrike Fiebiger, Ines Puschendorf, Gernot Reifenberger, and the staff of the animal research facility of the Charité – University Medicine Berlin for excellent technical assistance and animal breeding.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2017. 02430/full#supplementary-material

FIGURE S1 | Percentages of CD8+ cells in intestinal and systemic compartments of secondary abiotic and recolonized mice. Secondary abiotic mice were generated by broad-spectrum antibiotic treatment and perorally recolonized by gavage. Subsequently, lymphocytes from small intestinal and colonic lamina propria, MLN and spleen were isolated, and analyzed by flow cytometry as described in "Materials and Methods." The percentages of the CD8+ lymphocyte population within the (A) small intestine, (B) colon, (C) MLN and (D) spleen of naive conventional mice (N), secondary abiotic mice (ABx) and mice re-associated with either E. coli (Ec), L. johnsonii (Lj) or complex intestinal microbiota by FMT on d7 and d28 post-recolonization are depicted. Columns represent means +SD. Significance levels (p-values) determined with one-way ANOVA test followed by Tukey post-correction test for multiple comparisons are indicated. Significant differences as compared to secondary abiotic mice are indicated by asterisks ( <sup>∗</sup>p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data were pooled from three independent experiments.

FIGURE S2 | Absolute cell numbers of CD8+ cells in intestinal and systemic compartments of secondary abiotic and recolonized mice. Secondary abiotic mice were generated by broad-spectrum antibiotic treatment and perorally recolonized by gavage. Subsequently, lymphocytes from small intestinal and colonic lamina propria, MLN and spleen were isolated, and analyzed by flow cytometry as described in "Materials and Methods." The concentrations of CD8+ lymphocytes in the (A) small intestine, (B) colon, (C) MLN and (D) spleen of naive conventional mice (N), secondary abiotic mice (ABx) and mice re-associated with either E. coli (Ec), L. johnsonii (Lj) or complex intestinal microbiota by FMT on d7 and d28 post-recolonization are depicted. Columns represent means +SD. Significance levels (p-values) determined with one-way ANOVA test followed by Tukey post-correction test for multiple comparisons are indicated. Significant differences as compared to secondary abiotic mice are indicated by asterisks (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data were pooled from three independent experiments.

FIGURE S3 | CD8+ memory/effector T cells in intestinal and systemic compartments of secondary abiotic and recolonized mice. Secondary abiotic mice were generated by broad-spectrum antibiotic treatment and perorally recolonized by gavage. Subsequently, lymphocytes from small intestinal and colonic lamina propria, MLN and spleen were isolated, and analyzed by flow cytometry as described in "Materials and Methods." The proportions of CD8+ memory/effector cells (CD8+CD44hi, gated on CD8+ cells) in the (A) small intestine, (B) colon, (C) MLN and (D) spleen of naive conventional mice (N), secondary abiotic mice (ABx) and mice re-associated with either E. coli (Ec), L. johnsonii (Lj) or complex intestinal microbiota by FMT on d7 and d28 post-recolonization are depicted. Columns represent means +SD. Significance levels (p-values) determined with one-way ANOVA test followed by Tukey post-correction test for multiple comparisons are indicated. Significant differences as compared to secondary abiotic mice are indicated by asterisks (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data were pooled from three independent experiments.

FIGURE S4 | Activated DCs in intestinal and systemic compartments of secondary abiotic and recolonized mice. Secondary abiotic mice were generated by broad-spectrum antibiotic treatment and perorally recolonized by gavage.

Subsequently, lymphocytes from small intestinal and colonic lamina propria, MLN and spleen were isolated, and analyzed by flow cytometry as described in "Materials and Methods." The frequencies of activated DCs (CD86+, gated on CD4-CD8- live cells) in the (A) small intestine, (B) colon, (C) MLN and (D) spleen of naive conventional mice (N), secondary abiotic mice (ABx) and mice re-associated with either E. coli (Ec), L. johnsonii (Lj) or complex intestinal

#### REFERENCES


microbiota by FMT on d7 and d28 post-recolonization are depicted. Columns represent means +SD. Significance levels (p-values) determined with one-way ANOVA test followed by Tukey post-correction test for multiple comparisons are indicated. Significant differences as compared to secondary abiotic mice are indicated by asterisks (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data were pooled from three independent experiments.


mediated sensing of lipopolysaccharide from commensal Escherichia coli. Gut 56, 941–948. doi: 10.1136/gut.2006.104497


response to commensal enteric bacteria. Clin. Vaccine Immunol. 19, 1207–1217. doi: 10.1128/CVI.00282.12



syndrome in infants: a double-blind placebo-controlled trial. Allergy 60, 494–500. doi: 10.1111/j.1398-9995.2004.00514.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Ekmekciu, von Klitzing, Neumann, Bacher, Scheffold, Bereswill and Heimesaat. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Modulatory Influence of Segmented Filamentous Bacteria on Transcriptomic Response of Gnotobiotic Mice Exposed to TCDD

Robert D. Stedtfeld<sup>1</sup> \*, Benli Chai<sup>2</sup> , Robert B. Crawford3,4, Tiffany M. Stedtfeld<sup>1</sup> , Maggie R. Williams<sup>1</sup> , Shao Xiangwen<sup>1</sup> , Tomomi Kuwahara<sup>5</sup> , James R. Cole<sup>2</sup> , Norbert E. Kaminski3,4, James M. Tiedje<sup>2</sup> and Syed A. Hashsham1,2 \*

<sup>1</sup> Department of Civil and Environmental Engineering, East Lansing, MI, United States, <sup>2</sup> Center for Microbial Ecology, Michigan State University, East Lansing, MI, United States, <sup>3</sup> Institute for Integrative Toxicology, Michigan State University, East Lansing, MI, United States, <sup>4</sup> Department of Pharmacology and Toxicology, Michigan State University, East Lansing, MI, United States, <sup>5</sup> Department of Molecular Bacteriology, Institute of Health Biosciences, University of Tokushima Graduate School, Tokushima, Japan

Environmental toxicants such as 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), an aryl hydrocarbon receptor (AhR), are known to induce host toxicity and structural shifts in the gut microbiota. Key bacterial populations with similar or opposing functional responses to AhR ligand exposure may potentially help regulate expression of genes associated with immune dysfunction. To examine this question and the mechanisms for AhR ligand-induced bacterial shifts, C57BL/6 gnotobiotic mice were colonized with and without segmented filamentous bacteria (SFB) – an immune activator. Mice were also colonized with polysaccharide A producing Bacteroides fragilis – an immune suppressor to serve as a commensal background. Following colonization, mice were administered TCDD (30 µg/kg) every 4 days for 28 days by oral gavage. Quantified with the nCounter <sup>R</sup> mouse immunology panel, opposing responses in ileal gene expression (e.g., genes associated with T-cell differentiation via the class II major histocompatibility complex) as a result of TCDD dosing and SFB colonization were observed. Genes that responded to TCDD in the presence of SFB did not show a significant response in the absence of SFB, and vice versa. Regulatory T-cells examined in the mesenteric lymph-nodes, spleen, and blood were also less impacted by TCDD in mice colonized with SFB. TCDD-induced shifts in abundance of SFB and B. fragilis compared with previous studies in mice with a traditional gut microbiome. With regard to the mouse model colonized with individual populations, results indicate that TCDD-induced host response was significantly modulated by the presence of SFB in the gut microbiome, providing insight into therapeutic potential between AhR ligands and key commensals.

Keywords: TCDD, segmented filamentous bacteria, gnotobiotic mice, regulatory T-cells, gut dysbiosis, host microbe response, B. fragilis

#### Edited by:

Yves Renaudineau, University of Western Brittany, France

#### Reviewed by:

Dane Parker, Columbia University, United States Jessica Lynn Humann, Florida A&M University, United States

#### \*Correspondence:

Syed A. Hashsham hashsham@egr.msu.edu Robert D. Stedtfeld stedtfel@msu.edu

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 10 June 2017 Accepted: 23 August 2017 Published: 07 September 2017

#### Citation:

Stedtfeld RD, Chai B, Crawford RB, Stedtfeld TM, Williams MR, Xiangwen S, Kuwahara T, Cole JR, Kaminski NE, Tiedje JM and Hashsham SA (2017) Modulatory Influence of Segmented Filamentous Bacteria on Transcriptomic Response of Gnotobiotic Mice Exposed to TCDD. Front. Microbiol. 8:1708. doi: 10.3389/fmicb.2017.01708

## INTRODUCTION

fmicb-08-01708 September 5, 2017 Time: 16:57 # 2

Dysbiosis of certain key immune modulating commensals can influence host disposition to disease and environmental exposure (Snedeker and Hay, 2012). Specific individual bacterial groups can also revive or influence differentiation of T-cells observed in germ free or antibiotic-treated mice (Faith et al., 2014; Sefik et al., 2015; Honda and Littman, 2016). For example, multiple immune activating strains isolated from mice and human stool including segmented filamentous bacteria (SFB) induce Th<sup>17</sup> cells in animal studies (Gaboriau-Routhiau et al., 2009; Ivanov et al., 2009; Atarashi et al., 2015; Tan et al., 2016). Alternatively, bacteria such as Bacteroides fragilis promote the expression of Treg cells through polysaccharide A (PSA) production (Troy and Kasper, 2010). Dysregulation of Treg/Th<sup>17</sup> cells can lead to various disease outcomes (Hand and Belkaid, 2010; Ivanov and Littman, 2010). For example, Helicobacter pylori succeeds in colonizing the gut by causing changes in both Treg and Th<sup>17</sup> cells (Kao et al., 2010). Thus imbalanced levels of T-cells can influence protection against pathogens or autoimmune diseases (Fantini et al., 2007; Peck and Mellins, 2010).

Attention to environmental exposure such as dioxins and other persistent organic pollutants has increased due to possible contributions to autoimmune diseases (Hertz-Picciotto et al., 2008), among others such as developmental disorders (Lee et al., 2007), obesity (Ibrahim et al., 2011), and diabetes (Taylor et al., 2013). Mediated in part through the aryl hydrocarbon receptor (AhR), 2,3,7,8-tetrachlorodibenzop-dioxin (TCDD, a porotype for studying the bioactivity for AhR) and other dioxins promote multiple toxic health effects including immune suppression (Kerkvliet et al., 2002, 2009). Thus, AhR ligands are also critically linked to the balance of Treg/Th<sup>17</sup> cells (Ho and Steinman, 2008; Mustafa et al., 2008; Quintana et al., 2008; Veldhoen et al., 2008; Chmill et al., 2010; Marshall and Kerkvliet, 2010; Veldhoen and Duarte, 2010; Zhang et al., 2010). For example, AhR ligands have been shown to abrogate inflammation caused by Crohn's disease (Benson and Shepherd, 2011), which may in part be initially caused by commensal dysbiosis. While not completely understood, the pleiotropy type relationship between AhR modulating environmental toxicants, host, and gut commensals may potentially be employed to therapeutically modulate health.

Emerging studies also suggest that environmental toxicants such as TCDD (Lefever et al., 2016; Stedtfeld et al., 2017b) and other AhR ligands interact with gut commensals (Hubbard et al., 2015; Zhang et al., 2015; Murray et al., 2016). Previous murine studies with a traditional gut microbiome found that TCDD induced structural shifts in key bacterial populations; with increasing abundance of SFB (Bhaduri, 2015; Stedtfeld et al., 2017a) and decreasing abundance of Bacteroidetes in response to TCDD (Lefever et al., 2016). Known opposing T-cell host responses to SFB and exposure to TCDD suggest that expansion of SFB could potentially abrogate or lessen TCDD-induced toxicity and differentiation of regulatory T-cells (Marshall et al., 2008; Ivanov et al., 2009). It was also unknown if SFB response was due to structural shifts in other bacterial populations (e.g., decreased abundance in Bacteroidetes), or due to TCDD-induced shifts in the host.

To answer these questions and examine the modulatory potential of an immune activating bacteria in animals exposed to TCDD, gnotobiotic mice were colonized with and without SFB. Mice were also colonized with immune suppressing (Round and Mazmanian, 2010; Troy and Kasper, 2010; Atarashi et al., 2011) PSA producing B. fragilis to serve as a commensal background. A separate group of mice was also mono-colonized with SFB or non-colonized to further verify the modulatory potential.

## MATERIALS AND METHODS

## Animal Models and Bacterial Cocktails

Germ-free female C57BL/6 mice were bred and maintained at the Germ-Free Mouse Facility housed in the Unit for Laboratory Animal Medicine at the University of Michigan (Ann Arbor, MI, United States) and maintained in germ-free isolators. Mice were orally colonized with bacteria 4–6 weeks after birth (Supplementary Figure S1). TCDD dosing started 4 weeks after colonization. A previously described TCDD dosing regimen of 30 µg/kg (AccuStandard, New Haven, CT, United States) by oral gavage once every 4 days for 28 days (Fader et al., 2015; Nault et al., 2015) was used. Mice were dosed by oral gavage with 0.1 ml of sesame oil vehicle control (Sigma–Aldrich, St. Louis, MO, United States) or TCDD in sesame oil vehicle.

Results shown in this study are based on the following two experiments and animal numbers: Experiment 1 consisted of untreated (vehicle) with B. fragilis mono-colonization (n = 4), TCDD treated with B. fragilis mono-colonization (n = 4), an untreated (vehicle) with co-colonization of SFB and B. fragilis groups (n = 4), and TCDD-treated with co-colonization of both groups (n = 4).

To further verify modulation potential of SFB, experiments were replicated in the absence of B. fragilis including untreated (vehicle) uncolonized (UC; n = 4), TCDD-treated UC (n = 4), untreated (vehicle) with SFB mono-colonization (n = 4), and TCDD-treated with SFB mono-colonization (n = 4). One untreated mouse mono-colonized with SFB and one co-colonized treated mouse died prior to sacrifice. Mice had access to sterile water and food ad libitum. All animals received humane care in compliance with the animal use protocol approved by the University of Michigan, U of M Animal Welfare Assurance (A3114-01). Handling of blood and tissue exposed to TCDD was carried out as per MSU's Environmental Health and Safety approved protocols under AUF numbers 02/14-030-00.

Bacteroides fragilis (DSM 2151) used for colonization was grown in Brucella broth (AS-105, Anaerobe Systems, Morgan Hill, CA, United States). Candidatus Savagella SFB-mouse-Japan, isolated as described previously (Kuwahara et al., 2011), was used for SFB groups. SFB was provided through an MTA between MSU and Kagawa University, Japan (No. AGR2015- 00006 Kagawa University) and used as per approved protocols for handling BSL2 organisms. Prior to the association of bacteria into germ-free mice, qPCR was used to estimate abundance of bacteria and confirmed by Sanger sequencing of the 16S rRNA gene to

ensure correct bacterial species. qPCR reactions included 1 ng of DNA extracted from fecal pellets, 18 µl Master Mix, and 1 µl of 10 mM primer mix in a 20 µl reaction. PCR conditions were as previously described (Bouskra et al., 2008). Briefly, cycling consisted of an initial 95◦C for 5 min, 40 cycles of 95◦C for 55 s, 60◦C for 55 s, and 72◦C for 1.5 min. Amplicons from three biological replicates (used for oral gavage) were purified using the Qiagen PCR purification kit and sequenced using the 96-capillary electrophoretic ABI 3730xl platform. Sequenced samples showed 99% identity with published mouse SFB sequences in the NCBI database (Accession Numbers: CP008713.1, AP012209.1, and AP012202.1). B. fragilis colonized groups were also inoculated with additional commensals including Faecalibacterium prausnitzii DSM 17677, Ruminococcus bromii Strain VPI 6883, R. obeum DSM 25238, Butyrivibrio fibrisolvens DSM 3071, and Eubacterium rectale DSM 17629 to serve as a background; however, only B. fragilis and SFB colonized to detectable levels in matrices tested.

## Collection of Fecal Pellets and Isolation of Tissue

Fresh fecal pellets were collected and analyzed every 8 days to ensure colonization, no contamination of unwanted bacterial groups, and examine the influence of TCDD on bacterial abundance. Fecal pellets were placed in RNA/DNA stabilizer (Zymo Research Corp, Irvine, CA, United States) and stored in a −20◦C freezer. Whole blood, spleen, mesenteric lymph nodes, and intestinal tissues were collected at the time of sacrifice. Mice were weighed prior to sacrifice. Two intestinal sections were removed and used for subsequent analysis, including the cecum and the ileum. For the ileum, a 0.25 cm segment proximal to the cecum was removed and stored for all animals. For the cecum, sections of content and tissue were both removed and placed in the same vial. Intestinal tissue samples were immediately placed on RNA stabilizer and stored at −80◦C.

### Transcriptomic Response Analysis with qPCR and nCounter <sup>R</sup>

RNA was extracted from ileum and cecum of all mice to evaluate the transcriptome response of SFB and B. fragilis in response to TCDD and the response of host immune cells. In detail, the PureLink RNA Mini Kit with Trizol (12183018A, Ambion/Thermo Fisher Scientific, Waltham, MA, United States) was used to extract RNA from mouse tissue and content. Two additional steps were used to digest DNA including the DNase 1 (Invitrogen/Thermo Fisher Scientific, Waltham, MA, United States) and Turbo DNase I Kit (Life Technologies/Thermo Fisher Scientific, Waltham, MA, United States). After each step, isolated RNA was quantified using a Qubit (Life Technologies), and assessed for purity using the Nanodrop ND-1000 UV– Vis spectrophotometer (Nanodrop Products, Wilmington, DE, United States).

2,3,7,8-Tetrachlorodibenzo-p-dioxin-induced expression of both bacterial members was monitored in the ileum and cecum using qPCR. For mRNA analysis of bacteria, qPCR primers were designed from sequences available for the colonized organisms. Genes targeting SFB functions were selected based on potential interaction with host immunity (Pamp et al., 2012) including the putative hemolysin A producing gene (BAK56433.1), and the putative rubrerythrin producing gene (BAK56170.1). Primers targeting B. fragilis gene clusters wcfQ (CAH07088.1) responsible for biosynthesis of PSA (Troutman et al., 2014) were selected due to known influence in signaling naïve T-cells differentiation into Treg cells (Troy and Kasper, 2010). Primers were designed as previously described (Stedtfeld et al., 2008). Briefly, qPCR primers were designed from sequences downloaded from NCBI using Primer Express (Applied Biosystems) default settings; with a maximum length of 150 bases, and a theoretical melting temperature of 59◦C. NCBI BLAST analysis (Altschul et al., 1997) was used to check theoretical specificity of designed primers against the GenBank database (Supplementary Table S1). cDNA was synthesized via random primers as instructed in the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, Waltham, MA, United States). qPCR was performed using a custom SmartChipTM (Wafergen Biosystems, Fremont, CA, United States) with species-specific 16S rRNA gene and functional gene primers (Supplementary Table S1). Briefly, Wafergen's Custom SmartChipTM array allows for 5,184 qPCR reactions with 100 nl volumes to be run in parallel. Sample/primers were dispensed into the SmartChipTM using a Multi-sample Nano-dispenser (Wafergen Biosystems, Fremont, CA, United States). PCR cycling conditions and initial data processing were performed as previously described (Wang et al., 2014; Stedtfeld et al., 2016). Amplification reactions on the SmartChipTM consisted of 1× LightCycler 480 SYBR <sup>R</sup> Green I Master Mix (Roche Inc., United States), nuclease-free PCR-grade water, 50 ng/µl cDNA template per sample, and 0.5 µM of each forward and reverse primer. Thermal cycling included an initial denaturation at 95◦C for 3 min, followed by 40 cycles of denaturation at 95◦C for 30 s, and annealing at 60◦C for 1 min. Each primer/sample combination was tested in triplicate. Genetic copies were estimated as previously described (Looft et al., 2012). Negative controls with no templates were also included on the array.

2,3,7,8-Tetrachlorodibenzo-p-dioxin-induced ileal expression of host maker genes related to immune function was quantified using the nCounter <sup>R</sup> mouse immunology panel. In detail, 200 ng of digested RNA was submitted per sample using the nCounter <sup>R</sup> (NanoString Technologies, Seattle, WA, United States). NanoString's nCounter <sup>R</sup> mouse immunology panel contains probes targeting 547 immunology-related mouse genes (Reis et al., 2011).

## qPCR of gDNA from Fecal Pellets

Genomic DNA was also extracted from fecal pellets 18–20 days after colonization and 21 days after initial treatment (following the sixth dose of TCDD) using the PowerSoil Extraction Kit (MoBio). qPCR was performed using species-specific 16S rRNA gene primers (Supplementary Table S1). Assays were performed in parallel using the Wafergen SmartChipTM with the same reaction conditions described above for cDNA; however, only 0.1 ng/µl of gDNA was used per sample. Results showed

colonization of SFB and B. fragilis in the groups expected to contain these populations.

## Flow Cytometry of Whole Blood, Splenocytes, and Mesenteric Lymph Nodes

Regulatory T-cell responses were measured in mesenteric lymph nodes, blood, and spleen. Trunk blood was collected into heparin-treated tubes followed by red blood cell lysis using a standard ACK lysing protocol (BD Biosciences, San Jose, CA, United States). Whole blood leukocytes were washed and counted, 1–2 × 10<sup>6</sup> leukocytes were stained with antibodies for analysis by flow cytometry. Spleens and mesenteric lymph nodes were isolated and made into single-cell suspensions by removing the capsule and connective tissue via mechanical disruption. Cells were then washed and 1–2 × 10<sup>6</sup> lymphocytes stained with antibodies for analysis by flow cytometry. Subsequently, Fc receptors on peripheral blood, lymph node, and spleen-derived leukocytes were blocked with an anti-mouse CD16/CD32 (BD Biosciences, San Jose, CA, United States), and the cells were stained for extracellular proteins using the following antibodies (Biolegend, San Diego, CA, United States) CD3, CD4, CD25, and NK1.1. Samples were incubated for 20 min at 4◦C, and washed twice with FACS buffer (1× HBSS containing 1% BSA and 0.1% sodium azide). Next, cells were fixed with Cytofix (BD Biosciences, San Jose, CA, United States) and resuspended in FACS buffer. For intercellular protein staining, cells that were previously fixed after surface staining were permeabilized with FoxP3 Perm Buffer (Biolegend, San Diego, CA, United States) by incubating the cells in the perm buffer for 10 min, followed by the addition of the antibodies (FoxP3 and IL-17F (Biolegend, San Diego, CA, United States) directly to the cells and incubated for an additional 30 min at room temperature. Cells were washed four times in the Perm Buffer and after the last wash cells were re-suspended in FACs buffer. Cells were analyzed using a FACSCanto II Flow Cytometer (BD Biosciences, San Jose, CA, United States) and the data analyzed with Flowjo software (ver. 8.8.7 Treestar Software, Ashland, OR, United States), with a gating strategy as shown in Supplementary Figure S2. The percentage of Th<sup>17</sup> cells was near the detection limit in measured matrices, as the background ranged from 0.001 to 0.08% in whole blood samples for IL17 staining.

#### Statistical Analysis and Host/Bacterial Functional Gene Annotation

qPCR was analyzed for absolute abundance based on mass of tissue used to extract DNA and extraction yield, or normalized for relative expression of 16S rRNA gene specific to SFB or B. fragilis.

For data obtained using the nCounter <sup>R</sup> , expression counts were normalized to the geometric mean of multiple housekeeping genes including Rpl19, Ppia, G6pdx, Tubb5, Alas1, Tbp Gusb, Hprt, Gapdh (all had %CV <50) with an algorithm developed by NanoString Technologies. Genes with a maximum normalized expression count of 10 or less among all colonized groups were removed. The Shapiro–Wilk test was used to evaluate normality of data. If necessary, the variance between compared groups was corrected with the Geisser–Greenhouse method. For comparison of more than two groups, one-way ANOVA followed by a multiple comparison Sidak test was performed. When normal distribution was satisfied, a Student's t-test was used for comparing differences between two groups. Otherwise, the nonparametric Mann–Whitney test was used. Statistical analysis and some plots were generated using Prism (version 7 for Windows; GraphPad Software, San Diego, CA, United States), additional plots were rendered using Excel or Cytoscape v. 3.3.0 (Institute for Systems Biology, Seattle, WA, United States). Genes were deemed significantly different if p < 0.05.

Genes that responded to TCDD and colonization were analyzed for enriched functions using the database for annotation, visualization, and integrated discovery (DAVID) v6.8 (Huang et al., 2009a,b). Gene ontology-biological processes (GO-BPs) were used for clustering the enrichment analysis. Functional groups with an enrichment score (ES)≥1.3 were considered significantly enriched, representing the −log scale geometric mean p < 0.05.

## RESULTS

### Transcriptomic Response of Ileal Immune Cells

The influence of SFB on TCDD-induced transcriptomic response was measured using the nCounter <sup>R</sup> mouse immunology panel (**Figure 1** and Supplementary Table S2). Comparisons were made between groups of mice (Supplementary Figure S1) that remained uncolonized (UC), mice mono-colonized with B. fragilis (B), mice mono-colonized with SFB (SFB), and mice co-colonized with both bacteria (SFB+B) to examine: (i) influence of TCDD in presence of SFB (SFB+B, **Figure 1A**) and (ii) influence of TCDD in absence of SFB (UC and B. fragilis groups, **Figure 1B**). The influence of SFB colonization was determined by comparing vehicle dosed SFB+B vs B groups; and the influence of B. fragilis was determined by comparing vehicle dosed UC vs B groups. Results showed that 11 genes were significantly influenced by TCDD in the presence of SFB (**Figure 1A**). Nine out of 11 of these genes had an opposing response between TCDD and SFB colonization, in terms of transcript regulation. The two genes with similar responses between SFB colonization and TCDD (cfd, Il2rb) had a conflicting response between TCDD and colonization of B. fragilis. The influence of SFB on TCDDinduced host response was further verified in comparing UC with SFB mono-colonized groups (Supplementary Figure S3).

Nine genes were influenced by TCDD in the absence of SFB (**Figure 1B**), all of which were no longer significantly up/down regulated in the presence of SFB. All nine of the genes influenced by TCDD in the absence of SFB were up/down regulated in the same direction as SFB colonization. Influence of SFB colonization in **Figure 1B** was measured by comparing SFB vs UC groups. TCDD-induced responses in groups with and without SFB exhibited opposing modulation of genes associated with immune cell function.

A majority of genes that responded to TCDD in presence/absence of SFB are related to T-cell differentiation

including: Il1β, Ciita, H2-Eb1, and H2-Aa which were significantly upregulated in response to SFB, and were downregulated in response to TCDD. These genes encode the class II, major histocompatibility complex (MHC II), transactivator, which is required for SFB induction of Th<sup>17</sup> cells (Lecuyer et al., 2012; Goto et al., 2014). The Il1β gene, which has also been shown to substitute TGFβ in differentiation of T-cells (Ghoreschi et al., 2010), was also upregulated with SFB and downregulated with TCDD (Supplementary Figure S3). In the group without SFB, TCDD induced downregulation of genes related to T-cells (Traf1) and activation of NF-κB (Tlr3) and genes related to regulation of IgA and IgM (Fcamr, Pdcd1) homeostasis (Kawamoto et al., 2012; Ouchida et al., 2012).

Colonization with SFB had a greater overall influence on immune gene expression compared to TCDD (**Figures 1A,B**, **2A**). In total, 93 genes were influenced by SFB colonization; and 83 of these were up-regulated. To identify the function of genes that responded to colonization and TCDD, the DAVID v6.8 (Huang et al., 2009a,b) was used. Seven and zero significant clusters were observed in response to SFB and B. fragilis colonization, respectively (**Figure 2B**). Three enriched clusters functionally associated with T-cell differentiation and inflammatory response (**Figure 2C**) were identified from the genes that responded to TCDD in one or more groups.

## Response of Regulatory T Cells

2,3,7,8-Tetrachlorodibenzo-p-dioxin-induced responses to Treg cells in the mesenteric lymph node, blood, and spleen were less impacted in the presence of SFB, which is known to promote differentiation of cytokines toward Th<sup>17</sup> cells (**Figure 3**). TCDD-induced differentiation of Treg was greater in the group mono-colonized with PSA producing B. fragilis. The percent of CD3<sup>+</sup> and CD4<sup>+</sup> in the spleen and mesenteric lymph nodes of groups with SFB was also less influenced by TCDD (**Table 1**). Colonization and TCDD also tended to influence Th<sup>17</sup> cells in an expected manner; however, Th<sup>17</sup> measurements were near the limit of detection (Supplementary Figure S4).

## Response of Colonized Bacteria

qPCR analysis showed that TCDD significantly influenced the abundance of SFB and B. fragilis in the absence of other gut commensals (**Figure 4**). In detail, the absolute abundance of SFB was significantly higher in response to TCDD (**Figure 4B**). Increased SFB abundance was also verified with gDNA extracted from fecal pellets, and using assays targeting the SFB fliC functional gene (Supplementary Table S3). Overall, a 2.8-fold higher level of SFB (p = 0.038) was observed in response to TCDD. In contrast, the absolute abundance of B. fragilis, measured using species-specific rplB gene primers, significantly (p = 0.029) decreased 2.1-fold in mice dosed with TCDD (**Figure 4A**). The cecal and ileal abundance of B. fragilis and SFB did not differ among co-colonized or monocolonization groups (Supplementary Table S3). Thus, bacterial abundances in both groups of mice were plotted and analyzed concurrently.

FIGURE 2 | nCounter <sup>R</sup> mouse immunology panel analysis and functional response. (A) Genes in ileal tissue that responded to SFB colonization. Line color and direction indicate upregulated (red) and downregulated (blue) genes. Abbreviations include mice that were mono-colonized with SFB (SFB), mono-colonized with B. fragilis (B) and co-colonized (SFB+B). (B,C) Functional clusters identified using the DAVID using nCounter <sup>R</sup> mouse immunology panel analysis with gene expression of ileal tissue showing (B) clusters of genes that responded to colonization with SFB, and (C) clusters of genes that responded to TCDD in presence and absence of SFB. Functional clusters with scores =1.3 were included as significantly enriched.

percent and error bars represent standard error in presence/absence of SFB. P-values are shown between vehicle- (Veh) and TCDD-treated groups.

## Response of Select Functional Genes from Colonized Bacteria

The expression of select functional SFB and B. fragilis genes was also quantified via qPCR to examine responses putative to host interaction. In the ileum of the SFB colonized group, expression of the hemolysin A producing genes was downregulated in response to TCDD (**Figure 4D** and Supplementary Table S3). Hemolysin A has been shown to decrease the ability of host cells to phagocytize bacteria and undergo chemotaxis in vitro (Cavalieri and Snyder, 1982). Host macrophage recruiting genes were downregulated in the jejunum of mice dosed with TCDD in previously described studies (Fader et al., 2015). Down-regulation of this gene was only observed in the SFB mono-colonized group, as the expression was below the limit of detection in the group that also had B. fragilis.

Expression of the wcf gene involved in PSA production by B. fragilis was also measured. In detail, the wcfQ gene was significantly upregulated in the cecum of the B. fragilis colonized mice (**Figure 4C**) in response to TCDD. B. fragilis produce PSA as a control mechanism to induce regulatory T-cells and create a more self-tolerable growth environment (Round et al., 2011).

## DISCUSSION

Comparative analysis between gnotobiotic mice colonized with and without SFB was performed to examine modulatory responses to TCDD. Noting the reductionist approach of the mouse model colonized with individual populations, the presence of SFB had an opposing response to TCDD in terms of host ileal immune gene expression, and helped modulate levels of TCDDinduced regulatory T-cells in blood, spleen, and mesenteric lymph nodes.

#### Opposing Influence of SFB Colonization and TCDD on Host

Transcriptomic response fell into two categories including: (i) only responding to TCDD in opposition to SFB colonization and (ii) only responding to TCDD in the group without SFB. In the first category, genes influenced by TCDD would not have differed if had they not been expressed via colonization of SFB. This is evident in that none of the MHC II genes were differentially expressed in response to TCDD in groups without SFB. Within the second category, SFB reduced impact of TCDD on host immune gene expression. Verification of SFB influence in groups without B. fragilis further verifies opposing responses with TCDD (Supplementary Figure S3).

A majority of ileal immune genes were down-regulated with TCDD, with gene expression corresponding with differentiation of regulatory T-cells. In the presence of SFB, only two genes were significantly upregulated in response to TCDD (Pparg and cfd) in the group with background B. fragilis. The peroxisome proliferator-activated receptor-γ (Pparg) is a nuclear acceptor that inhibits NFκB activity. The downregulation of this gene has been associated with ulcerative colitis (Dubuquoy et al., 2003), and its increased expression correlates with anti-inflammatory responses, which is akin to what has been observed with TCDD (Benson and Shepherd, 2011). The genes that responded to TCDD in the absence of SFB (e.g., Il12rb2, Cd36) also regulate the number of regulatory T-cells and suppression of an inflammatory response (Zhao et al., 2008; Cecil et al., 2009).

#### Influence of T-Cell Response in Health

2,3,7,8-Tetrachlorodibenzo-p-dioxin-induced response of regulatory T-cells in the mesenteric lymph nodes was less impacted in groups colonized with SFB, extending to blood and spleen matrices beyond the gut (**Figure 3**). Differences in regulatory T-cell responses among the colonized groups further indicate the influence of gut commensals on host disposition to TCDD. The impact of TCDD on regulatory T-cells has been shown to influence susceptibility to infection in the gut (Thigpen et al., 1975). However, the modulatory influence of SFB or Th<sup>17</sup> inducers isolated from human stool (Tan et al., 2016) may help modulate susceptibility to bacterial infection and abrogate TCDD-induced response.

### TCDD-Induced Host Response Favors SFB

2,3,7,8-Tetrachlorodibenzo-p-dioxin-induced down-regulation of host immune genes that were activated with SFB colonization may provide a more favorable environment for SFB proliferation. This is evident in opposing functional inflammatory responses. Previous studies have found that mice deficient in mucosal

TABLE 1 | Body weight and T-cell number (CD3<sup>+</sup> and CD4+) in spleen, blood, and mesenteric lymph nodes (lymph) after TCDD (30 µg/kg) or vehicle (sesame oil) treatment of mice by oral gavage once every 4 days for 28 days.


Numbers indicate mean and standard error. Abbreviations identify mice that remained uncolonized (UC) or were mono-colonized with B. fragilis (B) and co-colonized (SFB+B). Vehicle dosed mice (Veh) and TCDD dosed mice (TCDD). <sup>∗</sup>Significant difference between TCDD and vehicle-dosed groups (p < 0.05).

immunity maintainers displayed a 10-fold increase in SFB abundance in fecal pellets (Upadhyay et al., 2012). Studies have also shown that mice lacking an enzyme critical to the production of IgA had an overgrowth of SFB despite the presence of other commensal organisms (Suzuki et al., 2004). TCDD also tended to reduce expression of the flagella receptor Toll-like receptor 5 gene TLR5 gene 2.0-fold (p = 0.06) in the ileum of UC mice. Increased levels of flagella producing bacterial members have previously been observed in mice lacking the TLR5 gene (Cullender et al., 2013). As such, TCDD appears to repress host immune cell gene expression in the gut, permitting proliferation of SFB. Downregulation of mRNA from measured SFB functional genes putative to evasion from host defense has similar implications.

A similar shift in SFB abundance was also observed in mice with a traditional gut microbiome; however, it was unknown if this was due to the expansion of other bacterial populations or changes in nutrients caused by commensal fluctuations. In detail, previous studies described an increased ratio of Firmicutes to Bacteroidetes in response to TCDD (Lefever et al., 2016; Stedtfeld et al., 2017a), and higher levels of Proteobacteria (Stedtfeld et al., 2017b). The expansion in these predominantly flagella producing groups (Lozupone et al., 2012) may in part influence the relative levels of Bacteroidetes and SFB. However, similar shifts in both bacteria in the gnotobiotic mice indicate this response can occur solely via TCDD-induced modulation of the host.

### TCDD-Induced Host Response Influences B. fragilis

The abundance of B. fragilis decreased in response to TCDD in the gnotobiotic mice. While the reduced abundance of Bacteroidetes was previously observed in TCDD-dosed mice with a traditional gut microbiome (Bhaduri, 2015; Lefever et al., 2016), the causation for this dysregulation is unclear. Previous studies performed by Round et al. (2011) suggest that B. fragilis favors an immunosuppressed environment, and that expression of B. fragilis PSA is a defense or colonization mechanism to increase differentiations of Treg cells. In accordance, it would seem that PSA-producing genes should be downregulated to balance the levels of Treg cells induced by TCDD. However, the upregulation of mRNA genes clusters related to the production of PSA indicates that B. fragilis are in a state of distress in response to TCDD. In addition, TCDD did not significantly influence the abundance of Bacteroidetes or other key commensals when tested in vitro in this (Supplementary Figure S5) or previously

FIGURE 4 | Bacteroides fragilis and SFB abundance and selected functional gene expression based on isolated RNA. (A) Absolute abundance of B. fragilis expression measured with specific rplB gene primers in cecum content from respective groups. (B) Absolute abundance of SFB expression using species-specific 16S rRNA gene primers from ileum of respective groups. (C) PSA biosynthesizing wcfQ gene normalized expression, measured with cecum content from respective groups. (D) SFB functional genes putative to interaction with host immunity (hemolysin A producing gene) normalized expression, measured with ileal RNA from SFB mono-colonized mice. SFB functional genes in co-colonized group are not included as the genes were expressed below detection limit in two or more biological replicates. Bars represent mean normalized abundance. Open dots indicate mono-colonized groups, closed dots indicate co-colonized groups. Abbreviations include vehicle dosed (Veh) and TCDD treated (TCDD).

described studies (Lefever et al., 2016). Collectively, shifts in B. fragilis function and abundance also appear to be the result of TCDD-elicited changes in host.

## Relevance of TCDD Dose

fmicb-08-01708 September 5, 2017 Time: 16:57 # 9

Considering a half-life of 7–11 days for TCDD (Gasiewicz et al., 1983; Birnbaum, 1986), mice in this study had an estimated concentration of 60.5–90.7 µg/kg TCDD at sacrifice. One previously measured human exposure event resulted in comparable levels; with lipid-adjusted blood levels of 56 µg/kg TCDD in children near a 1976 plant accident in Seveso, Italy (Bertazzi et al., 1993). In addition, the lowest observable effect level of TCDD on inhibiting an immune response (IgM secretion) was lower in human (0.3 nM) than mice (3 nM) in lymphocytes tissue tested in vitro (Wood et al., 1993). We have also observed significantly higher levels of SFB at doses ≥1.0 µg/kg TCDD in other studies (Bhaduri, 2015). Thus, similar responses to those described here may occur with 10- to 100-fold less TCDD, warranting additional experiments. While environmental levels of TCDD are typically lower than that used in this study and are thought to be decreasing in the environment, cumulative exposure to other AhR ligands such as polyaromatic hydrocarbons and flame retardants is increasing (Lim et al., 2008; Wiseman et al., 2011).

## TCDD versus Other AhR Ligands

The AhR ligand response demonstrated in this study differs from those published by Zhang et al. (2015). In detail, their study with wild-type mice orally dosed with 2,3,7,8 tetrachlorodibenzofuran (TCDF), a less potent AhR activator, heightened host inflammation, reduced levels of SFB, and increased abundance of Bacteroidetes over Firmicutes. However, our results showed reduced levels of host immune gene expression, which has also been previously described in the jejunum in response to TCDD (Fader et al., 2015) and AhR activation (Monteleone et al., 2013). Dissimilar responses may be due to differences in AhR ligand potency, which has been reported in a study comparing TCDD and 6-formylindolo[3,2 b]carbazole (Farmahin et al., 2016). Future studies addressing AhR activity and influence to host and gut microbes in response to different ligands throughout the gut are necessary (Zhang et al., 2017).

#### Differences in Murine and Human Models

Before inferring if TCDD and presence/absence of SFB would have a similar influence on humans compared to the murine model, both the host and microbial phenotype must be considered. Differences in host phenotype have been described previously (Kreisman and Cobb, 2011), with varied gene expression in response to TCDD depending on the measured tissue and host species (Kovalova et al., 2017). In regards to microbiota, recent meta-analysis has shown that 90 and 89% of bacterial phyla and genera, respectively, are shared between mice and the human gut microbiota (Krych et al., 2013), with the most abundant bacterial species being common in both. Studies examining the presence of SFB in human microbiome have conflicting results (Klaasen et al., 1993; Qin et al., 2010; Caselli et al., 2013; Yin et al., 2013; Shukla et al., 2015), which may be due to variability in the methods or matrices used for measuring SFB. However, immune-activating bacterial groups isolated from human gut microbiome appear to have a similar response to SFB in the capacity to differentiate T-cells in mice (Tan et al., 2016). Commercially available probiotic consortia have also been shown to contain Th<sup>17</sup> cell inducers in mice (Tan et al., 2016), highlighting the therapeutic potential of immuneactivating bacteria to modulate host response to TCDD.

Collectively, the presence of SFB significantly influenced host response to TCDD in the matrices measured. TCDD exposure also appeared to provide a more favorable environment for SFB. B. fragilis also responded to TCDD-induced shifts in the host, expressing higher levels of the wcfQ gene, a PSA-producing gene cluster. TCDD did not appear to influence abundance of these bacterial groups in vitro in TCDD-dosed batch reactors (Supplementary Figure S5), further suggesting that the bacterial response of SFB and B. fragilis is in part due to TCDD induced influences on the host.

Follow-up studies are underway to further elucidate causation of functional dysbiosis with B. fragilis. This will include the analysis of additional genes related to host interaction and immune tolerance. Further identification of key bacterial species and functional genes may facilitate attempts to manipulate the intestinal microbiome toward a protective and healthy state. Future studies will also examine the ability of other human bacterial commensals to influence TCDD-induced response in the absence of SFB. These studies may lead to new therapeutic approaches to treat intestinal pathogens and resulting autoimmune diseases.

## AUTHOR CONTRIBUTIONS

RS, TS, and SH wrote the manuscript; RS, SH, JT, NK, RC, JC, and TK contributed to experimental design; and SH, BC, RC, TS, RS, SX, and MW aided in collection of tissue, measurements, and data analysis.

## FUNDING

This work was supported by the National Institute of Environmental Health Sciences Superfund Basic Research Program (NIEHS SBRP P42ES04911) with contributions from Project 1, 4, 5, and Core-B.

## ACKNOWLEDGMENT

We would like to thank Kathryn Eaton and her laboratory team at the University of Michigan Germ Free Facility.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmicb.2017. 01708/full#supplementary-material

#### REFERENCES

fmicb-08-01708 September 5, 2017 Time: 16:57 # 10


in the coordinated maturation of gut helper T cell responses. Immunity 31, 677–689. doi: 10.1016/j.immuni.2009.08.020




differentiation of regulatory T cells via activation of aryl hydrocarbon receptor. Invest. Ophthalmol. Vis. Sci. 51, 2109–2117. doi: 10.1167/iovs.09-3993


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Stedtfeld, Chai, Crawford, Stedtfeld, Williams, Xiangwen, Kuwahara, Cole, Kaminski, Tiedje and Hashsham. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Human Gut Symbiont *Roseburia hominis* Promotes and Regulates Innate Immunity

*Angela M. Patterson1†‡, Imke E. Mulder1‡, Anthony J. Travis1 , Annaig Lan1 , Nadine Cerf-Bensussan2,3, Valerie Gaboriau-Routhiau2,3,4, Karen Garden1 , Elizabeth Logan1 , Margaret I. Delday1 , Alistair G. P. Coutts1 , Edouard Monnais1 , Vanessa C. Ferraria1 , Ryo Inoue5 , George Grant1,6 and Rustam I. Aminov1,7\**

*1Rowett Institute of Nutrition and Health, University of Aberdeen, Aberdeen, United Kingdom, 2INSERM, UMR1163, Lab Intestinal Immunity, Paris, France, 3Université Paris Descartes-Sorbonne Paris Cité and Institut Imagine, Paris, France, 4Micalis Institute, INRA, AgroParisTech, Université Paris-Saclay, Jouy-en-Josas, France, 5Kyoto Prefectural University, Kyoto, Japan, 6School of Medicine, Medical Sciences and Nutrition, University of Aberdeen, Aberdeen, United Kingdom, 7 Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia*

#### *Edited by:*

*Laurel L. Lenz, University of Colorado Denver School of Medicine, United States*

#### *Reviewed by:*

*Erguang Li, Nanjing University, China Ricardo Silvestre, Instituto de Pesquisa em Ciências da Vida e da Saúde (ICVS), Portugal*

#### *\*Correspondence:*

*Rustam I. Aminov rustam.aminov@abdn.ac.uk*

#### *†Present address:*

*Angela M. Patterson, Quadram Institute Biosciences, Norwich Research Park, Norwich, United Kingdom and Norwich Medical School, University of East Anglia, Norwich, United Kingdom*

*‡ These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 17 May 2017 Accepted: 04 September 2017 Published: 26 September 2017*

#### *Citation:*

*Patterson AM, Mulder IE, Travis AJ, Lan A, Cerf-Bensussan N, Gaboriau-Routhiau V, Garden K, Logan E, Delday MI, Coutts AGP, Monnais E, Ferraria VC, Inoue R, Grant G and Aminov RI (2017) Human Gut Symbiont Roseburia hominis Promotes and Regulates Innate Immunity. Front. Immunol. 8:1166. doi: 10.3389/fimmu.2017.01166*

Objective: *Roseburia hominis* is a flagellated gut anaerobic bacterium belonging to the *Lachnospiraceae* family within the Firmicutes phylum. A significant decrease of *R. hominis* colonization in the gut of ulcerative colitis patients has recently been demonstrated. In this work, we have investigated the mechanisms of *R. hominis*–host cross talk using both murine and *in vitro* models.

Design: The complete genome sequence of *R. hominis* A2-183 was determined. C3H/ HeN germ-free mice were mono-colonized with *R. hominis*, and the host–microbe interaction was studied using histology, transcriptome analyses and FACS. Further investigations were performed *in vitro* and using the TLR5KO and DSS-colitis murine models.

Results: In the bacterium, *R. hominis*, host gut colonization upregulated genes involved in conjugation/mobilization, metabolism, motility, and chemotaxis. In the host cells, bacterial colonization upregulated genes related to antimicrobial peptides, gut barrier function, toll-like receptors (TLR) signaling, and T cell biology. CD4+CD25+FoxP3+ T cell numbers increased in the *lamina propria* of both mono-associated and conventional mice treated with *R. hominis.* Treatment with the *R. hominis* bacterium provided protection against DSSinduced colitis. The role of flagellin in host–bacterium interaction was also investigated.

Conclusion: Mono-association of mice with *R. hominis* bacteria results in specific bidirectional gene expression patterns. A set of genes thought to be important for host colonization are induced in *R. hominis*, while the host cells respond by strengthening gut barrier function and enhancing Treg population expansion, possibly *via* TLR5-flagellin signaling. Our data reveal the immunomodulatory properties of *R. hominis* that could be useful for the control and treatment of gut inflammation.

Keywords: *Roseburia*, T lymphocytes, immune tolerance, inflammatory bowel disease, flagellin, TLR5

#### INTRODUCTION

The human gut microbiota consists of more than 500–1,000 different phylotypes the majority of which belong to the Bacteroidetes and Firmicutes bacterial phyla (1). Successful symbiotic relationships arising from bacterial colonization of the human gut yield a wide variety of metabolic, structural, protective, and other beneficial functions. The immunological importance of the gut

**128**

microbiota is also well recognized; it is particularly apparent in germ-free (GF) animals that have an impaired immune system, which, however, can be functionally reconstituted by the introduction of gut commensal bacteria (2–4).

In sharp contrast to the production of secretory intestinal IgA, which is mainly driven by microbial colonization *per se* (5, 6), the development and differentiation of T cells require colonization by specific commensal bacteria. Some species of clostridia, such as segmented filamentous bacteria (SFB), appear to be potent inducers of differentiation and maturation of intestinal Th1, Th17, and Treg cell lineages (7, 8). Recent studies have demonstrated that the clostridia clusters IV and XIVa and the Altered Schaedler Flora can induce *de novo* generation of Treg cells, while mono-colonization with *Bacteroides fragilis* can correct the Th1/Th2 imbalance in germ-free mice by promoting the expansion of Treg cells (4, 9, 10). The effects of commensal bacteria on T cell differentiation pathways are variable and may be influenced by a specific array of toll-like receptors (TLR) ligands associated with particular bacteria (11). For instance, the Treg-enhancing effects of *B. fragilis* are implemented *via* TLR2 signaling by polysaccharide A (12).

Dramatic changes in microbiota composition affecting the balance of symbionts and pathobionts have been documented in gastrointestinal disorders such as inflammatory bowel disease. Crohn's disease (CD), for example, is characterized by a greater relative abundance of the Proteobacteria and a reduction of other bacteria such as the Bacteroidales and Clostridiales (13–17). A specific decrease of *Roseburia* spp*.* in patients with CD has also been noted (16, 18). Interestingly, this microbial dysbiosis is also associated with imbalances in T effector cell populations. The gut microbiota alterations in ulcerative colitis (UC) have not been characterized to the same extent. Only recently, a study of UC patients has demonstrated that *Roseburia hominis*, together with *Faecalibacterium prausnitzii*, is significantly decreased in this disease (16, 19). Both species display an inverse correlation with disease activity. Previous studies have shown anti-inflammatory effects of *F. prausnitzii* colonization on the host (20). Much less is known about *R. hominis*, which is a member of the clostridia cluster XIVa.

The complete DNA sequence and annotation of the *R. hominis* genome has been previously described (21). In this paper, bacterial and host transcriptome responses to *R. hominis* colonization were investigated*.* Bacterial responses included expression of genes involved in colonization of and adaptation to the environment of the murine gut, while host responses to colonization included genes of immunity and gut function.

## MATERIALS AND METHODS

#### Bacterial Cultures

*Roseburia hominis* A2-183T (=DSM 16839T = NCIMB 14029T) and other *Roseburia* species were maintained and grown on synthetic YCFA media as described before (22). *Escherichia coli* and *Salmonella enterica* strains were cultivated in LB medium. All *Roseburia* culture manipulations were performed in a MACS-MG-1000 anaerobic workstation (Don Whitley Scientific) under an atmosphere of 80% N2, 10% CO2 and 10% H2 at 37°C.

## Animals, Experimental Design, and Sampling

Germ-free animal experiments were performed in the gnotobiotic rodent breeding facility of INRA (ANAXEM platform, Institut Micalis, INRA, Jouy-en-Josas, France). The GF C3H/ HeN male mice were allocated into the control (*N* = 8) and treatment (*N* = 10) groups and caged individually. *R. hominis* A2-183T (=DSM 16839T = NCIMB 14029T) was grown anaerobically at 37°C in YCFA media. At days 0, 1, and 2, animals in the treatment group were given 109 colony-forming units (CFU) of *R*. *hominis* culture by gavage, while control animals were given 100 µL of YCFA medium. The ileal, ascending colonic, and cecal samples were collected at days 14 and 28. GF TLR5KO (C57BL/6 genetic background, *N* = 3) and C57BL/6 (*N* = 3) animals (Jouy-en-Josas), and conventional TLR5KO (C57BL/6 genetic background, *N* = 6) and Boy/J (C57BL/6 congenic, *N* = 6) animals (Medical Research Facility, University of Aberdeen) were gavaged with the live cultures of *R. hominis* to evaluate its functional importance. Boy/J (C57BL/6 B6 Cd45.1) mice were the background controls for conventional TLR5K0. They carry a CD45.1 pan leukocyte marker but are otherwise equivalent to C57BL/6 wild-type mice (23, 24). Twelve female C57BL/6 (6 weeks old) mice were used to evaluate the effect of *R. hominis* during DSS-induced mild colitis. Three mice were dosed daily by direct administration of a culture of *R. hominis* for a total of 8 days. Six mice were dosed with culture medium. Exposure to DSS was for 4 days (from day 3 to day 6). Three *R. hominis*-dosed and three control mice were offered sterile water containing DSS (20 g/L) over this period. On day 7, these mice were switched back to sterile water with no DSS. The mice were euthanased and dissected aseptically on day 9. Twenty-two female C57BL/6 mice (6 weeks old) were used to evaluate the effect of *R. hominis* during DSSinduced pathological colitis. After the acclimatization period of 7–10 days, the mice were dosed daily with 50 µL (109 CFU) of *R. hominis* in growth culture medium for 14 days. Control animals were given growth culture medium. From day 9, mice were given DSS (MW 50 kDa, 30 g/L) in their drinking water for 6 days. The animals were euthanized on day 15 and tissue sampling was performed as described earlier. The management and experimental procedures with animals were approved by the respective Local Ethical Review Committees.

#### Transcriptome Analyses

Bacterial RNA from the mouse cecum contents was isolated and labeled with dCTP-Cy3 or dCTP-Cy5 during cDNA synthesis. PCR products amplified from an *R. hominis* small fragment library (6,000 clones) were used to create duplicate microarrays on glass slides using a MicroGrid II TAS array-spotting robot (BioRobotics). Microarray hybridization was performed in a GeneTAC hybridization station (Genomic Solutions). Dye-swap and separate RNA purification protocols were used to avoid potential biases.

Murine RNA was extracted from the ileum and ascending colon tissues and hybridized to the GeneChip "NuGO Mouse Array" and GeneChip "Mouse Genome Array" (Affymetrix). Data analysis was performed using R1 and Bioconductor.2 The microarray data were submitted to NCBI GEO (Gene Expression Omnibus) with accession number GSE25544.

The *R. hominis*-specific primers 5′-CCCACTGACAGAGTA TGTAATGTAC-3′ and 5′-GCACCACCTGTCACCAC-3′ were used for qPCR analyses of fecal samples to validate the efficiency of colonization. The analyses were performed using a 7500 Fast Real-Time PCR System (Applied Biosystems) with a Power SYBR Green PCR Master Mix (Applied Biosystems). RT-qPCR analyses were performed on the ileal and colonic RNA samples for host gene expression studies using a QuantiFast SYBR Green PCR Kit (Qiagen) and QuantiTect Primer Assays (Qiagen).

#### Immunofluorescence and Histology

FISH analyses were performed with neutral buffered, formalinfixed gut tissue sections (2 µm thick) with 16S rRNA probes, specific for the domain of Bacteria (Cy3-labeled Eub338) and specific for the *R. hominis* strain A2-183 (FITC-labeled GTACATTACATACTCTGTCAGTG). Bacteria were visualized at x630 magnification. Immunolocalization of *R. hominis* flagellin was examined in methanol-fixed colon content smears using rabbit antisera anti-FlaA1 or anti-FlaA2 (Covalab) and Alexa donkey anti-rabbit 488 (Molecular Probes). T cell markers were examined on sequential intestinal tissue sections (8 µm) using Ly6G-FITC, CD3-FITC, isospecific IgG (BD Biosciences), and double-labeled FoxP3 Alexa Fluor 594 (Abcam) and CD3-FITC. Intestinal tissue sections (4 µm) were also stained with hematoxylin/eosin. A complete transverse cross-section of the colon from each animal was visualized at ×200 magnification. Each field of view was scored from 0 to 4 according to the method based on Ref. (25).

#### Cell Experiments

Caco-2 and HT29 cells were grown at 37°C in a 75% humidified atmosphere with 5% CO2 in 24-well plates. Before any treatment, cells were washed twice with Hanks' Balanced Salt Solution and kept in DMEM supplemented with l-glutamine, selenium, and transferrin for 24 h. Epithelial cells were harvested after incubation for 2 or 4 h and then stored in RNAlater for further analyses.

Bone marrow cells were harvested from the femur and tibia of C3H/HeN and C57BL/6 mice. Bone marrow-derived dendritic cells (BMDCs) were generated by Flt3L or GM-CSF culture, stimulated with 100 ng/mL flagellin, and analyzed by flow cytometry. Supernatants were collected and tested for IL-10 and IL-12 by cytometry bead arrays (BD Bioscience).

Cells from the intestine and mesenteric lymph nodes (MLNs) were isolated as described previously (26), with minor modifications. In brief, cell suspensions were incubated with 100 U/mL of collagenase VIII (Sigma-Aldrich) in RPMI supplemented with 20% of FBS at 37o C for 20 min (MLN) or 60 min (intestine). Single cell suspensions were analyzed by flow cytometry. For mixed leukocyte reactions, naïve CD4<sup>+</sup> T cells were purified from the major lymph nodes and spleens of OTII transgenic mice using MACs Miltenyi CD4<sup>+</sup> T cell Isolation Kit as per manufactures instruction. Purified CD4<sup>+</sup> T cells were co-incubated with flagellin activated Flt3L-derived BMDCs in the absence or presence of OVA323–339 peptide (1 µg/ml). After 5 days, T cell differentiation was determined by flow cytometry.

#### Flow Cytometry

Intestinal lamina propria cells and MLN cells were labeled with CD4-FITC, CD25-APC (eBioscience), CD8-APC-Cy7, CD3- PerCP (Biolegend), and B220-BV570 (BD Biosciences). Labeling of intracellular FoxP3 was performed after extracellular staining and fixation/permeabilization of cells. GM-CSF-derived dendritic cells were labeled with CD11b-PerCP Cy5.5 (BD Biosciences), CD11c-PE-Cy7, I-A/I-E-APC-Cy7, CD80-PE, CD86-APC, CD8- FITC, and B220-BV570 (Biolegend). Flt3L-derived dendritic cells were labeled with CD11c-PE-Cy7, CD11b- or Siglec-H-PerCP Cy5.5 (Biolegend), I-A/I-E-APC-Cy7, CD317-PE, CD40-Alexa Fluor 647, CD103-FITC, and B220-BV570. Coculture T cells were labeled with CD4-PerCP Cy5.5 (BD Biosciences), CD127- PE-Cy7, CD73-APC, CD195 (CCR5)-FITC, CD62L-BV570, and CD25-APC-Cy7 (Biolegend).

#### Statistical Analyses

Spot intensities on the microarrays were log-transformed and Loess normalization was applied to the microarray results. RT-qPCR data were transformed on a base-2 logarithmic scale and analyzed by one-way analysis of variance with a significance threshold of *P* < 0.05. Statistical significance between the treatment groups was evaluated using Student's *t*-test.

More detailed protocols are available in SI Materials and Methods in Supplementary Material.

#### RESULTS

#### *R. hominis* Responds to the Gut Environment by Upregulating Chemotaxis, Motility, and Mobilization Genes

Genome analysis (**Figure 1A**) identified about 5% of all genes were related to chemotaxis and motility function including four different flagellin genes of which one located within the flagellar operon is associated with flagellar motility.

A biomass of *R. hominis* was given by gavage to GF C3H/HeN mice on three consecutive days. *R. hominis*-colonized both the ileum and colon but was found in much higher numbers in the colon, up to 1 × 1010 bacteria/g feces (Figure S1 in Supplementary Material) and was found to be closely associated with the colonic mucosa.

Differential gene expression in the bacterium in response to association with the host was investigated using the *R. hominis* microarray. Bacterial RNA was isolated from three different experimental conditions to distinguish between the effects of the gut environment and animal dietary components, (i) *in vivo,* from the cecum of mono-associated mice; (ii) *in vitro*, from bacteria grown in the presence of dietary components; and (iii) *in vitro*, from bacteria grown in culture.

<sup>1</sup>http://www.r-project.org.

<sup>2</sup>http://www.bioconductor.org.

(*N* = 4) identified with the Affymetrix microarray. Bar graphs represent the number of genes up- or downregulated in mice after 14 and 28 days of colonization. (C) Heatmap generated from differentially expressed genes with functional significance between GF and *R. hominis*-colonized mice at 14 and 28 days.

Colonization of GF mice with *R. hominis* correlated with the increased host gene expression, which was highest in the colon, particularly at day 28 post-colonization (**Figure 1B**). The number of differentially expressed genes in the ileum was very low at day 28, consistent with the reduced bacterial numbers. Differential transcriptome responses in the two tissue sites and time points were illustrated by a clear separation of significant transcripts in the heatmap analysis (**Figure 1C**).

Differentially expressed genes were identified *(in vivo* vs*. in vitro*) with RT-qPCR validation performed on 42 differentially expressed genes (Tables S1 and S2 in Supplementary Material). The *mobA*- and *mobL*-like genes that are involved in conjugation/ mobilization transfer showed strong upregulation *in vivo* (**Figure 2A**). Other subsystems induced by the gut environment included membrane transport, in particular magnesium transport, and motility and chemotaxis including multiple methyl-accepting chemotaxis proteins and genes of the flagellar operon (**Figure 2B**). *R. hominis* possesses multiple flagellin genes: *flaA1*, *flaA2*, *flaA3*, and *flaB.* The expression of *flaA1*and *flaA2* in the murine gut environment was verified by Western–blot and immunocytochemistry (**Figure 2C**).

## *R. hominis* Affects Host Innate Signaling Pathways

Genes involved in innate immunity and gut barrier function of the host were significantly induced by *R. hominis* colonization [**Figure 1C**; NCBI GEO (Gene Expression Omnibus), accession number GSE25544]. The GO process "innate immune response" (GO:0045087) was upregulated and included the TLR-related genes *Tlr5*, *Tlr1*, and *Vnn1*. The upregulation of *Tlr5* was of particular interest, given the presence of flagellar genes in the genome of *R. hominis* and the expression of the corresponding proteins in the gut of mono-colonized mice. The flagellins may have a role in this innate signaling pathway to mediate innate and adaptive immune responses. Other innate immunity genes affected by *R. hominis* colonization included the antimicrobial peptides *Defb37*, *Pla2g3*, *Muc16*, and *Itln* and the gut barrier function genes *Sprr1a*, *Cldn4*, *Pmp22*, *Crb3*, and *Magi3.* Innate immunity genes that were upregulated in the ileum in response to *R. hominis* included *Defcr20*, *Pcgf2*, *Ltbp4, Igsf8*, and *Tcfe2a*. We also found a negative regulation of the NF-κB pathway (GO:0043124) by *R. hominis*, which may contribute to the immune homeostasis by downregulating this inflammatory cascade.

Figure 2 | Responses of *Roseburia hominis* to the gut environment. (A) RT-qPCR quantification of *R. hominis* transcripts involved in conjugation/mobilization transfer. *R. hominis* RNA samples: *in vivo*—isolated from mono-associated animals; *in vitro*—isolated from YCFA medium-grown cultures; *in vitro* + diet—isolated from cultures grown on YCFA medium with addition of murine chow. (B) RT-qPCR quantification of *R. hominis* transcripts involved in motility and chemotaxis. *R. hominis* RNA samples: *in vivo*—isolated from mono-associated animals; *in vitro*—isolated from YCFA medium-grown cultures; *in vitro* + diet—isolated from cultures grown on YCFA medium with addition of murine chow. (C) Western blot of *R. hominis* proteins grown *in vitro* in the presence of UV-irradiated standard murine chow. The membrane was immunostained with affinity-purified 1 antibody, Fla2 specific antiserum, and anti-DNA gyrase A antibody (lane 1: no diet, lane 2: 0.01 g diet/10 mL of *R. hominis* culture, lane 3: 0.02 g diet/10 mL, lane 4: 0.05 g diet/10 mL, lane 5: 0.1 g diet/10 mL, lane 6: 0.2 g diet/10 mL, lane 7: 0.5 g diet/10 mL, and lane 8: 1 g diet/10 mL). Electron microscopy of *R. hominis* showing flagella (black arrows). (D) Immunocytochemistry with FlaA1 and FlaA2 specific antisera performed on *R. hominis* from the luminal contents of mono-colonized mice and on *R. hominis* grown *in vitro*. Original magnification is ×1,000. RT-qPCR results are presented as fold change. Statistical significance: \**P* < 0.05, \*\**P* < 0.01, and \*\*\**P* < 0.001.

## *R. hominis* Directs T Cell Pathways

Immune response was a major pathway induced by *R. hominis* at day 28 in the ascending colon of mono-associated mice. The pathways significantly affected in this category were mostly involved in T cell function, including IL-10 signaling and regulation of T cell function by CTLA-4 (Table S3 in Supplementary Material). The Ly6 gene family was also induced in the ascending colon. In particular, the GPI-anchored gene product of *Ly6g6c* was upregulated 25-fold, and the related gene *Ly6g6e* was upregulated twofold at day 28. The majority of hematopoietic cells, including neutrophils and plasmacytoid dendritic cells, express one or more members of the Ly6 family. The increased presence of Ly6G<sup>+</sup> cells in *R. hominis*-colonized mice was confirmed by immunocytochemistry (**Figure 3A**).

While the pathways of T cell regulation were significantly influenced by association with *R. hominis*, the corresponding effects on T cell differentiation required further investigation. We therefore assessed the number of double-positive CD3+FoxP3+ cells in the colonic lamina propria of monoassociated C3H/HeN mice. A significant increase in the number of regulatory T cells (CD3<sup>+</sup>FoxP3<sup>+</sup> cells) was detected in the germfree C3H/HeN and C56BL/6 mice colonized with *R. hominis* (**Figures 3A,B**).

Conventional C3H/HeN and C57BL/6 mice were gavaged with *R. hominis* for 14 days to determine the impact of the bacterium on Treg cells in the *lamina propria* of animals with a normal microbiota. This treatment resulted in a small but significant increase in the population of CD3<sup>+</sup>CD4<sup>+</sup>CD25<sup>+</sup>FoxP3<sup>+</sup> T cells in the C3H/HeN (*P* = 0.04) and a smaller increase in the C57BL/6 mice (*P* = 0.05) when compared to non-colonized mice (**Figure 3C**). While the population of Tregs was induced in conventional mice after colonization with *R. hominis* under homeostatic conditions, it was unknown if this induction could be maintained under conditions of mild colitis. We therefore investigated the impact of the bacterium on Treg cells in the lamina propria of animals that were given low dose DSS. FoxP3-tagged mice of C57BL/6 background were colonized with *R. hominis* for 2 days before treatment with low dose of DSS for 4 days followed by 2 days of pure drinking water. DSS treatment alone significantly increased the population of CD3<sup>+</sup>CD4<sup>+</sup>CD25<sup>+</sup>FoxP3<sup>+</sup> T cells (*P* = 0.02) (**Figure 3D**). When the animals were additionally treated with *R. hominis*, there was a further significant increase in the population of CD3<sup>+</sup>CD4<sup>+</sup>CD25<sup>+</sup>FoxP3<sup>+</sup> T cells (*P* = 0.02) versus DSS (**Figures 3D,E**).

## The Role of *R. hominis* Flagellins in Modulation of T Cell Differentiation

The influence of bacteria on the differentiation of T cells may reflect a specific array of TLR ligands displayed by a particular bacterium. TLR5KO (GF and conventional) and WT murine strains were gavaged with the bacterium to evaluate the functional importance of *R. hominis* and its flagellins. Analysis of the differentially expressed genes in TLR5KO and WT Boy/J mice colonized with *R. hominis* revealed that although T cell pathways were still influenced by the colonization event in TLR5KO mice, responses were more related to IL-4, IL-5, IL-6, IL-9 pathways and not to IL-10 and CTLA-4 (Table S4 in Supplementary Material). Furthermore, in contrast to mono-associated C3H/HeN and C57BL/6 animals (**Figure 3B**) and conventional Boy/J animals (**Figures 4B,C**), the numbers of double-positive CD3<sup>+</sup>FoxP3<sup>+</sup> cells in the *lamina propria* of TLR5KO mice were not increased due to *R. hominis* treatment (**Figure 4A**).

A set of flagellins from pathogenic and commensal bacteria, including recombinant *R. hominis* flagellins, was used to compare their effect on activation of signaling responses in IECs and BMDCs. Caco-2 cells were treated with the identical concentrations of different bacterial flagellins (**Figure 5A**). The flagellin of pathogenic *S. enterica* serotype Enteritidis (SE) induced a larger gene panel than the flagellins of commensal *E. coli* K12 or *R. hominis* FlaA1. *E. coli* and *R. hominis* induced a strong response and, with a similar gene complement between the two, formed a separate clade distinct from SE. In contrast, FlaA2 was generally neither pro-inflammatory nor did it activate the conserved gene signature (*IL8*, *CXCL1*, *CXCL2*, and *CXCL10*) that was induced by the other recombinant bacterial flagellins. In fact, the set of genes affected was closer to that of the control cells and formed a cluster separated from the other flagellins. FlaA1 from *R. hominis* induced different responses in Flt3L- and GM-CSFderived BMDCs compared to SE and K12 (**Figures 5B,C**). In particular, FlaA1 was uniquely able to activate Flt3L-expanded DCs, with the upregulation of I-A/I-E and CD40 and production of IL-10 by BMDCs from both C3H/HeN and C57BL/6 mice. The IL-10/IL-12 ratio was particularly elevated in C57BL/6 DCs (**Figure 5D**), which were found to be CD103<sup>+</sup>Siglec-H<sup>+</sup>. To test whether flagellin activated BMDCs could differentially modulate naïve T cell differentiation, mixed leukocyte cultures of CD4<sup>+</sup> OTII cells with flagellin stimulated Flt3-derived BMDC were performed. The data show that the production of IL-10 and IL-12 by the flagellin stimulated Flt3-derived BMDC ratio triggered naïve T cell differentiation toward a Tregs profile (**Figure 5E**). In contrast to FlaA2, FlaA1 significantly increased the Treg population in the T cell cocultures.

## *R. hominis* Attenuates Colitis in DSS-Treated Mice

The effects of *R. hominis* on innate and adaptive immunity prompted us to test its therapeutic efficacy using the murine DSS model (Figure S2 in Supplementary Material; **Figure 6**). The treatment group was dosed daily with *R. hominis* (~50 μL, ca. 109 CFU) *via* gavage for a period of 14 days while the untreated group received the same amount of bacterial growth medium. The two groups were then given DSS in drinking water (MW 50 kDa, 30 g/L) beginning from day 8 onward. On day 15, DSS-treated mice without *R. hominis* addition had a strong elevation of a panel of pro-inflammatory biomarkers compared to control mice, with gene induction levels ranging from 4- to 49-fold (**Figure 6A**). Induction of pro-inflammatory genes was significantly lower in the *R. hominis*-treated mice compared to the control group, indicating a strong therapeutic benefit of oral administration of *R. hominis*. On day 15, severe inflammation was detected in the ascending colon of DSS-treated mice without an *R. hominis* supplement, while the colonic mucosa in the *R. hominis*-supplemented group displayed only low-level inflammation, consistent with the reduced expression of pro-inflammatory genes (**Figures 6B,C**). Treatment with *R. hominis* also alleviated the weight loss in DSS-treated mice (Figure S3 and Table S5 in Supplementary Material).

## *R. hominis* Colonization Influences Body Composition and Expression of Satiety Genes

In the next set of experiments, we investigated metabolic changes in the host and bacterium due to host–microbe association. Mice mono-associated with *R. hominis* displayed significant

Figure 3 | Induction of FoxP3+ Treg cells by *Roseburia hominis.* (A) Immunofluorescence analysis of lamina propria cells labeled with (i) anti-Ly6G, (ii) anti-CD3, and (iii) anti-CD3/anti-FoxP3 in the ascending colon of germ-free (GF) and *R. hominis*-treated C3H/HeN mice. Data are shown as the number of positive cells per field of view in GF (*N* = 7–8) and *R. hominis*-treated mice (*N* = 8–10). Original magnification ×630. Statistical significance: \**P* < 0.05. (B) Immunofluorescence analysis of lamina propria cells in the ascending colon labeled with anti-CD3 and anti-FoxP3 in germ-free and *R. hominis* mono-colonized C3H/HeN (*N* = 8) and C57BL/6 (*N* = 3) mice. (C) Flow cytometry analysis of FoxP3+ Treg cells in the lamina propria of conventional mice treated for 14 days with *R. hominis*: C3H/HeN mice (\**P* = 0.04 between control and *R. hominis* treatment) and C57Bl/6 mice (\*\**P* = 0.05 between control and *R. hominis* treatment)*.* (D) Flow cytometry analysis of FoxP3+ Tregs in lamina propria of FoxP3-tagged conventional mice treated for 8 days with *R. hominis*, with 4 days of low dose DSS treatment or untreated. Statistical significance: \**P* < 0.005 control versus DSS treatment and \*\**P* = 0.02 DSS treatment versus DSS<sup>+</sup> *R. hominis* treatment. (E) Flow cytometry plots of FoxP3+ Treg cells in the colonic lamina propria of DSS-treated mice and DSS<sup>+</sup> *R. hominis*-treated mice. Plots are representative of three experiments. The gating strategy for the FACS plots was as follows: FSC/SSC (removal of cell debris), live/dead-SSC, singlets (FSC-A/FSC-H), CD3+CD8−, and CD25+FoxP3+. The percentages were calculated as CD25+FoxP3+ cells of total CD4+ T cells in each preparation. CD4+ cells were derived from the gating strategy of CD3+CD8−.

metabolic changes compared to control GF mice. In particular, the GO processes "negative regulation of response to food" (GO:0032096), "negative regulation of appetite" (GO:0032099), and "regulation of catecholamine secretion" (GO:0050433) were all downregulated in the ascending colon after colonization by *R. hominis* [NCBI GEO (Gene Expression Omnibus), accession

number GSE25544]. The genes involved in these processes were *Agt*, *Cartpt*, *Cck*, and *Cxcl12*, with the corresponding mRNA changes induced ranging from 2- to 12-fold. *Cck*, in particular, plays a major role in digestion and satiety as a hunger suppressant. *Gcg* expression was also downregulated at this gut site.

To establish whether the transcriptional changes of these genes have any physiological relevance in terms of body composition, analyses of dry carcass weight and lipid composition were performed. Dry carcass weights of *R. hominis*-associated mice were significantly heavier than those of GF animals (**Figure 7A**). Carcass lipid analysis showed that the total adiposity was also significantly higher in *R. hominis*-colonized animals at day 14 (**Figure 7B**).

Significant metabolic changes were also detected on the other side of the host–microbe interaction spectrum. In particular, expression of *R. hominis* genes involved in butyrate production was upregulated by the gut environment (**Figure 7C**). These were the genes encoding acetyl-CoA acetyltransferase, 3-hydroxyacyl-CoA dehydrogenase, butyryl-CoA dehydrogenase, and phosphoenolpyruvate carboxykinase. The stimulatory effect of host epithelial cells on production of acetate and butyrate by *R. hominis* was further confirmed in separate *in vitro* experiments (Figure S4 in Supplementary Material). Bacterial cells incubated with Caco-2 and HT29 cells produced significantly higher amounts of these metabolites compared to *R. hominis* control cells incubated under the same conditions but without the epithelial cells.

#### DISCUSSION

A long-term co-evolution of host–microbe interaction has likely driven the selection of functionally important bacterial species in the gut, the majority of which are not encountered in other ecosystems. There is only limited information available at present about the contribution of individual members of the microbial community to intestinal functions, in particular to the development and regulation of mucosal immunity.

The specific functions of certain intestinal bacteria such as *B. fragilis* and SFB have been investigated in the mouse gut to define their individual contributions to T cell biology, and these bacteria have been shown to be potent inducers of Tregs and Th17 cells (4, 7, 8). Members of the *Clostridium* clusters IV and XIVa have recently received more attention. For instance, their presence in the altered Schaedler flora and the contribution of a

Figure 5 | Effects of *Roseburia hominis* flagellin FlaA1 on intestinal epithelial cells and murine bone marrow-derived dendritic cells. (A) Heatmap of differentially expressed genes in Caco-2 cells treated with different bacterial flagellins: from *Salmonella enterica* serotype Enteritidis (SE), *Escherichia coli* K12 (K12), and *R. hominis* (FlaA1 and FlaA2). (B) Expression of CD40; I-A/I-E and CD103 by CD11c+B220+CD317+ Flt3L-derived dendritic cells from conventional C3H/HeN, control (blue) and after 24 h incubation with recombinant flagellins (SE, K12, FlaA1, FlaA2; all in green) determined by flow cytometry. Histogram represents data from three experiments. (C) Frequencies of Flt3L- and GM-CSF-derived dendritic cells from conventional C3H/HeN gated on CD11c+B220+CD317+ cells and CD11c+CD11b+B220− cells treated with different recombinant flagellins (SE, K12, FlaA1, and FlaA2). Results are shown as the percentage of the total, live and singlet cells, mean ± SD from three experiments. (D) The level of IL-10 and IL-12 cytokines was measured by cytometry bead arrays in supernatants from control (unstimulated DCs; *N* = 3) and FlaA1-treated DCs (*N* = 3) derived from C3H/HeN and C57BL/6. Data are presented as mean ± SD. Statistical significance: \*\*\**P* < 0.001. (E) Flagellin stimulated Flt3-derived dendritic cells were cocultured with CD4+ T cells for 5 days, and T cell were stained for the expression of T cell differentiation markers. Graph indicates the percentage of activated Treg cells. Statistical significance: \*\**P* < 0.01.

mixed culture of *Clostridium* spp. to T cell differentiation has also been noted (9, 10).

We report here the successful and stable mono-association of *R. hominis*, a strictly anaerobic flagellated member of the Firmicutes phylum, in the gut of GF mice. The transcriptional responses of *R. hominis* following mono-colonization could be attributed to both the host gut environment and diet. These included chemotaxis and motility subsystems. The role of motility and the flagellar apparatus in host colonization is well understood for pathogenic bacteria but less is known

presented as a fold change compared to control. Statistical significance: \*\*\**P* < 0.001. (B) Histopathology tissue scoring (25) presented as mean percentage of fields of view at a given grade. DSS treatment significantly altered all fields of view at grades 0, 2, 3, and 4. *R. hominis* significantly reduced the % fields of view for grade 4 pathology (*P* = 0.02) and increased the % fields of view for grade 0. Data obtained on day 15 are presented as mean ± SD. (C) Ascending colon (hematoxylin/eosin stained) of control, DSS<sup>+</sup> *R. hominis*-treated and DSS-treated animals. Images shown are the representative fields of view for each treatment group with the allocated score in parenthesis. Single black arrow head indicates infiltrating cells, and the double headed arrow indicates transmural infiltration. L, lumen. Original magnification ×200. Scale bar represents 50 µm.

about the role of flagellins in commensal bacteria. Genes encoding cell motility functions, and flagellins in particular, are generally expressed at variable and low abundance rates in the gut microbiome (27). In addition, only a subset of human gut microbiomes contains detectable flagellin genes (28). We showed that *R. hominis* flagellins are expressed in the mouse intestinal environment. Flagellin gene expression appears to be partly dependent on the presence of certain dietary constituents. Previous work has shown that flagellin gene expression of the related species *Roseburia inulinivorans* is indeed substrate dependent; its expression is higher in the presence of starch compared to glucose, inulin, or fructan (29).

The presence of *R. hominis* in the gut induces genes involved in promoting gut barrier function and innate immunity. Tight junctions, gap junctions, and adherens junctions operate to limit bacterial translocation to the subepithelial layer of the gut, and these functions may be promoted by intestinal bacteria (30). Both CD and UC are characterized by the loss of barrier function and the integrity of tight junctions. Interestingly, dysbiosis of the gut microbiota in UC and CD is associated with a significant reduction of *R. hominis* and *F. prausnitzii* (16, 19). Our results concerning the active contribution of *R. hominis* to gut barrier function support the view that its loss in patients with CD or UC may have significant consequences. Tight junction complexes can be activated by a number of other commensal and probiotic bacteria (31), potentially contributing to the amelioration of the "leaky gut" condition in these diseases.

*Roseburia hominis* induced the expression of genes such as *Ly6g6c*, as well as pathways involved in the regulation of T cells. The most affected T cell pathways included those related to IL-10, ICOS, and CTLA-4, which are all involved in the differentiation

(*N* = 8) and treatment (*N* = 10) groups. At days 0, 1, and 2, animals in the treatment group were dosed by ~109 gavage, while control animals were given 100 µL of YCFA medium alone. Dry body weight and lipid content of carcasses were analyzed. (A) Weight of dry carcasses of *R. hominis*-associated mice was significantly heavier compared to GF animals. (B) Carcass lipid analyses showed that the total adiposity was also significantly higher in *R. hominis*-associated animals at day 14. Data are presented as mean ± SD. (C) RT-qPCR analysis of genes involved in butyrate metabolism.

of Treg cells. Significant increases in colonic CD3<sup>+</sup>CD4<sup>+</sup>CD2 5<sup>+</sup>FoxP3<sup>+</sup> cells were observed in mice either mono-colonized with *R. hominis* or in conventional mice fed a supplement of live *R. hominis* culture. Our findings complement recent reports on Treg differentiation effects by other *Clostridium* species (10, 32). We have shown here that the single bacterial strain *R. hominis* A2-183 is able to promote mucosal T cell expansion and impact T cell differentiation in mono-associated, conventional and DSS-treated mice.

Members of the *Roseburia* genus are some of the most prevalent motile bacterial species in the normal healthy human intestinal microbiota (33). Flagellins are potent immunomodulatory proteins: flagellated bacteria, such as *R. hominis*, could interact actively with the host immune system. Another important consequence of motility among the *Roseburia* species is their ability to penetrate the mucus layer and attach to the host gut epithelial cell surfaces (34). This property is considered to be a very desirable characteristic enhancing the probiotic potential of bacteria (35). Close proximity to the host cells allows commensal gut bacteria to exert potent physiological effects, potentially reversing metabolic disorders and controlling inflammation, gut barrier function, and gut peptide secretion.

Flagellin signaling by many pathogenic bacteria through the host's TLR5 receptors induces a strong pro-inflammatory response, mainly driven by the activation of the NF-κB signaling pathway (36). However, the numerical prevalence of flagellated commensal bacteria makes a strong pro-inflammatory scenario seem unlikely. These bacteria are capable of flagellin/ TLR5 signaling, which is important for host defense and disease protection, because deletion of TLR5 results in colitis (37). The presence of flagellin genes in the SFB genomes (38, 39) may explain the potent induction of Th17 cells by these bacteria (7, 8).

Differential expression of *R. hominis* flagellins *in vitro* has effects on IECs and BMDCs. The panel of flagellins tested affected IECs and BMDCs differently, although all the flagellin structures included the conserved Arg90 domain associated with flagellins that bind and activate TLR5 (40), which suggests that other sequence/structural properties might account for the signaling responses mediated by FlaA1 and FlaA2. Certain commensal flagellin structures may help to direct immune tolerance responses through TLR5 expressed on either CD103<sup>+</sup>DC or Treg subsets (23, 41, 42). The significance of flagellin-TLR5 signaling in Treg responses induced by *R. hominis* was further investigated using TLR5KO mice. We showed that *R. hominis*-induced CD3<sup>+</sup>CD4<sup>+</sup>CD25<sup>+</sup>FoxP3<sup>+</sup> cells were significantly lower in TLR5KO mice, indicating that TLR5/flagellin signaling is an important mediator for the expansion of this T cell subset. The immunomodulatory effects of *R. hominis* were shown in DSS-treated mice. Although the DSS model of colitis is generally a T cell-independent model, we showed that *R. hominis* induced regulatory T cell populations in these animals. Other anti-inflammatory molecules such as butyrate may also contribute to immune tolerance and limit tissue damage (43, 44). Another effect of short-chain fatty acids produced by the bacterium is the improved food conversion efficiency. The dietary constituents such as polysaccharides that are poorly utilized by the host are degraded by the bacterium and used by the host resulting in a better energy recovery from the diet. Differential expression of satiety genes may be another factor contributing to the heavier weight and higher adiposity in *R. hominis*-colonized animals.

In summary, gut colonization with *R. hominis* caused the induction of specific subsets of genes on both the bacterial and host sides of the interaction. In this study, we have focused on signaling amongst other molecular mechanisms to explain the nature of this cross talk. TLR5/flagellin signaling could, potentially, drive the expansion of Treg cells *via* a TLR5-dependent mechanism, and our results suggest a potential therapeutic benefit of *R. hominis* in the treatment of UC, which is characterized by the loss of this bacterium from the gut (16).

#### ETHICS STATEMENT

The management and experimental procedures with animals were approved by the respective Local Ethical Review Committees at Institut Micalis, INRA, Jouy-en-Josas, France and Medical Research Facility, University of Aberdeen, United Kingdom.

## AUTHOR CONTRIBUTIONS

AP, IM, and RA designed research; AP, IM, AT, AL, NC-B, VG-R, KG, EL, MD, AC, EM, VF, RI, GG, and RA performed research; AP, IM, AL, GG, and RA analyzed data; AP, IM, GG, and RA wrote the manuscript.

## ACKNOWLEDGMENTS

The authors would like to thank Andrew Gewirtz (Georgia State University, Atlanta, GA, USA) and Adam Cunningham (University of Birmingham, Edgbaston, Birmingham, UK) for generously providing the germ-free and conventional TLR5KO mice, respectively. The authors also wish to acknowledge a significant contribution to the inception of the work, experimental work, interpretation of results, and writing of the manuscript by Denise Kelly, as well as the support from RESAS (Rural and Environmental Science and Analytical Services) of the Scottish Government. EM and VF were supported by the Marie Curie Initial training network Fellowships funded by the EU (grant #215532). The authors acknowledge a generous support of Prof. Harry Flint and Dr. Sylvia Duncan in the microbiological part of this work. The skilled technical support of Gillian Campbell and Pauline Young at RINH Genomics is also gratefully acknowledged.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu.2017.01166/ full#supplementary-material.

## REFERENCES


extensive auxotrophy and adaptations to the intestinal environment. *Cell Host Microbe* (2011) 10(3):260–72. doi:10.1016/j.chom.2011.08.005


44. Furusawa Y, Obata Y, Fukuda S, Endo TA, Nakato G, Takahashi D, et al. Commensal microbe-derived butyrate induces the differentiation of colonic regulatory T cells. *Nature* (2013) 504(7480):446–50. doi:10.1038/ nature12721

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Patterson, Mulder, Travis, Lan, Cerf-Bensussan, Gaboriau-Routhiau, Garden, Logan, Delday, Coutts, Monnais, Ferraria, Inoue, Grant and Aminov. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Free Fatty acids Profiles are related to gut Microbiota signatures and short-chain Fatty acids

*Javier Rodríguez-Carrio1 , Nuria Salazar1 , Abelardo Margolles1 , Sonia González2 , Miguel Gueimonde1 , Clara G. de los Reyes-Gavilán1 \* and Ana Suárez3*

*1Department of Microbiology and Biochemistry of Dairy Products, Instituto de Productos Lácteos de Asturias (IPLA-CSIC), Villaviciosa, Asturias, Spain, 2Area of Physiology, Department of Functional Biology, University of Oviedo, Oviedo, Asturias, Spain, 3Area of Immunology, Department of Functional Biology, University of Oviedo, Oviedo, Asturias, Spain*

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Mario M. D'Elios, University of Florence, Italy Francois-Pierre Martin, Nestle Institute of Health Sciences, Switzerland Arne Yndestad, Oslo University Hospital, Norway*

*\*Correspondence:*

*Clara G. de los Reyes-Gavilán greyes\_gavilan@ipla.csic.es*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 07 April 2017 Accepted: 29 June 2017 Published: 24 July 2017*

#### *Citation:*

*Rodríguez-Carrio J, Salazar N, Margolles A, González S, Gueimonde M, de los Reyes-Gavilán CG and Suárez A (2017) Free Fatty Acids Profiles Are Related to Gut Microbiota Signatures and Short-Chain Fatty Acids. Front. Immunol. 8:823. doi: 10.3389/fimmu.2017.00823*

A growing body of evidence highlights the relevance of free fatty acids (FFA) for human health, and their role in the cross talk between the metabolic status and immune system. Altered serum FFA profiles are related to several metabolic conditions, although the underlying mechanisms remain unclear. Recent studies have highlighted the link between gut microbiota and host metabolism. However, although most of the studies have focused on different clinical conditions, evidence on the role of these mediators in healthy populations is lacking. Therefore, we have addressed the analysis of the relationship among gut microbial populations, short-chain fatty acid (SCFA) production, FFA levels, and immune mediators (IFNγ, IL-6, and MCP-1) in 101 human adults from the general Spanish population. Levels of selected microbial groups, representing the major phylogenetic types present in the human intestinal microbiota, were determined by quantitative PCR. Our results showed that the intestinal abundance of *Akkermansia* was the main predictor of total FFA serum levels, displaying a negative association with total FFA and the pro-inflammatory cytokine IL-6. Similarly, an altered FFA profile, identified by cluster analysis, was related to imbalanced levels of *Akkermansia* and *Lactobacillus* as well as increased fecal SCFA, enhanced IL-6 serum levels, and higher prevalence of subclinical metabolic alterations. Although no differences in nutritional intakes were observed, divergent patterns in the associations between nutrient intakes with intestinal microbial populations and SCFA were denoted. Overall, these findings provide new insights on the gut microbiota–host lipid metabolism axis and its potential relevance for human health, where FFA and SCFA seem to play an important role.

Keywords: free fatty acids, microbiota, *Akkermansia*, short-chain fatty acids, serum lipids, subclinical metabolic alterations

#### INTRODUCTION

Free fatty acids (FFA) are lipid species released from the adipose tissue and several cell types upon lipolysis. Apart from their classical roles in energy supply or as structural components, FFA are emerging as active players of a number of biological processes. FFA can affect gene expression of macrophages (1), adipocytes (2), or endothelial cells (3). In addition, FFA can modulate the production of chemokines and cytokines (3–5), the expression of genes coding for adhesion molecules (6, 7) and they give rise to pro-inflammatory and inflammation pro-resolving lipid-derived species (8).

Therefore, a growing body of evidence emphasizes a role for FFA as common mediators between metabolic conditions and the immune system. FFA have been proposed as a mechanistic explanation for the relationship among obesity, inflammation, altered glucose homeostasis, and cardiovascular disease (9). Similarly, some FFA have been reported as regulators of systemic metabolism homeostasis in mice (10). Importantly, quantitative and qualitative differences among the effects of individual FFA have been revealed (11). Thus, not only increased levels but also an altered FFA pool composition may be associated with the risk of developing a range of disorders in which the immune system plays a role (12–17). However, the underlying causes of the altered FFA pool composition remain unknown.

Compelling evidence from recent years has shed some light on the effects that gut microbiota can exert on the host health. Alterations in the composition of the intestinal microbiota have been related with the development of metabolic disorders both in mice and in human studies (18), whereas some dietary factors are also known to modulate these microbial communities (19). Recent studies indicate that the gut microbiota is also involved in the host energy metabolism by regulating the absorption of nutrients, local production of hormones and immune mediators, fat storage, and gut permeability (20–22). However, to what extent host lipid metabolism is associated with the intestinal microbial populations, and whether the gut microbiota could be related to FFA levels, remains largely unknown. Short-chain fatty acids (SCFA) are produced in the gut by the metabolic activity of the intestinal microbiota as catabolic end-products from the fermentation of undigested dietary components, mainly complex carbohydrates. These compounds are pivotal in the interactions between the host and intestinal microbial populations (23), but their actual links with the metabolism of fatty acids in humans are controversial. In addition, since most studies have been performed with relatively low sample sizes and mainly with *a priori* established conditions, evidence for their potential relevance in healthy adults is lacking in the literature. However, due to their pivotal role in shaping the host metabolism, it is tempting to speculate that the intestinal microbiota could be related to FFA levels in healthy populations.

Based on these lines of evidence, we hypothesize that alterations in the composition and metabolic activity of the gut microbiota may be associated with altered levels of FFA, which can be in turn be related with some mediators of inflammation. With the aim to gain insight into the joint impact of these players in human health, we have studied the relationships among selected microbial populations, fecal SCFA, serum FFA levels, and composition, as well as inflammatory mediators. In this way, the main objective of the present study was to evaluate whether specific gut microbial signatures can be related to altered levels of FFA species in adults from the general population.

#### MATERIALS AND METHODS

#### Ethical Approval

Ethical approval for this study was obtained from the Institutional Review Board (Comité de Ética de Investigación Clínica del Principado de Asturias) in compliance with the Declaration of Helsinki. All participants were informed and gave a signed informed consent prior their inclusion in the study.

#### Subjects

A group of 101 individuals was recruited from the general population in the Asturias region, northern Spain. Exclusion criteria were the diagnosis of chronic or immune-mediated diseases, recent infections, cancer, or altered metabolic conditions, as well as the current usage (within the previous 6 months) of immunomodulatory drugs, metabolic agents, probiotics, or antibiotics. Demographical parameters of the analyzed population are summarized in **Table 1**.

Table 1 | Description of the study population.


*Variables are summarized as median (interquartile range), mean* ± *SD, or n (%), as appropriated, unless otherwise stated.*

Subjects were asked to participate in this study and, upon acceptance, an overnight fast blood sample was drawn by venipuncture in tubes without anticoagulant. After blood clotting, serum was collected and stored at −80°C until further analyses. In addition, basic serum biochemical analyses were performed by standardized procedures. Subjects were considered to exhibit a subclinical metabolic alteration if they meet any of the following criteria: fasting glucose levels higher than 100 mg/dl or that of triglycerides over 150 mg/dl. These objectives cut offs were obtained from clinical national guidelines.

#### Total FFA Assessment

Total FFA serum levels were quantified by means of a colorimetric assay using a commercial kit (NEFA kit half-microtest, Roche Life Sciences, Penzberg, Germany) following the protocol from the manufacturer. Final absorbance was measured at 546 nm, and detection limit was 0.02 mM.

#### Individual FFA Quantification

Free fatty acids were analyzed after a methyl-*tert*-butylether (MTBE)-based extraction protocol as previously described (24) with slight modifications. Briefly, 100 µl serum samples were spiked with 5 µl of internal standard (600 ppm heptadecanoic acid). Then, protein precipitation was performed by adding 200 µl methanol chromasolv grade (Sigma Aldrich, MO, USA), and tubes were vortexed for 30 s. Then, 1,200 µl MTBE chromasolv grade (Sigma) was added, vortexed, and an incubation in an ultrasound water bath at 15°C was performed for 30 min. Finally, organic phase was separated after the addition of 200 µl milliQ water and a centrifugation step for 7 min at 5,000 rpm (15°C). After collecting the upper layer, the extraction protocol was repeated once with 100 µl MetOH, 500 µl MTBE, and 100 µl milliQ H2O.

Lipid extracts were dried in a miVac centrifugal evaporator (Genevac Ltd., UK) and redissolved in 100 µl of water:acetonitrile 38:62.

For the determination of the fatty acids, a Dionex Ultimate 3000 HPLC system (Thermo Scientific, Bremen, Germany), consisting of a high pressure binary pump, an autosampler and a column oven, was used. The column was a Zorbax Eclipse Plus C18, 50 mm × 2.1 mm, 1.8 μm from Agilent. Mobile phases A and B were water and acetonitrile, respectively, both containing 0.1% of formic acid. Fatty acids separation was carried out by the following gradient program: 62% B (held for 4.5 min) followed by a linear increase up to 100% B in 10 min (held for 1 min). The column temperature was set at 45°C and the injection volume was 2 µl.

Mass detection was performed using a Bruker Impact II q-ToF mass spectrometer with electrospray ionization, operated in the negative mode. The settings of the mass spectrometer were as follows: spray voltage, 4.5 kV; drying gas flow 12 l/min; drying gas temperature 250°C; nebulizer pressure 44°psi.

For quantitation, calibration curves for each compound were prepared by proper dissolution of the pure standards in methanol to encompass the expected concentration of the analytes in the sample. The calibration ranges were as follows: 0.4–12.5 µg/ml for eicosapentaenoic (EPA) and γ-linolenic; 1.2–37.5 µg/ml for docosahexaenoic acid (DHA) and linolenic; 2.3–75 µg/ml for arachidonic (AA) and palmitoleic; 3.9–125 µg/ml for linoleic; and 7.8–250 µg/ml for oleic, palmitic, and stearic. A good linearity was observed in all cases (*r*<sup>2</sup> > 0.994). Heptadecanoic acid, added at a level of 30 µg/ml, was used as internal standard to compensate for possible biases during the sample preparation step.

#### Quantification of Cytokines

Serum levels of IFNγ, MCP-1, and IL-6 were determined by immunoassays using commercial kits (IFNγ OptEIA kit) from BD Biosciences (NJ, USA) and MCP-1 and IL-6 mini-EDK kits from Peprotech (NJ, USA), following manufacturer's instructions. Detection limits were 0.58, 8, and 5.8 pg/ml for IFNγ, MCP-1, and IL-6, respectively.

#### Nutritional Assessments

Dietary intake was assessed by means of an annual semi quantitative validated food frequency questionnaire including 160 items (25). Trained dieticians asked about cooking practices, number and amount of ingredients used in each recipe, as well as enquiring about menu preparation (e.g., type of oil used and type of milk) and other relevant information to the study. During an interview, subjects were asked item by item whether they usually ate each food and, if so, how much they usually ate. For this purpose, three different serving sizes of each cooked food were presented in pictures to the participants so that they could choose from up to seven serving sizes (from "less than the small one" to "more than the large one"). For some of the foods consumed, amounts were recorded in household units, by volume, or by measuring with a ruler. Methodological issues concerning dietary assessment have been detailed elsewhere (25). Food intake was analyzed for energy, macronutrients, and total fiber content by using the nutrient Food Composition Tables developed by the Centro de Enseñanza Superior de Nutrición Humana y Dietética (26).

#### Anthropometric Measures

Height was measured using a stadiometer with an accuracy of ±1 mm (Año-Sayol, Barcelona, Spain). The subjects stood barefoot, in an upright position and with the head positioned in the Frankfort horizontal plane. Weight was measured on a scale with an accuracy of ±100 g (Seca, Hamburg, Germany). Quetelet index was calculated using the formula: weight (kg)/height (m2 ).

#### Analysis of Fecal Microbiota

Fecal samples from individuals participating in the study were collected at home in sterile containers, kept at 4°C in the domestic refrigerator (maximum 2–3 h from deposition) and frozen at −80°C until analyses just right on arrival to the laboratory, as previously reported (27, 28). One gram of fecal sample was employed for DNA extraction by using the QIAamp DNA stool mini kit (Qiagen, Hilden, Germany) as previously described (28). Quantification of bacterial groups (*Akkermansia, Bacteroides* group, *Bifidobacterium*, *Clostridium* cluster XIVa, *Lactobacillus* group, and *Faecalibacterium*) in feces was performed with a 7500 Fast Real Time PCR System (Applied Biosystems, Foster City, CA, USA) using SYBR Green PCR Master Mix (Applied Biosystems) as previously described (29–31). These microbial groups include the main representatives of the human intestinal microbiota, which all together represent over 95% of the total bacteria in the human intestine (32, 33). For qPCR analysis, 1 µl of template fecal DNA (~5 ng) and 0.2 µM of each primer were added to the reaction mixture (25 µl). PCR cycling was as follows: an initial cycle of 95°C 10 min, 40 cycles of 95°C 15 s, and 1 min at the appropriate primer temperature. We compared the *C*<sup>t</sup> values obtained from a standard curve constructed as previously indicated (30) to estimate the number of cells. Standard cultures, primers, and annealing temperatures used for qPCR were the same as those recently reported (31).

## Analysis of Short-Chain Fatty Acids in Fecal Samples

Analysis of short chain fatty acids (SCFA) was performed by gas chromatography to determine the concentrations of acetate, propionate, and butyrate. One gram of fecal samples was weighed, diluted 1:10 in sterile PBS, and homogenized in a LabBlender 400 stomacher (Seward Medical, London, UK) at full speed for 4 min. Supernatants were then obtained by centrifugation (10,000 *g*, 30 min, 4°C), filtered through 0.2-µm filters, mixed with 1/10 of ethyl butyric (2 mg/ml) as an internal standard, and stored at −80°C until analysis. A gas chromatograph 6890N (Agilent Technologies Inc., Palo Alto, CA, USA) connected to a mass spectrometry (MS) 5973N detector (Agilent Technologies) and to a flame ionization detector was used for identification and quantification of SCFA.

Data were collected using the Enhanced ChemStation G1701DA software (Agilent). Samples (1 µl) were directly injected into the gas chromatograph equipped with an HP-Innowax capillary column (60-m length by 0.25-mm internal diameter, with a 0.25 µm film thickness; Agilent) using He as gas carrier and a constant flow rate of 1.5 ml/min. The temperature of the injector was kept at 220°C, and the split ratio was 50:1. Chromatographic conditions were as follows: initial oven temperature of 120°C, 5°C/min up to 180°C, 1 min at 180°C, and a ramp of 20°C/min up to 220°C to clean the column. In the MS detector, the electron impact energy was set at 70 eV. The data collected were in the range of 25–250 atomic mass units (at 3.25 scans/s).

SCFA were identified by comparison of their mass spectra with those held in the HP-Wiley 138 library (Agilent) and by comparison of their retention times with those of the corresponding standards (Sigma Aldrich, St. Louis, MO, USA). The peaks were quantified as relative abundances with respect to the internal standard. The concentration (in mM) of each SCFA was calculated using the linear regression equations (*r*<sup>2</sup> ≥ 0.99) from the corresponding standard curves obtained with six different concentrations.

#### Statistical Analysis

The study design was a cross-sectional, one-group, observational analysis. Continuous variables were summarized as median (interquartile range) or mean ± SD, whereas *n* (%) was used for the categorical ones. Mann–Whitney *U* or *t*-tests were used to assess statistical differences, when appropriate. Correlations were analyzed by Spearman's rank test. After univariate correlation analyses, multiple regression analyses were conducted to assess the strength of the association between these variables including other (continuous) variables as potential confounders. Multiple regression analyses were also performed to identify the main predictors of a candidate independent variable. When coefficients of determination correspond to multiple regression analyses they are indicated as R2 . For each analysis, dependent and independent variables were indicated and the β coefficient, *B* coefficient with 95% confidence intervals (CIs), and *p*-values were computed. The association between two categorical variables was first studied by χ2 tests in univariate models and then was analyzed by multiple logistic regression models to include potential confounders as independent variables. For this analysis, odds ratio (OR) and 95% CI were computed. Principal component analysis (PCA) with Varimax rotation was performed to reduce sample dimensionality and potential collinearity effects. The number of components retained was based on eigenvalues (>1), and loadings greater than 0.5 were used to identify the variables comprising a single component. For cluster analysis, squared Euclidean distances were calculated, and Ward's Minimum Variance Method was used to identify clusters minimizing the loss of information. R package *heatmap.2* was used to generate heatmaps for visualization purposes. Some statistical analyses were performed independently in each cluster to evaluate whether differences among the studied variables independently arise on each cluster. The statistical approach is summarized in **Figure 1**. SPSS 19.0, R 3.0.3, and GraphPad Prism 5.0 for Windows were used.

## RESULTS

#### FFA and Microbial Populations

Total FFA serum levels and fecal microbial groups were analyzed in 101 human adults from the general population (**Table 1**). High total FFA serum levels were considered as a surrogate marker of impaired lipid metabolism. Then, the associations between FFA levels and intestinal microbial populations were determined by univariate and multivariate analyses adjusted for socio-demographical [age, gender, and body mass index (BMI)] and nutritional parameters (total energy and intakes of carbohydrates, lipids, proteins, and fiber). Interestingly, only *Akkermansia* abundance was correlated with FFA serum levels (*r* = -0.383, *p* < 0.001) (**Figure 2**). Moreover, this association remained statistically significant in a multiple regression analysis adjusted for age, gender, BMI, microbial groups analyzed, total energy as well as carbohydrates, lipids, and proteins intake as potential confounders (**Table 2**). Although a slight correlation between total FFA and the serum levels of the pro-inflammatory cytokine IL-6 was observed (*r* = 0.240, *p* = 0.020), no effect was evidenced when IL-6 was introduced in the previous multiple regression analysis.

Next, we aimed to evaluate whether this finding was due to a general effect on total FFA levels or it could be also associated with individual FFA species. To this end, associations between *Akkermansia* and the levels of individual FFA (γ-linolenic, palmitic, oleic, stearic, linoleic, palmitoleic, linoleic, AA, EPA, and

(left) indicates groups analyzed and sample sizes, whereas statistical methods for analysis and stratification of the study population are indicated at the left side.

DHA) were assessed. *Akkermansia* abundance was negatively associated with some FFA [stearic (*r* = −0.218, *p* = 0.039), palmitic (*r* = −0.321, *p* = 0.002), oleic (*r* = −0.261, *p* = 0.013), palmitoleic (*r* = −0.297, *p* = 0.004), linoleic (*r* = −0.272, *p* = 0.010), and γ-linolenic (*r* = −0.232, *p* = 0.028)], but no association was observed with EPA, DHA, AA, or linoleic levels. As important differences among FFA exist, individual FFA were grouped according to their chemical structure: saturated (SFA), monounsaturated (MUFA), w3-poly-unsaturated (w3-PUFA), and w6-PUFA. Interestingly, *Akkermansia* exhibited a strong negative association with SFA (*r* = −0.314, *p* = 0.003) as well as with

Table 2 | *Akkermansia* as predictor of free fatty acids (FFA) serum levels.


*The associations between FFA serum levels (as dependent variable) and microbial populations were studied by multiple lineal regression analysis including sociodemographical parameters and nutritional intakes as potential confounders (all included as independent variables in the model).* β *and B coefficients [with 95% confidence intervals (CI)] and p-values are calculated for each parameter within the same model. R2 (model)* = *0.303. The p-value highlighted in bold represent statistically significant differentes.*

MUFA and w6-PUFA to a lower degree (*r* = −0.266, *p* = 0.011 and *r* = −0.244, *p* = 0.020, respectively), whereas no association was observed for w3-PUFA (*r* = −0.104, *p* = 0.330). Thus, our findings point to a negative association between *Akkermansia* abundance and levels of SFA, MUFA, and w6-PUFA, all of them pro-inflammatory. Since some collinearity was noted among individual FFA levels, a PCA was carried out to avoid potential biases. PCA was then conducted with the serum levels of specific FFA, given that all of them exhibited communalities higher than 0.5, thus supporting the appropriateness of this analysis. The Kaiser–Meyer–Olkin test provided a good adequacy of the data (0.858) as the Bartlett test of sphericity (*p* = 10<sup>−</sup>144) did. PCA identified two main components (explaining 72.4% of the total variance): C1 (including loadings from γ-linolenic, palmitic, oleic, stearic, linoleic, palmitoleic, linoleic, AA; explaining 59.81% of the total variance) and C2 (including EPA and DHA; explaining 13.09% of the total variance). Notably, *Akkermansia* was negatively associated with C1 (*r* = −0.308, *p* = 0.003), thus supporting an opposite relationship between the levels of this microorganism and those of the saturated, monounsaturated, and w6-PUFA retained in this component, whereas no association was found with anti-inflammatory w3-PUFA (C2). These results from the PCA strengthen our previous findings on the separated analysis of FFA according to their chemical structure. Moreover, since PUFA are present in lower levels, and taking into account that nine FFA loaded on C1 whereas only two did in C2, this association was studied by regression analyses including total FFA serum levels as confounder to rule out the possibility that decreased FFA abundance may bias the analysis. Interestingly, *Akkermansia* levels remained associated with C1 after performing this adjustment (β [95% CI], *p*: −0.286 [−1.999, −0.315], *p* = 0.008), whereas no association was observed with C2 (−0.052 [−0.152, −0.092], *p* = 0.625), thereby confirming our previous findings. In addition, the analysis of inflammatory mediators showed that *Akkermansia* was negatively associated with IL-6 serum levels (*r* = −0.233, *p* = 0.032), whereas IL-6 levels and C1 were in turn associated (*r* = 0.220, *p* = 0.032). No associations were found for *Akkermansia* with IFNγ (*r* = 0.056, *p* = 0.604) or MCP-1 (*r* = −0.028, *p* = 0.802) serum levels.

Therefore, all these results suggest a relationship between intestinal *Akkermansia* and serum FFA levels in humans. Moreover, *Akkermansia* was negatively associated with specific FFA species, mainly those saturated and/or with an attributed pro-inflammatory role, such as MUFA and w6-PUFA. This finding is in line with the negative association observed between *Akkermansia* and IL-6 serum levels.

#### Associations between FFA Profiles, Microbial Populations, and SCFA

Our results from the PCA analysis and the differences found in the associations with *Akkermansia* and IL-6 levels point to the existence of different FFA profiles among individuals. To address this hypothesis, an unsupervised cluster analysis was performed for the individual FFA levels. Interestingly, this approach confirmed the existence of two independent groups of individuals based on FFA levels, thereafter referred to as cluster I (*n* = 62) and cluster II (*n* = 39) (**Figure 3A**).

Although all FFA species were higher in cluster II, mainly saturated and/or pro-inflammatory fatty acids were associated with this group, whereas a strikingly different pattern was observed for EPA and DHA, which seem not to play a role in clusters definition on our analysis (**Figure 3B**). No significant differences in age, gender or BMI were observed between individuals from clusters I and II (**Table 3**). Nutrient intakes were then compared between clusters in order to evaluate whether distinct daily intakes could account for the differences observed in FFA clusters. Interestingly, no differences in daily intakes were found (**Table 4**), hence supporting that additional factors other than demographic and nutritional parameters may underlie the differences found between FFA clusters.

Notably, important differences in intestinal microbial populations and fecal SCFA levels were observed between clusters (**Table 5**). Cluster II was characterized by decreased levels of *Akkermansia* and increased *Lactobacillus* group counts as well as higher acetate, propionate and total fecal SCFA concentrations than cluster I. Moreover, increased total FFA [0.48 (0.24) vs 0.26 (0.13) mM] and IL-6 [752.40 (1,155.22) vs 411.77 (636.85) pg/ml], but not IFNγ (*p* = 0.252) or MCP-1 (*p* = 0.541) serum levels were also registered in cluster II (**Figure 3C**).

Finally, some statistical analyses performed independently within each cluster revealed particular interesting features for each group, thereby stressing the existence of important differences between groups. On the one hand, IL-6 serum levels were negatively correlated to *Akkermansia* abundance (*r* = −0.312, *p* = 0.034) and acetate (*r* = −0.344, *p* = 0.014) in subjects included in cluster I, whereas no associations were observed in cluster II, characterized by higher IL-6 and acetate levels. Conversely, acetate and total fecal SCFA were strongly associated with higher total FFA serum levels in subjects included in cluster II (*r* = 0.459, *p* = 0.006; and *r* = 0.410, *p* = 0.016, respectively). Similarly, total energy intake was positively associated in this group with total SCFA production (*r* = 0.438, *p* = 0.008), but not in their cluster I counterparts (*r* = −0.044, *p* = 0.748), hence supporting a cross talk between host and microbial metabolisms, in which SCFA are involved.

Overall, our findings support the existence of homeostatic relationship between microbial populations, host lipid metabolism and inflammatory mediators. Thus, imbalanced microbial populations and/or increased fecal SCFA production could be related to an altered lipid metabolism as well as a serum IL-6 shift.

#### Analysis of the Associations among Impaired FFA Profile, Subclinical Metabolic Alterations, and Microbial Populations

Then, we addressed in our group of adult subjects, whether an altered gut microbiota composition and/or serum inflammatory mediators may have a negative impact on the host metabolism and if a clinical relevance of these players could be expected.

Although no significant differences in serum levels of glucose, cholesterol or triglycerides were found between groups, individuals within cluster II were more likely to exhibit hyperglycemic (>100 mg/dl) or hypertriglycemic (>150 mg/dl) states than those within cluster I (22/39 vs 22/62, *p* = 0.039 and 7/39 vs 4/62, *p* = 0.070). Consequently, metabolic alterations (either hyperglycemia or hypertriglycemia, as stated in the Methods section) were more prevalent in cluster II (23/39 vs 24/62, *p* = 0.047).

The strength of these associations was further analyzed by multiple logistic regression. By selecting subclinical metabolic

specific FFA serum levels (columns). Colors in the vertical bar at the right of the heatmap identify clusters I (black) and II (gray). Tiles are colored based on serum FFA concentrations, red and blue indicating low or high levels, respectively. (B) Biplot obtained in the cluster analysis of specific FFA serum levels. Red arrows represent the vectors showing the associations among the original variables entered and the outcome of the cluster analysis. Numbers represent the final cluster assigned to each subject (1: cluster I, 2: cluster II). Whereas saturated and mainly pro-inflammatory FA were closely related, EPA and DHA were not involved in clusters definition. (C) Comparison of serum levels of FFA, IL-6, IFNγ, and MCP-1 between subjects of both clusters. Boxes represent median and interquartile range, whereas whiskers represent minimum and maximum values. Differences were assessed by Mann–Whitney *U* tests.



*Demographical parameters from the whole population (n* = *101) and from the study participants classified according to clusters computed from individual FFA levels are shown. Differences between clusters were analyzed by Mann–Whitney U or* χ*<sup>2</sup> tests. Variables are expressed as mean* ± *SD, median (interquartile range) or n, as appropriate.*

alterations as the dependent variable and adjusting the model for confounders (age, gender, BMI, total energy, as well as lipids, carbohydrates, protein, and fiber intakes, all introduced as independent variables), we found that individuals within cluster II were more likely to exhibit metabolic alterations (OR [95% CI], *p*: 3.064 [1.127, 8.330], 0.028). Interestingly, this association remained significant when the microbial groups were entered in the model, appearing *Akkermansia* abundance also associated with metabolic alterations (*p* = 0.029). This highlights the interplay between altered FFA profile, microbial populations and impaired host metabolism.

Moreover, additional observations reinforced the finding commented just above. Since our initial results point to a cross

#### Table 4 | Nutritional parameters according to free fatty acids (FFA) clusters.


*Dietary intakes from the whole population (n* = *101) and from the study participants classified according to clusters computed from individual FFA levels are shown. Differences in daily intakes between clusters were analyzed by Mann–Whitney U tests (unadjusted) or linear regression models to adjust for confounders (adjusted). Energy was adjusted by gender and age, whereas the rest of the nutrients were adjusted by gender, age, and energy. Variables are expressed as mean* ± *SD.*

Table 5 | Intestinal microbial populations, fecal short-chain fatty acid (SCFA), and serum cytokine levels according to free fatty acids (FFA) clusters.


*Microbial populations and SCFA fecal levels and serum cytokines from the whole population (n* = *101) and from the study participants classified according to clusters computed from individual FFA levels are shown. Differences in microbial populations, fecal SCFA, and serum cytokine levels between clusters were analyzed by t or Mann– Whitney U tests. The p-values highlighted in bold represent statistically significant differences. Variables are expressed as mean* ± *SD or median (interquartile range).*

talk between host and microbial metabolism, we further deepen into this idea focusing on possible differential associations between microbial groups or SCFA with nutrient intakes. In subjects with subclinical metabolic alterations, *Lactobacillus* abundance was correlated with carbohydrates intake in both cluster I (*r* = 0.603, *p* = 0.004) and II (*r* = 0.765, *p* < 0.001), but with fiber intakes only in cluster I (*r* = 0.609, *p* = 0.003). No associations were observed in those subjects free of metabolic alterations. On the other hand, the association between total SCFA and total FFA levels mirrored this situation, a positive correlation being found in individuals with metabolic alterations in both cluster I (*r* = 0.524, *p* = 0.018) and II (*r* = 0.432, *p* = 0.045), but not in those individuals without subclinical metabolic alterations.

In sum, these results support an association between microbial populations, nutritional factors, and impaired host metabolism. Subjects exhibiting an altered FFA profile, which also display imbalanced *Akkermansia* and *Lactobacillus* populations, were more frequently associated with subclinical pathogenic metabolic states. Divergent patterns in the associations between dietary intakes with microbial populations and SCFA were also observed.

#### Microbial Imbalanced Populations As Predictors of Impaired FFA Levels

Finally, although *Akkermansia* abundance was identified as the only microbial predictor of serum FFA levels in the whole population analyzed (*n* = 101), striking differences in *Akkermansia* levels were evidenced between clusters I and II. Since the results commented just above suggest that altered microbial populations and their interactions with nutrient intakes may underlie impairment in the host metabolism and that SCFA can have relevance in such interactions, we included all these parameters (microbial populations, SCFA, dietary intakes, age, gender, and BMI) as independent variables in a multivariate regression analysis, FFA levels being selected as the dependent variable, after stratifying our populations by the clusters obtained in the previous analysis.

Interestingly, whereas only age (*p* = 0.038) and BMI (*p* = 0.050) were identified as predictors of FFA levels in individuals within cluster I (*R*<sup>2</sup> (model) = 0.407), *Lactobacillus* abundance (*p* = 0.032), fiber intake (*p* = 0.021), SCFA (acetate *p* = 0.008, propionate *p* = 0.018, and butyrate *p* = 0.003), and gender (*p* = 0.008) were associated with FFA levels in subjects within cluster II (*R*<sup>2</sup> (model) = 0.646).

In summary, different factors were found to be associated with total FFA levels depending on the levels of *Akkermansia*. Above a threshold of *Akkermansia* levels, anthropometric factors such as BMI and age were the main predictors of serum FFA. However, when *Akkermansia* abundance is low, factors other than the anthropometric characteristics (as imbalanced microbial populations or SCFA production) seem to impact the FFA pool.

## DISCUSSION

Despite the important research advances in recent years, the links between human metabolism and gut microbiota are far from being completely understood, especially in the field concerning the lipid metabolism. The present study addresses a multilevel analysis of this scenario, by assessing different surrogate biomarkers of the lipid metabolism, in addition to some relevant intestinal microbial groups and SCFA production as well as inflammatory mediators; the study was performed in a sample of adult subjects from the general population in order to gain some insight into the relationship among these parameters and their potential impact on the human health. Our results revealed an association between the abundance of the intestinal microorganism *Akkermansia* and circulating FFA. This association was restricted to a specific group of FFA species, *Akkermansia* and serum IL-6. In addition, cluster analyses revealed that imbalanced intestinal microbial groups and levels of SCFA production may be related to an impaired FFA profile, pro-inflammatory and saturated fatty acids mainly hallmarking this group. Interestingly, individuals who exhibit these features were more likely to show metabolic alterations. Therefore, the results herein presented provide valuable information on the gut microbiota-host metabolism axis and its involvement in human health.

A major finding of our study was the association between *Akkermansia* and serum FFA levels. *Akkermansia* has been reported to participate in the maintenance of gut integrity and energy harvest by the host (34). It has been confirmed that *Akkermansia* has a causative role in lowering body fat mass and in glucose homeostasis in mice models (35, 36), although evidence in humans is limited. Our results demonstrate for the first time an association between *Akkermansia* and serum FFA in a cohort of human adults from the general population. Interestingly, we were able to cluster these healthy adults into two independent groups on the basis of serum individual FFA species, one of these groups showing an increased prevalence of subclinical metabolic alterations. In our study *Akkermansia* was negatively associated mainly with saturated FFA, in line with the differences that have been previously reported by other authors between lard-like and fish oil-enriched diets in the gut microbiota of mice (36, 37). Moreover, differences in *Akkermansia* abundance found in our study between FFA clusters were associated in turn with striking differences in total FFA levels. This observation supports a gradual relationship between *Akkermansia* abundance and FFA serum levels and therefore aligns with the concept proposed by other authors (36, 38) that a "threshold" for *Akkermansia* levels may exist. In line with this hypothesis, below certain levels of *Akkermansia* abundance, gut barrier, and other functions developed by this microbe may become insufficient, thus promoting a shift from a healthy toward a pathologic-prone status. As previously commented, different parameters were found to be predictors of total FFA serum levels depending on the *Akkermansia* levels. Therefore, low *Akkermansia* levels may render the host metabolism more sensitive to a number of factors which can lead to an imbalanced FFA profile and, potentially, altered metabolism, hence supporting this notion.

*Akkermansia* is known to reside within the mucus layer of the intestine, thus contributing to strengthen the intestinal wall (34). Then, it is feasible that decreased *Akkermansia* levels may lead to a compromised barrier function and increased gut permeability, hence promoting metabolic endotoxemia, which has been related to the development of obesity and associated disorders (35, 39). Interestingly, we have found that decreased *Akkermansia* levels were associated with elevated IL-6 serum concentrations and impaired FFA profile, and the subpopulation with such profile exhibited a bias toward the enrichment in pro-inflammatory and saturated fatty acids, as well as an increased prevalence of metabolic disturbances, thus supporting those previous findings. Therefore, these results suggest that *Akkermansia* may be linked to the inflammatory milieu by modulating the FFA profile in the host.

Previous studies have revealed reduced levels of *Akkermansia* in patients suffering from inflammatory bowel disease or other metabolic impaired conditions (40, 41), although some controversy exist on the role of *Akkermansia* in these disorders. Interestingly, a recent study reported decreased counts of *Akkermansia* in pre-diabetic individuals (42), suggesting an early involvement of this microorganism in metabolic disorders. Thus, we decided to focus on subclinical metabolic features (suggestive of an impaired metabolism), which can be detected in healthy subjects, in order to improve our knowledge on the potential role of *Akkermansia* as an early marker of impaired metabolic conditions in the general population. These metabolic features were related to an increased risk of metabolic complications in the long-term (43–48). Thus, these subjects showing such subclinical alterations could be classified as "at risk" population according to the scientific literature. Based on this approach, it may be hypothesized that imbalanced microbial populations can underlie the subclinical stage of some metabolic conditions. Actually, the lack of association in our study between *Akkermansia* and other clinical hard end-points, such as obesity or related immune mediators, such as MCP-1 or IFNγ, may support a very early role of these features. However, long-term studies are warranted.

Importantly, we found an opposite behavioral pattern between *Akkermansia* and *Lactobacillus* groups, the latter being overrepresented in subjects with impaired FFA profile and increased prevalence of metabolic traits. Although classically regarded as "beneficial" microbes, some recent evidence highlights a positive association between *Lactobacillus* and BMI (49, 50) as well as with some pathological outcomes (51–53). In fact, a slight increase in the relative abundance of *Lactobacillales* has been reported in the context of the inflammatory condition of Behçet syndrome (54), related to an altered SCFA production and aberrant immune responses. Thus, these pieces of evidence point to a *Lactobacillus* within-group heterogeneity with relevance for the human health, as suggested by other authors (40, 55). In agreement with this hypothesis our findings on the interactions between *Lactobacillus* abundance and nutrient parameters, evidenced different association patterns as depending on the metabolic "health" status of the subject. Controversy observed in mice studies and clinical interventions in humans is also consistent with this idea. A recent work shows opposite patterns upon mucin usage as substrate by *Lactobacillus* and *Akkermansia*

(56), thereby suggesting that trophic interactions may underlie, at least in part, these opposite trends of both microorganisms. Actually, a negative effect on *Akkermansia* levels in the gut upon administration of a probiotic mixture was observed when *Lactobacillus*, but not other probiotic bacteria, were added to the mixture (57). Conversely, *Akkermansia* has been shown to increase the production of the antimicrobial peptide RegIIIγ by colonic epithelial cells (35). This peptide specifically targets Gram-positive bacteria, thus potentially accounting for the opposite trends between *Akkermansia* and *Lactobacillus* groups. In sum, it is feasible that *Akkermansia* may modulate the gut environment and some intestinal microbial populations through several mechanisms (58).

human health may be suggested, SCFA production playing an important role.

It has been reported that *Akkermansia* and *Lactobacillus* exhibited diverging trends in twins discordant for metabolic syndrome (40). Our results herein presented are in line with these findings and suggest that the altered composition of the intestinal microbiota may be found in subclinical stages in healthy subjects. A recent study in mice revealed that whereas the colonization by *Akkermansia muciniphila* shifted the intestinal mucosa gene expression profile toward gene pathways involved in immune tolerance and metabolic homeostasis, colonization by *Lactobacillus plantarum* resulted in an overexpression of genes involved in the metabolism of fatty acids, lipoprotein lipase being one of the most up-regulated genes (59). An enhanced enzymatic activity of this enzyme may contribute to explain the striking increase in serum FFA found in association with increased *Lactobacillus* abundance in our study. Actually, overexpression of lipoprotein lipase gene has been related to fatty acid accumulation and insulin resistance (60, 61). On the other hand, the balanced immune responses promoted upon *Akkermansia* colonization may also account for the increased IL-6 levels found in our study in subjects with diminished abundance of *Akkermansia*. This is also in accordance with the positive effect on the induction of regulatory T cells reported in mice administered *Akkermansia* (62). It must be noted that IL-6 can promote a number of pleiotropic functions other than triggering inflammation. However, since we have focused our analysis in subjects with no previous diagnosis of chronic or immune-mediated conditions, our approach allowed us to gain insight into the relationships between *Akkermansia*, IL-6 and low-grade inflammation in healthy states. Nevertheless, these associations cannot be directly translated into pathological frameworks. Therefore, a study of the gut microbiota composition— IL-6 axis in disease stages remains to be elucidated.

Modulation by *Akkermansia* of genes involved in lipid metabolism seems to be mediated, at least in part, by the production of SCFA (63). Imbalanced microbial populations can have an impact on SCFA production, which can elicit different responses in the host metabolism and immune system (64). Interestingly, several authors have found increased SCFA in obesity and metabolic syndrome [reviewed in Ref. (65)], acetate and propionate being especially relevant in such alterations (66, 67). The associations found in our study between acetate and IL-6 and that of the total SCFA with FFA serum levels strongly support the cross talk between gut microbiota, host metabolism and immune networks, highlighting a role for SCFA as important elements of this interplay.

The present study has a number of limitations that can be remarked. On the one hand, we have performed a targeted analysis of the gut microbial composition instead of a global profiling by 16S rRNA gene sequences analysis. As a consequence, whether a reduced diversity underlies the present findings cannot be concluded. Similarly, a number of selected FFA and the major SCFA species were chosen for the analysis, based on their outmost relevance in several biological processes. In addition, the same concerns apply to the analysis of inflammatory mediators. Finally, a more precise characterization of dietary intakes may be needed to precisely account for the exact contribution of short-term nutrition to the FFA levels.

In conclusion, we reported an association between some intestinal microbial populations and the characteristics of the FFA profile in healthy middle-aged subject. *Akkermansia* and *Lactobacillus* groups seem to be connected to the health metabolic status of the host, the interplay with nutritional parameters playing a potential role (**Figure 4**). Finally, our findings provide some evidence on the role of SCFA as mediators of the cross talk in the gut microbiota-host lipid metabolism axis. Although our results point to a very early role of an altered microbial composition in the further development of metabolic disorders, prospective and long-term studies are needed to accurately address this possibility.

#### AUTHOR CONTRIBUTIONS

All the authors listed made substantial contributions to the design of the work, analysis, or interpretation of the results obtained;

#### REFERENCES


participated in the study design and data interpretation, reviewed the manuscript, and approved the final version; and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. JR-C and NS performed most of the experimental procedures. SG was involved in the nutritional assessments, anthropometrical measurements, and collection of samples. AM, MG, CR-G, and AS provided biological samples and financial support. JR-C and AS drafted the manuscript.

#### ACKNOWLEDGMENTS

The authors acknowledge the excellent technical assistance of Ana M. Hernández-Barranco (IPLA) with the analysis of SCFA in fecal samples as well as of the staff of the scientific core facilities from the University of Oviedo (Unidad de Espectrometría y Espectrofotometría, Servicios Científico-Técnicos, Universidad de Oviedo). The authors also show their deepest gratitude to all the study volunteers. The authors acknowledge support of the publication fee by the CSIC Open Access Publication Support Initiative through its Unit of Information Resources for Research (URICI).

#### FUNDING

This work was funded through the Grant GRUPIN14-043 "Microbiota Humana, Alimentación y Salud" funded by "Plan Regional de Investigación del Principado de Asturias," Asturias, Spain and by the grants AGL2010-14952 from Spanish "Plan Nacional I+D+I" and by Biopolis SL within the framework of the e-CENIT Project SENIFOOD from the Spanish Ministry of Science and Innovation. JR-C and NS benefit from postdoctoral contracts supported by the Grant GRUPIN14-043 and by a Clarín regional contract cofinanced by the Marie Curie CoFund European Program, respectively. Regional grants received cofounding from European Union FEDER funds.


lupus erythematosus: a pilot study. *JPEN J Parenter Enteral Nutr* (2011) 35:198–208. doi:10.1177/0148607110386378


**Conflict of Interest Statement:** The authors declared no potential competing financial interests concerning this study. Funders had no role in study conception, design, analysis of the results, or decision to publish.

*Copyright © 2017 Rodríguez-Carrio, Salazar, Margolles, González, Gueimonde, de los Reyes-Gavilán and Suárez. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Dietary and Microbial Metabolites in the Regulation of Host Immunity

Naoko Shibata1,2,3, Jun Kunisawa1,3,4,5 \* and Hiroshi Kiyono1,6

<sup>1</sup> Division of Mucosal Immunology, Department of Microbiology and Immunology and International Research and Development Center for Mucosal Vaccines, The Institute of Medical Science, The University of Tokyo, Tokyo, Japan, <sup>2</sup> Department of Mucosal Immunology, School of Medicine, Chiba University, Chiba, Japan, <sup>3</sup> Laboratory of Vaccine Materials and Laboratory of Gut Environmental System, National Institutes of Biomedical Innovation, Health and Nutrition, Osaka, Japan, <sup>4</sup> Graduate School of Medicine, Graduate School of Pharmaceutical Sciences, Graduate School of Dentistry, Osaka University, Osaka, Japan, <sup>5</sup> Department of Microbiology and Infectious Diseases, Kobe University Graduate School of Medicine, Kobe, Japan, <sup>6</sup> Department of Immunology, Graduate School of Medicine, Chiba University, Chiba, Japan

Mucosal surfaces in the body, especially the intestine, are constantly exposed to trillions of microbiomes. Accumulating evidence has revealed that changes in the composition of the gut microbiome, especially that of the commensal bacteria population, are frequently associated with immunologic disorders. These changes coincide with changes in the production of certain dietary metabolites. Recent studies have uncovered the molecular and cellular mechanisms underlying the relationships among diet, commensal bacteria, and the host immune system. In this review, we describe how dietary and microbial metabolites modulate host immunity.

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Ricardo Silvestre, Instituto de Pesquisa em Ciências da Vida e da Saúde (ICVS), Portugal Helder Nakaya, University of São Paulo, Brazil

> \*Correspondence: Jun Kunisawa

kunisawa@nibiohn.go.jp

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 31 July 2017 Accepted: 23 October 2017 Published: 07 November 2017

#### Citation:

Shibata N, Kunisawa J and Kiyono H (2017) Dietary and Microbial Metabolites in the Regulation of Host Immunity. Front. Microbiol. 8:2171. doi: 10.3389/fmicb.2017.02171 Keywords: microbiome, metabolite, fatty acid, vitamin

## INTRODUCTION

Mucosal surfaces, especially that of the gastrointestinal tract, are constantly exposed to a wide variety of antigens including trillions of bacteria and fungi, as well as dietary components and their metabolites. The host immune system discriminates between harmful and beneficial antigens, simultaneously inducing immune responses to exclude harmful antigens while tolerating beneficial antigens to establish appropriate homeostatic conditions in the gut.

It has long been recognized that commensal bacteria regulate host responses, including immunity, but, due to the difficulty of culturing the intestinal microbiome, it has been challenging to obtain information. In addition, because dietary and microbial metabolites are generated in a complex network that includes the diet, host, and intestinal microbiome, it is difficult to analyze such metabolites. However, recent advances in genome-based analysis of bacteria has enabled direct analysis of the intestinal microbiome without culturing; this analysis has revealed that changes in the composition of the gut microbiome are associated with immunologic disorders and diseases (Browne et al., 2017). In addition, advances in high throughput metabolomics have allowed immunologists to investigate metabolites generated in the intestine and to discover that these changes in the microbiome composition coincide with changes in either useful or harmful metabolites involved in the regulation of immune responses. For example, commensal bacteria are

**Abbreviations:** AA, arachidonic acid; DC, dendritic cell; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; FAO, fatty acid oxidation; FP, formylpterin; GPR, G-protein-coupled receptor; IEC, intestinal epithelial cell; IL, interleukin; LCFA, long-chain fatty acid; MAIT, mucosal-associated invariant T cell; MR1, MHC-like protein 1; PD1, Protectin D1; PPAR, peroxisome proliferator-activated receptor; Rv, resolving; SCFA, short-chain fatty acid; TCA, tricarboxylic acid; TLR, toll-like receptor; TNF, tumor necrosis factor; TX, Thromboxane.

involved in the extraction, synthesis, and absorption of many nutrients and metabolites including short- and LCFAs, and vitamins (Hirata and Kunisawa, 2017).

In this review, we describe recent findings regarding the role of the gut microbiome–diet interaction in the generation of immunologically active metabolites.

## LONG-CHAIN FATTY ACIDS IN THE CONTROL OF IMMUNE RESPONSES

Fatty acids are an essential nutrient that is mainly obtained from the diet. Generally, dietary oils are composed of various LCFAs. LCFAs primarily serve as a source of energy and membrane components, but they are also actively involved in the regulation of immune responses by being metabolized into lipid metabolites (**Figure 1**) (Hirata and Kunisawa, 2017).

Studies to date have mainly focused on the lipid metabolites generated after absorption into the body. Among LCFAs, ω3 and ω6 LCFAs are essential FAs, meaning that they are not generated by mammals, including humans (Hirata and Kunisawa, 2017). These essential FAs are metabolized into bioactive lipid mediators through reactions that are mediated by several series of oxidative enzymes, such as cyclooxygenases, lipoxygenases, and cytochrome P450 monooxygenases (Arita, 2012). It is thought that ω3 LCFAs have anti-allergic and anti-inflammatory properties, whereas ω6 LCFAs have proinflammatory properties (Arita, 2012; Serhan, 2017). AA and linoleic acid are the major ω6 LCFA, whereas α-linolenic acid abundantly present in linseed or perilla oil, EPA and DHA abundantly present in fish oils are typical ω3 LCFAs. TX B2, an AA derivative, is a potent vasoconstrictor and platelet activator, whereas TXB3, an EPA derivative, has barely any physiological effects. Dietary supplementation with EPA has been shown to reduce TXB<sup>2</sup> levels in plasma and cellular fatty acid and to induce a shift to less reactive platelets and to reduce blood pressure in response to pressor hormones (Weber et al., 1986). In another study, dietary supplementation with ω3 LCFAs suppressed the ability of monocytes to synthesize inflammatory cytokines, such as interleukin-1 (IL-1) and TNF, which was accompanied by a decreased ratio of AA to EPA in the membrane phospholipids of the monocytes (Endres et al., 1989). Since the conversion of ω3 and ω6 LCFAs to bioactive lipid mediators shares the same series of enzymes, competition exists between AA and EPA/DHA for metabolism. This competition is one of the mechanisms through which EPA inhibits the inflammatory properties of AA.

In addition to the competition between ω3 and ω6 LCFAs, some lipid metabolites of ω3 LCFAs exhibit anti-inflammatory or anti-allergy effects (Arita, 2012; Kunisawa et al., 2015a). E-series Rv are derived from EPA, whereas D-series Rv and Protectin D1 (PD1) are derived from DHA. RvD1, RvD2, RvE1, and PD1 activate inflammation resolution programs by reducing levels of pro-inflammatory cytokines, including IL-6, IL-1β, IL-23, and TNFα, and neutrophil influx, and by promoting macrophage phagocytosis of apoptotic cells and inflammatory debris (Schwab et al., 2007; Spite et al., 2009; Chiang et al., 2012).

In addition to these studies, we previously reported that allergic diarrhea was ameliorated when mice were maintained on linseed oil rich in ω3 α-linolenic acid (Kunisawa et al., 2015a). The α-linolenic acid is metabolized to eicosapentaenoic acid, levels of which also increased in the intestine of mice maintained on linseed oil. When we performed lipidomics analysis, we identified 17,18-epoxy-eicosatetraenoic acid as an anti-allergic lipid metabolite derived from eicosapentaenoic acid (Kunisawa et al., 2015a). Given that synthetic 17,18-epoxy-eicosatetraenoic acid administration is sufficient to inhibit allergic diarrhea, this lipid metabolite appears to be effective in the control of intestinal allergy.

Although these lipid metabolites seem to be generated in the body, several lines of evidence indicate that commensal bacteria also express enzymes that participate in LCFA metabolism (**Figure 1**). Indeed, germ-free animals exhibit alterations of composition in their lipid metabolites (Claus et al., 2008; Martin et al., 2009; Chiang et al., 2012). In this regard, we identified conjugated linoleic acids, oxo FAs, and hydroxy FAs as microbial lipid metabolites (Kishino et al., 2013). These metabolites are formed in the intestine through the action of commensal bacteria, especially Lactobacillus plantarum (Kishino et al., 2013). Indeed, these lipid metabolites are abundant in the intestinal lumen of specific pathogen-free mice but are scarce in germ-free mice (Kishino et al., 2013). Of note, these lipid metabolites are detected not only in the intestinal lumen but also in the serum, suggesting that these microbial lipid metabolites act both in the intestinal lumen and elsewhere in the body.

Administration of 10-hydroxy-cis-12-octadecenoic acid, one of the microbial lipid metabolites, has been reported to ameliorate experimental colitis by enhancing tight junctions on epithelial cells. Signaling though LCFA receptors is mediated by GPR40, which suppresses TNFR2 gene expression and NF-κB via the MEK-ERK pathway (Miyamoto et al., 2015). LCFAs are also recognized by PPARs, which also modulate immune reactions and allergic diseases (Ricote et al., 1998; Fukui et al., 2009). For example, it was reported that probiotic bacteria in the intestine produce conjugated linoleic acids, which target PPAR-γ in macrophages to suppress the inflammatory response (Bassaganya-Riera et al., 2012). In other studies, Bacteroides thetaiotaomicron activated PPAR-γ, which decreased NF-κB–dependent IL-8 production (Kelly et al., 2004), and Enterococcus faecalis activated PPAR-γ1 in IECs thereby increasing the production of IL-10 in human newborn babies (Are et al., 2008). Although it remains unclear which types of lipids act as ligands for these PPRA receptors, these microbial lipids and lipid metabolites may prevent allergy and inflammation.

As mentioned before, RvD1, RvD2, RvE1, and PD1 participate in the activation of inflammation resolution programs (Schwab et al., 2007; Spite et al., 2009; Chiang et al., 2012). It is important to note that commensal bacteria are also involved in this process (Chiang et al., 2012). Given that germ-free mice have increased levels of endogenous RvD1 and PD1 in their colon (Chiang et al., 2012), these reports suggest that microbial suppression of RvD1 and PD1 production may be involved in the regulation of the host immune defense against invading pathogens.

#### SHORT-CHAIN FATTY ACIDS IN THE CONTROL OF IMMUNE RESPONSES

Short-chain fatty acids are present at high concentrations in the intestine as bacterial fermentation products of dietary indigestible polysaccharides such as cellulose (**Figure 2**) (Wong et al., 2006). Because mammals lack the enzymes to degrade polysaccharides, germ-free mice exhibit remarkably decreased amounts of SCFAs and increased amounts of indigestible oligosaccharide, a bacterial fermentation substrate (Hoverstad and Midtvedt, 1986).

Among SCFAs, acetate, propionate, and butyrate have been well studied. Accumulated evidence indicated that these SCFAs modify several cellular processes including gene expression, chemotaxis, differentiation, proliferation, and apoptosis, which affect various biological responses including the immune response (**Figure 2**) (Correa et al., 2016). Generally, SCFAs are recognized on the cell surface by G-protein coupled receptors (GPRs) such as GPR41, GPR43, GPR109A, and Olfr78 (Bolognini et al., 2016). SCFAs are also transported by monocarboxylate transporter-1 and the sodium-dependent monocarboxylate transporter-1 and by passive diffusion across the plasma membrane into the cytoplasm (Thwaites and Anderson, 2007).

Because SCFAs are found at high concentrations in the intestine, they are in direct contact with the IECs. IECs take up SCFAs through both passive and active mechanisms into their cytosol, where the SCFAs, especially butyrate, are used as a source of ATP for energy metabolism (den Besten et al., 2013). In addition to their role in energy metabolism, SCFAs enhance some immune surveillance functions of IECs by increasing the expression of certain antimicrobial peptides; for example, butyrate increases the expression of LL-37 and CAP-18 (Raqib et al., 2006), and modulating cytokine (CCL20 and IL-8) production (Iraporda et al., 2015). In addition, the activation of GPR43 and GPR109a by SCFAs in IECs reportedly resulted in an increase in the production of IL-18, a cytokine involved in the maintenance of epithelial integrity (Singh et al., 2014; Macia et al., 2015).

Short-chain fatty acids are also known to enhance the induction of Treg cells in the intestine. A key molecule involved in this event is the butyrate receptor GPR109a expressed on DCs. Indeed, in one study, GPR109a-deficient mice were found

to have reduced numbers of Treg cells (Singh et al., 2014). In another study, SCFAs were found to directly affect the preferential differentiation of T cells to Treg cells with concurrent enhancement of histone H3 acetylation in the Foxp3 locus by butyrate, or the GPR43-mediated pathway (Furusawa et al., 2013; Smith et al., 2013). Consistent with these findings, mice lacking either GPR43 or GPR109a show exacerbated symptoms in food-allergy models (Tan et al., 2016). In addition to GPR43 and GPR109a, Slc5a8, a Na+-coupled high-affinity transporter for SCFAs such as butyrate, plays an important role in the increased expression of indoleamine 2,3-dioxygenase 1 and aldehyde dehydrogenase in DCs, which results in the preferential induction of FoxP3<sup>+</sup> Treg cells (Gurav et al., 2015).

Besides their effects in the intestine, SCFAs exhibit several anti-allergic properties after absorption into the body. For example, mice fed a high-fiber, but not mice fed a low-fiber, showed increased levels of SCFAs in the blood and were protected against allergic inflammation in the lung (Trompette et al., 2014). In this case, propionate enhanced the generation of macrophage and DC precursors from bone marrow cells, which is dependent

whereas butyrate inhibits the IgA class switching of B cells. Propionate enhances the generation of regulatory type of macrophages and DCs from bone marrow cells.

on GPR41 but not GPR43. DCs and macrophages generated in this process are highly phagocytic but impaired in their ability to promote the effector function of Th2 cells, thereby blocking allergic responses (Trompette et al., 2014).

Short-chain fatty acid are also involved in the production of intestinal IgA. The acetate-GPR43 axis positively regulates this process. Indeed, supplementation of acetate promotes intestinal IgA production in a GPR43-dependent manner (Wu et al., 2017). As an underlying mechanism, it was demonstrated that acetate induces the expression of aldehyde dehydrogenase in DCs, which converts vitamin A to retinoic acid for the promotion of IgA production (Wu et al., 2017). In contrast to the positive effects of acetate on intestinal IgA production, butyrate has been reported to suppress class switching to IgA through the upregulation of miR-155, -181b, -361, -23b, -30a, and -125b in B cells, which silence AID and Blimp-1 (White et al., 2014). These findings collectively suggest that a balance among SCFAs controls the immunologic status quo in both humoral and cellular immunity.

Short-chain fatty acids regulate IEC functions controlled by TLRs. IECs express several TLRs, including TLR2/1, TLR4, TLR5, and TLR9, which mainly recognize bacterial lipoproteins, lipopolysaccharides, flagellin, and DNA, respectively (de Kivit et al., 2014). SCFAs have been reported to modify immune responses by altering TLR-induced inflammatory gene expression through the inhibition of histone deacetylases (Lin et al., 2015). For example, when IECs were incubated with butyrate or propionate and then stimulated with flagellin, a ligand for TLR5 (Oh et al., 2014), the expression levels of proinflammatory cytokines such as TNF-α were upregulated, whereas those of chemotactic chemokines, such as IL-8 and monocyte chemotactic protein-1, were downregulated (Lin et al., 2015). A similar regulation pathway, mediated by TLR ligands and SCFAs, has been observed in hematopoietic cells (Mirmonsef et al., 2012). Given that many TLR ligands are present in not only pathogenic but also commensal bacteria and that SCFAs are produced by these bacteria via fermentation, the combination of TLR ligands and SCFAs seems to be an essential component of one of the commensalmediated immune regulation pathways. Of note, flagellininitiated TLR5 stimulation plays essential roles in the induction of immune responses against trivalent inactivated influenza vaccine-induced antibody responses (Oh et al., 2014). These reports indicate that the effectiveness of vaccination varies depending on the balance between commensal bacteria carrying flagellins and commensal bacteria contributing to the synthesis of SCFAs.

#### VITAMIN B FAMILY IN THE CONTROL OF IMMUNE RESPONSES

Mammals do not have biosynthetic pathways for vitamins and therefore must obtain vitamins externally. Of course, diets are a general source of vitamins, but commensal bacteria also produce vitamins and simultaneously consume dietary vitamins. Therefore, both diet and commensal bacteria determine the vitamin contents in the intestine (**Figure 3**).

Vitamins are required for the maintenance of many biological responses by acting as antioxidants, transcription factors, and cofactors for metabolic enzymes in the generation, conversion, and digestion of fatty acids, nucleotides, carbohydrate, and amino acids. Immune cells require all of these processes for their development, differentiation, and activation and therefore vitamin deficiency is frequently associated with increased risk of infectious, allergic, and inflammatory diseases (**Figure 3**) (Kunisawa and Kiyono, 2015; Suzuki and Kunisawa, 2015).

Vitamin B3 (nicotinic acid) supplementation acts on monocytes to dampen TLR2- and TLR4-induced release of inflammatory mediators such as TNF-α, IL-6, and monocyte chemotactic protein-1 (Digby et al., 2012). These effects are mediated by GPR109a, a vitamin B3 receptor expressed on monocytes (Digby et al., 2012). Vitamin B3 also reduces macrophage production of pro-inflammatory cytokines, including IL-1, IL-6, and TNF-α, in a murine model of atherosclerosis (Lipszyc et al., 2013). Collectively, these findings suggest that vitamin B3 exerts its anti-inflammatory properties by modulating immune cells.

The vitamin B complex also contributes to energy metabolism. Currently, the metabolic processes in immune cells are recognized as the emerging field of immunometabolism (Buck et al., 2017; Mills et al., 2017). The core function of metabolic pathways is the synthesis or degradation of sugars, fatty acids, nucleic acids, or proteins, coupled to the consumption or generation of ATP by oxidative phosphorylation or glycolysis. Changes in these metabolic pathways are frequently associated with immune cell functions such as cell proliferation and the production of cytokines, chemokines, and antibodies (Buck et al., 2017; Mills et al., 2017). Recent advances in our understanding of immunometabolism have revealed that quiescent or regulatory-type cells such as naïve T and B cells, Treg cells, and M2 macrophages use anabolic pathways for energy generation from the TCA cycle, such as FAO, whereas activated or inflammatory cells (e.g., Th1, Th2, Th17, IgA-producing plasma cells, M1 macrophages) use catabolic pathways and shift to glycolysis for energy generation (Buck et al., 2017; Kunisawa, 2017; Mills et al., 2017).

In metabolic energy pathways, vitamin B2 and its active forms (e.g., flavin adenine dinucleotide) function as cofactors for various enzymatic reactions such as the TCA cycle and FAO (Huskisson et al., 2007). FAO is necessary for the generation of acetyl-CoA, which enters the TCA cycle in mitochondria to produce energy. Studies using rodent models have shown that vitamin B2 deficiency reduces the activity of acyl-CoA dehydrogenases, which participate in the dehydrogenation step of FAO, and that vitamin B2 supplementation rescues the activity of these enzymes (Sakurai et al., 1982). As with vitamin B2, vitamin B1 (also known as thiamine) and its derivatives (e.g., thiamine pyrophosphate) acts as a cofactor for several enzymes such as pyruvate dehydrogenase and α-ketoglutarate dehydrogenase that are involved in TCA cycle (Frank et al., 2007). In agreement with the importance of energy metabolism in immune system, mice maintained on diets lacking the vitamin B complex show impaired immunity (Kunisawa et al.,

FIGURE 3 | Modification of energy metabolism and immune responses by vitamins. Vitamins are involved in the maintenance of immunological homeostasis in the gut through the regulation of energy metabolism, cellular function, and differentiation. Glucose enters the cell and is metabolized to pyruvate, which is then converted to lactate (glycolysis) or enters the TCA cycle for energy generation. In the TCA cycle, citrate can be transferred out of the mitochondria into the cytoplasm, where it is used for fatty acid synthesis (FAS). Fatty acids can be degraded in the mitochondria and enter the TCA cycle to produce energy (FAO). Vitamins B1 and B2 act as cofactors for enzymes involved in the TCA cycle. Vitamin B2 also acts as a cofactor for enzymes involved in FAO. Vitamin B3 inhibits the production of inflammatory cytokines such as TNF-α, IL-6, IL-1, MCP-1 from macrophages and monocytes. Vitamins B2 and B9 are metabolized by commensal bacteria and converted to 6-hydroxymethyl-8-D-ribityllumazine and 6-FP, respectively. 6-hydroxymethyl-8-D-ribityllumazine activates MAIT cells, whereas 6-FP binds to MR1 without activating MAIT cells.

2015b; Suzuki and Kunisawa, 2015; Hosomi and Kunisawa, 2017).

## CONCLUSION

In addition to their direct effects on immune cells and energy metabolism, microbial metabolites of some B vitamins act as ligands for immune cells, especially mucosal-associated invariant T (MAIT) cells. Activated MAIT cells produce IL-17 and IFN-γ and help exclude infectious bacteria (Bourhis et al., 2013). Because MAIT cells are found in several inflammatory tissues in brain and kidney cancer patients, accompanied by reduced numbers of MAIT cells in the circulating blood, it is thought that MAIT cells infiltrate the inflammatory tissues and contribute to inflammatory disease (Bourhis et al., 2013). MAIT cells are innate-like T cells that recognize MHC-like protein 1 (MR1) restricted presentation. Intriguingly, the microbial metabolite of vitamin B2 (riboflavin) 6-hydroxymethyl-8-D-ribityllumazine has been shown to activate MAIT cells (Kjer-Nielsen et al., 2012; Patel et al., 2013). Similarly, a microbial vitamin B9 metabolite, 6-formyl pterin (6-FP), has been shown to bind to MR1; however; unlike 6-hydroxymethyl-8-D-ribityllumazine, 6-FP cannot activate MAIT cells (Kjer-Nielsen et al., 2012; Patel et al., 2013). In line with these findings, a previous study suggests that acetyl-6-FP, an analog of 6-FP, acts as antagonist of MR1 in the inhibition of MR1-dependent MAIT cell activation (Eckle et al., 2014). Taking into account that the kinds and levels of biosynthesized B vitamins differ among bacteria (Magnusdottir et al., 2015), these findings collectively suggest that the balance between the amount of vitamin B2 and B9 as well as the type of bacteria and its metabolism are critical factors in determining MAIT cell activation.

There are many immune cells in the intestine, where biological communication between dietary components and microorganisms produces numerous kinds of metabolites. These dietary or microbial metabolites affect cell composition and function and alter energy metabolism. Accumulating evidence has revealed the importance of these changes in cell composition, function, and energy metabolism in the control of host immune responses and the subsequent incidence of inflammatory, allergic, and infectious diseases. It should be noted that not all of the results obtained in the rodent models can be directly translated to humans, because their diets and commensal bacteria differ greatly. In this context, it is important to utilize cohort studies and clinical data that include the effects of antibiotic treatment. In addition, it will be important to apply novel and/or improved technologies such as single-cell omics, whole-genome sequencing, nontarget metabolomics, and bioinformatics in future research. Understanding the integrated mechanisms by which diet, microbiota, and metabolites influence the function of the immune system would be an interesting topic for future study that could provide new strategies for the control of immune and infectious diseases.

## AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### ACKNOWLEDGMENTS

fmicb-08-02171 November 5, 2017 Time: 18:19 # 7

This review article contains results obtained from our studies that were supported at least in part by grants from the Core Research for Evolutional Science and Technology Program of the Japan Agency for Medical Research and Development (HK); by the Ministry of Education, Culture, Sports, Science, and Technology of Japan (Grants-in-Aid for Scientific Research S [HK; 23229004], for Scientific Research B [JK; 26293111], for Scientific Research on Innovative Areas [JK; 16H01373], for Young Scientists B [NS; 17K17686], and for a JSPS Research Fellow [NS; 17J07480]); by the Practical Research

#### REFERENCES


Project for Allergic Diseases and Immunology (Research on Allergic Diseases and Immunology) and for the Research on Development of New Drugs (Adjuvant Database Project) from the Japan Agency for Medical Research and Development (AMED; JK, and HK) and by the Ministry of Health, Labour and Welfare (MHLW) (JK); by the Science and Technology Research Promotion Program for Agriculture, Forestry, Fisheries, and Food Industry (JK); and by the Astellas Foundation for Research on Metabolic Disorders (JK), the Public Health Research Foundation for Public Health Science (NS), the Terumo Foundation for Life Science and Arts (JK), and the Suzuken Memorial Foundation (JK).

tumor necrosis factor by mononuclear cells. N. Engl. J. Med. 320, 265–271. doi: 10.1056/NEJM198902023200501



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Shibata, Kunisawa and Kiyono. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Amino Acids As Mediators of Metabolic Cross Talk between Host and Pathogen

*Wenkai Ren1,2, Ranjith Rajendran3 , Yuanyuan Zhao1 , Bie Tan4 , Guoyao Wu5 , Fuller W. Bazer5 , Guoqiang Zhu2 \*, Yuanyi Peng6 , Xiaoshan Huang7 , Jinping Deng1 \* and Yulong Yin1 \**

*1Guangdong Provincial Key Laboratory of Animal Nutrition Control, Institute of Subtropical Animal Nutrition and Feed, College of Animal Science, South China Agricultural University, Guangzhou, China, 2Jiangsu Co-Innovation Center for Important Animal Infectious Diseases and Zoonoses, Joint International Research Laboratory of Agriculture and Agri-Product Safety of Ministry of Education of China, College of Veterinary Medicine, Yangzhou University, Yangzhou, China, 3School of Medicine, College of Medical, Veterinary and Life Sciences (MVLS), University of Glasgow, Glasgow, United Kingdom, 4Laboratory of Animal Nutrition and Health and Key Laboratory of Agro-Ecology, Institute of Subtropical Agriculture, The Chinese Academy of Sciences, Changsha, China, 5Department of Animal Science, Texas A&M University, TAMU, College Station, TX, United States, 6Chongqing Key Laboratory of Forage & Herbivorce, College of Animal Science and Technology, Southwest University, Chongqing, China, 7Changsha Medical University, Changsha, China*

#### *Edited by:*

*Yves Renaudineau, Université de Bretagne Occidentale, France*

#### *Reviewed by:*

*Ricardo Silvestre, Instituto de Pesquisa em Ciências da Vida e da Saúde (ICVS), Portugal Miguel Prudêncio, Instituto de Medicina Molecular (IMM), Portugal*

#### *\*Correspondence:*

*Guoqiang Zhu yzgqzhu@yzu.edu.cn; Jinping Deng dengjinping@scau.edu.cn; Yulong Yin yinyulong@isa.ac.cn*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 08 November 2017 Accepted: 05 February 2018 Published: 27 February 2018*

#### *Citation:*

*Ren W, Rajendran R, Zhao Y, Tan B, Wu G, Bazer FW, Zhu G, Peng Y, Huang X, Deng J and Yin Y (2018) Amino Acids As Mediators of Metabolic Cross Talk between Host and Pathogen. Front. Immunol. 9:319. doi: 10.3389/fimmu.2018.00319*

The interaction between host and pathogen decidedly shapes the outcome of an infection, thus understanding this interaction is critical to the treatment of a pathogen-induced infection. Although research in this area of cell biology has yielded surprising findings regarding interactions between host and pathogen, understanding of the metabolic cross talk between host and pathogen is limited. At the site of infection, host and pathogen share similar or identical nutritional substrates and generate common metabolic products, thus metabolic cross talk between host and pathogen could profoundly affect the pathogenesis of an infection. In this review, we present results of a recent discovery of a metabolic interaction between host and pathogen from an amino acid (AA) metabolism-centric point of view. The host depends on AA metabolism to support defensive responses against pathogens, while the pathogens modulate AA metabolism for its own advantage. Some AA, such as arginine, asparagine, and tryptophan, are central points of competition between the host and pathogen. Thus, a better understanding of AA-mediated metabolic cross talk between host and pathogen will provide insight into fruitful therapeutic approaches to manipulate and prevent progression of an infection.

#### Keywords: amino acids, arginine, asparagine, metabolism, infection

**163**

**Abbreviations:** AA, amino acids; ADI, arginine deiminase; ADR, arginine deprivation response; AhR, aryl hydrocarbon receptor; ALR, AIM2 (absent in melanoma 2)-like receptor; AnsA, asparaginase; AnsP2, asparagine transporter; AP-1, activator protein-1; ATF, activating transcription factor; CAT, cationic amino acid transporter family; CLRs, C-type lectin receptors; DCs, dendritic cells; EHEC, enterohaemorrhagic *Escherichia coli*; ETEC, enterotoxigenic *Escherichia coli*; GABA, gammaaminobutyric acid; GAS, group A *Streptococcus*; GCN2, general control nonderepressible 2; GDAR, glutamate-dependent acid resistance; HBV, hepatitis B virus; HCMV, human cytomegalovirus; IDO, indoleamine 2,3-dioxygenase; iNOS, inducible nitric oxide synthase; IL, interleukin; MAPK, mitogen-activated protein kinase; MPK2, mitogen-activated protein kinase 2; mTORC1, mechanistic target of rapamycin complex 1; NF-κB, nuclear factor-κB; NLRs, nucleotide-binding oligomerization domain-like receptors; NO, nitric oxide; NOS, NO synthase; PAMPs, pathogen-associated molecular patterns; PCV2, porcine circovirus type 2; PERK, protein kinase-like endoplasmic reticulum kinase; PRRs, pathogen recognition receptors; ROS, reactive oxygen species; RLRs, retinoic acid-inducible gene-I-like receptors; SCV, *Salmonella*-containing vacuole; SNAT2, sodium-dependent neutral amino acid transporter 2; STAT, signal transducer and activator of transcription; TLRs, toll-like receptors.

#### INTERACTION BETWEEN HOST AND PATHOGEN

The interaction between host and pathogen has a profound effect on the outcome of an infection. The host senses the presence of the pathogen through recognition of pathogen-associated molecular patterns, which are highly conserved. Cells of the host recognize unique molecules present on pathogens *via* pathogen recognition receptors, including toll-like receptors (TLR), C-type lectin receptors, nucleotide-binding oligomerization domain-like receptors, retinoic acid-inducible gene-I-like receptors, and AIM2 (absent in melanoma 2)-like receptor (ALR) (1, 2). Pathogen recognition by immune cells of the host results in activation of a common set of cell signaling pathways, including nuclear factor-κB (NF-κB), activator protein-1, and mitogen-activated protein kinase (MAPK). These signaling pathways modulate the immune responses of the host against the pathogen that include production of proinflammatory cytokines/chemokines, migration of neutrophils, and secretion of antibodies (1–3). However, the pathogen is usually equipped to evoke countermeasures to inhibit the host's immune responses. For example, pathogenic *Escherichia coli* inhibit the activation of NF-κB through its pathogenicity factors, such as NleH1 and NleB (4–6).

Knowledge about host–pathogen interactions is obviously critical for understanding the pathogenesis of infection; however, it generally overshadows knowledge of metabolic cross talk between host and pathogen. At the site of infection, which can be regarded as a closed system, host and pathogen share similar nutritional substrates and generate common metabolic products. The host depends upon nutritional substrates to support its immune responses against the pathogen, while the pathogen is also highly dependent on nutritional substrates for its physiology because most pathogens are unable to synthesize some nutritional substrates. For example, *Plasmodium falciparum* has completely lost its capacity for *de novo* biosynthesis of amino acids (AA), thus it depends primarily on AA scavenging from the host and through the catabolism of hemoglobin (7, 8). Usually, the host experiences significant metabolic alterations after the infection by a pathogen (9–11), and a slight change in metabolism at the site of infection will remarkably shape the outcome of an infection. For example, the host experiences a significant change in glucose metabolism to support immune responses against pathogens, such as activation of T cells and monocytes, activation of inflammasome signaling, and production of IL-1 β (12–14). The abundances of glucose and α-glucan in the host affect global gene expression of *Streptococcus suis*, including the virulence factor amylopullulanase, which promotes epithelial cell adherence (15). Fucose from the host's intestinal microbiota affects the metabolism of enterohaemorrhagic *Escherichia coli* and its expression of virulence genes for intestinal colonization (16). In essence, the host modulates the availability of nutritional substrates or metabolic products to effect the progression of pathogen-induced infection, while the pathogen uses the same or similar substrates to sense the anatomical location and the physiological status of the host to adapt (17). For example, the host achieves a metal-limited environment during infection by expressing calprotectin which chelates manganese; however, *Acinetobacter baumannii* coordinates transcription of a manganese transporter to facilitate manganese accumulation and overcome the manganese limitation resulting from expression of calprotectin (18). Indeed, there is fierce competition for trace elements and metabolic precursors between pathogen and host. Therefore, the host experiences a significant alteration in metabolism during infection, including metabolism of glucose, fatty acid, and AA (19–21). Evidence for metabolic cross talk between a pathogen and its host was highlighted in a recent review by Olive and Sassetti (17); however, a number of key areas involving AA interactions between pathogen and its host require further in-depth research. In this review, we examine metabolic interactions between host and pathogen from an AA metabolism-centric point of view.

#### AA AFFECTS THE IMMUNE SYSTEM OF THE HOST

Numerous reviews indicate that AA metabolism shapes the host's physiology, including growth, reproduction, and immunity. AA metabolism affects the physiology of the host by serving as an energy source for cells (e.g., lymphocytes, fibroblasts, and enterocytes), a basic substrate for protein synthesis, a substrate for production of regulatory molecules [e.g., nitric oxide (NO), polyamines, and creatine], a regulator for cell signaling pathways [e.g., mechanistic target of rapamycin complex 1 (mTORC1), MAPK, and NF-κB], and a regulator for host metabolism and intestinal microbiota (22, 23). Recent compelling results indicate that AA have a significant influence on immune responses of the host. For example, arginine or glutamine effect activation of the innate immune system, such as TLR signaling, secretory immunoglobulin A (SIgA), and Paneth antimicrobials, as well as activation of intestinal cell signaling pathways, such as NF-κB, MAPK, and PI3K–Akt (24–26). Indeed, glutamine promotes intestinal secretion of SIgA through the intestinal microbiota, and involving both T cell-dependent and T cell-independent pathways (27). These investigations revealed that AA influence the innate immunity of the host; however, the importance of AA in adaptive immunity during infection is not known. AA [e.g., leucine, glutamine, and gamma-aminobutyric acid (GABA)] are of critical importance in mediating T cell function, including activation and differentiation of T cells, especially for Th1 and Th17 cells (20, 28, 29). For example, extracellular serine is required for optimal T cell expansion even though the concentration of glucose is sufficient to support activation, bioenergetics, and effector functions of T cells (30). Mechanistically, the influence of AA on the immune system of the host may largely depend on mTORC1 signaling since AA-induced mTORC1 signaling is required for differentiation of Th17 cells and their expression of IL-17 (20). Collectively, available results indicate that AA are of critical importance for shaping the immune functions of the host, including innate immunity and adaptive immunity.

## AA EFFECT GROWTH AND EXPRESSION OF VIRULENCE FACTORS BY PATHOGENS

A pathogen requires AA to support its physiological functions, and alterations in AA availability have remarkable effects on growth of a pathogen and its expression of virulence factors. For example, the addition of asparagine to Dulbecco's Modified Eagle Medium promotes activation of the streptococcal invasion locus (*sil*) in Group A *Streptococcus* (GAS) (31). Dietary content of glutamine significantly affects the burden of *Pasteurella multocida* and the expression of its virulence factors (32). A high content of glutamine increases the bacterial burden in all tissues analyzed, including the heart, liver, spleen, lung, and kidney (32). Glutamine promotes the expression of virulence factors in the lung, including ompA, pm0442, pm0979, plpE, and hasR, and expression of pm0442, plpE, and hasR in the spleen (32). Similarly, glutamine acts as an on/off switch for the induction of virulence genes of *Listeria monocytogenes* (33). There is no virulence gene transcription by *L. monocytogenes* when concentrations of glutamine in macrophages are below the threshold, while there is maximum transcription of virulence genes when concentrations of glutamine exceed threshold concentrations of glutamine (33). Inactivation of GlnPQ (a l-glutamine high affinity ABC transporter) results in complete arrest of glutamine uptake, a dramatic reduction in expression of virulence genes, and attenuated virulence in a mouse infection model (33). These results indicate the importance of AA, especially glutamine, in the growth of a pathogen and its expression of virulence factors. AA metabolism is also of critical importance for pathogens to overcome defensive responses of the host. For example, the host imposes manganese and zinc starvation during *Staphylococcus aureus* infection, impairing glycolysis in *S. aureus* because manganese and zinc are essential for the activitiy of certain glycolytic enzymes in *S. aureus* (34). Glucose and other sugars are the preferred carbon source utilized by *S. aureus* to generate energy, and impaired glycolysis decreases the growth of *S. aureus* (34, 35). However, *S. aureus* overcomes this deficiency by shifting away from metabolism of sugars as an energy source to the metabolism of AA for energy and to reduce demand for manganese and zinc (34). Collectively, AA significantly affect the growth of pathogens and their expression of virulence factors.

Assimilation of local nutrients is also essential for fungal pathogens to establish an infection in their mammalian host. AA metabolism is crucial to the pathogenicity of major fungal pathogens such as *Candida albicans*. Metabolic adaptation to the microenvironments of the host is associated with fungal morphogenesis, cell wall remodeling, biofilm formation and stress responses, commensalism, all of which influence progression of infections. Metabolic adaptation is regulated by complex transcriptional networks such as the general control of AA metabolism (GCN response) in fungal species such as *Saccharomyces* and *Candida* (36). The exposure of *C. albicans* to macrophages or neutrophils induces expression of a cluster of genes required for nutrient assimilation and AA metabolism, as well as genes associated with hyphal growth (ECE1), adhesion (HWP1), and adaptation to oxidative stress (SOD1 and CAT-1) (37–39). *Candida glabrata* also increases the biosynthesis of both arginine and lysine in response to their internalization by macrophages (40). AA catabolism stimulates hyphal morphogenesis in *C. albicans* (41). Moreover, our recent transcriptomic analyses revealed the critical role of metabolism of AA, such as arginine, proline, aspartate, and glutamate metabolism, on biofilm formation by *C. albicans* (42). The aspartate aminotransferase gene is a common member of these AA pathways and it is significantly upregulated in isolates have a high capacity for producing biofilm (42).

During host tissue damage and invasion, *C. albicans* secretes various aspartic proteases, which liberate AA from host proteins (43). Released AA form peptides that are then taken up by *C. albicans via* dedicated oligopeptide transporters (Opt1–8) and other membrane AA permeases (44). Moreover, during a glucose deficiency *C. albicans* exploits AA as a carbon source, excreting excess nitrogen in the form of ammonia, potentially altering the pH of the host environment, and thereby triggering hyphal development (41). This process can contribute to neutralization of the acidic environment in phagosomes of macrophages (45, 46). In *Cryptococcus neoformans* upregulation of expression of AA transporters is a survival mechanism within macrophages (47). Overall, in fungal species, there is tight coordination of nutrient sensing and metabolic pathways *via* transcriptional circuitry, which regulates the global activation of AA biosynthesis in response to the dynamic nature of local niches of infection.

### AA METABOLIC ALTERATION AFTER INFECTIONS

There are significant metabolic alterations in AA at the site of infection or even other anatomic sites during an infection. For example, a *C. neoformans* infection perturbs the content of cysteine in the human lung epithelial cell line (A549) after 6 h (48). *Plasmodium yoelii* infection alters concentrations of AA in plasma of infected mice, including increases in 10 AA (valine, leucine, tyrosine, phenylalanine, EOHNH2, histidine, proline, aspartate, glutamate, alanine) and decreases in five AA (citrulline, cysteine, methionine, 1-MeHis, and arginine) (49). *P. yoelii* infection also increases 21 AA in the liver of mice, including threonine, asparagine, and arginine (49). Similarly, *P. yoelii* infection affects the abundance of AA in red blood cells (8). Using NMR (nuclear magnetic resonance)-based metabolomics to study white spot syndrome virus infection in crayfish gills, the virus: (1) increases glutamate, alanine, and methionine at 1 h postinfection; (2) increases alanine, tryptophan, histidine, tyrosine, and methionine at 6 h postinfection; and (3) increases the abundances of alanine, valine, leucine, isoleucine, glutamate, glutamine, phenylalanine, tyrosine, threonine, and methionine at 6 h postinfection in crayfish hepatopancreas (50). Enterotoxigenic *Escherichia coli* (ETEC) infection reduces concentrations of isoleucine in the serum of piglets (9). In the jejunum, the abundances of six AA changes after an ETEC infection, which includes decreases in the abundances of glutamine, asparagine, citrulline, and ornithine, and increases in the abundances of glycine and GABA (51). Changes in concentrations of AA in serum of mice infected with porcine circovirus type 2 (PCV2) have been studied systematically using isotope dilution liquid chromatography (LC)-mass spectrometry methods (52). PCV2 infection increases concentrations of proline, ornithine, and methionine in serum on day 3 postinfection, while concentrations of aspartate, arginine, proline, lysine, valine, isoleucine, and leucine decreases on day 7 postinfection, and there is no effect of PCV2 infection on concentrations of AA in serum on either day 10 or 14 postinfection (52). Those results indicate significant changes in concentrations of AA in serum of the host, but the exact mechanism for the changes is not known. However, changes in concentrations of AA in serum after PCV2 infection may result from changes in either AA metabolism or AA transport. Among seven AA transporters (i.e., Slc6a14, Slc6a20, Slc7a5, Slc7a6, Slc7a7, Slc7a8, and Slc7a9) responsible for the transport of these altered AA, PCV2 infection decreases the expression of *Slc7a5* and *Slc7a6* in the jejunum on day 7 postinfection (52).

## AA AS MEDIATORS OF METABOLIC CROSS TALK BETWEEN HOST AND PATHOGEN

A compelling example of extensive AA-dependent metabolic communication between host and pathogen occurs during infection with *Salmonella* or *Shigella* (**Figure 1**). Those infections rapidly induce a state of AA starvation in epithelial cells, which reduces abundances of leucine/isoleucine in cytosol, and inhibits activation of mTORC1, but induces the general control nonderepressible 2-dependent stress response pathway to promote expression of activating transcription factor 3 (ATF3) (53) (**Figure 1**). This starvation is induced through aseptic

between the host and *Salmonella* or *Shigella*. *Salmonella* (A, 1–2 h postinfection) or *Shigella* infection induces a rapid state of AA starvation through aseptic membrane damage (AMD), which inhibits the activation of mechanistic target of rapamycin complex 1 (mTORC1). mTORC1 negatively controls the autophagy response, which is responsible for targeting and degradating *Salmonella* or *Shigella.* However, *Salmonella* escapes from the autophagy-mediated degradation through replenishment of intracellular AA pools to reactivate mTORC1 signaling at a later phase of the infection (B).

membrane damage, but not protein metabolism because chloramphenicol has little effect on the expression of ATF3 while inhibiting synthesis of bacterial proteins (53). AA starvation may inhibit the host's innate immune responses and adaptive immune responses against pathogens, but activate host cell autophagy responses which are normally inhibited by mTORC1 (54) (**Figure 1**). Autophagy is a highly conserved cellular process that triggers nutrient recycling to sustain essential metabolic functions during nutrient or energy deprivation (55). Autophagy involves degradation and recycling of cellular constituents, such as dysfunctional organelles or macromolecular complexes (55). Autophagy also promotes targeting and degradation of intracellular bacteria (53, 56, 57), such as *Shigella* or *Salmonella,* and has important roles in the pathogenesis of extracellular bacterial infections, such as ETEC (58, 59). However, *Salmonella* could escape from autophagy-mediated degradation through the replenishment of intracellular AA pools to recruit and reactivate mTORC1 signaling at the surface of the *Salmonella*-containing vacuole (SCV) (53) (**Figure 1**).

A small change in cell and tissue content of AA after an infection has substantial effects on the ultimate outcome of the infection, and both host and pathogen can influence the availability of AA (e.g., asparagine, arginine, and tryptophan) and amounts of their metabolic products (e.g., NO, polyamines, kynurenine) at the site of infection to their respective advantages. The discussion will now focus on a few specific AA, arginine, asparagine, and tryptophan, which affect competition between host and pathogen, and strategies used by the pathogen to compete with the host for these AA, although the pathogen may also compete with the host for other AA.

#### Arginine

Utilization of arginine by both host and pathogen represents a metabolic bottleneck which is critical in determining the outcome of a pathogenic infection (17). The two predominant pathways for arginine metabolism are *via* NO synthase (NOS) for NO production and *via* arginase and ornithine decarboxylase for production of polyamines (putrescine, spermidine, and spermine). NO is an antimicrobial molecule, while polyamines are essential for the proliferation of pathogens (e.g., *Leishmania*). During infection, macrophages increase expression of inducible nitric oxide synthase (iNOS) to produce NO from arginine for antimicrobial purposes (60). However, the activity of iNOS in macrophages is highly dependent on the abundance of arginine in host cells (e.g., macrophages) and this influences the fate of an infection (61, 62). Arginase competes with iNOS for arginine; therefore, many pathogens exploit this to block NO production by increasing expression of arginase to limit arginine availability for metabolism *via* iNOS (61–63). One compelling example of the extensive competition for arginine between a host and a pathogen occurs during infections with *Leishmania* (**Figure 2**). Upon *Leishmania* invasion, infected macrophages activate cytotoxic pathways in an attempt to kill the pathogen, including the induction of NO biosynthesis from arginine; however, *Leishmania* infection upregulates arginase I activity in macrophages, thereby decreasing the availability of arginine for metabolism *via* iNOS (**Figure 2**) (64, 65). Mechanistically, the increase in Th2 cytokines

available for synthesis of NO *via* NOS by macrophages. Mechanistically, the increase in Th2 cytokines during *Leishmania* infection, such as interleukin (IL)-4 and IL-5, induces the expression of arginase I in macrophages to produce polyamines. *Leishmania* directly activates signal transducer and activator of transcription-6 (STAT6) to increase the expression of arginase I in macrophages. *Leishmania* also uses the arginine transporter (amino acid permease 3, AAP3) (b) to compete with macrophages for arginine. During macrophage invasion by *Leishmania*, there is a coordinated arginine deprivation response mediated *via* the mitogen-activated protein kinase 2 (MPK2) cell signaling pathway, which upregulates arginine transport away from the macrophage. *Leishmania* uses polyamines to produce trypanothione that neutralizes effects of reactive oxygen species (ROS) released from macrophages. (B) *Streptococcus pyogenes* uses the arginine deiminase (ADI) pathway to limit available arginine for NO production by macrophages. This pathway includes the ArcD antiporter, which concomitantly transports ornithine out and arginine into the cell, enzymes which convert arginine to ornithine, CO2, NH3, and one molecule of adenosine triphosphate (ATP). The NH3 can be utilized to buffer against acid stress.

during *Leishmania* infection, such as IL-4, IL-10, and transforming growth factor-beta, induces the expression of macrophage arginase I (64), or *Leishmania* directly activates signal transducer and activator of transcription-6 to promote expression of arginase I (66). Although pathogenic *Leishmania* (*L. major* and *L. tropica*) induces increases in arginase activity in susceptible BALB/c mice, and even in resistant C57BL/6 mice, there is no change in arginase activity at the site of infection and in the draining lymph nodes of either strain of mice during non-pathogenic *Leishmania* (*L. tarentolae*) infection (67).

Arginase in *Leishmania* can also subvert antimicrobial activity of macrophages by diverting arginine away from iNOS. *Leishmania* lacking arginase have poorer survival in mouse macrophages, and the decrease in intracellular survival is abrogated in iNOS-deficient macrophages (68). l-ornithine, the product of arginase, is used for the production of polyamines required for growth of *Leishmania* (64). In addition, *Leishmania* uses spermidine to produce trypanothione which neutralizes reactive oxygen species (ROS) released by macrophages (69) (**Figure 2**). Interestingly, an a comparison of arginase genes in pathogenic *Leishmania* (*L. major* and *L. tropica*) and non-pathogenic *Leishmania* (*L. tarentolae*) revealed that amino acid sequences of arginase from the pathogenic and non-pathogenic *Leishmania* are 98.6 and 88% identical to the reference gene in *L. major* Friedlin, respectively, and that the activity of arginase is greater in pathogenic than non-pathogenic *Leishmania* (67). *Leishmania*, *P. falciparum* infection also causes a rapid depletion of arginine through activation of arginase (8). Patients with *P. falciparum* infection have lower concentrations of l-arginine in their plasma and exhale less NO than controls, while l-arginine infusion increases concentrations of l-arginine in their plasma, exhaled NO, and other clinical indices without important side effects in *P. falciparum-*infected patients with malaria (70). Thus, a pathogen may affect the activity of arginase in the host and, therefore, arginine availability to cells of the infected host.

*Leishmania* also competes with macrophage for arginine by depleting arginine from phagolysosomes *via* changes in an arginine transporter (amino acid permease 3, AAP3) (**Figure 2**) (71). During macrophage invasion, there is a coordinated arginine deprivation response induced in *Leishmania* through a mitogen-activated protein kinase 2-mediated signaling pathway which upregulates expression of arginine transporters in *Leishmania* and arginine transport from the macrophage reduces available arginine for synthesis of NO *via* NOS by macrophages (71). *Salmonella* infection promotes the expression of arginine transporters in macrophages, including the cationic amino acid transporter members (CAT)-1 and 4 (61, 62). CAT-1 is preferentially localized in the cell membrane of uninfected macrophages, while CAT-1 is localized in close proximity to SCV in infected macrophages, which promotes the usage of arginine from macrophages by *Salmonella via* the arginine permease lysine–arginine–ornithine-binding periplasmic protein ArgT (61, 62). Interestingly, intracellular *Mycobacterium bovis* BCG also promotes the colocalization of CAT-1 to the intracellular BCG in macrophages (61, 62). Thus, *M. bovis* BCG may enhance expression of CAT-1 and CAT-2B for uptake of arginine into macrophages without a significant increase in NO production by infected macrophages (72, 73).

Pathogens also compete with macrophages for arginine *via* the arginine deiminase (ADI) pathway. The ADI pathway is critically important for *Streptococcus pyogenes* infection because *S. pyogenes* uses this pathway to increase the production of cellular energy and provide protection against acidic stress (**Figure 2**) (74, 75). This pathway includes the ArcD antiporter, which transports ornithine out and concomitantly brings arginine into the cell, and ArcA (ADI), ArcB (ornithine carbamoyl transferase), and ArcC (carbamate kinase) which convert arginine to ornithine, CO2, NH3, and one molecule of adenosine triphosphate (ATP) (75). NH3 can be utilized to buffer against acidic stress by reacting with protons from the fermentative metabolism of *S. pyogenes* and the accumulation of lactic acid. The ADI pathway mediates consumption of arginine and limits the availability of arginine for NO<sup>⋅</sup> production by macrophages infected with *S. pyogenes* (**Figure 2**) (75).

Overall, during infection, the host uses arginine as a substrate to produce NO to protect itself against the pathogen, while, in parallel, the pathogen uses various mechanisms including arginase, arginine transport and the ADI pathway to deplete arginine and/or divert arginine away from host cells that produce NO (**Table 1**).

#### Asparagine

Asparagine is an important source of nitrogen for pathogens, especially GAS that is an extracellular pathogen infecting the human throat and skin. GAS creates endoplasmic reticulum

Table 1 | Strategies used by pathogens to compete with the host for arginine, asparagine, and tryptophan.


stress (ERS) in host cells through its streptolysin toxins which causes transmembrane pores on host cells to permit extracellular calcium influx into the cytosol to dysregulate intracellular calcium (31, 84) (**Figure 3**). The ERS response activates the protein kinase-like endoplasmic reticulum kinase, which induces activation of transcription factor 4 and an increases expression of asparagine synthetase for production of asparagine (31, 76, 77) (**Figure 3**). Asparagine promotes GAS proliferation and changes the expression of about 17% of the GAS genes (including those for virulence, growth, and metabolism) partly through the twocomponent system TrxSR (31) (**Figure 3**). Asparaginase, which hydrolyzes asparagine into aspartate, blocks GAS growth in human blood and GAS proliferation in a mouse model of human bacteremia (31).

Asparagine is also essential for an intracellular pathogen (e.g., *Mycobacterium tuberculosis*) to resist acidic stress in a phagosome. Asparagine supports *M. tuberculosis* resistance to acidic stress by serving as substrate for production of the weak base ammonia, which reacts with protons in the phagosome to form ammonium ions (**Figure 3**) (78, 85, 86). *M. tuberculosis* expression of the asparagine transporter (AnsP2) increases markedly during infection to capture asparagine from macrophages (78). *M. tuberculosis* also expresses asparaginase (AnsA), which hydrolyzes asparagine into aspartate and ammonia (78). AnsA can also be secreted into the lumen of the phagosome of a macrophage through the type VII secretion system to hydrolyze asparagine for the production of aspartate and ammonia (78).

Besides its required role for pathogen survival and infection, asparagine is required for T cell activation. For example, the absence of asparagine in medium abolishes T cell activation induced by anti-CD3 and anti-CD28 antibodies, and also inhibits T cell blastogenesis and IL-2 secretion (80). Mechanistically, asparagine may affect T cell functions through mTORC1 which is critically important for T cell activation and differentiation, especially for Th1 and Th17 cells (9). Thus, the pathogen could inhibit the immune response of the host by inducing starvation of asparagine in host cells. For example, *Salmonella* Typhimurium infection induces asparagine deprivation by hydrolyzing asparagine *via S.* Typhimurium asparaginase, resulting in inhibition of mTOR signaling, Myc expression, T cell activation, and immune responses in the host (79–81). Using gas chromatography-mass spectrometry based metabolomics, ETEC was found to induce asparagine deprivation in the jejunum of piglets (51), but not in serum (9). This decrease may be responsible for inhibition of immune responses in the jejunum of piglets after ETEC infection (87). An ETEC infection inhibits activation of the NF-κB and MAPK pathways in the jejunum, and expression of TLR and other indicators associated with intestinal immunity, including phospholipase A2, lysozyme, polymeric immunoglobulin receptor, and mucin 2 (87, 88). However, in response to the ETEC infection, piglets try to alleviate asparagine deprivation by upregulating asparagine synthetase in the jejunum. Those results are based on the use of isobaric tags for relative and absolute quantitation (iTRAQ) combined with multi-dimensional LC and MS analysis (87). Interestingly, piglets with diarrhea have a lower content of asparagine in the jejunum than control piglets (51), whereas resistant and control piglets have similar contents of

asparagine in the jejunum, and piglets that have recovered from ETEC-induced diarrhea restore their asparagine content, compared to piglets that did not recover from diarrhea (manuscript submitted).

Collectively, available results indicate that the competitive utilization of asparagine by pathogens *via* asparagine synthetase, asparagine transporter, and asparaginase (**Table 1**) has important influences on the pathogenesis of infection and the outcome of an infection.

## Tryptophan

Tryptophan is required for optimal immune responses, such as T cell proliferation (**Figure 4**). For example, activated T cells placed under tryptophan-deficient conditions undergo a mid-G1 arrest in cell cycle progression (89, 90). The tryptophan pool diminishes due to its conversion into kynurenine by indoleamine 2,3-dioxygenase (IDO) (**Figure 4**). Various regulators, including interferon (IFN)-α, IFN-γ, tumor necrosis factor-α, and prostaglandins, regulate the expression of IDO (91, 92). IDO has an immunoregulatory function in some situations, such as pregnancy, chronic infections, and tumors, by inhibiting T cell proliferation, increasing T cell apoptosis, and altering the balance of Th1 and Th2 cells (93, 94). The effect of IDO on immune responses may largely depend on availability of its substrate tryptophan and breakdown products of tryptophan (i.e., kynurenines). Kynurenine enhances the generation of regulatory T cells (Treg) through its interaction with aryl hydrocarbon receptors (95–97) (**Figure 4**). Kynurenine also promotes apoptosis of human and mouse neutrophils, and inhibits production of ROS (82) (**Figure 4**). Thus, a pathogen, such as *Clostridium difficile*, promotes activation of IDO to deplete the tryptophan pool of the host and diminish the immune responses of the host against the pathogen (**Figure 4**). A *C. difficile* infection induces the expression of IDO1 (4.7-fold) and production of kynurenine (about eightfold) in mouse intestinal lamina propria (82), while inhibition of tryptophan catabolism in IDO1-knockout mice increases the percentage of IFN-γ-expressing neutrophils and clearance of *C. difficile* from mice (82). *Mycobacterium avium* subsp. *paratuberculosis* infection promotes the expression of IDO in human monocytes in the gut and draining lymph nodes of sheep, and in peripheral blood cells of sheep and cattle, coincident with a decrease in amounts of tryptophan in those cells (83). HIV infection is associated with increased tryptophan catabolism (i.e., a high ratio of kynurenine to tryptophan in plasma and IDO1 activation), expansion of Tregs, and depletion of Tc17/mucosa-associated invariant T cells, while combination antiretroviral therapy in HIV-infected patients decreases tryptophan catabolism (98) (**Figure 4**). Further experiments demonstrated that HIV-induced IDO1 activation may be responsible for acute and progressive numeric loss of CD4<sup>+</sup> T-helper cells and functional impairment of T-cell responses during an HIV infection (99). Mechanistically, the N-terminal domain of the HIV-1 transactivator regulatory protein (Tat) induces the initial expression of IDO, while increased IFN-γ

during HIV-1 infection enhances the expression of IDO in monocyte-derived dendritic cells (DC) (100) (**Figure 4**).

Tryptophan is also essential for the growth of some pathogens, thus IDO has an antimicrobial activity by depleting cells of tryptophan. IDO is an anti-hepatitis B virus (HBV) effector based on evidence that patients with acute hepatitis B, but not hepatic flare, have robust activation of IDO (101). Further, patients with acute hepatitis who eventually clear HBV have increased IDO activity at the time of peak alanine aminotransferase activity, but patients with a hepatic flare-up have less IDO activity (101). Indeed, IDO promotes HBV suppression in Huh7 cells that is abrogated in IDO-knockout cells, but restored by re-induction of IDO in the cells (101). Inhibition of IDO in mice increases mortality and parasite burdens during *Toxoplasma gondii* infections (102). Depletion of tryptophan by high IDO activity causes *Chlamydia trachomatis* to lose infectivity and decrease transcriptional activity, but these outcomes are reversed in response to replenishment of cellular tryptophan pools (103). The suppression of IDO in the replication of intracellular pathogens also occurs in response to other pathogens, such as HIV (104), herpes simplex virus (105), and *Chlamydia pneumonia* (106). In parallel, some pathogens have evolved to synthesize tryptophan to protect them from local tryptophan depletion due to the increased expression of IDO during infection. *M. avium* subsp. *paratuberculosis* has the ability for tryptophan biosynthesis because it expresses genes for *trpA* (tryptophan synthase subunit alpha, MAP1307), *trpB* (tryptophan synthase subunit beta, MAP1306), *trpC* (indole-3-glycerol-phosphate synthase, MAP1305), *hisA* (phosphoribosyl isomerase A, MAP1297), *trpD*

(anthranilate phosphoribosyltransferase, MAP1931c), and *trpE* (anthranilate synthase component I, MAP1303), which are all involved in the biosynthesis of l-tryptophan (83).

Collectively, tryptophan is required for both host and pathogen and IDO has biologically important roles in the interaction between host and pathogen, especially intracellular pathogens. IDO not only dampens protective mechanisms for host immunity to increase pathogen burdens but also suppresses replication of pathogens to limit the spread of infection. Thus, the dominant nature of IDO (i.e., antimicrobial or immunoregulatory) is pathogen-specific.

#### Other AA

Our understanding of other AA involved in metabolic cross talk between host and pathogen lags far behind our understanding of the roles for arginine, asparagine, and tryptophan. There are sporadic reports on metabolism and regulation of other AA during infection. For example, human cytomegalovirus (HCMV) infection causes host cells to use glutamine to produce the tricarboxylic acid cycle intermediate-alpha ketoglutarate due to the removal of citrate from the citric acid cycle to increase fatty acid biosynthesis (107, 108). Thus, infection may induce changes in the energy substrate from glucose to glutamine, which is the major source of ATP in HCMV infected human fibroblasts, but not in uninfected cells (107). Glutamine deprivation inhibits the formation of infectious virions, while alpha ketoglutarate supplementation rescues ATP production and viral growth under conditions of glutamine deprivation (107). Also, glutamine uptake by the sodium-dependent neutral amino acid transporter 2 is required for motility and migration of *T. gondii* infected DCs (109).

Glutamate has a central role for bacteria due to its involvement in a wide range of metabolic processes. The glutamate-dependent acid resistance (GDAR) system is one of the best-characterized systems in commensal and pathogenic bacteria, including *E. coli*, *Shigella flexneri*, *L. monocytogenes*, and *Lactococcus lactis* (110). GDAR couples the glutamate decarboxylase(s) GadA and/or GadB, and glutamate:γ-aminobutyrate exchanger GadC (110). GadC has important roles for intracellular multiplication and virulence of *Francisella* because the Δ*gadC* mutant *F. novicida* exhibits impaired multiplication in macrophages, as well as in liver and spleen of mice (111). The Δ*gadC* mutant *F. novicida* is more sensitive to oxidative stress than the wild-type strain, but similar sensitivities to oxidative stress in a glutamate deficient medium (111). Further, Δ*gadC* mutant *F. novicida* loses the capacity to escape from the phagosomal compartment of infected macrophages due to a defect in its ability to neutralize ROS produced in the phagosomal compartment (111).

Thus, AA other than arginine, asparagine, and tryptophan can mediate interactions between a host and pathogen, and influence the outcome of the infection.

### MODULATION OF AA DURING AN INFECTION

Amino acids are associated with interactions between a host and its pathogens and profoundly influence the outcome of infection. Therefore, AA therapy can be used to manipulate the progression of an infection. For example, a decrease in asparagine blocks GAS growth in human blood and GAS proliferation in a mouse model of human bacteremia (31). Glutamine deprivation inhibits formation of infectious virions of HCMV (107). In PCV2 infected mice with a reproductive defect, dietary supplementation with functional AA (e.g., arginine, glutamine, and proline) enhances host defense responses that reduce replication of PCV2 and improve reproductive performance (112–114). In *P. multocida*infected animal models, dietary supplementation with arginine, glutamine, and proline promotes immune responses and decreases bacterial replication and their expression of virulence factors, as well as mortality (115–117). Collectively, AA influence the outcome of an infection by modulating the activation of host defensive responses and the growth of the pathogen or its expression of virulence factors. However, such knowledge should be transferred carefully in its application as a prophylactic measurement against pathogens. It is difficult to inhibit growth of a pathogen by deprivation of AA without inhibiting the immune responses of the host and it seems impossible to improve the immune responses of the host through replenishing of AA without promoting the growth of the pathogen. Thus whether adding or depleting one particular AA to favor the host to clear a pathogen depends on the specific metabolic situation associated with the pathogenic infection. Probiotics offer a promising surrogate to regulate the AA status of hosts. For example, the use of multi-species probiotics (*Bifidobacterium bifidum* W23, *Bifidobacterium lactis* W51, *Enterococcus faecium* W54, *Lactobacillus acidophilus* W22, *Lactobacillus brevis* W63, and *Lactococcus lactis* W58) can increase concentrations of tryptophan in serum and reduce the incidence of upper respiratory tract infections in individuals (118). *L. lactis*-derived GABA also modifies the expression of IL-17 in the intestine during ETEC infection (51). It would be of great interest to use genetically engineered probiotics to compete with pathogens in acquiring essential AA.

## CONCLUSION

There is an obligatory and extensive metabolic cross talk between host and pathogens. Upon infection, the host alters metabolism to support defensive responses against the pathogen, while the pathogen uses metabolic cues to sense its anatomical position and the immune status of the host (17). The host has the ability to alter AA metabolism after an infection by a pathogen (9, 52). Thus, AA influence immune responses of the host against a pathogen, such as the function of innate immune cells (e.g., macrophage), the activation and differentiation of T cells and the production of antibodies by B cells (27, 51, 112, 115, 116, 119). Also, AA play an important role in the physiology and virulence of pathogens. Thus, a change in AA metabolism at the site of infection will influence the outcome of an infection. Some AA, such as arginine, asparagine, and tryptophan, are central to competition between host and pathogen. However, the importance of other AA, such as glutamine, proline, and GABA, in the metabolic cross talk between a host and pathogen, must be explored further. For example, there is a significant change in GABA signaling during ETEC infection that affects the expression of IL-17 in the intestine during ETEC infection (51). Also, strategies and mechanisms used by pathogens to compete with the host for AA remain to be discovered (see **Table 1**). Understanding metabolic cross talk involving AA between host and pathogen will offer significant insight into pathogenic infections and reveal novel treatments to prevent and cure infections by modulating the abundance of AA and/or the metabolism of those AA.

## AUTHOR CONTRIBUTIONS

WR, GZ, and JD conceived this study. WR, RR, and YZ wrote the manuscript. BT, YP, and XH provided critical discussion in manuscript preparation. FB, GW, and YY revised the manuscript.

## ACKNOWLEDGMENTS

The authors' profound admiration and respect go to researchers in this field and in their laboratories for their dedication and hard work. The authors apologize to scientists whose work in this field has not been cited owing to space limitations. This study was supported by the National Key R & D Program (2016YFD0501201), the National Natural Science Foundation of China (31330075, 31372326, 31110103909, 31272463), Key Programs of frontier scientific research of the Chinese Academy of Sciences (QYZDY-SSW-SMC008), and the Priority Academic Program of Development Jiangsu High Education Institution.

## REFERENCES


42. Rajendran R, May A, Sherry L, Kean R, Williams C, Jones BL, et al. Integrating *Candida albicans* metabolism with biofilm heterogeneity by transcriptome mapping. *Sci Rep* (2016) 6:35436. doi:10.1038/srep35436

43. Naglik JR, Challacombe SJ, Hube B. *Candida albicans* secreted aspartyl proteinases in virulence and pathogenesis. *Microbiol Mol Biol Rev* (2003) 67(3):400–428, table of contents. doi:10.1128/MMBR.67.3.400-428.2003


arginine availability towards intracellular *Salmonella* growth. *PLoS One* (2010) 5(12):e15466. doi:10.1371/journal.pone.0015466


suppression of activation-induced T cell metabolic reprogramming. *J Leukoc Biol* (2016) 99(2):387–98. doi:10.1189/jlb.4A0615-252R


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2018 Ren, Rajendran, Zhao, Tan, Wu, Bazer, Zhu, Peng, Huang, Deng and Yin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Variation of Carbohydrate-Active Enzyme Patterns in the Gut Microbiota of Italian Healthy Subjects and Type 2 Diabetes Patients

Matteo Soverini, Silvia Turroni, Elena Biagi, Sara Quercia, Patrizia Brigidi, Marco Candela and Simone Rampelli\*

Unit of Microbial Ecology of Health, Department of Pharmacy and Biotechnology, University of Bologna, Bologna, Italy

The human gut microbiota (GM) has been associated, to date, with various complex

#### Edited by:

Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia

#### Reviewed by:

Maryam Dadar, Razi Vaccine and Serum Research Institute, Iran Mario M. D'Elios, University of Florence, Italy

> \*Correspondence: Simone Rampelli simone.rampelli@unibo.it

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 31 July 2017 Accepted: 11 October 2017 Published: 24 October 2017

#### Citation:

Soverini M, Turroni S, Biagi E, Quercia S, Brigidi P, Candela M and Rampelli S (2017) Variation of Carbohydrate-Active Enzyme Patterns in the Gut Microbiota of Italian Healthy Subjects and Type 2 Diabetes Patients. Front. Microbiol. 8:2079. doi: 10.3389/fmicb.2017.02079 functions, essentials for the host health. Among these, it is certainly worth noting the degradation of the so-called microbiota-accessible carbohydrates (MACs), which the GM breaks down through specific enzymes, referred to as carbohydrate-active enzymes (CAZymes). This degradation constitutes the first step in the production of short-chain fatty acids (SCFAs), key microbial small molecules having multiple healthpromoting effects for the host organism. The decline in MAC dietary intake in urban Western populations forced the shrinkage of CAZyme repertoire in the GM, as shown by the literature comparing the microbiome layout between Western urban citizens and traditional rural populations. Even if this reduction in GM functional complexity has been associated with the onset of the so-called "diseases of civilization," only few information regarding the CAZyme variation within Western populations has been provided to date, and its connections with diet and health are still unexplored. In this scenario, here we explore the GM-encoded CAZyme repertoire across two Italian adult cohorts, including healthy lean subjects consuming a Mediterranean diet and obese patients affected by type 2 diabetes, consuming a high-fat diet. In order to impute the CAZyme panel, a pipeline consisting of publicly available software – QIIME, FragGeneScan and HMMER – was specifically implemented. Our study highlighted the existence of robust clusters of bacterial species sharing a common MAC degradation profile in the Italian GM, allowing the stratification of the individual GM into different steady states according to the carbohydrate degradation profile, with possible connections with diet and health.

Keywords: gut microbiota, microbiota-accessible carbohydrates, CAZymes, co-abundance groups, type 2 diabetes

## INTRODUCTION

The large ensemble of bacteria that stably inhabit the distal part of our gastrointestinal tract, namely the gut microbiota (GM), is of vital importance for many physiological functions of our organism, exerting a key role in several biological processes, such as nutrition, immune function, and the regulation of the central nervous system (Neish, 2009; Hooper et al., 2012;

Nicholson et al., 2012; Honda and Littman, 2016; Sharon et al., 2016). As a strategic partner of the human holobiont, the GM expands our metabolic potential by adding degradative functions, thus enabling the metabolism of a wide range of complex polysaccharides, otherwise indigestible, including exogenous carbohydrates introduced with diet (Schnorr et al., 2014). Indeed, despite the wide distribution of complex polysaccharides in edible plants (Lattimer and Haub, 2010), our genome is astonishingly poor in genes coding for carbohydrate-active enzymes (CAZymes) (Gill et al., 2006), i.e., enzymes specifically devoted to polysaccharide degradation. Conversely, the gut microbiome (i.e., the cumulative genome of the GM) encodes for a much wider and diversified arsenal of CAZymes (El Kaoutari et al., 2013), allowing for the catabolism of the vast array of complex polysaccharides that reach the colon undigested. These molecules are converted to short-chain fatty acids (SCFAs), microbial metabolites with multiple roles in human physiology (Koh et al., 2016). This inter-kingdom cross-feeding is the basis of the mutualistic relationship we share with our intestinal microbial counterpart, which thus emerges as a key player in the co-metabolism of complex carbohydrates in our gut. In this scenario, the dietary polysaccharides that are metabolically available to gut microbes have been specifically defined as microbiota-accessible carbohydrates (MACs) (Sonnenburg and Sonnenburg, 2014) and their abundance in the host diet, and consequently their availability within the gut, has proved to be crucial for microbiota–host homeostasis.

The MAC availability in the host diet is among the aspects that have deeply changed in recent human evolutionary history, with the transition from Paleolithic hunting-gathering to Neolithic rural populations and contemporary Westernized societies (Quercia et al., 2014; Schnorr et al., 2014; Sonnenburg and Sonnenburg, 2014; Obregon-Tito et al., 2015). In particular, along with Westernization, we have witnessed a progressive reduction of the MAC content and diversity in the diet, with the transition to foods high in refined simple sugars. According to Sonnenburg et al. (2016), this reduction of dietary MACs forced the progressive impoverishment of the CAZyme repertoire in the human microbiome, compromising the overall metabolic plasticity of the human meta-organism. This was confirmed by gut metagenome studies comparing the CAZyme repertoire between traditional populations consuming high-fiber diets and Western urban citizens, which highlighted a relevant reduction of the microbial CAZyme diversity in the latter (King et al., 2012; Bhattacharya et al., 2015; Rampelli et al., 2015). Such wide GM comparative surveys provided important information on the GM-host co-evolutionary dynamics but they did not allow for a precise discrimination of the impact of individual covariates (e.g., diet, health, medication) on the GM functional repertoire. Aiming to fill this gap in knowledge, here we compared the imputed CAZyme repertoire from two cohorts of Western urban adults from Italy, overcoming the intrinsic variation of the CAZyme profiles according to geography and ethnicity (Soverini et al., 2016). In particular, we analyzed the GM structure and metadata from16 healthy subjects following a Mediterranean diet and 40 obese, type 2 diabetic (T2D) patients consuming a high-fat low-MACs diet from two previously published studies (Schnorr et al., 2014; Candela et al., 2016). Our findings highlighted the existence of a potentially limited number of well-balanced host–microbe symbiotic configurations, with a possible connection to diet and health status.

#### MATERIALS AND METHODS

#### Determination of the Pan-microbiome from Italian Healthy Subjects

The publicly available 16S rRNA sequencing data of the fecal samples of 16 Italian healthy subjects from Schnorr et al. (2014) were downloaded from the MG-RAST website<sup>1</sup> and taxonomically characterized to the species level using the QIIME pipeline (Caporaso et al., 2010), with blastn as an assignment method and the HMP gastrointestinal 16S rRNA dataset as reference sequences<sup>2</sup> . The detected species were considered part of the so-called Italian "pan-microbiome," i.e., the virtual entity gathering the vast majority of bacterial species present in the GM of the Italian population. The assembled reference genomes of these bacterial species were downloaded from the NCBI genome section<sup>3</sup> . Then, to characterize the CAZyme repertoire of these microorganisms, the CAZyme identification pipeline developed by Soverini et al. (2016) was applied. Briefly, ORFs were extracted from the assembled genomes using FragGeneScan 1.16 (Rho et al., 2010). From the translated ORFs, the CAZyme-coding sequences were detected using the hmmscan tool of the HMMER software package (Robert et al., 2011) and the dbCAN CAZyme database (Yin et al., 2012). The outputs were further processed by the script hmmscan-parser.sh<sup>4</sup> , selecting only the ORFs that showed a minimum identity of 30% to the query sequences and an alignment length of at least 100 residues.

#### Identification of CAZyme Co-abundance Groups within the Italian Pan-microbiome

The CAZyme profiles were used to generate CAZy co-abundance groups (CCGs), which were conceived as groups of bacterial species sharing a similar CAZyme profile. In brief, the CCGs were generated by applying hierarchical Ward-linkage clustering based on Spearman correlation coefficients to the abundances of glycosyl-hydrolase (GH) and auxiliary activity (AA) families detected in the bacterial genomes. Permutational multivariate analysis of variance (function "adonis" of the vegan package in R) was used to determine whether CCGs were significantly different from each other. CAZymes were also manually classified for their ability to degrade specific substrates by consulting the publicly available CAZy database<sup>5</sup> . Specifically, we evaluated the ability to degrade different types of MACs: resistant starch (RS),

<sup>1</sup>http://metagenomics.anl.gov/linkin.cgi?project=mgp8810

<sup>2</sup>http://hmpdacc.org/HM16STR/healthy/

<sup>3</sup>https://www.ncbi.nlm.nih.gov/genome/

<sup>4</sup>https://github.com/carden24/Bioinformatics\_scripts/blob/master/hmmscanparser.sh

<sup>5</sup>http://www.cazy.org

non-digestible carbohydrates (NDC), non-starch polysaccharides (NSP), and mucins/glycoproteins (M/G). When more than one activity was found, we selected the most relevant one, i.e., the one with the highest abundance of genes involved in the degradation of a given substrate.

### Assessment of Redundant Patterns of CAZymes in Italian Healthy Subjects and Type 2 Diabetes Patients

To explore CAZyme profiles in the Italian population in health and disease, we integrated the dataset used to determine the pan-microbiome with the 16S rRNA sequences of the GM from 40 patients affected by type 2 diabetes (Candela et al., 2016). The sequences were downloaded from MG-RAST<sup>6</sup> and analyzed using QIIME (Caporaso et al., 2010) and the HMP database, as described above for healthy subjects. The CAZyme profile of each GM was obtained by quantifying the relative abundance of each CCG, as a sum of the relative contribution of component bacterial species. We then grouped the subjects using hierarchical Wardlinkage clustering based on Spearman correlation coefficients. Separation between clusters was tested using the permutational multivariate analysis of variance (function "adonis" of vegan). All statistical analyses were computed in R version 3.1.3 using R studio version 1.0.36 with packages vegan and made4.

#### RESULTS

### The Italian Pan-microbiome from Healthy Adults

The analysis of the publicly available 16S rRNA sequences of stool samples from 16 Italian healthy adults (aged 20–40 years, mean 32 years) consuming a standard Mediterranean diet (Schnorr et al., 2014) led to the identification of a total of 98 bacterial species, present at least once in the samples (**Supplementary Table S1**). Faecalibacterium prausnitzii was the most represented species, with an average relative abundance of 11 ± 0.09% (standard deviation of the mean). Eubacterium rectale (7 ± 0.06%), Ruminococcus bromii (6 ± 0.05%), and Bifidobacterium adolescentis (6 ± 0.06%) occurred as co-dominant species. Subdoligranulum variabile (3 ± 0.04%), Ruminococcus champanellensis (3 ± 0.02%), Clostridium asparagiforme (2 ± 0.01%), Bacteroides vulgatus (2 ± 0.03%), Coprococcus eutactus (2 ± 0.03%), and Roseburia intestinalis (1.8 ± 0.01%) were ancillary species present to a lower extent. In terms of prevalence, F. prausnitzii, E. rectale, Butyricicoccus pullicaecorum, R. intestinalis, Ruminococcus spp., Flavonifractor plautii, Blautia obeum, Dorea formicigenerans and Anaerostipes hadrus were present in all samples of the dataset, whereas Methanobrevibacter smithii, Bifidobacterium catenulatum, Lactobacillus ruminis, Weissella paramesenteroides, and Sutterella parvirubra were found in less than three samples, being among the least prevalent microorganisms in the analyzed Italian microbiomes.

## CAZyme Repertoire in the Bacterial Species of the Italian Pan-microbiome

Reference genomes from the bacterial species included in the Italian pan-microbiome were retrieved from the NCBI database, and the respective CAZyme-coding sequences were identified. The number of these sequences varied widely between the pan-microbiome components. In particular, Odoribacter laneus, Lactobacillus salivarius, and F. plautii showed the lowest amounts of CAZyme-coding genes, while Eubacterium biforme, Dialister succinatiphilus, and D. formicigenerans the highest (see Supplementary Figure S1 for retrieving the number of CAZymecoding sequences for each bacterial species).

To identify common patterns of MAC-degrading enzymes among the different species of the Italian pan-microbiome, we determined co-abundance associations between the bacterial CAZyme profiles and then clustered them based on similarity. Four robust CAZy Co-abundance Groups (CCGs) were identified in the Italian pan-microbiome, each one including GM species that share a similar CAZyme profile, describing the pattern of CAZyme variation within the Italian GM species (p-value < 0.001, permutation test with pseudo F ratios). CCGs were named according to the most abundant species in each group as follows: S. variabile (CCG1), E. rectale (CCG2), R. bromii (CCG3), and F. prausnitzii (CCG4) (**Figure 1**). Interestingly, F. prausnitzii and R. bromii CCGs included species with the highest amount of CAZyme-coding sequences, such as D. succinatiphilus, D. formicigenerans, and E. biforme. On the other hand, CCG1 (S. variabile) and CCG2 (E. rectale) included species with a lower amount of CAZyme genes, such as O. laneus and F. plautii. It is important to note that the bacterial genome size is not predictive of the number of encoded CAZymes (El Kaoutari et al., 2013). For example, Bacteroides oleiciplenus, Bacterioides cellulosilyticus and Bacteroides ovatus, the three species with the largest genome size in the Italian pan-microbiome (Supplementary Figure S1), were found to be included within the S. variabile group and show the lowest richness in the CAZyme repertoire.

Finally, we specifically evaluated the distribution of CAZyme families involved in the degradation of different types of MACs across the four CCGs (**Figure 2**). According to our findings, E. rectale and F. prausnitzii groups showed the highest number of sequences encoding for CAZymes involved in the degradation of non-digestible carbohydrates and non-starch polysaccharides, while the R. bromii group was the least rich in these enzymatic functions (p-value < 0.05, Wilcoxon rank sum test). Interestingly, the ability to degrade RS was evenly spread among the different CCGs.

## CAZyme Distribution in the Gut Microbiome of Italian Individuals: Associations with the Diet and Health Status

We next explored the variation of CCG profiles in Italian subjects, also in association with the dietary pattern and health conditions. To this aim, the dataset of analyzed individuals was integrated

<sup>6</sup>http://metagenomics.anl.gov/mgmain.html?mgpage=project&project= mgp17675

with publicly available16S rRNA gene sequences from 40 Italian obese patients suffering from type 2 diabetes and consuming a high-fat low-MACs diet (Candela et al., 2016). The addition of these samples did not result in an increase in the size of the Italian pan-microbiome, which still consisted of the previously identified 98 species. To compare the CCG distribution among individual gut microbiomes, for each CCG, the sum of the abundances of constituent microorganisms was calculated, and

the personal profile of CCG abundance in the GM was computed for each subject. According to a clustering analysis of the individual CCG abundance profiles, subjects could be clustered into four groups (p-value < 0.001, permutation test with pseudo F ratio), each including individuals that shared a comparable profile of the four CCGs, and therefore potentially a similar carbohydrate-degrading functional pattern. We defined these redundant patterns as CAZyTypes (CTs), i.e., clusters of different GM configurations with a similar carbohydrate-degrading profile (**Figure 3**). Interestingly, CT2 and CT4 were more represented in healthy individuals consuming a Mediterranean diet, while CT1 and CT3 occurred with higher frequency in obese T2D patients consuming a high-fat low-MACs diet. By Principal Coordinates Analysis (PCoA), we represented the CCG variation patterns in individual microbiomes in a two-dimensional space (Supplementary Figure S2), highlighting the separation of subjects based on both the CCG abundance and health status (p-value < 0.001 permutation test with pseudo F ratio).

Looking at the CCG distribution in the Italian population, we found that the most prevalent CCG was F. prausnitzii group, with the co-presence of E. rectale group as an ancillary CCG in the health-prevalent CT2 and CT4 (**Figure 3**). On the other hand, S. variabile group and R. bromii group were sporadically represented in the subjects of all CTs, but generally present in lower abundance in healthy subjects. Indeed, only three healthy subjects out of 16 showed a relative abundance of the S. variabile and R. bromii groups higher than the average contribution in the overall dataset. Conversely, a higher abundance of these two CCGs was found in T2D patients (18 out of 40).

#### DISCUSSION

Our work aimed at characterizing the frame of the CAZyme variation in the GM meta-community from 56 Italian subjects, of whom 16 were healthy lean adults consuming a standard Mediterranean diet (Schnorr et al., 2014) and 40 were obese T2D patients consuming a high-fat low-MACs diet (Candela et al., 2016). The pan-microbiome of the studied population included a total of 98 bacterial species and, in healthy subjects, it was dominated by F. prausnitzii, E. rectale, R. bromii and B. adolescentis, which also emerged as the most prevalent species in the GM from Italian healthy adults. When compared with previously characterized GM pan-genomes, such as that from the Chinese population (Zhang et al., 2015), the Italian one showed some peculiarities, i.e., the presence of E. rectale and Bifidobacterium and the absence of Phascolarctobacterium within the core community. Although both studies are based on relatively small cohorts, and a more extensive screening at the population level is needed, these data seem to suggest a certain level of country specificity in the gut microbiome structure, which may contribute to the immunological and metabolic peculiarities of the populations.

According to our findings, the bacterial species belonging to the Italian pan-microbiome showed two different types of MAC-degrading profiles, essentially characterized by a high or low content of glycosyl-hydrolase-coding sequences, respectively. As expected, the CAZyme distribution in the various species of the Italian GM was heterogeneous, and the absolute number of CAZymes was independent from the genome size. However, it should be mentioned that, for each identified species, the analysis of the CAZyme content was performed by using the type strain reference genome deposited in the NCBI database and therefore our classification was blind with respect to the possible strain-level functional variability in the CAZyme profile.

The GM species of the Italian pan-microbiome were successfully clustered into four CCGs according to the similarity of the CAZyme pattern: S. variabile (CCG1), E. rectale (CCG2), R. bromii (CCG3), and F. prausnitzii (CCG4). Interestingly, each of the identified CCGs was characterized by a peculiar structure in terms of CAZyme content. In particular, F. prausnitzii and R. bromii groups were the most enriched CCGs in terms of represented CAZyme functions, whereas F. prausnitzii and E. rectale groups were the most equipped CCGs in terms of CAZymes specifically involved in the breakdown of nondigestible carbohydrates and non-starch polysaccharides (i.e., xylans, pectins, and mannans). These observations suggest that the Italian pan-microbiome is diversified in at least

four patterns of carbohydrate degradation, raising several open questions related to: (i) the major determinants of the co-evolutionary processes underlying this differentiation; (ii) the relative contribution of host genetics, lifestyle and diet as drivers of this functional convergence; (iii) the ultimate connections between the observed CCGs and the host metabolic phenotype.

When exploring the quantitative distribution of CCGs in the individual microbiota from all subjects analyzed, we observed four robust clusters of subjects sharing a similar CCG profile, termed from CT1 to CT4. In particular, the CT1 and CT3 clusters included CCG4 (F. prausnitzii group) as the most prevalent CCG, being present in all individuals, and CCG3 (R. bromii group) and CCG1 (S. variabile group) as less prevalent, ancillary, and generally mutually exclusive groups. Conversely, CT4 was dominated by both CCG4 (F. prausnitzii group) and CCG2 (E. rectale group), which equally shared the ecosystem. Finally, CT2 showed CCG2 (E. rectale group) as the most prevalent group and CCG4 (F. prausnitzii group) as ancillary and less prevalent group, except for three subjects that were dominated by CCG1 (S. variabile group). These observations are indicative of a different ecological behavior for the diverse CCGs. Indeed, while CCG4 (F. prausnitzii group) appears to co-exist with all other CCGs, CCG2 (E. rectale group) and the CCGs 1 (S. variabile group)/3 (R. bromii group) are mutually exclusive. Confirming this, none of the CTs showed the simultaneous presence of CCG2 and CCG1 and/or CCG3. Taken together, these data suggest that the GM-host co-evolution process has resulted in the establishment of four well-defined functional steady states, i.e., the four CTs, each determined by the CCG propensity to share the same gut environment, and each conferring to the host a specific pattern of CAZymes.

In order to explore possible associations of these CTs with the host diet and health, we explored their variation in Italian healthy adults consuming a Mediterranean diet and obese T2D patients consuming a high-fat low-MACs diet. Interestingly, according to our data, most healthy subjects belonged to the CTs 2 and 4, which were characterized by the simultaneous presence of CCG4 (F. prausnitzii group) and CCG2 (E. rectale group). Conversely, the great majority of obese T2D patients belonged to CT1 and CT3, where CCG2 (E. rectale group)

was substituted by CCG3 (R. bromii group) and/or CCG1 (S. variabile group). Although caution is needed in interpreting results, the analysis presented here suggests that a high-fat low-MACs diet, in the context of metabolic deregulation, such as obesity and T2D, could force changes in the GM CTs, supporting the presence of CCG1 (S. variabile group) and/or CCG3 (R. bromii group) to the detriment of CCG2 (E. rectale group). Interestingly, compared to CCG1 and CCG3, the CCG2 showed higher levels of enzymes involved in the degradation of non-digestible carbohydrates and non-starch polysaccharides, which are indeed abundant MACs in the Mediterranean dietary regimen. Though preliminary, our data highlight a possible adaptive or maladaptive nature for each of the four CT steady states that describe the Italian pan-microbiome. Indeed, the steady states CTs 2 and 4, that were generally associated with healthy hosts, seem to be the result of an adaptive microbiomehost co-evolution process, in which the interplay between diet, gut microorganisms and the host can contribute to overall metabolic health. On the contrary, the CTs 1 and 3 that were associated with T2D and a high-fat low-MACs diet, may result from a maladaptive microbiome–host process, in which this type of diet has led to the selection of CT steady states able to contribute to metabolic and/or immunological deregulation. Supporting these hypotheses, patients suffering from Behcet's syndrome, a systemic inflammatory condition, showed a specific GM functional dysbiosis accompanied by a decreased production of SCFAs (Consolandi et al., 2015). Although the CAZyme profile has not yet been explored in patients with this disease, we can speculate that the observed reduction in SCFA biosynthesis could be the result of a maladaptive transition to CT1 or CT3, which would

REFERENCES


diminish the GM potential to provide the host with essential metabolites, such as butyrate, crucial to support immunological health.

## CONCLUSION

Our findings highlighted the existence of specific and welldefined GM functional layouts (CAZyTypes, CTs) for what concerns the ecosystem capacity to metabolize MACs, and support the hypothesis that the human GM has the ability to reconfigure its own CAZyme functional layout in response to dietary changes, with possible implications for the host health and metabolic regulation.

### AUTHOR CONTRIBUTIONS

SR and MS: conceived, designed and performed the analysis; SR, MS, MC, and ST: wrote the manuscript; EB, SQ, and PB: revised and edited the draft. All authors discussed the results, commented on the manuscript and approved the final version.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02079/full#supplementary-material

TABLE S1 | Prevalence data and average abundance of the identified 98 bacterial species in the dataset.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Soverini, Turroni, Biagi, Quercia, Brigidi, Candela and Rampelli. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Modulation of Immunological Pathways in Autistic and Neurotypical Lymphoblastoid Cell Lines by the Enteric Microbiome Metabolite Propionic Acid

*Richard E. Frye1,2\*, Bistra Nankova3 , Sudeepa Bhattacharyya1,2, Shannon Rose1,2, Sirish C. Bennuri 1,2 and Derrick F. MacFabe <sup>4</sup>*

*1Department of Pediatrics, University of Arkansas for Medical Sciences, Little Rock, AR, United States, 2Autism Research Program, Arkansas Children's Research Institute, Little Rock, AR, United States, 3New York Medical College, Valhalla, NY, United States, 4Kilee Patchell-Evans Autism Research Group, Alberta Children's Hospital Research Institute, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada*

#### *Edited by:*

*Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia*

#### *Reviewed by:*

*William Parker, Duke University, United States Joao Luiz Mendes Wanderley, Universidade Federal do Rio de Janeiro, Brazil*

*\*Correspondence:*

*Richard E. Frye rfrye@phoenixchildrens.com*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 31 July 2017 Accepted: 14 November 2017 Published: 22 December 2017*

#### *Citation:*

*Frye RE, Nankova B, Bhattacharyya S, Rose S, Bennuri SC and MacFabe DF (2017) Modulation of Immunological Pathways in Autistic and Neurotypical Lymphoblastoid Cell Lines by the Enteric Microbiome Metabolite Propionic Acid. Front. Immunol. 8:1670. doi: 10.3389/fimmu.2017.01670*

Propionic acid (PPA) is a ubiquitous short-chain fatty acid which is a fermentation product of the enteric microbiome and present or added to many foods. While PPA has beneficial effects, it is also associated with human disorders, including autism spectrum disorders (ASDs). We previously demonstrated that PPA modulates mitochondrial dysfunction differentially in subsets of lymphoblastoid cell lines (LCLs) derived from patients with ASD. Specifically, PPA significantly increases mitochondrial function in LCLs that have mitochondrial dysfunction at baseline [individuals with autistic disorder with atypical mitochondrial function (AD-A) LCLs] as compared to ASD LCLs with normal mitochondrial function [individuals with autistic disorder with normal mitochondrial function (AD-N) LCLs] and control (CNT) LCLs. PPA at 1 mM was found to have a minimal effect on expression of immune genes in CNT and AD-N LCLs. However, as hypothesized, Panther analysis demonstrated that 1 mM PPA exposure at 24 or 48 h resulted in significant activation of the immune system genes in AD-A LCLs. When the effect of PPA on ASD LCLs were compared to the CNT LCLs, both ASD groups demonstrated immune pathway activation, although the AD-A LCLs demonstrate a wider activation of immune genes. Ingenuity Pathway Analysis identified several immune-related pathways as key Canonical Pathways that were differentially regulated, specifically human leukocyte antigen expression and immunoglobulin production genes were upregulated. These data demonstrate that the enteric microbiome metabolite PPA can evoke atypical immune activation in LCLs with an underlying abnormal metabolic state. As PPA, as well as enteric bacteria which produce PPA, have been implicated in a wide variety of diseases which have components of immune dysfunction, including ASD, diabetes, obesity, and inflammatory diseases, insight into this metabolic modulator may have wide applications for both health and disease.

Keywords: mitochondrial disease, autism, propionic acid, short-chain fatty acids, microbiome, inflammation, epigenetics, histone deacetylase inhibitor

## INTRODUCTION

The human microbiome represents a diverse ecosystem of microbes housed in the human body. Microbial cells outnumber the cells in the human body by a factor of 10 and microbial genes out number human genes by a factor of over 100 (1–3). There is a particular focus on the enteric (gut) microbiota since it represents about 99% of the human microbiome (4). The importance of the enteric microbiome in relation to human health and disease has been recognized since it appears to influence the immune system (5), metabolic processes (6), gene expression (7, 8), the nervous system (9, 10), and behavior (9, 10). Disruption of the enteric microbiome has been implicated in a wide range of human diseases including depression and anxiety (11), gastrointestinal disorders (12), inflammatory airway disease (13), diabetes (14–16), obesity (17, 18), atopic disease (5), and neurodegenerative conditions (19). The enteric microbiome may be particularly important early in life around the time of birth as it has been linked to early brain development and behavior (9, 10, 20) and disruption and/or treatments (i.e. early antibiotics) early in life can influence the development of childhood diseases, particularly atopic disease (9, 10).

The mechanism in which the enteric microbiome modulates particular effects on the host is not completely clear, although several mediators are potential vehicles for such influence. Such mediators include lipopolysaccharides, peptidoglycans, shortchain fatty acids (SCFAs), neurotransmitters and gaseous molecules (21–23). We are particularly interested in SCFAs because of their role as both mediators of physiology and mitochondrial fuels. SCFA are particularly intriguing as they are derived as a consequence of fermenting carbohydrates and some proteins, and also present naturally or as an additive in many foods, in particular wheat and dairy. Thus, dietary variations can have a larger influence on their production (19, 24, 25). Of the SCFAs, propionic acid (PPA) has been of key interest because it has several links to autism spectrum disorder (ASD), a disorder which affects as many as ~2% of children in the United States. What is intriguing about ASD is that the etiology is largely unknown but is strongly influenced by both genetic and environmental factors (26, 27).

The enteric microbiome is a major environmental factor that may contribute to the etiology of ASD (2, 9, 10, 28). First, several factors which may have a direct effect on health through disruption of the microbiome are associated with increased risk of developing ASD, including dietary alteration, environmental exposures that disrupt enteric microbiome bacteria content and diversity, being born by C-section delivery which reduces maternal transfer of enteric and vaginal bacteria, increased antibiotic use which can destroy key bacteria in the enteric microbiome, formula feeding and early hospitalization (2, 9, 28). Second, specific bacteria, such as *Clostridia* spp., a major SCFA producer, have been repeatedly reported to be overrepresented in the ASD microbiome (29, 30). Third, exposure to PPA has been demonstrated in several animal models to result in the development of ASD-like behaviors and physiological changes to the brain similar to those found in ASD are seen in adult rats acutely exposed to PPA (24, 25, 31) and in juvenile rats systematically exposed to PPA pre- and postnatally (32–34).

Although the mechanism by which PPA influences host function is still unclear, data from the animal model of PPA induced ASD demonstrates neuroinflammation and electrophysiological disturbances as well as disruptions in lipid, mitochondrial and redox metabolism (24, 25, 31). We have performed a series of studies to demonstrate that changes in mitochondrial metabolism similar to those found in the animal model exposed to PPA are also found in humans. For example, we found that the unique pattern of biomarkers of mitochondrial dysfunction found in the PPA rodent model was also found in a subset of children with ASD (28, 35, 36). We also demonstrated that PPA modulates mitochondrial respiration in lymphoblastoid cell lines (LCLs) derived from children with ASD differently than LCLs derived from age and gender matched typically developing control LCLs (37).

PPA also could induce changes in host physiology through modulation of the immune system. The animal models of PPA induced ASD behavior demonstrates neuroinflammation but inflammatory mediators induced by PPA in human ASD cells has not been investigated. In this study, we investigate whether PPA can differentially regulate immune genes using our LCL model of ASD. We have developed a cell line model of ASD in which LCLs derived from individuals with autistic disorder (AD) are classified into two groups: those with normal mitochondrial function (AD-N) and those with atypical mitochondrial function (AD-A) (38–40). The AD-A LCLs have respiratory rates approximately twice that of control and AD-N LCLs and are very sensitive to *in vitro* increases in reactive oxygen species (ROS) (38–40). We recently demonstrated that this atypical increase in mitochondrial function characteristic of AD-A LCLs was associated with more severe repetitive behaviors in the children from which these LCLs were derived (40). In this way, we believe that the AD-A LCLs may represent a more severe ASD phenotype. Given the connection between metabolism and immune system (41), we hypothesize that the AD-A LCLs will demonstrate a greater activation of immune genes with PPA exposure as compared to the control and AD-N LCLs.

## MATERIALS AND METHODS

## LCLs and Culture Conditions

Lymphoblastoid cell lines were derived from white males diagnosed with AD chosen from pedigrees with at least other 1 affected male sibling (i.e., multiplex family) [mean (SD) age 7.3 (3.5) years]. These LCLs were obtained from the Autism Genetic Resource Exchange (Los Angeles, CA, USA) or the National Institutes of Mental Health (Bethesda, MD, USA) center for collaborative genomic studies on mental disorders. In our previous studies (37, 39, 40, 42–44), these LCLs where categorized into two different types of AD LCLs; ones with atypical mitochondrial respiration (AD-A) and those with normal respiration (AD-N). These metabolic groupings have been shown to be consistent and repeatable in our previous studies (37, 39, 40, 42–44). Eight pairs of AD-N and AD-A LCLs were age and gender matched to control LCLs. The sample size chosen was based on our previous studies. Control (CNT) LCLs were derived from healthy white male donors with no documented behavioral or neurological disorder and with no first degree relative suffering from any medical disorder that might involve mitochondrial dysfunction [mean (SD) age 7.5 (3.3) years]. CNT LCLs were obtained from Coriell Cell Repository (Camden, NJ, USA). Due to low availability of CNT LCLs which fit our criteria, a single CNT LCL line was paired with two AD LCL lines in one case (see **Table 1**). Also two AD-A LCLs were paired twice with AD-N LCLs. On average,


Table 1 | Lymphoblastoid cell lines used in this study.

*Three types of cell lines were used with two types of autistic disorder (AD) cell lines, characterized in our previous studies, and one type of control cell line.*

cells were studied at passage 12, with a maximum passage of 15. Genomic stability is very high at this low passage number (45, 46). Cells were maintained in RPMI 1640 culture medium with 15% FBS and 1% penicillin/streptomycin (Invitrogen, Grand Island, NY, USA) in a humidified incubator at 37°C with 5% CO2.

## PPA Exposure

Each group of LCLs were cultured with PPA 1 mM for 24 or 48 h or left untreated (0 mM). This concentration was selected because it provided optimal metabolic activation in our previous studies (37). The sodium propionate was buffered with sodium bicarbonate in the culture medium to prevent changes in pH which could cause changes in influx of PPA (47). As PPA is mostly disassociated at physiological pH, the effects of the PPA treatment are most likely a combination of both PPA and propionate.

#### Expression Studies

Total RNA samples from each LCL group were pooled together and after DNase treatment and purified using RNeasy Mini Kit (Qiagen Sciences, MD, USA) as described in our previous studies (48). The cDNA synthesis and microarray analyses were performed at Keck Affymetrix GeneChip Resource at Yale, New Haven, CT, USA (NIH Neuroscience Microarray Consortium) as previously described (48).

## Analytic Approach

Analysis of variance was conducted between the exposure conditions and different cell types. Genes showing expression of at least ≥2.0-fold were exported for functional annotation to several pathway analysis packages including Ingenuity Pathway Analysis (IPA) and Panther software. For the initial comparison of the effect of PPA for each exposure time on a particular LCL type, the statistical significance of the comparison was not considered as there was only an *N* of 1 for each example. When the ASD LCL types were compared to controls, the two PPA exposure times were combined and the genes selected not only showed a difference in expression of at least ≥2.0-fold but also a *p* < 0.05.

## RESULTS

## The Effect of PPA on Gene Expression for Each LCL Type

The change in gene expression resulting from 1 mM exposure to PPA for 24 and 48 h was determined for each LCL type separately. Table S1 in Supplementary Material demonstrates the number of genes up- and downregulated more than 2.0-fold for each LCL type.

The CNT LCLs demonstrated no upregulation or downregulation of known genes with 24 h PPA exposure and only one gene upregulated and downregulated with 48 h PPA exposure. Only the downregulated gene was associated with immune function. Panther analysis demonstrated no overrepresentation of immune genes associated with PPA exposure in CNT LCLs.

Exposure of AD-N LCLs to PPA for 24 h demonstrated no upregulated genes and downregulation of several immune genes including two major histocompatibility complex genes. Exposure of AD-N LCLs to PPA for 48 h demonstrated upregulation of two microRNA genes not known to be involved in immune function and downregulation of the gene for complement C4B. Panther analysis demonstrated overrepresentation of genes associated with major histocompatibility complex antigen with 24 h PPA exposure in AD-N LCLs (see **Table 2**).

Exposure of AD-A LCLs to PPA for 24 or 48 h demonstrated upregulation of several genes related to immune function, particularly several genes associated with immunoglobulin production and one gene related to activation of proinflammatory caspases. Downregulation of the gene for complement C4B was found for 24 h exposure and no genes were downregulated for 48 h exposure. Panther analysis demonstrated overrepresentation of many immune processes and proteins as result of PPA exposure to AD-A LCLs for 24 and 48 h, demonstrating that PPA did significantly activate immune processes for AD-A LCLs (**Table 2**).

### Comparison of PPA Effect on ASD LCLs as Compared to Control LCLs

To better understand how PPA exposure affects ASD LCLs differently than control LCLs, gene expression was compared between CNT LCLs and each ASD LCL group independently. Both the 24- and 48-h PPA exposure data was combined since the previous analysis demonstrated little difference between the changes in gene expression with these two different exposure durations. Table S2 in Supplementary Material outlines the genes that were upregulated or downregulated with PPA exposure for each ASD LCL group as compared to CNT LCLs. **Table 3** demonstrates the biological processes identified by the differential gene expression for AD-N and AD-A LCLs as compared to CNT LCLs. The major processes identified are also represented in **Figure 1**. Biological



process was the only Panther analysis used as it was the most robust for representing the difference in pathway activation.

This analysis suggests that both the AD-N and AD-A LCLs demonstrate change in immune genes as compared to CNT LCLs. Both AD-A and AD-N LCLs demonstrate an upregulation in genes associated with immunoglobulin production and adaptive immune responses without any downregulation in genes involved in these processes. AD-A LCLs demonstrate both upregulation and downregulation of genes involved in a wider variety of immune responses as compared to AD-N LCLs, including phagocytosis, complement system activation, B cell regulation, and B cell receptors. This suggests that AD-A LCLs may have a wider network of immune genes activated as compared to AD-N LCLs as well as CNT LCLs.

**Table 4** represents the top canonical pathways (*p* < 0.01) identified by IPA for the comparison between the AD-A and CNT LCLs. As we see, many of these processes are involved in immune activation and immune disorders. IPA also identified the top upstream regulators as RUNX3, ONECUT1, SNAI2, STAT5A, and TCF7. Interestingly, as will be discussed below, these genes are regulatory of both developmental and immune processes.

#### DISCUSSION

In this study, we examined the effect of PPA, a SCFA produced by enteric bacteria that are overrepresented in the ASD gut, on transformed B cells (LCLs) derived from children with ASD as well as controls. We examined two types of LCLs derived from children with ASD, those with mitochondrial dysfunction (AD-A) and those found to have mitochondrial function similar to controls (AD-N). We hypothesized that PPA would activate immune pathways in ASD LCLs since the PPA animal model of Table 3 | Biological processes panther overrepresentation analysis of genes differentially expressed in autism cell lines as compared to control cell lines.


*Numbers represent number of genes associated with the identified biological process.*

ASD demonstrates neuroinflammation and immune activation, including increased GFAP immunoreactivity in the hippocampus, increased activation of microglia, and increased interleukin (IL)-6 (24, 25, 31). We further hypothesized that the AD-A LCLs would have a greater enhancement of immune pathways since this is a more severe ASD phenotype and since optimal mitochondrial function is required for appropriate immune function and response (41).

Exposure to PPA for either 24 or 48 h resulted in upregulation in genes associated with immune system activation in AD-A LCLs, particularly genes involved in immunoglobulin production. This effect was not seen in CNT or AD-N LCLs. In fact, there was a decrease in major histocompatibility complex antigen genes in AD-N LCLs exposed to PPA for 24 h. We then compared the effect of PPA on ASD LCLs as compared to the effect of PPA on CNT LCLs. We found that both the AD-N and AD-A LCLs demonstrated changes in gene expression as compared to the control LCLs with a significant change in genes related to immune pathways almost exclusively. Although the AD-N LCLs

Table 4 | Top canonical pathways identified using Ingenuity Pathway Analysis (IPA).

B cell development T helper cell differentiation Primary immunodeficiency signaling Graft-versus-host disease signaling Calcium-induced T lymphocyte apoptosis IL-4 signaling Altered T cell and B cell signaling in rheumatoid arthritis Antigen presentation pathway

demonstrated activation of immune pathways, the AD-A LCLs demonstrated a wider range of genes and processes involved in immune pathways. In addition, IPA analysis of AD-A LCL gene expression changes identified canonical pathways almost exclusively related to immune function.

Several of the genes identified by the IPA analysis are involved in regulation of the immune system and may be linked to ASD. Several genes are linked to regulation of T cells. TCF7 is a T lymphocyte-specific enhancer of the CD3-Epsilon T cell antigen receptor complex. Interestingly TCF7 expression may be regulated by beta-catenin (49). This is intriguing since betacatenin has been shown to be dysregulated in an animal model of ASD (50). STAT5 is induced in response to T cell activation with cytokines, most notably IL-2, and is believed to be involved in the effect of IL-2 in the immune response and may be involved in the suppression of IL-3 production. This is interesting as IL-2 is produced by neurons and astrocytes, is important in brain development and normal brain physiology and has been implicated in neurodegenerative disease, cognitive dysfunction and has been linked to ASD (51). RUNX3 is also important in immune system function as well as neuronal development. RUNX3 is essential during thymopoiesis where it modulates the development of CD8 T cells, thus having an important role in immune system development through lineage specification (52). Interestingly, RUNX3 is involved in the TNF-beta signaling cascade (53), a cytokine whose dysregulation has been correlated with ASD severity (51). RUNX3 appears to have an important role in the development of proprioceptive afferent neurons in mice, resulting in ataxia (54), a neurological finding that is not uncommon in ASD. Other genes identified are related to B cell function. SNAI2 is an evolutionarily conserved zinc finger transcription factor which plays an important role in prenatal fetal development, most notably the development of neural crest-derived cells and adipocytes (55). SNAI2 is also involved in regulation of B cells and can promote the aberrant survival and malignant transformation of mammalian pro-B cells otherwise slated for apoptotic death (56) and has antiapoptotic effects (57).

In conclusion, ASD is being recognized as having a very strong immune component to its etiology (58). Several models of ASD demonstrate immune dysregulation, including prenatal exposure to immune challenges (59, 60). In fact two animal models have been developed to parallel prenatal exposure to autoantibodies (61), including fetal brain antibodies (62) and antibodies to the folate transporter (63, 64). The microbiome is being recognized as important in the etiology of neurodevelopmental disorders (9, 10), potentially through modulation of the immune system (65) through enteric metabolites (65) including SCFAs like PPA (24, 25, 31). It is important to note the effects of SCFA on gene expression and inflammation are complex, and include histone deacetylase activity, activation of free fatty acid G-coupled receptor and mitochondrial inflammatory signaling cascades, which may or may not be mutually reinforcing. Furthermore, we do not yet know if the effects found in our LCL model also occur in patients, as many effects of SCFA, in particular PPA and butyrate, are dose and tissue dependent, and have different effects at key developmental time periods (9, 10, 24, 31, 48, 66, 67). Nonetheless, this study provides insight into the mechanism in which the microbiome may influence the immune system to result in disease and demonstrates the predisposition of certain cells to be sensitive to microbiome metabolites. It also may lead to further reevaluation of the widespread use of PPA in agriculture and the food industry (24, 31). Certainly, further research is needed in this area to better define the role of the microbiome and microbial metabolites in immune modulation and disease.

#### AUTHOR CONTRIBUTIONS

The conception and design of the work was agreed upon by all authors as was the drafting and final approval of the manuscript. BN, SR, and SB were involved in laboratory analysis. SB was involved in data analysis. RF, DM, SB, SR, and SB were involved in interpretation of data.

#### ACKNOWLEDGMENTS

We thank the autism families that participated in the Autism Genetic Research Exchange and the studies at the National Institutes of Mental Health. We would also like to express our utmost thanks to David Patchell-Evans, for his tireless devotion to persons with autism, and his daughter, Kilee Patchell-Evans. Our heartfelt thanks go out to countless parents and caregivers of persons with autism who have shared their stories.

#### FUNDING

This research was supported by the Arkansas Biosciences Institute (Little Rock, AR, USA), The Jonty Foundation (St Paul, MN), The Autism Research Institute (San Diego, CA), the Gupta Family

### REFERENCES


Foundation (Atherton, CA) and the Jager Family Foundation (Chicago, IL) to REF, and GoodLife Children's Charities, Autism Canada and Autism Research Institute to DFM.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://www.frontiersin.org/article/10.3389/fimmu.2017.01670/ full#supplementary-material.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Frye, Nankova, Bhattacharyya, Rose, Bennuri and MacFabe. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Secretome of Intestinal Bacilli: A Natural Guard against Pathologies

Olga N. Ilinskaya<sup>1</sup> \*, Vera V. Ulyanova<sup>1</sup> , Dina R. Yarullina<sup>1</sup> and Ilgiz G. Gataullin<sup>2</sup>

<sup>1</sup> Department of Microbiology, Kazan Federal University, Kazan, Russia, <sup>2</sup> Department of Surgery and Oncology, Regional Clinical Cancer Center, Kazan, Russia

Current studies of human gut microbiome usually do not consider the special functional role of transient microbiota, although some of its members remain in the host for a long time and produce broad spectrum of biologically active substances. Getting into the gastrointestinal tract (GIT) with food, water and probiotic preparations, two representatives of Bacilli class, genera Bacillus and Lactobacillus, colonize epithelium blurring the boundaries between resident and transient microbiota. Despite their minor proportion in the microbiome composition, these bacteria can significantly affect both the intestinal microbiota and the entire body thanks to a wide range of secreted compounds. Recently, insufficiency and limitations of pure genome-based analysis of gut microbiota became known. Thus, the need for intense functional studies is evident. This review aims to characterize the Bacillus and Lactobacillus in GIT, as well as the functional roles of the components released by these members of microbial intestinal community. Complex of their secreted compounds is referred by us as the "bacillary secretome." The composition of the bacillary secretome, its biological effects in GIT and role in counteraction to infectious diseases and oncological pathologies in human organism is the subject of the review.

#### Edited by:

Wesley H. Brooks, University of South Florida, United States

#### Reviewed by:

Marisa Mariel Fernandez, Instituto de Estudios de la Inmunidad Humoral (CONICET-UBA), Argentina Andrey Tatarenkov, University of California, Irvine, United States

#### \*Correspondence:

Olga N. Ilinskaya ilinskaya\_kfu@mail.ru

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 25 July 2017 Accepted: 17 August 2017 Published: 01 September 2017

#### Citation:

Ilinskaya ON, Ulyanova VV, Yarullina DR and Gataullin IG (2017) Secretome of Intestinal Bacilli: A Natural Guard against Pathologies. Front. Microbiol. 8:1666. doi: 10.3389/fmicb.2017.01666 Keywords: Lactobacillus, Bacillus, gastrointestinal tract, metabolites, secretome

## INTRODUCTION

The human gut microbiota consists of about 1500 microbial species which constitute 10<sup>12</sup> bacteria per gram of stool (Browne et al., 2016). The gastrointestinal microbiota of healthy human adults consists primarily of bacteria belonging to phyla Firmicutes and Bacteroidetes, and to a lesser extent to phyla Actinobacteria and Proteobacteria (Dethlefsen et al., 2008; Yang and Jobin, 2014). Density and composition of microbiota varies along both the length of the gut and the cross-section (Nava and Stappenbeck, 2011; Tropini et al., 2017). Changes in nutrients, availability of oxygen, and presence of immune effectors in local microenvironment determine species variation and abundance (Donaldson et al., 2016). The most dominant taxa have the highest stability in the gut (Martí et al., 2017). Rapidly dividing facultative anaerobes from Lactobacillaceae and Enterobacteriaceae dominate in small intestine, while saccharolytic representatives of Bacteroidales and Clostridiales orders are abundant in the large intestine (Donaldson et al., 2016; Tropini et al., 2017). Mucin-utilizing species of Akkermansia and Bacteroides are followed by aerotolerant Proteobacteria and Actinobacteria in direction to the

**Abbreviations:** CSF, competence and sporulation factor; EVs, extracellular vesicles; GIT, gastrointestinal tract; NO, nitric oxide; NOS, nitric oxide synthase; RNases, ribonucleases; SCFAs, short-chain fatty acids.

epithelium (Tropini et al., 2017). Proteobacteria and Firmicutes are found in crypts and represent the stock for reseeding the colon because they are protected from the luminal flow (Tropini et al., 2017). At least 50–60% of the bacterial genera from the intestinal microbiota of a healthy individual produce spores which facilitate both microbiota persistence and transmission (Browne et al., 2016). The majority of gut bacteria are transient populations which pass through the lumen of the lower GIT (Tropini et al., 2017).

Composition of gut microbiota varies among individuals with geographic provenance, gender, age, diet, malnutrition, and intake of probiotics or antimicrobial agents (Panda et al., 2014; Haro et al., 2016; Odamaki et al., 2016; Maffei et al., 2017; Martí et al., 2017; Singh et al., 2017). Alterations in the composition of the gut microbiota and reductions in microbial diversity lead to different disorders such as inflammatory conditions of the intestine (inflammatory bowel disease, irritable bowel syndrome, colorectal cancer) (Gagnière et al., 2016; Ghoshal et al., 2017; Rapozo et al., 2017), type 2 diabetes, obesity, anorexia nervosa, forms of severe acute malnutrition, cardiovascular diseases (atherosclerosis, hypertension, heart failure) (Tang et al., 2017), neurobehavioral diseases (autism spectrum disorder, major depression) (Clemente et al., 2012). The microbiota is increasingly recognized for its ability to maintain homeostasis in health and disease influencing host appetite, function of the nervous system and several complex host behaviors (Sharon et al., 2016; Winek et al., 2016; van de Wouw et al., 2017; Vuong et al., 2017). A healthy gut microbiota can be defined by the presence of the various microbial species that enhance metabolism, resistance to infection and inflammation, prevention against cancer and autoimmunity.

Lactobacilli are historically considered as integral part of human intestinal microbiota. Today, a large body of evidence indicates that only a small number of Lactobacillus species are true autochthonous inhabitants of the mammalian intestinal tract and that most lactobacilli present are allochthonous members derived from food or oral cavity (Reuter, 2001; Walter, 2008). Lactobacillus spp. content of fecal samples is characterized by temporal fluctuations and lack of stability (Walter et al., 2001; Vanhoutte et al., 2004). Attempts to divide lactobacilli into resident and transient ones are hardly legitimate since the style of nutrition significantly affects their contents in the intestine. Lactobacilli together with enterococci dominate in the duodenum and in the jejunum (Reuter, 2001; Hammes and Hertel, 2015), although their absolute number increases along the intestine from duodenum to colon (Derrien and van Hylckama Vlieg, 2015). However, they constitute only a minor fraction within the human adult fecal microbiota, i.e., around 0.01 to 0.6% of total bacterial counts (Harmsen et al., 2002; Matsuda et al., 2009).

Metagenomics widely used for study of gut microbiota is unable to detect bacteria at concentrations less than 10<sup>5</sup> bacteria per gram (Lagier et al., 2015). A culturing approach that uses high-throughput culture conditions in combination with matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry and 16S rRNA sequencing for taxonomic identification and referred to as culturomics has allowed significant increase in a number of bacteria discovered in human GIT including species known in humans but not in the gut, species previously considered unculturable as well as new species. Considerable part of them is represented by Firmicutes (Lagier et al., 2016) including Bacillus species (Lagier et al., 2015; Mourembou et al., 2016; Senghor et al., 2017).

For a long time, the representatives of Bacillus genus, unlike the species of Lactobacillus, were not considered as a part of the normal intestinal microbiome. Being isolated from feces, Bacillus spp. as soil microorganisms were considered transient. Recent studies show that they are present in the GIT in the amounts significantly higher than what can be explained by their ingestion with food only. Bacillus spp. (B. pumilus, B. licheniformis, B. clausii, B. subtilis, B. megaterium, B. mediterraneensis, B. thuringiensis) have been isolated from the healthy human GIT, where they are well-adapted and are more colonizing than transient (Fakhry et al., 2008; Alou et al., 2016; Lopetuso et al., 2016). In environment, the vegetative forms of Bacillus are present usually near decomposing plants and in their rhizosphere. In the soil they exist mainly in the form of spores, which germinate in the digestive tract of humans and animals. Germination of Bacillus spores in the human small intestine and transient colonization should be considered as a part of the life cycle of human-associated Bacillus species (Hong et al., 2009). In the GIT, spores not only germinate but also are formed again from vegetative cells during a time shorter than in the laboratory (Tam et al., 2006; Ghelardi et al., 2015).

Thus, it can be concluded that GIT microbiota including Bacillus and Lactobacillus species undergoes constant dynamic change. In our opinion, distinguishing between the resident and transient intestinal inhabitants is less relevant issue compared to the study of molecular pool released by them. Most of bacteria absorbed in the body can supplement the gastrointestinal microbiome (Derrien and van Hylckama Vlieg, 2015). The challenge of identifying the "spheres of influence" of the transient microbiota in the human body has not been solved, and has not even been formulated, with the exception of some aspects of the pathogen entry into the body. Its bottleneck is the lack of data on the complexes secreted by this microbiota, their components, functions and the interaction between components, namely, the composition and biological role of the secretome.

We consciously narrowed the spectrum of secretome producers observed here to two representatives of the Bacilli class, genera Lactobacillus and Bacillus, due to their wide distribution and high probability of entering the human body. The ingestion of microorganisms occurs with food, water and bacterial probiotics. Facultative aerobic bacilli represent a smaller proportion of the intestinal microbiota than anaerobic bacteria (Rajilic-Stojanovi ´ c and de Vos, 2014 ´ ), but they actively influence the microbial community of GIT and also the whole organism thanks to the great diversity of secreted compounds. Our studies of biopsies taken during surgical intervention in patients with diagnosed colorectal cancer revealed the presence of Bacilli closely associated with intestinal epithelium, traditionally considered as transient ones (Siraj et al., 2015). Since secretory components can be studied only in culturable microorganisms, the insufficiency of genomic analysis of intestinal microbiota and the transition to functional analysis became evident (Derrien and

Ilinskaya et al. Bacilli Secretome

van Hylckama Vlieg, 2015; Lagier et al., 2015). In this regard, the identification, characterization, and elucidation of the functional role of the components secreted by the minority of the intestinal microbial community, namely representatives of the Bacilli class in GIT, is an actual task. The modern concept of a gut–brain axis (McKay et al., 2017) must be detailed and refined taking into account the spectra of compounds secreted by Bacilli, namely low- and high-molecular components of the secretome and EVs, which affect the whole body and shape human health.

### BENEFICIAL EFFECTS OF Lactobacillus AND Bacillus

Historically, species of Lactobacillus and Bacillus are found in the traditional fermented food products possessing beneficial properties for the intestinal function (Nithya et al., 2012; Satish Kumar et al., 2013; Lee et al., 2016; Sornplang and Piyadeatsoontorn, 2016; Marco et al., 2017) and are widely used as components of commercially available probiotics: DE111 (Deerland Enzymes, United States), Enterogermina (Sanofi Winthrop, Italy), Biosubtyl (Biophar, Vietnam), Biosporin (Biopharma, Russia), BioSpora (Klaire Labs, United States), Blicheni and Zhengchangsheng (Northeast Pharmaceutical Group, China), GanedenBC 30 (Ganeden, United States), Lactobacterin (Microgen, Russia), HOWARU or DR20 (Danisco, United States), Yakult (Yakult, Japan), PCC (Probiomics, Australia). In food industry, lactobacilli are applied as starter cultures in the production of fermented milk products, cheese, sausages, bread, kimchi, pickles, and yogurts, the latter accounting for the largest share of sales (Giraffa et al., 2010; Tamang et al., 2016). The administration of probiotics has been shown to favorably alter the intestinal microbiota balance, enhance intestinal integrity and motility, inhibit the growth of harmful bacteria and increase resistance to infections (Tamang et al., 2016).

As a part of GIT microbiota Bacilli participate in metabolism of dietary components, xenobiotics and drugs helping to maintain intestinal homeostasis and host health (Jandhyala et al., 2015; Rowland et al., 2017). The beneficial effect of probiotics on GIT is mediated by influence on composition, diversity and function of the intestinal microbiota as well as whole human organism. Probiotics suppress pathogenic bacteria and favor beneficial ones via competition for nutrients, especially for shared limited resource like iron, competitive attachment to the epithelium, formation of substrates for growth, production of waste products and antimicrobial compounds, strengthening of the barrier function of the epithelium, and modulation of innate immunity (Thomas and Versalovic, 2010; Bermudez-Brito et al., 2012; Stubbendieck and Straight, 2016). For example, consumption of B. coagulans was shown to increase beneficial groups of bacteria in the gut of 65–80 years old humans and production of anti-inflammatory cytokines (Nyangale et al., 2015).

The efficacy of Lactobacillus and Bacillus in the prevention and/or treatment of intestinal diseases such as diarrhea, colitis, irritable bowel syndrome, irritable bowel disease, and colorectal cancer was demonstrated (Camilleri, 2006; Sazawal et al., 2006; Rafter et al., 2007; Pillai and Nelson, 2008; Ghouri et al., 2014; Urgesi et al., 2014; Choi et al., 2015; Matsuoka and Kanai, 2015; Majeed et al., 2016; Zhang et al., 2016). In particular, treatment of colorectal colitis in mice with probiotic B. subtilis restored balance in gut microflora: beneficial species of Bifidobacterium, Lactobacillus, and Butyricicoccus spp. were increased, while gut damage-promoting species of Acinetobacter sp., Ruminococcus sp., Clostridium spp., and Veillonella sp. were decreased (Zhang et al., 2016). B. subtilis also retained gut barrier integrity, decreased the endotoxin concentration and reduced gut inflammation (Zhang et al., 2016; Bene et al., 2017). Sporulation of B. subtilis plays a major role in the development of GALT – gut lymphoid tissue associated with the gastrointestinal mucosa - and in the diversity of the primary antibody population ("preimmune" repertoire) in rabbits (Rhee et al., 2004). Bacillus spp. like other strains isolated from human stool were able to bind the human norovirus strains, the cause of acute viral gastroenteritis and foodborne diseases, around the outer cell surfaces and pili structures (Almand et al., 2017). The interaction between virus and bacteria is hypothesized to help the host immune system to better recognize infectious particles.

The ratio between the two major phyla inhabiting the human GIT, Firmicutes and Bacteroidetes, reflects the GIT status during the life and diseases. It is significantly decreased in infants and elderly individuals as compared to adults (0.4, 0.6, and 10.9, respectively) (Mariat et al., 2009) and lowers upon antibiotic-associated diarrhea, coeliac disease, Crohn's disease, and ulcerative colitis (Ott et al., 2004; Panda et al., 2014; Carding et al., 2015; Quagliariello et al., 2016). A decrease in populations of Ruminococcus and Lactobacillus was observed in a rat model of colorectal cancer (Zhu et al., 2014). Microbial content of the patients with diagnosed colorectal cancer and healthy individuals differed significantly. Firmicutes and Fusobacteria were overrepresented whereas Proteobacteria were under-represented in patients. In addition, Lactococcus and Fusobacterium exhibited a relatively higher abundance while Pseudomonas and Escherichia– Shigella were reduced in cancerous tissues compared to adjacent non-cancerous ones (Gao et al., 2015). Bacilli were shown to decrease quantitatively upon type 2 diabetes (Sankar et al., 2015). The possibility of using probiotics in the therapy of diseases, namely allergy, asthma, diabetes, cardiovascular diseases is discussed (Ebel et al., 2014).

Viability is by definition a prerequisite for probiotic effectiveness as it is essential for colonization of intestinal mucosa, displacement of pathogens and immunomodulation. Viable bacteria demonstrate adhesive and antagonistic properties and produce a large number of extracellular enzymes and biologically active compounds (Shobharani and Halami, 2014). Nevertheless, there is increasing evidence that isolated bacteriaderived molecules and surface components (e.g., cell wall components, cell wall associated proteins, S-layer proteins) potentiate probiotic benefits attributed earlier to viable probiotic bacteria (Lahtinen, 2012; Ruiz et al., 2014).

Despite more than a century of active use of probiotics, initiated by I. Mechnikov in 1907, the majority of modern reviews assessing the effectiveness of these drugs confirm the need for

further studies to determine the exact mechanisms of positive effects of probiotics on the human body. Both live probiotic bacterial cells and their metabolites can be useful in treatment of intestinal diseases (Okamoto et al., 2012). The molecular basis for the effectiveness of probiotics remains unexplored or only partially studied. The future study aimed at deciphering the mechanisms that determine the probiotic properties of bacteria will certainly allow expanding the areas of scientifically proven probiotic use in medicine.

## MICROBIOTA–HUMAN METABOLIC INTERACTION

Between the gut microbiota and host organism there is an extremely complex relationship that affects the human metabolism, immunity and health (Marchesi et al., 2016). This crosstalk is mediated by nutrients, metabolites, antimicrobial compounds. It was demonstrated that the psychological and physical stress of a host affects its gut microbiota and, in turn, gut microflora can modulate host's mood and appetite (Sandrini et al., 2015). Gut microbiota is regulated by the host through production of non-specific antimicrobial peptides such as defensins (Nakamura et al., 2016), secreted IgA which provides the selection and the maintenance of the commensal bacteria (Fransen et al., 2015), and miRNAs specifically regulating bacterial transcripts and affecting bacterial growth (Liu and Weiner, 2016). It was proved that host genetic background affects the composition and function of the gut microbiota, altering the production of microbial metabolites and intestinal inflammation (Lamas et al., 2016). For example, the microbiota of mice deficient in caspase recruitment domain family member 9 (CARD9) failed to metabolize tryptophan that increased host susceptibility to colitis (Lamas et al., 2016).

Microbial species are recognized by host's immune system. Commensal bacteria have immunomodulatory properties that allow them inducing tolerogenic immune responses against themselves and contributing to host protective immune responses against pathogens (Bene et al., 2017; Guo et al., 2017; Shi et al., 2017). It is known that probiotics affect key signaling pathways, such as NFκB and MAPK, through the pattern-recognition receptors (TLR, NOD) (Bermudez-Brito et al., 2012) enhancing the production of anti-inflammatory cytokines (Nyangale et al., 2015) and reducing the emergence of proinflammatory ones (Selvam et al., 2009). The gut microbiota is able to influence host antigen production by human monocytederived dendritic cell populations in a species-specific manner (Bene et al., 2017).

Gut microbiota affects host physiology by releasing bioactive metabolites including antibiotics, enzymes, vitamins and amino acids (choline, methionine, vitamin B), minerals (cobalt, iodine, selenium, and zinc) and energy metabolites (SAM, acetyl-CoA, NAD+, α-KG, and ATP), SCFAs (acetate, propionate, butyrate, caproate, and valerate), neurotransmitters [acetylcholine, dopamine, noradrenaline, serotonin, and γ-aminobutyric acid (GABA)], hormones, bacterial antigens, pathogen-associated molecular patterns, and toxins (Donia et al., 2014; Luber and Kostic, 2017). These molecules enter host circulation thereby mediating the link between the gut and other organs (brain, lung, liver, muscle) (Shukla et al., 2017) and modulate physiological pathways and even behavior (van de Wouw et al., 2017; Vuong et al., 2017). The influence of gut microbiota on the epigenetic regulation of host genes via DNA methylation and histone modifications has been demonstrated (Ye et al., 2017). Contactindependent metabolic exchange helps signal dispersal among neighboring cells as well as its blockage when needed.

Bacteria produce a lot of chemically diverse metabolites with poorly understood function. Bacillus species are among the most frequent producers of bioactive secondary metabolites (800 compounds), while lactobacilli produce 100s of compounds (Bérdy, 2005). Known to date, the results of the study of representatives of the Bacilli class colonizing the human GIT mostly refer to ascertaining their positive, less often negative, impact on the body. We tried to systematize the available knowledge about compounds and complexes produced by these bacteria which serve as effectors triggering certain processes in the body (**Figure 1**), and identify the stage responsible for actually registered "influence." Bacilli introduced into the GIT through the consumption of fermented food do integrate the resident microbiome (Derrien and van Hylckama Vlieg, 2015) and contribute to its regulatory and health promoting action producing a variety of substances ranging from low molecular weight regulatory agents to proteins and peptides with antimicrobial and antitumor effects. Metabolites secreted by bacteria form a coat around the cells which contributes to nutrient supply, communication, and protection from damage caused by direct interaction with other species or their metabolites. Due to diffusion, the concentration of extracellular metabolites decreases with distance. To ensure that secreted components will reach their targets bacteria utilize membranous vesicles for their transportation.

## TOP-APPRECIATED COMPONENTS OF Bacilli SECRETOME

It is well-appreciated that complex of **enzymes** (proteases, amylases, cellulases, lipases) secreted by Bacilli aid in digestion of food components in GIT (Khochamit et al., 2015; Keller et al., 2017). Bile salt hydrolases of lactobacilli reduce blood cholesterol and diminish the risk for cardiovascular diseases (Patel et al., 2010; Kumar et al., 2012). Recently, an antagonistic role of L. johnsonii La1 extracellular bile salt hydrolase against intestinal protozoan parasite Giardia duodenalis was revealed (Travers et al., 2016). Other enzymes like N-acylated homoserine lactone (AHL)-lactonase help to modulate the microbiota content by decreasing the number of quorum-sensing pathogenic bacteria in the GIT through direct disruption of their signal molecules (Vinoj et al., 2014; Zhou et al., 2016). Moreover, many enzymes are involved in the formation of metabolites which possess their own biological activities. For instance, during fermentation of milk and other proteinaceous products lactobacilli are able to release biologically active peptides with angiotensin I-converting enzyme (ACE)-inhibitory activity. Among these antihypertensive

FIGURE 1 | Compounds secreted by the representatives of Lactobacillus and Bacillus mediating their beneficial effects in the GIT. Top-appreciated compounds are shown in upper part of panels, under-appreciated compounds – in lower parts. Both Lactobacillus and Bacillus are minor part of GIT microbiota. Of them, Lactobacillus spp. dominate quantitatively secreting a few compounds, while Bacillus spp. are less abundant but produce a variety of secreted substances with a wide spectrum of activities. Representatives of the Bacilli class were isolated from colon epithelia biopsy of the patients with diagnosed colorectal cancer. Atomic force microscopy images of L. plantarum (photo provided by Dr. Dina Yarullina) and B. pumilus (photo kindly provided by Dr. Galina Yakovleva) were obtained in air (contact mode) of stationary phase cells that were deposited on glass and dried prior analysis. Bacteria were identified using MALDI-TOF technique and 16S RNA sequencing. Representative AFM images show nanoscale structures on the cell surface and around the cells attributed to EVs.

peptides β-casein-derived tripeptides (lactotripeptides) are most studied (Hayes et al., 2007; Fekete et al., 2015).

**Short-chain fatty acids** are formed upon dietary carbohydrates fermentation by both Bacilli genera, studied in this review. The most common SCFA is lactate followed by acetic, propionic, butyric acids and minor isobutyrate, 2-methylpropionate, valerate, isovalerate, hexanoate. SCFAs are one of the most important gut microbial products affecting a range of host processes including energy utilization, host– microbe signaling, and control of colonic pH. Decrease of a luminal pH creates an environment favoring beneficial species like Faecalibacterium prausnitzii and inappropriate for many others bacteria and yeasts (Nyangale et al., 2015). SCFAs positively influence the gut motility (Yang and Chiu, 2017) and intestinal secretion (Bhattarai et al., 2017), inhibit proliferation of tumor cells by apoptosis induction, stimulate production of insulin-like growth factor 1 promoting bone growth and remodeling (Yan et al., 2016), cause epigenetic modifications, regulate blood pressure and inflammation (Natarajan and Pluznick, 2014). The multifaceted roles of SCFAs nominate them for the key molecular link between diet, the microbiome and health.

Lactate produced by Lactobacillus provides an unfavorable environment for the growth of many pathogenic bacteria, it also acts as a permeabilizer of the Gram-negative bacterial outer membrane, thus increasing the susceptibility of pathogens

Ilinskaya et al. Bacilli Secretome

to antimicrobial molecules, e.g., bacteriocins or host lysozyme (Alakomi et al., 2000). Strains of B. licheniformis and B. coagulans also ferment different sugars with formation of lactic acid (Wang et al., 2011; Nyangale et al., 2015). Butyrate is the local energy source for colonocytes (LeBlanc et al., 2017); also it plays an important role in maintenance of the gut barrier function through stimulation of tight junction integrity and mucin production (Peng et al., 2009; Jung et al., 2015). The SCFAs produced by the human gut microbiota are transported from the gut lumen with the bloodstream to a variety of different organs, where they are used in lipid and energy metabolism, particularly by the hepatocyte cells of the liver, which use propionate for gluconeogenesis, whilst acetate and butyrate are mostly involved in lipid biosynthesis (den Besten et al., 2013a). Besides, SCFAs appear to exert regulatory effects on gluconeogenesis and lipogenesis mediated by peroxisome proliferator-activated receptor gamma (PPARγ) (den Besten et al., 2015) and protein kinases, such as AMP-activated protein kinase (Peng et al., 2009; den Besten et al., 2015) or mitogen-activated protein kinases (MAPK) (Jung et al., 2015). SCFAs have been reported to represent the natural ligands for free fatty acid receptors 2 and 3 (FFAR 2/3) (former G protein-coupled receptors, GPR43 and GPR41), involved in the regulation of lipid and glucose metabolism and found on a wide range of cell types, including enteroendocrine and immune cells (den Besten et al., 2013b). Moreover, as far as these receptors are expressed on neurons of the peripheral, autonomic and somatic nervous systems, SCFAs can modulate neuronal activity and visceral reflexes (Nøhr et al., 2015). SCFAs are considered as promising for the prevention and treatment of the metabolic syndrome, certain types of cancer, bowel disorders, such as ulcerative colitis, Crohn's disease, and antibiotic-associated diarrhea (den Besten et al., 2013b; Ríos-Covián et al., 2016).

**Hydrogen peroxide (H**2**O**2**)** production by lactobacilli has been suggested to be a non-specific antimicrobial defense mechanism. L. jensenii, L. crispatus, L. gasseri, and L. acidophilus are the most common H2O2-producing lactobacilli inhabiting the human intestine and are often applied as probiotic supplements in the food industry (Martín and Suárez, 2010; Hertzberger et al., 2014). In gastrointestinal environment, SCFAs and bacteriocins have been considered as key antimicrobial factors, whereas the impact of H2O<sup>2</sup> production remains underappreciated. However, H2O2, like other reactive oxygen species, exerts strong cytotoxicity against microorganisms. Although H2O<sup>2</sup> itself is not highly reactive, it can readily diffuse across cellular membrane and through the Fenton reaction form highly reactive hydroxyl radicals, which cause oxidative damage to major biological macromolecules, e.g., oxidation of protein thiols, peroxidation of lipids, DNA base damage, and strand breakage of nucleic acids (Imlay, 2003; Valko et al., 2005). The role for H2O<sup>2</sup> in the anti-Salmonella activity of L. johnsonii NCC533, the human intestinal isolate and a probiotic strain, was revealed in vitro (Pridmore et al., 2008). L. delbrueckii VI1007 produces at least three growthinhibiting factors, other than lactic acid, one of which has been identified as H2O<sup>2</sup> (Van De Guchte et al., 2001). H2O<sup>2</sup> may contribute to the maintenance of the normal microbiota. Especially for the vaginal microbiota, strong evidence exists that colonization with H2O2-producing lactobacilli is associated with lower rates of bacterial vaginosis and HIV acquisition (Wilks et al., 2004; Balkus et al., 2012). Moreover, H2O<sup>2</sup> may exert immunomodulatory properties. It was showed that H2O2, produced by L. crispatus M247, acts as a signal transducing molecule activating peroxisome proliferator activated receptor γ (PPAR-γ), which plays a central role in regulation of intestinal inflammation and homeostasis (Voltan et al., 2008). L. johnsoniiderived H2O<sup>2</sup> has been shown to affect the activity of indoleamine 2,3-dioxygenase, an important immune modulator, both in vitro and in the rat model of type 1 diabetes (Valladares et al., 2013).

**Poly P,** a linear polymer of over 700 phosphate residues, is synthesized by Lactobacillus with the help of polyphosphate kinase (Alcántara et al., 2014). It suppresses the oxidantinduced intestinal permeability inducing cytoprotective heat shock proteins in mouse small intestine through activation of integrin β1-p38 MAPK pathway (Segawa et al., 2011). Poly P was shown to improve the inflammation grade and survival rate in mice model of colitis (Segawa et al., 2011) and to inhibit viability of colon cancer cells via apoptosis through activation of the ERK pathway (Sakatani et al., 2016).

**Indole** derivatives formed from tryptophan by Lactobacillus cells act on the aryl hydrocarbon receptor in intestinal immune cells increasing IL-22 production which, in turns, beneficially impacts the immune system, enhances antifungal resistance and protection of mucosa from damage (Zelante et al., 2013; Lamas et al., 2016; Etienne-Mesmin et al., 2017). The main inhibitory neurotransmitter in the brain, **GABA,** is produced with the help of glutamate decarboxylase expressed by multiple strains of Lactobacillus (Barrett et al., 2012; Yunes et al., 2016).

Bacilli synthesize B-group **vitamins** including folate and biotin during the fermentation of foods in GIT and can exchange them, thereby enabling the survival of organisms that do not synthesize those (Magnúsdóttir et al., 2015).

A number of **peptide and lipopeptide antibiotics and bacteriocins** are produced by Bacilli both involving ribosomes and non-ribosomally (Zacharofa and Lovitt, 2012; Sumi et al., 2015; Zhao and Kuipers, 2016). These structurally diverse compounds suppress the growth of competing species and pathogens through different mechanisms primarily connected to membrane permeabilization (Fiedler and Heerklotz, 2015; Shobharani et al., 2015). Antimicrobial peptides of Bacilli were shown to be active against pathogenic bacteria such as Staphylococcus aureus, methicillin resistant S. aureus, Clostridium perfringens, Klebsiella sp., and common food spoilage bacteria such as B. cereus, Escherichia coli, Listeria monocytogenes, Pseudomonas aeruginosa, Aeromonas sp., Serratia marcescens, Pasteurella haemolytica, Salmonella enteritidis, and S. gallinarum (Ahmadova et al., 2013; Martinez et al., 2013; Beric et al., 2014 ´ ; Ayed et al., 2015; Khochamit et al., 2015; Shobharani et al., 2015; Collins et al., 2016; Lim et al., 2016; Chauhan et al., 2017; Perez et al., 2017). Bacteriocins attract great interest with regard to their potential use as food preservatives (De Vuyst and Leroy, 2007; Kaškoniene et al., 2017 ˙ ) and are regarded as a promising alternative to prevent gastrointestinal infections (Dobson et al., 2012).

Lactobacillus produces a number of bacteriocins usually active against closely related Gram-positive bacteria which are likely to reside in the same ecological niche. Most Lactobacillus bacteriocins are small, heat-stable cationic peptides which form pores in the cytoplasmic membrane of sensitive bacteria and thus cause leaking of target cells (Oscariz and Pisabarro, 2001). Other bacteriocins interrupt production of peptidoglycan or act by interfering with essential enzyme activities in susceptible bacteria (Servin, 2004). Bacteriocins from Lactobacillus are generally recognized as being inactive against Gram-negative organisms. However, it has been reported that bacteriocin from L. plantarum TN635 is active against Salmonella enterica ATCC43972, Pseudomonas aeruginosa ATCC 49189, Hafnia sp. and Serratia sp. (Smaoui et al., 2010). Moreover, a small bacteriocin (<6.5 kDa) produced by L. acidophilus IBB 801 and designated as acidophilin 801, displayed bactericidal activity against E. coli Row and Salmonella panama 1467 (Zamfir et al., 1999). Bacteriocin OR-7 produced by L. salivarius NRRL B-30514 resulted in reduction of Campylobacter jejuni colonization in chicken GI tracts when was added into feed. Interestingly, OR-7 had high sequence similarity to acidocin A, which was previously identified from L. acidophilus and had activity only to Gram-positive bacteria (Stern et al., 2006).

Bacillus is considered to be the second most important bacteriocin producer following lactic acid bacteria which differs from the latter by broad antimicrobial spectrum (Abriouel et al., 2011; Ayed et al., 2015). Bacteriocins and bacteriocin-like inhibitory substances produced by Bacillus exhibit antibacterial activity toward Gram-positive and Gram-negative bacteria as well as fungi, however, activity against Gram-positives is comparatively higher (Hyronimus et al., 1998; Rey et al., 2004; Arias et al., 2013; Beric et al., 2014 ´ ; Chopra et al., 2014; Barbosa et al., 2015; Shobharani et al., 2015; Lee et al., 2016; Lim et al., 2016; Liu et al., 2017; Perez et al., 2017). Species of Bacillus differ by their antimicrobial potential (Perez et al., 2017). Non-ribosomal peptide antibiotics produced by Bacillus (bacitracin, proticin, lichenicidin, bacillaene) are essential for the protection of these bacteria from predation and antibiotics produced by other species (Rey et al., 2004; Barger et al., 2012; Alvarez-Ordóñez et al., 2014; Müller et al., 2014). Bacillus spp. were shown to produce a mixture of different lipopeptides with antimicrobial activities (Huang et al., 2006). B. subtilis produces surfactins, fengycins and iturins in a ratio of 6:37:57 (Fiedler and Heerklotz, 2015; Perez et al., 2017). The less abundant surfactins unlike other types of Bacillus lipopeptides exhibit a broad range of antimicrobial activities and possess antiviral action (Huang et al., 2006). They protect bacilli against extracellular antibiotic-containing vesicles of other species (Brown et al., 2014) and inhibit phospholipase A2 resulting in subsequent downregulation of pro-inflammatory cytokines and upregulation of anti-inflammatory cytokines (Selvam et al., 2009).

Probiotic effect of B. subtilis was shown to be connected to **competence and sporulation factor**, a small quorum-sensing peptide involved in bacteria communication, proliferation and sporulation (Okamoto et al., 2012). CSF activates the Akt and p38 MAPK pathways and exerts its anti-inflammatory effect by downregulation of pro-inflammatory mediators (IL-4, IL-6, and CXCL-1), the upregulation of anti-inflammatory IL-10, and the induction of cytoprotective heat shock protein Hsp27 in the intestinal epithelia (Okamoto et al., 2012). The similar effects were observed for two peptides secreted by B. megaterium isolated from human ileal biopsies of healthy volunteers (Di Luccia et al., 2016). Effects of CSF depend on its uptake by an organic cation transporter-2 in intestine which helps the host to monitor and respond to changes in the behavior or composition of colonic microbiota (Fujiya et al., 2007).

#### UNDER-APPRECIATED COMPONENTS OF Bacilli SECRETOME

**Extracellular and surface-associated proteins** secreted by commensal bacteria play an important role in gut colonization and persistence. Moreover, some of them can interact directly with mucosal cells, activating signaling pathways that lead to different cytokine secretion and gene expression profiles (Tsilingiri and Rescigno, 2013). For instance, two secreted proteins p75 and p40 (also known as Msp1 and Msp2) of L. rhamnosus GG have been demonstrated to prevent cytokineinduced cell apoptosis by activating the antiapoptotic protein kinase B and by inhibiting the pro-apoptotic MAPK (Yan and Polk, 2002; Yan et al., 2007), reduce TNF induced epithelial damage in the colon and as a result promote epithelial homeostasis (Yan et al., 2007). Homologs of genes that encode for p40 and p75 were also found in the genomes of L. casei and L. rhamnosus; the proteins from L. casei BL23 were demonstrated to elicit similar host responses (Bäuerl et al., 2010).

**Secreted hydrolytic enzymes** contribute to probiotic effects of Bacilli due to their ability to decompose food polymers releasing digestive discomfort. However, accumulating data indicate that these proteins might be involved in a complex interaction with host and its microbiota. Hydrolases demonstrate direct antimicrobial activity. Proteases, glycoside hydrolases and DNases participate in dispersal of bacterial biofilms and inhibition of biofilm formation (Chen et al., 2013; Nguyen and Burrows, 2014; Watters et al., 2016; Fleming and Rumbaugh, 2017). Extracellular nuclease, NucB, from B. licheniformis, was shown to digest extracellular DNA in biofilms of staphylococci and streptococci associated with chronic rhinosinusitis proving enzyme effectiveness in eradicating biofilms of multidrugresistant bacteria (Shields et al., 2013). Secretion of lowmolecular-weight guanyl-preferring ribonucleases (RNases) is a distinct feature of some Bacillus species (Ulyanova et al., 2016). A well-studied representative of these RNases, binase from B. pumilus, has manifested antitumor (Ulyanova et al., 2011; Cabrera-Fuentes et al., 2013; Mitkevich et al., 2013) and antiviral activities (Shah Mahmud and Ilinskaya, 2013; Ilinskaya and Shah Mahmud, 2014; Shah Mahmud et al., 2016, 2017; Müller et al., 2017). KRAS which has mutations in about 40% of patients with colorectal cancer (Prior et al., 2012) was shown to be a direct target for antitumor binase (Ilinskaya et al., 2016). 2<sup>0</sup> ,30 -cGMP generated by binase upon RNA cleavage (Sokurenko et al., 2015)

may also contribute to modulation of host cells physiology, since it is able to duplicate its counterpart 3<sup>0</sup> ,50 -cGMP (Boadu et al., 2001) and can exhibit its own regulatory functions. Extracellular cGMP enhances extracellular adenosine and reduces uric acid levels which may render tissue protective effect upon injury (Jackson et al., 2013). Guanylate cyclase which catalyzes the synthesis of cGMP from GTP is represented in the cell membranes along the intestine. The cGMP signaling regulates intestinal fluid and electrolyte balance, epithelial homeostasis, mucosal barrier integrity, visceral sensation through ERK and AKT pathways (Han et al., 2011; Lin et al., 2012; Hannig et al., 2014; Lan et al., 2016). Increased cGMP levels in the colon epithelium activate antioxidant gene expression (Wang et al., 2017). cGMP expression is significantly decreased upon ulcerative colitis and colon cancer (Lin et al., 2012; Lan et al., 2016; Pattison et al., 2016). cGMP and ways for enhancement of its production are considered for treatment of irritable bowel syndrome by decreasing of gastrointestinal pain and abdominal sensory symptoms (Lan et al., 2016) and as a tool for tumor suppression (Pattison et al., 2016).

Recently, gut Firmicutes were shown to produce **peptide aldehydes**, cell-permeable protease inhibitors with a half-life of hours, which target cathepsins in the host lysosome blocking immune recognition of these mutualistic species and enabling them to reside in gut epithelial (Guo et al., 2017).

**Nitric oxide** is a well-known ubiquitous molecular mediator produced in mammals by the NOS isoforms at a catalytic site comprising a heme associated with a biopterin cofactor. Genome sequencing has shown the presence of genes encoding for proteins that are highly homologous to the oxygenase domain of mammalian NOS in bacteria, including those of the class Bacilli: S. aureus (Bird et al., 2002; Salard et al., 2006), B. subtilis (Adak et al., 2002), B. anthracis (Midha et al., 2005; Salard et al., 2006), Geobacillus stearothermophilus (Sudhamsu and Crane, 2006), L. fermentum (Morita et al., 1997), and L. plantarum (Adawi et al., 1997; Iarullina et al., 2006; Iarullina and Ilinskaia, 2007; Yarullina et al., 2015). So, intestinal Bacilli have NOS that is evolutionary related to the mammalian enzymes. Moreover, as bacteria have the most ancient version of NOS, it was hypothesized that Eukaryotes acquired NOS from bacteria by horizontal gene transfer (Gusarov et al., 2013). Recently, the conservation of NOS-derived NO-heme receptor signaling between bacteria and mammals was proved (Kinkel et al., 2016). NO as reactive oxygen molecule is widely considered as important participant in the immune system of different organisms to confront microbial infections. Thus, inhibition of bacterial NOS has the potential to improve the efficacy of antimicrobials used to treat infections by Gram-positive pathogens S. aureus and B. anthracis possessing this enzyme (Holden et al., 2015). Commensal microbiotaderived NO has been shown to influence host physiology. NO synthesized by L. plantarum takes part in the regulation of intestinal motility in rat (Yarullina et al., 2016). Being a signaling molecule, NO released by B. subtilis in the intestine of Caenorhabditis elegans initiates a signaling cascade that results in the induction of 65 genes, including hsps and several other genes that have been implicated in longevity and stress resistance (Gusarov et al., 2013). Involvement of bacterial NO in human cardiovascular system is under investigation (Cabrera-Fuentes et al., 2016).

**Ferrichrome** of L. casei was identified as a tumor-suppressive molecule on colon cancer cells which induces apoptosis via activation of c-jun N-terminal kinase (JNK) (Konishi et al., 2016). Many siderophore-binding proteins were found in EVs of B. subtilis (Dubois et al., 2009; Brown et al., 2014). Siderophores can endow bacilli advantage in competition for low-available iron with pathogenic bacteria. Iron cations are potent crosslinkers of the biofilm matrix (Chen and Stewart, 2002) and their chelation causes dispersal of biofilms (Sobke et al., 2012).

Over the last decade, **extracellular vesicles** have emerged as prominent vehicles of biological signals. Intense research on that topic revealed that EVs play important roles in bacterial physiology and pathogenesis, ranging from secretion and delivery of biomolecules (for example, toxins, DNA, or quorum sensing molecules) over stress response and biofilm formation to immunomodulation and adherence to host cells (Roier et al., 2016). Both Bacillus and Lactobacillus species were reported to produce EVs, spherical membranous structures of 20–150 nm in diameter (Brown et al., 2014, 2015; Avila-Calderón et al., 2015; Behzadi et al., 2017; Li et al., 2017). EVs are formed both in planktonic cultures and bacterial biofilms where they help to maintain biofilm cohesion. The quantity of EVs varies with the strain (Brown et al., 2014), conditions and stage of growth (Kim et al., 2016). Thus, EVs can be produced by Bacilli in GIT.

The EVs are enriched with proteins, lipids, nucleic acids, and metabolites which exhibit biological activities. Release of vesicular cargo is achieved by direct intercellular transfer mediated by the membranes fusion (Kim et al., 2016; Stubbendieck and Straight, 2016) or by production of special molecules like lipopeptide surfactin which disrupts EVs unspecifically (Brown et al., 2014). Therefore, proteins with specific biological activities can be directly delivered inside EVs into other cells ensuring their penetration. In Gram-positive bacteria, proteins secreted via specific pathways are believed to be important for nutrient acquisition, detoxification, competitive survival, and communication (Brown et al., 2015). Recent findings support the importance of EVs for interaction of bacteria with each other and the host cells (Kim et al., 2016).

Extracellular vesicles carry hydrolytic enzymes for nutrient acquisition from extracellular complex substrates or key nutrients to feed sibling cells and contain specific agents for antagonizing competing species. In EV important for survival compounds are protected from damage retaining activities much longer and can be transported in concentrated amounts for long distances from producing cells. Among these compounds are antibiotics and hydrolytic enzymes including peptidoglycandegrading hydrolases (Mashburn and Whiteley, 2005; Alves et al., 2016; Stubbendieck and Straight, 2016). EVs isolated from B. subtilis contain proteins which are mostly associated with metabolic pathways including biosynthesis of secondary metabolites (Brown et al., 2014; Kim et al., 2016). Proteins with oxidoreductase and nucleotide binding activities are abundant in vegetative EVs, while proteins with hydrolytic, nucleic acid binding, and structural activity are predominant in sporulating EV (Kim et al., 2016). In EVs of sporulating B. subtilis superoxide

dismutase, alkaline phosphatase III, polyketide synthase PKsM (associated with antibiotic activity) were identified (Kim et al., 2016). Sunl protein which confers self-immunity to antibiotic sublancin and many siderophore-binding proteins were found in EVs of B. subtilis (Dubois et al., 2009; Brown et al., 2014). The targeted lysis of EVs by surfactin of B. subtilis is hypothesized to guard bacilli from alien EVs and disrupt cell signaling by means of EVs in competing populations (Stubbendieck and Straight, 2016). EVs were also shown to adsorb phages (Biller et al., 2014).

Interaction of EVs with the host is specific to the microorganism from which the EVs were produced and is based on the lipid content and cargo of the EVs (Brown et al., 2015). Gram-positive bacterial EVs are composed of various fatty acids which might have a positive effect on host organism (Rivera et al., 2010). EVs were shown to elicit protective immune response in host (Vargas et al., 2015). For example, treatment of C. elegans with EVs originated from L. plantarum WCFS1 led to increased transcription of host defense genes, cpr-1 and clec-60, and thus provided protection against vancomycinresistant Enterococcus faecium. Moreover, in human Caco-2 cells these EVs had similar effect, leading to the upregulation of REG3G, which is functionally similar to clec-60, and CTSB, the human ortholog of cpr-1 (Li et al., 2017). The EVs from B. lentus isolated from Korean soybean fermented food induced apoptosis of human colon carcinoma cells HCT116 (Yang et al., 2016). EVs derived from L. rhamnosus GG are likely to be implicated in the anti-cancer activity as they induce apoptosis in the hepatic adenocarcinoma cell line HepG2 via augmentation of the expression ratio between pro- and anti-apoptotic genes bax/bcl-2 (Behzadi et al., 2017).

#### CONCLUSION AND FURTHER PERSPECTIVES

Now, it has become clear that studies on phylotype profiling are limited to the identification of microbial constituents, where information is lacking about the molecular interaction of bacterial communities with the host. Lactobacilli are wellrepresented in the human GIT and secrete a number of compounds which have direct and indirect effects on the health of GIT and organism as a whole. Species of Bacillus genus are much less abundant but are capable of producing several times more extracellular molecules than lactobacilli. Many of

## REFERENCES


them still require exploration. Further deep studies are needed for better understanding of the complex interactions between human organism and its microbiota, clarification of the particular mechanisms underlying remarkable beneficial properties of probiotic Bacilli, and the specific action of innumerous secreted low- and high-molecular weight compounds and their vesicular transportation.

## AUTHOR CONTRIBUTIONS

The basis of this review was the experimental work of the group led by ONI, in which a variety of biological activities of bacilli and lactobacilli secretome was established. ONI developed the main idea of this review, collected literature data, verified the text and coordinated it with co-authors carrying out the general guidance of the review. The part of the review devoted to the properties of representatives of the genus Bacillus and the components of their secretome belongs to VVU. She structured the review and created the scheme summarizing the main idea of the review on the diversity of components secreted by the representatives of Bacilli class. DRY, being a specialist in the field of probiotic activity of lactobacilli, analyzed data on the influence of Lactobacillus and their metabolites on intestinal functions and microbiota, and host organism. IGG collected literature data on the microflora of the human intestine. On the basis of his analysis of epithelial biopsy samples obtained during the operations of patients with diagnosed colorectal cancer, a conclusion about the contribution of Bacilli class to the microbiota closely associated with the epithelium was made.

#### FUNDING

The research was performed within the Russian Government Program of Competitive Growth of Kazan Federal University and supported by the Russian Science Foundation grant No 14-14-00522.

## ACKNOWLEDGMENTS

Authors are thankful to Anna Makeeva for contribution to the Figure and to Galina Yakovleva for providing AFM image of B. pumilus.






AMP-activated protein kinase in Caco-2 cell monolayers. J. Nutr. 139, 1619– 1625. doi: 10.3945/jn.109.104638



alters tryptophan metabolite levels in BioBreeding rats. FASEB J. 27, 1711–1720. doi: 10.1096/fj.12-223339


growth. Proc. Natl. Acad. Sci. U.S.A. 113, E7554–E7563. doi: 10.1073/pnas.1607 235113


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Ilinskaya, Ulyanova, Yarullina and Gataullin. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Role of Lactobacillus reuteri in Human Health and Diseases

#### Qinghui Mu, Vincent J. Tavella and Xin M. Luo\*

Department of Biomedical Sciences and Pathobiology, Virginia-Maryland College of Veterinary Medicine, Virginia Tech, Blacksburg, VA, United States

Lactobacillus reuteri (L. reuteri) is a well-studied probiotic bacterium that can colonize a large number of mammals. In humans, L. reuteri is found in different body sites, including the gastrointestinal tract, urinary tract, skin, and breast milk. The abundance of L. reuteri varies among different individuals. Several beneficial effects of L. reuteri have been noted. First, L. reuteri can produce antimicrobial molecules, such as organic acids, ethanol, and reuterin. Due to its antimicrobial activity, L. reuteri is able to inhibit the colonization of pathogenic microbes and remodel the commensal microbiota composition in the host. Second, L. reuteri can benefit the host immune system. For instance, some L. reuteri strains can reduce the production of pro-inflammatory cytokines while promoting regulatory T cell development and function. Third, bearing the ability to strengthen the intestinal barrier, the colonization of L. reuteri may decrease the microbial translocation from the gut lumen to the tissues. Microbial translocation across the intestinal epithelium has been hypothesized as an initiator of inflammation. Therefore, inflammatory diseases, including those located in the gut as well as in remote tissues, may be ameliorated by increasing the colonization of L. reuteri. Notably, the decrease in the abundance of L. reuteri in humans in the past decades is correlated with an increase in the incidences of inflammatory diseases over the same period of time. Direct supplementation or prebiotic modulation of L. reuteri may be an attractive preventive and/or therapeutic avenue against inflammatory diseases.

#### Keywords: Lactobacillus reuteri, probiotic, microbiota, immune system, inflammatory diseases

## INTRODUCTION

Probiotics are defined as "live microorganisms which, when administered in adequate amounts, confer a health benefit on the host" by the World Health Organization. While the idea to use probiotics for health benefits is not new, the interest has significantly increased in recent years (Islam, 2016). This may be due, in part, to the increase in antibiotic resistance particularly in the treatment of diseases in the gastrointestinal (GI) system, as well as an increasing desire by the public for natural health promotants. Those probiotic microorganisms that have been shown to have beneficial properties include Lactobacillus spp., Bifidobacterium spp., Saccharomyces boulardii, Propionibacterium spp., Streptococcus spp., Bacillus spp., Enterococcus spp., and some specific strains of Escherichia coli (Kechagia et al., 2013).

There are certain criteria that a probiotic must have to be considered efficacious. These include the capacity to survive in the GI tract, a high resistance to gastric acids, the lack of any transferable antibiotic resistance genes, and the capacity to exert clear benefits in the

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Julio Galvez, Universidad de Granada, Spain Michael Gänzle, University of Alberta, Canada Teresa Zotta, Consiglio Nazionale Delle Ricerche (CNR), Italy

#### \*Correspondence:

Xin M. Luo xinluo@vt.edu

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 29 September 2017 Accepted: 04 April 2018 Published: 19 April 2018

#### Citation:

Mu Q, Tavella VJ and Luo XM (2018) Role of Lactobacillus reuteri in Human Health and Diseases. Front. Microbiol. 9:757. doi: 10.3389/fmicb.2018.00757

**206**

host (Montalban-Arques et al., 2015). Probiotics promote a healthy body through diverse mechanisms. A widespread generalization describing common mechanisms among studied probiotic genera includes colonizing resistance, producing acid, and short chain fatty acid (SCFA), regulating intestinal transit, normalizing perturbed microbiota, increasing enterocyte turnover, and competitive exclusion of pathogens (Hill et al., 2014). Though not widely observed, there are a lot of effects among specific probiotic species, some being strain specific. For instance, some probiotic strains can improve host food digestion by metabolizing bile salt or complementing the functions of missing digestive enzymes (Amara and Shibl, 2015; Shi et al., 2016).

Lactobacillus spp. are one of the most widely used probiotics and can be found in a large variety of food products throughout the world (Giraffa et al., 2010). The genus Lactobacillus comprises a large heterogeneous group of Gram-positive, nonsporulating, facultative anaerobic bacteria which include L. acidophilus, L. rhamnosus, L. bulgaricus, L. casei, and L. reuteri. This genus plays a very important role in food fermentation and can also be found in the GI system of humans and animals in variable amounts depending on the species, age of the host, or location within the gut (Duar et al., 2017).

Animal studies and preclinical results have shown that Lactobacilli may help in the prevention and treatment of numerous GI tract disorders. Among these disorders are enteric infections, antibiotic-associated diarrhea, necrotizing enterocolitis in preterm neonates, inflammatory bowel disease, colorectal cancer, and irritable bowel syndrome (Lebeer et al., 2008). Although the GI tract is the site where Lactobacilli are believed to show the most benefits, probiotic applications of some Lactobacillus strains at other sites of the body have been reported. These include the prevention and treatment of urogenital diseases and bacterial vaginosis in women, atopic disease, food hypersensitivity, and the prevention of dental caries (Lebeer et al., 2008).

One species of Lactobacillus, L. reuteri has multiple beneficial effects on host health such as prevention and/or amelioration of diverse disorders. L. reuteri was first isolated in 1962. It has been characterized as heterofermentative species that grows in oxygen-limited atmospheres and colonizes the GI tract of humans and animals (Kandler et al., 1980). The fact that it normally colonizes the GI tract may be the reason it confers great probiotic properties. This organism can withstand a wide variety of pH environments, employs multiple mechanisms that allow it to successfully inhibit pathogenic microorganisms, and has been shown to secrete antimicrobial intermediaries (Jacobsen et al., 1999; Valeur et al., 2004).

L. reuteri has been shown to be one of the truly indigenous bacteria of the human GI tract (Sinkiewicz, 2010). It naturally colonizes a wide range of vertebrates, including pigs, rodents, and chickens. In fact, it has gone through long-term evolution to diversify into host-adapted lineages (Oh et al., 2010; Walter et al., 2011). This organism is most typically found in the proximal digestive tract of the host (Frese et al., 2013). Several studies have assessed the safety of this organism in adults, children, infants, and even in an HIV-infected population (Wolf et al., 1998; Valeur et al., 2004; Weizman and Alsheikh, 2006; Mangalat et al., 2012; Jones et al., 2012a,c; Hoy-Schulz et al., 2016). The results showed that a dose as high as 2.9 × 10<sup>9</sup> colony-forming units (cfu)/day was still well tolerated, safe, and efficacious in humans. There have also been numerous articles enumerating the benefits of L. reuteri as a probiotic. These benefits include promoting health, reducing infections, improving feed tolerance, enhancing the absorption of nutrients, minerals, and vitamins, modulating host immune responses, promoting gut mucosal integrity, and reducing bacterial translocation (Tubelius et al., 2005; McFall-Ngai, 2007; Indrio et al., 2008; Spinler et al., 2008; Hou et al., 2015). In the current review, we will focus on the particular probiotic, L. reuteri, and discuss its beneficial functions in promoting health and preventing infections and diverse diseases.

## PROBIOTIC PROPERTIES OF L. reuteri

There are some prerequisites for becoming potential probiotics: to survive in low pH and enzyme-enriched environments, to adhere to epithelium for host-probiotic interaction, competition with pathogenic microorganisms, and most importantly, safety. L. reuteri meets all of these requirements. Here, additional probiotic properties of L. reuteri are discussed that contribute to its diverse beneficial effects on host health and disease prevention and/or amelioration (**Figure 1**).

#### Gut Colonization of L. reuteri

Built for digestion and absorption, some sites of the GI system have developed to be harsh for microorganism colonization. Examples of this can be seen in the low pH conditions caused by gastric acids and bile salts in upper small intestine. Thus, the very first step of colonizing the GI tract is to survive in such environments. Multiple L. reuteri stains are resistant to low pH and bile salts (Seo et al., 2010; Krumbeck et al., 2016). This resistance is believed to be at least partially dependent on its ability to form biofilms (Salas-Jara et al., 2016).

L. reuteri is capable of attaching to mucin and intestinal epithelia, and some strains can adhere to gut epithelial cells in a range of vertebrate hosts (Li et al., 2008; Hou et al., 2014, 2015). A possible mechanism for adherence is the binding of bacterial surface molecules to the mucus layer. Mucus-binding proteins (MUBs) and MUB-like proteins, encoded by Lactobacillalesspecific clusters of orthologous protein coding genes, serve as adherence mediators, or so-called adhesins (Roos and Jonsson, 2002; Kleerebezem et al., 2010; Gunning et al., 2016). The considerable diversity of MUBs among L. reuteri strains and the variation in the abundance of cell-surface MUBs significantly correlates with their mucus binding ability (Mackenzie et al., 2010). The strain-specific role of MUBs in recognizing mucus elements and/or their capability of promoting aggregation can explain the contribution of MUBs on the adherence of L. reuteri. Factors that mediate the attachment to the surfaces include multiple large surface proteins (Walter et al., 2005; Wang et al., 2008; Frese et al., 2011), MUB A (Jensen et al., 2014), glucosyltransferase A (GtfA) and inulosucrase (Inu) (Walter et al., 2008), and D-alanyl ester (Walter et al., 2007).

As L. reuteri that has colonized to the host GI tract can form biofilms, efforts have been made to study the regulation of L. reuteri biofilm secretion and its association with the adherence of bacteria to host GI epithelium. By doing in vitro biofilm assay, Water, J. et al. uncovered the contribution of GtfA and Inu in the biofilm formation of L. reuteri TMW1.106 (Walter et al., 2008). The in vivo biofilm formation of L. reuteri strains seems to be dependent on the host origin of the strains. In one study, nine L. reuteristrains isolated from different hosts (human, mouse, rat, chicken, and pig) were given to germ-free mice and the biofilms were evaluated after 2 days. Interestingly, only rodent strains were able to form biofilms and adhere to the forestomach epithelium, although the luminal populations were comparable among strains of different origins (Frese et al., 2013). Another study by the same authors showed that a specialized transport pathway (the SecA2-SecY2 system) was unique in the rodent and porcine strains (Frese et al., 2011). By using a rodent strain L. reuteri 100-23, they compared extracellular and cell wall-associated proteins between the wild-type strain and the secA2 mutant. Only one surface protein, L. reuteri 70902, was absent in the secA2 mutant. In vivo colonization studies showed that the absence of L. reuteri 70902 leads to almost completely eliminated biofilm formation. This strongly suggests that L. reuteri 70902 and the SecA2-SecY2 system are key factors regulating biofilm production from L. reuteri 100-23 in germ-free mice (Frese et al., 2013). Another group investigated the role of two-component systems bfrKRT and cemAKR in in vitro biofilm formation of L. reuteri 100-23 (Su and Ganzle, 2014). They found the deletion of certain genes in the operons enhanced the adherence and

biofilm formation. However, the contribution of the bfrKRT and cemAKR to in vivo biofilm formation remains to be elucidated. The role of exopolysaccharide (EPS) in assisting colonization was also examined with L. reuteri 100-23. The production of EPS was eliminated due to a mutation of the fructosyl transferase (ftf) gene (Sims et al., 2011). After administration to Lactobacillus-free mice, compared to the wild-type strain, the colonization of the ftf mutant in the forestomach and cecum was largely impaired. This indicates EPS production can enhance the colonizing ability of strain 100-23 in the gut. Interestingly, L. reuteri RC-14 has been demonstrated to be able to penetrate mature E. coli biofilm and become part of it (McMillan et al., 2011). Recently, L. reuteri was delivered as a biofilm on microsphere and such delivery was found to promote the adherence of L. reuteri to intestinal epithelium and enhance its probiotic property (Olson et al., 2016; Navarro et al., 2017).

### Production of Metabolites With Health-Promoting Effect

The antimicrobial and immunomodulatory effects of L. reuteri strains are linked to their metabolite production profile. Here, we discuss a few well-studied metabolites with regard to the probiotic potential of L. reuteri.

#### Reuterin

Most L. reuteri strains of human and poultry lineage are able to produce and excrete reuterin, a well-known antimicrobial compound (Talarico et al., 1988; Talarico and Dobrogosz, 1989; Cadieux et al., 2008; Jones and Versalovic, 2009;

Mishra et al., 2012; Greifova et al., 2017). Reuterin is a mixture of different forms of 3-hydroxypropionaldehyde (3-HPA) (Talarico and Dobrogosz, 1989). It is known that L. reuteri can metabolize glycerol to generate 3-HPA in a coenzyme B12-dependent, glycerol dehydratase-mediated reaction (Talarico and Dobrogosz, 1990; Chen and Chen, 2013). The production of 3-HPA has also been demonstrated in a few other bacterial species (Zhu et al., 2002; Raynaud et al., 2003; Yang et al., 2007). However, L. reuteri is unique in its ability to produce and secrete 3-HPA in a manner more than its bioenergetics requirement (Stevens et al., 2011). Moreover, the antimicrobial activity of reuterin seems to rely on the spontaneous conversion of 3-HPA to acrolein, a cytotoxic electrophile (Stevens and Maier, 2008; Engels et al., 2016). Reuterin can inhibit a wide range of microorganisms, mainly Gram-negative bacteria (Cleusix et al., 2007). Not surprisingly, most Lactobacillus species are resistant to reuterin, among which L. reuteri strains exert the most resistance (Jones and Versalovic, 2009; Mishra et al., 2012). In addition to its antimicrobial property, reuterin is able to conjugate heterocyclic amines, which also seems to be dependent on the formation of acrolein (Engels et al., 2016). This suggests acrolein is an essential compound in the activity of reuterin.

Apart from reuterin, several other antimicrobial substances, including lactic acid, acetic acid, ethanol, and reutericyclin, have been determined as products of some L. reuteri strains (Ganzle and Vogel, 2003; Burge et al., 2015; Gopi et al., 2015; Yang Y. et al., 2015; Greifova et al., 2017). With the synthesis of these substances, L. reuteri has been shown to be effective against a variety of GI bacterial infections. These infections include Helicobacter pylori, E. coli, Clostridium difficile, and Salmonella (Reid and Burton, 2002; Cherian et al., 2015; Abhisingha et al., 2017; Genis et al., 2017). One of the more notable illustrations of the efficacy of L. reuteri as a probiotic against infections is the use of L. reuteri to treat H. pylori. H. pylori infection is a major cause of chronic gastritis and peptic ulcers, as well as a risk factor for gastric malignancies (Franceschi et al., 2007; Lesbros-Pantoflickova et al., 2007; Park et al., 2007). The use of L. reuteri against H. pylori has been explored in many studies (**Table 1**). It has been suggested that L. reuteri works by competing with H. pylori and inhibiting its binding to glycolipid receptors (Mukai et al., 2002). The competition reduces the bacterial load of H. pylori and decreases the related symptoms (Lionetti et al., 2006; Francavilla et al., 2008). Some studies have shown that L. reuteri has the potential to completely eradicate H. pylori from the intestine (Ojetti et al., 2012). Importantly, L. reuteri is advantageous in the treatment of H. pylori as the supplementation eradicates the pathogen without causing the common side effects associated with antibiotic therapies (Francavilla et al., 2014).

A considerable amount of research has been done to determine the beneficial effects of L. reuteri against viruses and/or fungi. There is evidence showing the benefit of L. reuteri against pneumoviruses, circoviruses, rotaviruses, coxsackieviruses, and papillomaviruses (Shornikova et al., 1997a,b; Preidis et al., 2012; Ang et al., 2016; Brenner et al., 2016; Piyathilake et al., 2016; Karaffova et al., 2017). It has been suggested that L. reuteri ameliorates viral infection by regulating the microbiota and secreting metabolites that have antiviral components (Ang et al., 2016). Furthermore, some studies suggest that L. reuteri may have antifungal properties as well, where L. reuteri antagonizes, stops the growth of, and eventually kills various species of Candida (Jorgensen et al., 2017).

#### Histamine

A few strains of L. reuteri are able to convert the amino acid L-histidine, a dietary component, to the biogenic amine histamine (Diaz et al., 2016; Greifova et al., 2017). A human commensal bacterium, L. reuteri 6475 was used as the model strain for studying histamine in L. reuteri. J. Versalovic's group reported that L. reuteri 6475-derived histamine suppressed tumor necrosis factor (TNF) production from stimulated human monocytes (Thomas et al., 2012). This suppression was dependent on the activation of histamine H<sup>2</sup> receptor, increased intracellular cAMP and protein kinase A, and the inhibition of MEK/ERK signaling. The production of histamine and subsequent in vitro TNF-suppressing function are regulated by a complete chromosomal histidine decarboxylase (hdc) gene cluster, which contains hdcA, hdcB, and hdcP (Rossi et al., 2011; Thomas et al., 2012). The same group of researchers also found that oral administration of hdc<sup>+</sup> L. reuteri could effectively suppress intestinal inflammation in a trinitrobenzene sulfonic acid (TNBS)-induced mouse colitis model (Gao et al., 2015). Moreover, intraperitoneal injection of L. reuteri 6475 culture supernatant to TNBS-treated mice resulted in similar colitis attenuation. These results strongly indicate the involvement of L. reuteri metabolites, including histamine, in intestinal immunomodulation (Thomas et al., 2016). Further investigations showed that a gene called rsiR was necessary for the expression of hdc gene cluster in L. reuteri 6475 (Hemarajata et al., 2013). Inactivation of rsiR gene led to reduced TNF inhibition in vitro and diminished anti-inflammatory function in vivo. Additionally, both the in vitro TNF suppression and the in vivo anti-colitis effects appear to be regulated by a gene named folC2 (Thomas et al., 2016). Inactivation of folC2 gene resulted in suppression of the hdc gene cluster and diminished histamine production. Notably, histamine production by L. reuteri is highly strain-dependent, and most studies have been focused on strains of human origin (Mishra et al., 2012).

#### Vitamins

There are 13 essential vitamins for humans due to the inability of the human body to synthesize them (Linares et al., 2017). Like many other Lactobacillus spp., several L. reuteri strains are able to produce different types of vitamins, including vitamin B12 (cobalamin) and B9 (folate). As mentioned earlier, B12 is vital in reuterin production because the reduction of glycerol to 3-HPA requires a B12-dependent coenzyme. Up to now, at least 4 L. reuteri strains with various origins have been found to produce B12 (Taranto et al., 2003; Santos et al., 2008b; Sriramulu et al., 2008; Gu et al., 2015). Among these strains, L. reuteri CRL1098 and L. reuteri JCM1112 are the most studied (Morita et al., 2008; Santos et al., 2008a, 2011). In one study, the administration of L. reuteri CRL1098 together with a diet lacking vitamin B12 was shown to ameliorate pathologies in B12-deficient pregnant


female mice and their offspring (Molina et al., 2009). This clearly points to the potential application of L. reuteri in treating B12 deficiency. In addition to B12, folate can also be synthesized by some specific L. reuteri strains, including L. reuteri 6475 and L. reuteri JCM1112 (Santos et al., 2008b; Thomas et al., 2016).

#### Exopolysaccharide (EPS)

The EPS produced by L. reuteri is important for biofilm formation and adherence of L. reuteri to epithelial surfaces (Salas-Jara et al., 2016). In addition, EPS synthesized by L. reuteri is able to inhibit E. coli adhesion to porcine epithelial cells in vitro (Ksonzekova et al., 2016). More importantly, EPS-mediated blocking of adhesion also suppresses gene expression of proinflammatory cytokines that are induced by E. coli infection, including IL-1β and IL-6. Further in vivo experiments in piglets showed similar results in that EPS originated from L. reuteri prevented piglet diarrhea in bacterial infection by reducing the adhesion of E. coli (Chen et al., 2014). In addition, EPS of L. reuteri origin has been reported to suppress the binding of enterotoxigenic E. coli to porcine erythrocytes (Wang et al., 2010). EPS produced by rodent L. reuteri 100-23 was also demonstrated to induce Foxp3<sup>+</sup> regulatory T (Treg) cells in the spleen (Sims et al., 2011). In contrast, an L. reuteri 100-23 strain with the ftf mutation that eliminates EPS production from L. reuteri did not induce splenic Treg cells. This suggests that EPS is required for the L. reuteri-mediated induction of Treg cells and indicates the potential of using wild-type L. reuteri 100-23 to treat Treg deficiency.

#### L. reuteri-Mediated Modulation of Host Microbiota

Emerging evidence suggests that the host microbiota and immune system interact to maintain tissue homeostasis in healthy individuals (Kamada et al., 2013; Bene et al., 2017). Many diseases have been associated with perturbation of the microbiota (Mu et al., 2015), whereas restoration of the microbiota has been demonstrated to prevent or ameliorate several diseases (Scott et al., 2015). L. reuteri is able to influence the diversity, composition and metabolic function of the gut, oral, and vaginal microbiotas. These effects are largely strain-specific (Yang Y. et al., 2015; Garcia Rodenas et al., 2016; Galley et al., 2017; Su et al., 2017).

#### Gut Microbiota

Studies have shown the modulatory effects of L. reuteri on the microbiotas of rodents, piglets, and humans. One study assessed oral administration of a human-origin L. reuteri strain (DSM17938) to scurfy mice, which have gut microbial dysbiosis due to the foxp3 gene mutation. The results indicated that this strain of L. reuteri was able to prolong the lifespan of the mice and reduce multi-organ inflammation while remodeling the gut microbiota (He et al., 2017). Changes of gut microbiota included increases in the phylum Firmicutes and the genera Lactobacillus and Oscilospira. Notably, the disease-ameliorating effect of L. reuteri was attributed to the remodeled gut microbiota, though the community composition was still distinct from wild-type littermates. Further investigation showed that inosine production was enhanced by the gut microbiota upon L. reuteri administration. Through adenosine A2A receptor engagement, inosine can reduce Th1/Th2 cells and their associated cytokines. These results suggested that the L. reuteri – gut microbiota – inosine – adenosine A2A receptor axis serves as a potential therapeutic method for Treg-deficient disorders. Moreover, oral L. reuteri 6475 treatment led to a higher diversity of microbiota in both jejunum and ileum in an ovariectomy-induced bone loss mouse model (Britton et al., 2014). Specifically, there were more abundant Clostridiales but less Bacteriodales. However, whether or not the changed gut microbiota was directly associated with the prevention of bone loss requires further investigation. Furthermore, L. reuteri C10-2-1 has been shown to modulate the diversity of gut microbiota in the ileum of rats (Wang P. et al., 2016).

Compared to vaginally delivered infants, Cesarean (C)-section delivered infants display a higher abundance of Enterobacter but less Bifidobacterium in their gut microbiota (Garcia Rodenas et al., 2016; Nagpal et al., 2016). In one study, treating C-section

babies with L. reuteri DSM 17938 from 2 weeks to 4 months of age modulated the development of gut microbiota toward the community pattern found in vaginally delivered infants (Garcia Rodenas et al., 2016). The gut microbiota structure of vaginally born infants remained unaltered upon L. reuteri supplementation. In another study, treating infants with the same L. reuteri strain resulted in decreased anaerobic Gram-negative and increased Gram-positive bacterial counts in gut microbiota, whereas the abundances of Enterobacteriaceae and Enterococci were largely lowered by L. reuteri treatment (Savino et al., 2015b). The differences in infant age, duration of treatment, route of administration, and dosage may explain the differences in results from the two studies.

For human adults, L. reuteri NCIMB 30242 administered as delayed release capsules for 4 weeks was able to increase the ratio of Firmicutes to Bacteroidetes in healthy individuals (Martoni et al., 2015). This strain of L. reuteri is known to be able to activate bile salt hydrolase and its effect in increasing circulating bile acid has been reported (Jones et al., 2012b). The upregulation of circulating bile acid has been proposed as a reason for the modulated gut microbiota (Jones et al., 2012b). In type 2 diabetes patients, although 3 months of L. reuteri DSM 17938 supplementation did not significantly change the gut microbial structure, the disease outcome of L. reuteri treatment was highly correlated with the baseline gut microbiota structure of individuals (Mobini et al., 2017). Furthermore, the administration of L. reuteri DSM 17938 in cystic fibrosis (CF) patients rescued gut microbiota dysbiosis by reducing Proteobacteria while also enhancing the relative abundance of Firmicutes (del Campo et al., 2014). However, whether or not the modulated gut microbiota contributed to improved GI health in probiotic-treated CF patients needs to be explored further.

L. reuteri influences the gut microbial community in piglets in a strain-specific manner. For instance, oral L. reuteri ZLR003 administration was able to change both the diversity and the composition of the gut microbiota (Zhang et al., 2016). However, treatment with the I5007 strain did not affect colonic microbial structure in piglets (Liu H. et al., 2017). In another study, fodder fermented with L. reuteri changed the abundances of 6 different bacterial taxa, particularly the family Enterobacteriacae, in weanling pigs (Yang Y. et al., 2015). However, the major alterations including increased Mitsuokella and decreased a family under phylum Bacteroidetes could only be seen with L. reuteri TMW1.656 rather than L. reuteri LTH5794. TMW1.656 is a reutericyclin-producing strain while LTH5794 is not, suggesting the possible contribution of reutericyclin in modulating gut microbiota in piglets (Yang Y. et al., 2015).

#### Oral Microbiota

The phyla Firmicutes, Bacteroidetes, Fusobacteria, Proteobacteria, and Actinobacteria are most abundant in the human oral microbiome (Romani Vestman et al., 2015). In a randomized controlled trial, 12 weeks of daily consumption of two L. reuteri strains – DSM 17938 and PTA 5289 led to a shift in oral microbiota composition, though the bacterial species richness was not altered (Romani Vestman et al., 2015). The alterations disappeared 4 weeks after the treatments were terminated, suggesting the fast turnover of the oral microbiome. In another human study, oral L. reuteri treatment reduced the amount of periodontal pathogens in the subgingival microbiota, though no clinical impact was seen (Iniesta et al., 2012).

#### Vaginal Microbiota

Lactobacilli dominate the vaginal bacterial community in healthy women (Macklaim et al., 2015). One study showed that only 14 days of oral L. reuteri RC-14 administration could restore the normal vaginal flora in postmenopausal women (Petricevic et al., 2008). Interestingly, the relative abundance of Lactobacilli is largely decreased in bacterial vaginosis patients (Macklaim et al., 2015). A total of 4 weeks of oral capsule consumption of two Lactobacilli strains including L. reuteri RC-14 increased the relative abundance of Lactobacilli. A similar increase of Lactobacilli was seen when L. reuteri RC-14 was administered vaginally together with a L. rhamnosus strain (Bisanz et al., 2014). However, in pregnant women, 8 weeks of oral L. reuteri RC-14 treatment did not efficiently restore the normal vaginal microbiota (Gille et al., 2016). This suggests that L. reuteri RC-14 may not be able to act alone.

#### Role of L. reuteri in Immunomodulation

Lactobacillus reuteri is able to increase free secretory IgA (sIgA) levels in rats (Wang P. et al., 2016). However, the upregulation of sIgA was eliminated in vitamin A-deficient rats, suggesting that L. reuteri functions in a vitamin A-dependent manner. In pregnant women, the intake of L. reuteri did not alter the levels of total IgA or sIgA in breast milk (Bottcher et al., 2008). When it comes to the effect of L. reuteri in inducing salivary IgA, the results are controversial. Increased salivary IgA levels were reported in humans' chewing gum containing L. reuteri (Ericson et al., 2013). However, other studies showed that L. reuteri did not affect IgA concentration in saliva (Garofoli et al., 2014; Jorgensen et al., 2016; Braathen et al., 2017). The difference in the strains of L. reuteri used in the studies may explain the difference in results. Notably, an important commonality is that salivary L. reuteri-positive individuals have higher salivary IgA levels. Whether L. reuteri affects IgA levels by directly regulating B cells requires further investigations.

Many studies have shown that L. reuteri can induce anti-inflammatory Treg cells, which likely contributes to the beneficial effects of L. reuteri in a wide range of diseased and non-diseased conditions (**Table 2**). The Treg-inducing property of L. reuteri is largely strain-dependent. However, the anti-inflammatory effect of L. reuteri does not always rely on the induction of Treg cells. A good example is L. reuteri-mediated suppression of Th1/Th2 responses in Treg-deficient mice (He et al., 2017). Certain L. reuteri strains are able to reduce the production of many pro-inflammatory cytokines. For example, L. reuteri GMNL-263 can reduce serum MCP-1, TNF, and IL-6 levels in mice fed with high fat diet (Hsieh et al., 2016). Similar effects were observed in mice treated with heat-killed


IBD, inflammatory bowel disease; MLN, mesenteric lymph node.

GMNL-263. However, in some cases, the immunomodulatory effects of L. reuteri appear to rely on its metabolites, as the culture supernatant of L. reuteri BM36301 could reduce TNF production from human myeloid THP-1 cells (Lee et al., 2016). Interestingly, tryptophan catabolites of L. reuteri have been recognized as ligands for aryl hydrocarbon receptor (AhR). Through activating AhR, L. reuteri can promote local IL-22 production from innate lymphoid cells (ILCs) (Zelante et al., 2013). In addition, the derivatives of tryptophan generated by L. reuteri can induce the development of regulatory CD4+CD8αα<sup>+</sup> double-positive intraepithelial lymphocytes in an AhR-dependent manner (Cervantes-Barragan et al., 2017). Considering that AhR is ubiquitously expressed, L. reuteri and its metabolites may be able to influence many more types of immune cells beyond ILCs and T cells (Nguyen et al., 2013).

#### Neuromodulatory Capability of L. reuteri

The intestinal microbiota plays a role in the functions of the enteric nervous system (ENS) (Yoo and Mazmanian, 2017). Subjects with microbiota depletion exhibit an abnormal ENS state (Anitha et al., 2012; Brun et al., 2013, 2015; Yoo and Mazmanian, 2017). Antibiotic treatment reduces the number of neurons in the ENS. This may be related to the decrease in Glial cell line-derived neurotrophic factor (GDNF), which can be restored by TLR2 stimulation (Brun et al., 2013). Moreover, germ-free animals show defective ENS morphology and excitability, which can be reversed by microbiota colonization (McVey Neufeld et al., 2013; Collins et al., 2014). L. reuteri, specifically, can prevent visceral pain response mainly by reducing the enteric nerve activity during the colorectal distension pressure in mice (Kamiya et al., 2006; Ma et al., 2009). Interestingly, live, heat-killed, gamma-irradiated L. reuteri, or even the conditioned media all had a similar effect (Kamiya et al., 2006). L. reuteri can also produce gamma-aminobutyric acid (GABA), the major inhibitory neurotransmitter in the central nervous system (Su et al., 2011; Barrett et al., 2012; Pallin et al., 2016). However, the in vivo bioactivity of the produced GABA has not been addressed (Yoo and Mazmanian, 2017). Furthermore, L. reuteri can increase the excitability and the number of action potentials in rat colonic sensory neurons (Kunze et al., 2009). These distinct effects of L. reuteri may be due to the difference in target neurons (Lai et al., 2017).

#### Role of L. reuteri in Reversing the Leaky Gut

Physical, biochemical, and immunological barriers comprise the gut barrier function, which is required to block the entry of exterior antigens and toxins (Mu et al., 2017a). If any abnormalities occur in the intestinal barrier, the permeability may increase resulting in a leaky gut. Various probiotics are known for their abilities to enhance mucosal barrier function, of which L. reuteri is a well-known example (Mu et al., 2017a). In DSS-induced colitis, L. reuteri administration could reduce bacterial translocation from the GI tract to the mesenteric lymph nodes (MLN) (Dicksved et al., 2012). In addition, treatment of lupus-prone mice with a mixture of Lactobacillus species including L. reuteri led to a higher expression of tight junction (TJ) proteins in intestinal epithelial cells (Mu et al., 2017c). Subsequently, the translocation of pro-inflammatory molecules such as LPS was significantly suppressed, which correlated with attenuated disease. In addition to mouse studies, several strains of L. reuteri have been shown to possess the ability to modulate TJ protein expression and maintain intestinal barrier integrity in pigs (Yang F. et al., 2015; Wang Z. et al., 2016). Moreover, the ability of L. reuteri to decrease intestinal permeability has been seen in humans. In children with atopic dermatitis, where the impairment of intestinal barrier function has been positively correlated with disease pathogenesis (De Benedetto et al., 2011), treatment with L. reuteri DSM 12246 (and L. rhamnosus 19070-2) significantly reduced the frequency of GI symptoms while decreasing the lactulose to mannitol ratio (Rosenfeldt et al., 2004), which reflects the reversal of a leaky gut (Camilleri et al., 2010).

#### L. reuteri ATTENUATE HUMAN DISEASES

A growing body of evidences links microbiota and bacterial translocation with multiple diseases, including several autoimmune disorders (Mu et al., 2015, 2017a). Due to its strong modulatory effects on host microbiota and immune responses with almost no safety concerns, L. reuteri is a good candidate for disease prevention and/or treatment. Indeed, the therapeutic potential of various L. reuteri strains has been studied in diverse diseases and the results are promising in many cases.

TABLE 3 | Effects of L. reuteri on early-life diseases.

fmicb-09-00757 April 17, 2018 Time: 19:31 # 8


(Continued)

TABLE 3 | Continued


FAP, functional abdominal pain; NICU, neonatal intensive care unit; NEC, necrotising enterocolitis.

#### Early-Life Disorders

Taking advantage of the safety and tolerance of L. reuteri in infants and young children, a lot of efforts have been made to test the potential application of L. reuteri against disorders early in life (**Table 3**). In general, the results are promising. L. reuteri has been demonstrated beneficial in the prevention and/or treatment of many conditions including diarrhea, functional abdominal pain, caries, atopic dermatitis, allergy, feeding intolerance, and regurgitation. Infant colic, for example, has been the major therapeutic target of L. reuteri (**Table 3**). Infant colic is characterized by immoderate crying and affects 10–30% infants (Mi et al., 2015). The exact cause and efficient treatment of this condition have remained elusive. The clinical efficacy of L. reuteri DSM 17938 has been demonstrated as most of the clinical trials were successful (**Table 3**). The failure of some studies may be explained by the differences in the dosage of L. reuteri, the infant age when the studies initiated, or the baseline microbiota structure. It is worth mentioning that L. reuteri is naturally contained in human breast milk (Soto et al., 2014), though inconsistencies exist among individuals. The presence of L. reuteri in milk may complicate the results of studies that involved breastfeeding. When given during pregnancy, L. reuteri did not show a significant effect on allergy and eczema in infants after they were born (**Table 3**).

#### Systemic Lupus Erythematosus

The SLE is a multi-system autoimmune disease that involves both genetics and environment as the major disease causative factors (Tsokos, 2011; Edwards et al., 2017). The role of gut microbiota in SLE development was suggested by recent studies, and probiotics have been proposed as potential immunoregulators in SLE (Mu et al., 2015, 2017b; de Oliveira et al., 2017; Edwards et al., 2017; Esmaeili et al., 2017). We reported a significantly decreased level of Lactobacillaceae in lupus-prone MRL/lpr female mice compared to healthy control mice both before and after the disease initiated in MRL/lpr mice (Zhang et al., 2014). Moreover, we found that treatment with retinoic acid improved kidney disease in MRL/lpr mice, and that the improvement of lupus symptoms was associated with restoration of Lactobacilli. This suggests a possible beneficial effect of Lactobacilli in lupus. Therefore, we treated MRL/lpr mice with a mixture of five strains of Lactobacilli to determine their therapeutic benefit. As anticipated, increasing Lactobacilli in the gut improved renal function, reduced serum autoantibodies, and prolonged the survival of MRL/lpr mice (Mu et al., 2017c). Interestingly, L. reuteri and an uncultured Lactobacillus sp. accounted for > 99% of the observed effects. It suggests a central role of L. reuteri in attenuating lupus nephritis. Furthermore, we found that MRL/lpr mice had a "leaky" gut during disease progression, whereas Lactobacillus treatment enhanced the intestinal barrier function in these mice and subsequently decreased metabolic endotoxemia (Mu et al., 2017c). At the same time, the local and systemic pro- and anti-inflammatory network was rebalanced by Lactobacillus treatment. Specifically, IL-10 production was enhanced while the level of IL-6 was decreased systemically. Strikingly, the benefits of Lactobacilli were only observed in females and castrated males but not in intact males. Coincidently, the relative abundance of Lactobacilli in gut microbiota did not decrease as disease progressed in male MRL/lpr mice (Zhang et al., 2014). Consistent with our observations, daily consumption of L. reuteri BM36301 significantly lowered serum TNF level in females but not in males (Lee et al., 2016). The high serum level of testosterone in males may have led to the difference in the response to L. reuteri. Together, these results suggest possible interaction between sex hormones and gut microbiota in autoimmune disease development (Markle et al., 2013; Yurkovetskiy et al., 2013). Further investigation of this link is required. In another lupus mouse model, NZB/W F1, the administration of two L. reuteri strains, together with one L. paracasei strain, was shown to be effective in ameliorating lupus hepatitis (Hsu et al., 2017). Liver abnormalities, manifested as increased liver enzymes, portal inflammation and histopathological changes, have been observed in both lupus mouse models and SLE patients (Hsu et al., 2008; Grover et al., 2014). In this study, the oral L. reuteri treatment largely mitigated hepatic apoptosis and inflammation, suggesting a protective function of L. reuteri against lupusassociated liver disease (Hsu et al., 2017). The protection seems to rely on the capability of L. reuteri to increase antioxidant activity and reduce cytokines associated with more severe lupus,

such as IL-6 and TNF (Tzang et al., 2017). Interestingly, within these two L. reuteri strains, only GMNL-263 can significantly promote the differentiation of Treg cells, again emphasizing the uneven immunoregulatory abilities of different L. reuteri strains.

## Obesity

The correlation between gut microbiota and obesity is well documented (Okeke et al., 2014; Harakeh et al., 2016). The microbiota composition varies between lean and obese individuals, and a surprisingly high level of Lactobacillus spp. has been found in the microbiota of both obese adults and obese children (Armougom et al., 2009; Bervoets et al., 2013). Among different Lactobacillus spp., L. reuteri was specifically described to be associated with obesity (Million et al., 2012, 2013a). The association was further established when vancomycin-resistant L. reuteri in gut microbiota was determined as a body weight gain predictor during vancomycin treatment (Million et al., 2013b). However, in a randomized, double-blind and placebo-controlled clinical trial, the administration of L. reuteri JBD301 for 12 weeks significantly reduced body weight in overweight adults (Chung et al., 2016). Moreover, supplementation of infant formula with L. reuteri did not increase weight gain in infants (Braegger et al., 2011; Koleva et al., 2015). These conflicting results indicate that L. reuteri may influence the development of obesity in a strain-dependent manner. This hypothesis is partially verified in an animal study. In that study, three different strains of L. reuteri were used to test their influence on diet-induced obesity (Fåk and Bäckhed, 2012). It was demonstrated that only L. reuteri PTA 4659 efficiently reduced the body weight of mice fed with high-fat diet (HFD), whereas L. reuteri L6798-treated mice even gained some weight. The changes of adipose and liver weights were consistent with the body weight change.

In animal studies, several strains of L. reuteri have been reported to negatively regulate the development of obesity (Dahiya et al., 2017). In addition to the beneficial effect of L. reuteri JBD301 to human obese patients mentioned earlier, the favorable role of this strain of L. reuteri against weight gain was confirmed in HFD-fed mice (Chung et al., 2016). In HFD-induced obese mouse models, the beneficial role of L. reuteri GMNL-263 was also noted (Hsieh et al., 2016). Treatment with L. reuteri GMNL-263 reduced the body weight as well as the percentages of adipose tissue and liver to body weight. Interestingly, heat-killed GMNL-263 appeared to have a very similar beneficial function (Hsieh et al., 2016; Liao et al., 2016). L. reuteri 6475 has also been shown to be beneficial against obesity in mice (Poutahidis et al., 2013b). The function of L. reuteri 6475 was suggested to be largely dependent on its capability to induce Treg cells without changing the gut microbial ecology. Furthermore, the weight loss properties of some reagents have been attributed to their abilities to increase L. reuteri in mice. Polymannuronic acid, for example, was able to increase the relative abundance of L. reuteri and significantly reduce HFD-induced body weight gain (Liu F. et al., 2017). Whether the increase of L. reuteri is the cause of weight loss requires further investigation.

## Neurodevelopmental Disorder

Exposure to maternal obesity in utero increases the chance of neurodevelopmental disorders, such as autism spectrum disorder, in children (Connolly et al., 2016). In a recent mouse study, maternal HFD (MHFD) was shown to induce social deficits in the offspring (Buffington et al., 2016). The impaired social ability in GF mice was restored by fecal microbiota transplantation from offspring with maternal regular diet (MRD) but not MHFD, suggesting a potential role of microbiota in this process. Further analysis showed that the abundance of L. reuteri was reduced more than ninefold in the gut microbiome of MHFD vs. MRD offspring. The social defects in MHFD offspring were rescued by direct L. reuteri administration, suggesting an effect of L. reuteri in regulating neurodevelopment in MHFD mice. This regulatory function of L. reuteri was attributed to its capability to increase the level of oxytocin (Poutahidis et al., 2013a; Buffington et al., 2016). The results of these studies suggest a potential application of L. reuteri in the treatment of patients who suffer from neurodevelopmental disorders.

#### Stressor Exposure and Enteric Infection

The composition of gut microbiota shift when the host is exposed to stressors (Bailey et al., 2010; Galley et al., 2014). In C57BL/6 males, social stressors led to an altered intestinal microbiota composition, though there was no significant change in community diversity (Galley et al., 2014). Further analysis showed stressor-induced reductions in the families Porphyromonadaceae and Lactobacillaceae, especially in the genus Lactobacillus. Among Lactobacillus spp., L. reuteri was specifically measured and a lower abundance of L. reuteri was evident in stressor-exposed CD-1 mice but not C57BL/6 mice. In fact, the level of L. reuteri in C57BL/6 male mice was below the detection limit with or without stressor exposure (Galley et al., 2014). It is important to note that stressor exposure increased the severity of Citrobacter rodentium-induced inflammation in the gut (Bailey et al., 2010; Mackos et al., 2016). The colonization of C. rodentium was promoted by stressor exposure, which subsequently resulted in more severe colonic pathology and increased production of inflammatory cytokines and chemokines (Mackos et al., 2016). Further studies revealed that stressor-induced C. rodentium colitis was C-C motif chemokine ligand 2 (CCL2)-dependent. Interestingly, administration of L. reuteri ATCC 23272 was able to reverse stressor-induced C. rodentium infection, which also relied on CCL2 (Mackos et al., 2013, 2016). However, L. reuteri was not able to restore the gut microbiome altered by social stressors. This indicates that the beneficial effect of L. reuteri on stressor exposure and subsequent enteric infection is not microbiota-dependent (Galley et al., 2017).

#### CONCLUSION

There has been a decrease in the abundance of L. reuteri in humans in the past few decades likely caused by the modern

lifestyle (Antibiotic use, western diet, improved hygiene). Such decrease coincides with higher incidences of inflammatory diseases over the same period of time. While evidence is lacking to establish the correlation, it may be helpful to increase L. reuteri colonization and/or facilitate its probiotic functions as a new and relatively safe strategy against inflammatory diseases. In addition, through direct regulation or indirect modulation via the host microbiota, L. reuteri plays an impressive role in eliminating infections and attenuating both GI diseases and diseases in remote tissues. The safety and tolerance of L. reuteri has been proven by the numerous clinical studies. There are multiple L. reuteri strains with different host origins, and many of the probiotic functions of L. reuteri are strain-dependent. Therefore,

#### REFERENCES


it may be advantageous to combine different strains of L. reuteri to maximize their beneficial effects.

### AUTHOR CONTRIBUTIONS

QM and VT wrote the first draft of the review. XL edited and finalized the manuscript.

## FUNDING

This work was funded by NIH R15AR067418 and R01AR073240.

lean children: a cross-sectional study. Gut Pathog. 5:10. doi: 10.1186/1757-4749- 5-10


a pediatric outpatient setting. J. Pediatr. 91, 392–396. doi: 10.1016/j.jped.2014. 10.009





infection in preterm infants. Pediatrics 130, e1113–e1120. doi: 10.1542/peds. 2011-3584


effects on intestinal porcine epithelial cells challenged with enterotoxigenic Escherichia coli K88. J. Microbiol. Biotechnol. 26, 1018–1025. doi: 10.4014/jmb. 1510.10089


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Mu, Tavella and Luo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Bifidobacteria and Their Molecular Communication with the Immune System

Lorena Ruiz, Susana Delgado, Patricia Ruas-Madiedo, Borja Sánchez\* and Abelardo Margolles

Dairy Research Institute, Spanish National Research Council (Instituto de Productos Lácteos de Asturias – CSIC), Villaviciosa, Spain

Bifidobacterium represents a genus within the phylum Actinobacteria which is one of the major phyla in the healthy intestinal tract of humans. Bifidobacterium is one of the most abundant genera in adults, but its predominance is even more pronounced in infants, especially during lactation, when they can constitute the majority of the total bacterial population. They are one of the pioneering colonizers of the early gut microbiota, and they are known to play important roles in the metabolism of dietary components, otherwise indigestible in the upper parts of the intestine, and in the maturation of the immune system. Bifidobacteria have been shown to interact with human immune cells and to modulate specific pathways, involving innate and adaptive immune processes. In this mini-review, we provide an overview of the current knowledge on the immunomodulatory properties of bifidobacteria and the mechanisms and molecular players underlying these processes, focusing on the corresponding implications for human health. We deal with in vitro models suitable for studying strainspecific immunomodulatory activities. These include peripheral blood mononuclear cells and T cell-mediated immune responses, both effector and regulatory cell responses, as well as the modulation of the phenotype of dendritic cells, among others. Furthermore, preclinical studies, mainly germ-free, gnotobiotic, and conventional murine models, and human clinical trials, are also discussed. Finally, we highlight evidence supporting the immunomodulatory effects of bifidobacterial molecules (proteins and peptides, exopolysaccharides, metabolites, and DNA), as well as the role of bifidobacterial metabolism in maintaining immune homeostasis through cross-feeding mechanisms.

Keywords: bifidobacteria, Bifidobacterium, microbiota, immunomodulation, T cell response, PRRs, MAMPs

## EARLY COLONIZATION OF BIFIDOBACTERIA AND PROPER IMMUNE DEVELOPMENT

Microbiota establishment in newborns involves the assembly of a novel microbial community, a process that is dependent on several factors, including the mother's physiology (age, metabolic state, lifestyle, or even the potential transfer of microorganisms from mother to child before birth), mode of delivery, genetic background, environmental factors, type of feeding and early antibiotic use, among others (Hill et al., 2017). Similar results were found for preterm neonates, which are less abundantly colonized by bifidobacteria (Arboleya et al., 2012). Infant feeding is also a critical

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Douwe Van Sinderen, University College Cork, Ireland Julio Villena, Centro de Referencia para Lactobacilos, Argentina

> \*Correspondence: Borja Sánchez borja.sanchez@csic.es

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 11 August 2017 Accepted: 15 November 2017 Published: 04 December 2017

#### Citation:

Ruiz L, Delgado S, Ruas-Madiedo P, Sánchez B and Margolles A (2017) Bifidobacteria and Their Molecular Communication with the Immune System. Front. Microbiol. 8:2345. doi: 10.3389/fmicb.2017.02345

factor for bifidobacterial establishment in the gut, and breast-fed infants have been shown to possess higher levels of bifidobacteria than formula-fed infants (Yatsunenko et al., 2012); these high bifidobacterial levels decrease after breast milk cessation (Davis et al., 2016).

Pioneering studies revealed reduced levels of bifidobacteria in the gut microbiota composition of infants at high risk of atopic disease at 3 weeks and 3 months of age, and a higher incidence of atopic disease was found in this group of infants by the age of 1 year (Kalliomaki et al., 2001). Similarly, lower bifidobacterial levels were found in 3-month-old infants who later developed atopy at 2 years of age, or asthma at 4 years of age (Fujimura et al., 2016). All these data point to a critical role for bifidobacteria in the maturation of our immune system from gestation to childhood, suggesting that the low abundance of these early colonizers is associated with a deviated physiological state in infancy. Indeed, current evidence suggests a role of early life bifidobacteria establishment in programming future health. Therefore, it is of great importance to know the specific strains (and species) able to regulate immune responses, either directly or indirectly through the modulation of the gut microbiota, and the underlying mechanisms, in order to design dietary strategies focused on preventing immune-related disorders.

### STRAIN-SPECIFIC IMMUNOMODU-LATORY ACTIVITIES/IN VITRO AND IN VIVO MODELS OF STUDY

#### In Vitro Models

In vitro models have important limitations but they enable the preliminary screening of the effects that bacterial cells or fractions might have on different components of the immune response (Kobayashi et al., 2017). Most in vitro models based on immune cells employ peripheral blood mononuclear cells (PBMCs). In this way, whole cells of B. longum, B. breve, B. bifidum, and B. animalis subsp. lactis strains demonstrated capacity to induce dendritic cell (DC) maturation, and a species/straindependent T cell polarization response (Medina et al., 2007; López et al., 2010; Nicola et al., 2016). These studies revealed that, while B. animalis and many B. longum strains induced the production of the modulatory cytokine IL10 to varying degrees, the greatest strain-dependent differences were displayed in TNFα and INFγ production (**Figure 1**). Stimulation of PBMCs with subcellular fractions of bifidobacteria, including cytoplasmic, surface extracts, and supernatants, has also allowed the identification of molecular determinants of the elicited effects. For instance, a trypsin-labile cytoplasmic fraction of a B. bifidum strain was identified as the effector of CD8<sup>+</sup> T cell activation; and supernatants of B. breve BB99 and B. longum 1941 exerted a regulatory T cell induction (Mouni et al., 2009). PBMC models are thus useful to identify desirable immune profiles in probiotic strain screenings (Liu et al., 2016).

Other in vitro models differentiate DCs, a specialized type of antigen presenting cells, from monocytes. DCs are regarded as the main guardians of the intestinal mucosa and are important in initiating the microbiota–immune system cross-talk. Their pattern recognition receptors (PRRs) interact with specific microbial-associated molecular patterns (MAMPs), which orchestrated molecular cascades that will determine the nature of the immune response (Hoarau et al., 2008; Wittmann et al., 2013). In vitro differentiated DCs allowed the identification of specific domains of a B. bifidum surface protein and the exopolysaccharide (EPS) of B. longum 35624, as the effectors of the immune responses elicited by the strains (Guglielmetti et al., 2014; Schiavi et al., 2016). DC models have also been used to predict the anti-inflammatory potential of bifidobacterial strains/molecules in specific population groups; for instance, bifidobacteria improved antigen uptake and processing by DC from Crohn's disease patients (Strisciuglio et al., 2015).

Other in vitro models using immune cells employ murine splenocytes (Tanabe et al., 2008; Srutkova et al., 2015), macrophage-like cell lines (He et al., 2002; Lee et al., 2012; Mokrozub et al., 2015), or cells isolated from the gut-associated lymphoid tissues (GALT) (Hidalgo-Cantabrana et al., 2014), although they have not been widely used to examine the immunomodulation potential of bifidobacteria and thus their utility to predict immune responses is yet to be confirmed.

The immunomodulation potential of bifidobacteria has also been studied on enterocytes including Caco-2 or HT29 cell models (Bahrami et al., 2011; Chichlowski et al., 2012; Khokhlova et al., 2012; Arboleya et al., 2015; Sánchez et al., 2015; Luongo et al., 2017). Although the immune response of epithelial cells is much more limited than the one exerted by specialized immune cells, enterocytes are more directly exposed to the intestinal milieu and are considered to play a key role in initiating the bifidobacteria–host interactions.

Beyond that, co-culture systems employing both immune and intestinal cells have also been implemented to study microbial–host interactions and promise to overcome some of the limitations of single cell type models (Duell et al., 2011). However, few studies have used them on bifidobacteria (Pozo-Rubio et al., 2011). The application of organized multicellular systems like intestinal organoids for these kinds of studies is envisaged (Noel et al., 2017).

#### In Vivo Models

Germ-free (GF) and conventional in vivo models, including healthy and disease-induced models, have shed light on the immune modulation capability of bifidobacterial strains including live and heat-killed cells (Sugahara et al., 2017). Screening of a large collection of gut symbionts on GF mice identified a B. adolescentis strain which induces a robust Th17 response, albeit not inducing intestinal inflammation (Tan et al., 2016). However, immune responses may vary strongly depending on the health status of the host, as the human sera of Clostridium difficile patients were shown to be more reactive against B. longum extracts than that of healthy individuals (Górska et al., 2016).

In vivo models of intestinal diseases have demonstrated the potential of B. bifidum and B. animalis strains to restore immune markers and intestinal barrier in low chronic inflammation models (Philippe et al., 2011). Similarly, B. longum CECT

7347 attenuated the production of inflammatory cytokines and the CD4<sup>+</sup> T cell-mediated immune response in a gliadininduced enteropathy model (Laparra et al., 2012). Other disease models, described below, have been tested in literature. In foodallergy models, a vesicle-derived protein from B. longum (Kim et al., 2016) and a B. animalis (Ezendam et al., 2008) strain, administered during lactation, exerted immunomodulatory effects. In a gut model, B. longum strain 51-A reduced inflammation (Vieira et al., 2015). Finally, in obesity models, B. pseudocatenulatum restored the lymphocyte–macrophage balance and B. adolescentis IM38 improved high-fat-diet induced colitis inhibiting NF-κB activation (Moya-Pérez et al., 2015; Lim and Kim, 2017). Furthermore, the role of bifidobacteria in responsiveness to immunotherapy has recently been suggested. Accordingly, using tumor models in mice, Bifidobacterium administration was shown to improve tumor-specific immunity and response to therapy through augmented DC function, opening new avenues to exploit the bifidobacterial-immune dialog in the context of this disease (Sivan et al., 2015).

Finally, non-murine in vivo models, like pig models, are very attractive for the study of microbe–host interactions due to the similarities in the gastrointestinal function and development between pigs and humans. In this context, bifidobacterial administration in neonatal piglets has been shown to increase the production of intestinal IL-10 (Herfel et al., 2013), and to improve B and T cell responses following rotavirus vaccination (Vlasova et al., 2013; Kandasamy et al., 2014; Ishizuka et al., 2016). In addition, colonization with a combination of lactobacilli and bifidobacteria in non-vaccinated gnotobiotic piglets reduced the severity of rotavirus infection, while in vaccinated animals enhanced Th1 (Chattha et al., 2013). Thus, in vivo models closer to humans are valuable to study the immunomodulatory potential of certain strains, should be "particularly in the context of pig models in order to study pre-term birth and necrotizing enterocolitis (NEC)" (Oosterloo et al., 2014).

#### Humans

Different immunoreactive proteins from two B. longum strains have been identified in mono-colonized mice, rabbit, and human sera, revealing that the effects are strain and host specific (Górska et al., 2016) and emphasizing the need to further support in vitro immunomodulatory effects in clinical trials. A summary of human studies that focus on the immunomodulatory effects of bifidobacterial consumption in multiple disorders, in some of which gut microbial ecology dysbiosis and altered immune profiles coexist, is presented in **Table 1**.

#### MOLECULAR STRUCTURES DRIVING SPECIFIC IMMUNOMODULATORY EFFECTS

Findings from the last 10 years support the idea that bifidobacteria exert their beneficial effects on host health through the immunomodulatory action of some of their surfaceassociated molecules (Hoarau et al., 2006; Ewaschuk et al., 2008).

#### TABLE 1 | Bifidobacterium role on diseases with an immunological component.


Summary of observational and intervention studies in humans.

This is based on the interaction of a specific bifidobacteria molecule, a MAMP, with a PRR presents on the membrane of epithelial/immune cells, which mostly configures the cellular structure of the intestinal mucosa (Sutterwala and Flavell, 2009). Although mucosa itself is differently organized, depending on the gut section considered, bifidobacteria are thought to exert their immunomodulatory activity mainly in the colon and in the distal part of the ileum, where up to 46% of the Peyer's patches are located (Van Kruiningen et al., 2002). Scientific evidence has shown the presence of immunomodulatory compounds in bifidobacteria spent medium which are released during bacterial growth (**Figure 1**).

#### Proteins and Peptides

Bifidobacterial proteins are one of the targets of human immunoglobulins, notably IgA, which is secreted into the gut lumen in order to control the commensal microbiota populations. Up to six different extracellular proteins from the strains B. longum subsp. longum NCIMB 8809, B. bifidum LMG 11041<sup>T</sup> , and B. animalis subsp. lactis IPLA 4549 were recognized

by pooled sera from healthy individuals, or Inflammatory Bowel Disease (IBD) patients (Hevia et al., 2014). Perhaps the best known example of an immunomodulatory protein is the extracellular serpin secreted by B. longum subsp. longum. Serpin stands for serine protease inhibitor and includes different families that share the ability to bind and irreversibly inactivate proteases. The gene coding for serpin is not widely distributed among the genus Bifidobacterium, being present in up to nine species so far (Turroni et al., 2010a). More precisely, the targets of serpin secreted by B. longum are two important pro-inflammatory proteases: human neutrophil and pancreatic elastases (Ivanov et al., 2006), proteases that have been shown to induce the serpin gene through a two-component regulatory system (Alvarez-Martin et al., 2012). Limiting the local action of these proteases suggests a role of bifidobacteria in the maintenance of gut homeostasis.

Other well-known protein structures with an immunomodulatory action are pili, which self-assemble on the bifidobacteria surface in the form of filaments and have a primary function of adherence to the intestinal surface (Turroni et al., 2014). Lower levels of IL10 and higher levels of TNFα were detected in the murine cecum mucosa as a response to the presence of a Lactococcus lactis strain, genetically modified for producing B. bifidum pili. This response was not observed in the wild-type strain, suggesting a specific interaction of these structures with the gastrointestinal mucosa (Turroni et al., 2013). Another protein with an immunomodulatory effect is the peptidoglycan hydrolase TgaA, a surface-associated protein in B. bifidum, which was shown to induce IL2 production in monocyte-derived dendritic cell (MoDC), the key cytokine in Treg cell expansion (Zelante et al., 2012; Guglielmetti et al., 2014). Finally, our own work has revealed the presence of immunomodulatory peptides encrypted in the sequences of bifidobacteria proteins. In this sense, a peptide contained within the sequence of the protein translocase subunit SecA of B. longum DJ010A triggered a marked Th17 response when incubated with human PBMCs (Hidalgo-Cantabrana et al., 2017b).

#### EPSs

EPSs are carbohydrate polymers that are synthesized and exhibited in the bifidobacterial surface (Hidalgo-Cantabrana et al., 2014). Although the exact molecular mechanisms have not been described so far, EPSs have a great impact on the host immune function (Hidalgo-cantabrana et al., 2012). In a murine model, the EPS-producing strain B. breve UCC2003 was associated with increases in the mucosal levels of the proinflammatory IL12, INFγ, and TNFα which turned out to protect against Citrobacter infection (Fanning et al., 2012). Murine J77A.1 macrophages challenged with the EPS produced by strain B. longum BCRC 14634 increased the production of the antiinflammatory cytokine IL10 when compared to basal conditions, and when challenged with lipopolysaccharide, the presence of the EPS was linked to lower levels of the pro-inflammatory cytokine TNFα (Wu et al., 2010). It is noteworthy that the rhamnose-rich, high-molecular weight EPS isolated from the strain B. animalis subsp. lactis IPLA-R1 was able to increase IL10 production in a PBMC model and to decrease the TNFα production in human colonic biopsies (Hidalgo-Cantabrana et al., 2015). Moreover, the administration of strain IPLA-R1 to Wistar rats was associated with higher serum levels of TGFβ and lower serum levels of the pro-inflammatory interleukin IL6 (Salazar et al., 2014).

EPS produced by specific bifidobacteria strains have been shown as molecules able to prevent exacerbated proinflammatory responses. B. longum subsp. longum 35624 is a strain which has shown clinical efficacy in Irritable Bowel Syndrome, a human condition cursing with chronic mucosal inflammation (Altmann et al., 2016). The antiinflammatory effects elicited by this strain were shown to rely in its surface-associated EPS, which prevented expansion of the pro-inflammatory Th17 response compared to an exopolysaccharide-negative mutant derivative (Schiavi et al., 2016).

Finally, recent data on a mouse model of pathological cell shedding, EPS from B. breve UCC2003 appeared to confer protective effect through MyD88-dependent signaling (Hughes et al., 2017). Diversity of gene clusters responsible for EPS biosynthesis is high among bifidobacterial species/strains (not to mention variations in the level of EPS production) and this diversity may hold tremendous potential for strain-specific immune responses.

### DNA

Bifidobacteria possess genomes with high G+C proportions, and un-methylated CpG motifs derived from them can interact with the TLR 9 present on immune cells. Several publications have reported on the immunomodulatory activity of bifidobacterial DNA. CpG motifs have in one case been linked to a promotion of the Th1 response, dedicated to fight intracellular pathogens such as viruses (Ménard et al., 2010). Another work described an oligodeoxynucleotide derived from the B. longum BB536 strain able to inhibit anti-ovalbumin–IgE titres in a murine model of type-I allergic response after ovalbumin injection (Takahashi et al., 2006).

#### BIFIDOBACTERIAL METABOLISM TRIGGERS CROSS-FEEDING MECHANISMS THAT MAINTAIN IMMUNE HOMEOSTASIS IN THE GUT

Many efforts are currently being pursued to understand the metabolic fluxes within the gut ecosystem among bifidobacteria, other members of the gut microbiota and the human host (Hidalgo-Cantabrana et al., 2017a). A major metabolic contribution elicited by bifidobacteria from their host is represented by the breakdown of non-digestible, diet-derived glycans, and carbohydrates provided by the host known as host-derived glycans [mucins and human milk oligosaccharides (HMOs)] (Milani et al., 2015). Mucin is a host-produced glycan that constitutes one of the main barriers covering the gastrointestinal mucosa (Tailford et al., 2015). Among bifidobacteria, only members of B. bifidum species have been shown to efficiently metabolize mucin

(Ruas-Madiedo et al., 2008; Turroni et al., 2010b; Ruiz et al., 2011). HMOs are present in high concentrations in human colostrum and breast milk. Bifidobacteria, which dominate during early life, are among the best described gut bacteria with the ability to utilize HMOs. Several species possess glycosyl hydrolases that cleave specific linkages within the HMO molecules, the best characterized being those synthesized by B. bifidum (Ruiz et al., 2016). HMOs are preferentially fermented by B. bifidum and B. longum species which, together with B. breve, are the most abundant in breast-fed infant gut microbiota (Sela and Mills, 2010). Thus, the ability of these species to utilize these otherwise indigestible carbohydrates explains their abundance in breast-fed neonates (Zivkovic et al., 2011).

Metabolic cross-feeding mechanisms in the gut are commonly exploited by primary microbial degraders like bifidobacteria which, thanks to partial extracellular hydrolysis of specific complex carbohydrates (e.g., host-produced glycans), provide monosaccharides, oligosaccharides, or metabolites for other microbial gut inhabitants (De Vuyst and Leroy, 2011). As an example, B. bifidum PRL2010 is a strain specialized in the extracellular breakdown of host-glycans and, thus, in the release of simple sugars that can be utilized by other members of the (bifido)bacterial community (Turroni et al., 2016). The subsequent fermentative metabolism of these carbohydrates generates end-metabolites, such as acetate and lactate, which are the main end-products of the bifidobacteria catabolism. Acetate released in the gut by bifidobacteria is used as substrates for other microbial gut fermenters, mainly butyrate and propionate producers (Flint et al., 2015). The production of these two major short-chain fatty acid metabolites have been shown to have anti-inflammatory effects, and promotes and regulates the pool of colonic Treg cells (Arpaia et al., 2013; Smith et al., 2013). By inhibiting histone deacetylase activity in DC and T cells, butyrate acts in the differentiation of Treg cells, increasing the expression of the Treg marker FoxP3 (Furusawa et al., 2013). Signalization has been proposed to be mediated by the butyrate receptors in epithelial and immune cells named FFAR3 (free fatty acid

#### REFERENCES


receptor 3) and GPR109A (Ahmed et al., 2009; Remely et al., 2014).

#### CONCLUDING REMARKS

Bifidobacterial cells, their subcellular fractions, or specific molecules produced by these microorganisms, hold an important potential to trigger immunomodulatory responses involved in the maintenance of our healthy physiological state. However, these responses are poorly understood and need for more research on how this molecular communication between bifidobacteria and host cells is performed. Additionally, the increasing knowledge on the role played by different gut microbiota members, and the understanding of the cross-talk and cross-feeding interaction processes between bifidobacteria, the host, and the surrounding network of intestinal microbes, should facilitate the synergistic use of different intestinal microorganisms to modulate the immunological and inflammatory processes in a microbial dependent way.

#### AUTHOR CONTRIBUTIONS

AM and BS designed the structure of the mini-review. LR, SD, PR-M, BS, and AM wrote the manuscript and drafted the first version of the manuscript. All authors reviewed the final version of the manuscript.

#### ACKNOWLEDGMENTS

LR is a postdoctoral researcher supported by the Juan de la Cierva Postdoctoral Trainee Program of the Spanish Ministry of Economy and Competitiveness (MINECO; IJCI-2015-23196). BS and AM thanks MINECO for the funding of the project AGL2016-78311-R and PR-M for the project AGL2015-64901-R. SD is supported by a research contract associated to the project BIO2014-55019-JIN from the Spanish "Plan Estatal de I+D+i".

by HT-29 intestinal cell-lines in the presence of Bifidobacterium-treated infant microbiota. Benef. Microbes 6, 543–552. doi: 10.3920/BM2014.0111


attenuated human rotavirus vaccine and virulent human rotavirus infection in a neonatal gnotobiotic piglet disease model. J. Immunol. 191, 2446–2456. doi: 10.4049/jimmunol.1300678


in the neonatal piglet. Food Chem. Toxicol. 60, 116–122. doi: 10.1016/j.fct.2013. 07.020


fmicb-08-02345 November 30, 2017 Time: 16:11 # 7

differentially modulate systemic antibody responses to an oral human rotavirus vaccine in a neonatal gnotobiotic pig disease model. Gut Microbes 5, 639–651. doi: 10.4161/19490976.2014.969972


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Ruiz, Delgado, Ruas-Madiedo, Sánchez and Margolles. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fmicb-08-02345 November 30, 2017 Time: 16:11 # 9

# The Role of Lipoproteins in Mycoplasma-Mediated Immunomodulation

Alexei Christodoulides, Neha Gupta, Vahe Yacoubian, Neil Maithel, Jordan Parker and Theodoros Kelesidis\*

David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA, United States

Mycoplasma infections, such as walking pneumonia or pelvic inflammatory diseases, are a major threat to public health. Despite their relatively small physical and genomic size, mycoplasmas are known to elicit strong host immune responses, generally inflammatory, while also being able to evade the immune system. The mycoplasma membrane is composed of approximately two-thirds protein and one-third lipid and contains several lipoproteins that are known to regulate host immune responses. Herein, the immunomodulatory effects of mycoplasma lipoproteins are reviewed. A better understanding of the immunomodulatory effects, both activating and evasive, of Mycoplasma surface lipoproteins will contribute to understanding mechanisms potentially relevant to mycoplasma disease vaccine development and treatment.

#### Edited by:

Amy Rasley, Lawrence Livermore National Laboratory (DOE), United States

#### Reviewed by:

M. Victoria Delpino, National Scientific and Technical Research Council (CONICET), Argentina Hridayesh Prakash, All India Institute of Medical Sciences, India

> \*Correspondence: Theodoros Kelesidis tkelesidis@mednet.ucla.edu

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 22 October 2017 Accepted: 05 July 2018 Published: 31 July 2018

#### Citation:

Christodoulides A, Gupta N, Yacoubian V, Maithel N, Parker J and Kelesidis T (2018) The Role of Lipoproteins in Mycoplasma-Mediated Immunomodulation. Front. Microbiol. 9:1682. doi: 10.3389/fmicb.2018.01682 Keywords: mycoplasma, lipoproteins, immune system, immune modulation, inflammation

#### INTRODUCTION

Mycoplasmas cause a wide variety of human disease, such as asthma, bronchiectasis, and pelvic inflammatory diseases (Metaxas et al., 2015). Several species of mycoplasma such as Mycoplasma pneumoniae in the respiratory tract, M. genitalium in the genitourinary tract cause human disease (Metaxas et al., 2015). Mycoplasma infections can place a big burden on healthcare systems, with more than 100,000 annual hospitalizations arising from M. pneumoniae in the US alone (Atkinson et al., 2008). In addition, mycoplasma infections of the respiratory and urogenital tract can be chronic. A better understanding of the pathogenesis of mycoplasma infections will contribute to vaccine development and treatment.

A key characteristic of the pathogenesis of chronic mycoplasma infections is the cross-talk between mycoplasmas and the host immune system (Atkinson et al., 2008). An inflammation response induced by the host's immunity is one of the main characteristics of M. pneumoniae infection and contributes to clinical presentations. Elucidating the mycoplasma cellular components will help us understand how the relatively small, and biochemically simple, mycoplasmas are capable of eliciting strong and chronic immune responses (Himmelreich et al., 1996). Mycoplasmas can induce proinflammatory response through secreted toxins, surface antigens and other unclear mechanisms (Yeh et al., 2016). Considering that Mycoplasma lack cell walls, cell wall proteins do not interacting with the host's immune system (Pereyre et al., 2016). Analysis of the mycoplasmal cell membrane demonstrated that several lipoprotein surface antigens elicit strong immune responses (Muhlradt et al., 1997) through a distinct pathway than that seen with lipopolysaccharides (Rawadi and Roman-Roman, 1996). Unlike numerous other bacterial species, a large portion of mycoplasmal genome (6.68%) is dedicated to encoding various

lipoproteins (Dandekar et al., 2000; Hallamaa et al., 2006). The functionality of these genes is unknown, but they may drive adhesion and/or the initial stages of mycoplasmal infection since their products are cell surface proteins. Certain mycoplasmal lipoproteins have exonuclease activity that may be involved in the function of the ATP-binding cassette (ABC) transport system to import nucleic acid precursors (Schmidt et al., 2007). Little is known about the transcriptional regulation that occurs within these genes, especially since the mycoplasmal genome lacks many of the transcriptional regulators found in other bacteria (Himmelreich et al., 1996; Dandekar et al., 2000).

The lipoylation mechanism of lipoproteins in mycoplasmas is similar to that of other bacteria (Rottem, 2003). Bacterial lipoproteins play a major role in the cross-talk between bacteria and host immune responses. Bacterial lipoproteins contain a lipoylated amino-terminal cysteinyl residue that is usually N-acylated (Braun and Hantke, 1974). However, an N-acyltransferase gene has not been found in M. pneumoniae or M. genitalium genome (Himmelreich et al., 1996) and many mycoplasmal lipoproteins are not N-acylated [such as lipoproteins from M. gallisepticum (Jan et al., 1996; Muhlradt et al., 1997)]. Triacylation or diacylation of lipoproteins affects their recognition from different Toll like receptors (Jan et al., 1995). Thus, the unique structure of mycoplasmal lipoproteins can determine their immunomodulatory properties. Mycoplasma lipoproteins are all known to induce expression of proinflammatory cytokines including IL-1β, IL-6, and IL-2 (Garcia et al., 1998; Hoek et al., 2005). Several lipoproteins have been identified in mycoplasmas (**Table 1**).

Although Mycoplasma lipoproteins may function as immune-activators within their host, they also regulate mycoplasma colonization and translocation across mucosal membranes and facilitate host immune evasion (Citti et al., 1997; Chambaud et al., 1999). Understanding lipoprotein-induced immunomodulation will aid in developing novel treatments against Mycoplasmas (Martinez de Tejada et al., 2015). Herein, we review the evidence on the effects of mycoplasmal lipoproteins on host immunity (**Figure 1**). Given the complex immunomodulatory effects of mycoplasma lipoproteins we will dissect these effects by specific cell type (epithelial cells, neutrophils, myeloid cells, lymphocytes).

### Effects of Mycoplasmal Lipoproteins on Epithelial Cells

Epithelial cells are the main targets of M. pneumoniae and secrete cytokines in response to infections (Fahey et al., 2005). Mycoplasmas directly induce in vitro production of chemotactic cytokines such as IL-8 by bronchial epithelial cells (Chambaud et al., 1999; Peng et al., 2007; Chen et al., 2016). Mycoplasmal lipoproteins may also have a major role in regulation of the cytokine response in respiratory epithelial cells during M. pneumoniae infection (Eckmann et al., 1995; Yang et al., 2002). Similar to the role of respiratory epithelial cells in eliciting a response during M. pneumoniae infection, epithelial cells lining the vagina and cervix may induce pro-inflammatory


cytokines upon exposure to lipoproteins present in M. genitalium (McGowin et al., 2009). However, the exact mechanisms that contribute to differential effects of mycoplasma lipoproteins on epithelial compared to immune cells remain unclear. Epithelial cells are also able to have an effect on Mycoplasmas. Upon contact with human epithelial cells in-vivo, M. pneumoniae can undergo transcriptional changes in lipoprotein related genes (Hallamaa et al., 2008). Such an alteration in the genetic profile of the Mycoplasma upon coming into contact with epithelial cells represents the two-way dynamic existing between Mycoplasma and the host, where both parties are able to influence each other. Thus, the initial release of proinflammatory cytokines and chemokines by epithelial cells infected by Mycoplasmas can also induce early recruitment of immune cells such as neutrophils.

#### Effects of Mycoplasmal Lipoproteins on Neutrophils

Mycoplasma respiratory infections lead to the recruitment of polymorphonuclear leukocytes and subsequently monocytes/macrophages (M/M) and lymphocytes (Kruger and Baier, 1997). The attraction of leukocytes to the site of infection is controlled by chemokines, which are chemotactic cytokines. The M. pneumoniae/IL-8/neutrophil axis likely plays a vital role in the pathogenesis of MPP and mycoplasmas directly induce in vitro production of chemotactic cytokines such as IL-8 by bronchial epithelial cells (Chambaud et al., 1999; Peng et al., 2007; Chen et al., 2016). An M. fermentans lipoprotein fraction induces secretion of IL-8 from cultured human M/M (Chambaud et al., 1999; Peng et al., 2007; Chen et al., 2016). The expression of such chemokines allows for the recruitment of leukocytes, such as neutrophils, to the site of infection, and it is this influx of leukocytes that comes to partially contribute to the inflammatory response seen upon mycoplasma infection (Wu et al., 2007).

Except for IL-8 and neutrophil chemotaxis, mycoplasma lipoproteins also induce NETosis. Neutrophil extracellular traps (NETs) are a major component of the first line of defense against invading pathogens and they involve a pathogen facilitated cell death by which neutrophils can extrude chromatin structures loaded with anti-microbial molecules to eliminate pathogens (Gomez-Lopez et al., 2017). Experiments using M. agalactiae have shown that mycoplasmas are capable of inducing NETosis upon infection, and mycoplasma lipoproteins are the component primarily responsible for activation of these

pathways in neutrophils (Cacciotto et al., 2016). Furthermore, it has also been shown that lipoproteins induce Toll-like Receptor 2 (TLR2) signaling that plays a pivotal role in neutrophil NETosis (Cacciotto et al., 2016; Xu et al., 2017). Notably, NET formation ability decreases with patient age, and this may contribute to the susceptibility of older patients to mycoplasmal pathogens such as M. pneumoniae (Xu et al., 2017). More research is needed on the ability of lipoproteins to induce NETosis. Thus, mycoplasmal lipoproteins regulate neutrophilic host immune responses, which are the mainstay of pathogenesis of mycoplasmal infections. These early neutrophilic responses can also contribute to inflammatory responses from myeloid cells.

## Effects of Mycoplasmal Lipoproteins on Myeloid Cells

Mycoplasmas induce activation of cytolytic activity of macrophages, and stimulation of cytokines (interleukin [IL]-1, IL-6, tumor necrosis factor-alpha [TNF-α]). Although several mycoplasmal proteins can induce these proinflammatory responses (Chambaud et al., 1999; Peng et al., 2007; Chen et al., 2016), mycoplasmal lipoproteins seem to be a major instigator of production of proinflammatory cytokines by myeloid cells such as M/M and dendritic cells (Cole et al., 2005; He et al., 2009). Non-denaturing detergents enriched with mycoplasmal lipoproteins, have modulatory capacities (Cole et al., 2005; Liu et al., 2012). Synthetic lipopeptides such as the Mycoplasmaderived lipopeptide MALP-2 from M. fermentans (Muhlradt et al., 1997), FSL-1 (also known as Pam2C) from M. salivarium (Shibata et al., 2000), and MPPL-1 from M. pneumoniae (Into et al., 2007) were shown to have immunomodulatory properties. Activation of macrophages by mycoplasmal lipoproteins can occur at pico-molar concentrations of lipoproteins, making them extremely potent activators (Muhlradt et al., 1997).

In vitro and in vivo studies demonstrated that the M. fermentans-derived membrane component MALP-2 (macrophage-activating lipopeptide 2) is the main active component that induces these proinflammatory responses (Kaufmann et al., 1999). MALP-2 directly induced in vitro production of the proinflammatory cytokines TNF-α and IL-6 and the neutrophil-attracting CXC chemokines IL-8 and Growth-regulated Oncogene α (GRO-α) as well as the mononuclear leukocyte-attracting CC chemokines Monocyte chemoattractant protein-1 (MCP-1/CCL2), Macrophage inflammatory protein 1 alpha (MIP-1a/CCL3), and Macrophage inflammatory protein 1 beta (MIP-1b/CCL3) (Garcia et al., 1998; Rawadi et al., 1998). The MALP-2 molecule is a potent activator of human monocytes.

Mycoplasmal lipoproteins in M. hominis such as lipoprotein MHO\_4720 were recently shown to induce production of Interleukin-23 (IL-23), by dendritic cells through activation of inflammasome (Chen et al., 2017; Goret et al., 2017). Production of proinflammatory cytokines such as IL-23 has been shown to induce the expression of chemokines such as IL-17, allowing for effective containment of the infection (Wu et al., 2007). Additional in vivo studies have shown that many of the chemokines induced by mycoplasma infection in humans are conserved across species, such as macrophage inflammatory protein (MIP)-1β (Lam, 2002). Certain mycoplasmal lipoproteins, such as those from M. hyopneumoniae also elicit anti-inflammatory cytokines such as Interleukin-10 (Chambaud et al., 1999; Mollazadeh et al., 2017).

Inducing apoptosis of both monocytes and lymphocytes is a key step in allowing Mycoplasma to prolong their survival within their hosts, and has been documented in numerous Mycoplasma, with M. bovis being one of the most well-studied (Into et al., 2004; Jimbo et al., 2017). Mycoplasma lipoproteins induce apoptosis in both monocyte (ex. HL-60, THP-1) and lymphocyte cell lines (ex. MOLT-4) (Aliprantis et al., 1999; Into et al., 2002a, 2004) and thus directly contribute to immune evasion.

Thus, mycoplasmal lipoproteins regulate myeloid based host immune responses and inflammatory responses, which are key mediators of pathogenesis of mycoplasmal infections. These innate immune responses may also drive responses from adaptive immunity (such as lymphocytic responses).

### Effects of Mycoplasmal Lipoproteins on Lymphocytes

Mycoplasmas induce proliferation of T and B cells (Naot et al., 1979; Cole et al., 1990) through superantigens or unidentified factors. Various mycoplasmas also activate cytotoxic T lymphocytes (CTLs) and natural killer (NK) cells (Wayner and Brooks, 1984) and induce expression of major histocompatibility antigens on B cells (Ross et al., 1992), IL-2 by T cells, interleukin (IL)-1 by macrophages (Muhlradt et al., 1991) and interferons (IFNs) by lymphocytes (Arai et al., 1983). Collectively these mycoplasma-driven immunomodulatory effects contribute to the proliferation and maturation of both T and B lymphocytes as well as the activation of macrophages and NK cells (Arango Duque and Descoteaux, 2014).

During the early phases of infection, mycoplasmas usually induce an inflammatory and a humoral response preferentially directed against their membrane-bound, surface-exposed lipoproteins. Mycoplasmal lipoproteins such as spiralin induce T-cell-independent B-cell blastogenesis and secretion of proinflammatory cytokines (Brenner et al., 1997). Lipoproteins from M. fermentans and M. salivarium induce TLR2- and caspase-mediated apoptosis in lymphocytes (Into et al., 2004). The effects of mycoplasmal lipoproteins on lymphocytes may not be mediated by effects on TLR signaling on antigen presenting cells such as dendritic cells (Sawahata et al., 2011). Thus, mycoplasmal lipoproteins also regulate adaptive immune and function of both B cells and T cells.

## Effects of Mycoplasmal Lipoproteins on Evasion of Host Immunity

Although lipoproteins have a major role as immune system activators, they also facilitate evasion of the host's immune response, leading to chronic mycoplasma infections (Becker and Sander, 2016; Goret et al., 2016). Mycoplasma lipoproteins facilitate immune evasion through several mechanisms such

as creation of the mycoplasmal shield, lipoprotein antigenic variation, and protection from growth inhibiting host antibodies (Citti et al., 1997; Chambaud et al., 1999). Understanding the various modes of immune system evasion promises a better approach to treating mycoplasma infections.

## Lipoproteins as the Basis for Mycoplasmal Shield Against Host Immune Defense Mechanisms

The ability of mycoplasma to encode numerous forms of lipoproteins that can potentially be expressed on their surface allows mycoplasmas to avoid host immune responses (Simmons et al., 2004). Mycoplasmas are able to create a protective layer around themselves that can sterically hinder access of growth-inhibiting host antibodies and even macrophages (Citti et al., 1997; Simmons et al., 2004). In addition to facilitating Mycoplasmal protection, the lipoprotein shield also plays an important role in hemadsorption and mycoplasmal adherence to surfaces (Bolland and Dybvig, 2012; Xiong et al., 2016). Mycoplasmas that are minimally shielded may be highly adherent while maximally shielded mycoplasmas are less adherent (Bolland and Dybvig, 2012).

### Variable Surface Antigen (Vsa) Lipoproteins as Facilitators of Immune Evasion

In addition to the generation of a shield for protecting Mycoplasmas, lipoproteins are also able to protect Mycoplasma by undergoing variation. One major method by which lipoprotein variation in mycoplasmas can contribute to host immune evasion involves altering the length of these variable surface antigens (Vsa) (Xiong et al., 2016). Mycoplasmas use antigenic variation not only to evade immune responses but also to adapt to each host (Bencina et al., 2001). Vsa lipoproteins normally have a 242-amino acid N-terminus with a variable C-terminal domain that can contain up to 60 tandem repeating units which in themselves can range in size from 10 to 19 amino acids (Bencina et al., 2001). Each cell can only transcribe one Vsa gene at a time, with silent Vsa genes missing the sequences necessary to make the conserved N-terminal domain (Shen et al., 2000). To achieve a large variety of Vsa recombination of lipoproteins, mycoplasmas use DNA recombination (Dybvig et al., 1998; Shen et al., 2000). Although a single cell may only transcribe a single Vsa gene, during an infection subpopulations of cells are able to express varying Vsa lipoproteins (Pantoja et al., 2016).

Despite the great importance Vsa size variation plays in facilitating prolonged mycoplasmal survival within hosts, adaptive immunity is still able to decrease mycoplasmal survival, thus necessitating phase variation of the shield as well (Simmons and Dybvig, 2009). Thus, in addition to being able to conduct size variation of surface lipoproteins, mycoplasmas are also able to conduct phase (type) variation of their lipoproteins, hence the name Vsa lipoproteins (Dybvig et al., 1998; Chambaud et al., 1999). Phase variation in vivo is only seen upon the onset of an adaptive immune response, with no alterations occurring in Vsa expression if an organism lacks B and T cells (Gumulak-Smith et al., 2001). It is estimated that phase switching occurs at a frequency of around 10−<sup>3</sup> per CFU per generation (Rosengarten and Wise, 1990). In having the ability to undergo phase variation, mycoplasmas are able to evade host immune responses and indeed lead to chronic infections, similar to how Spirochetes might also use antigenic variation in order to establish a chronic infection (Chambaud et al., 1999; Christodoulides et al., 2017). Thus, Mycoplasmal lipoproteins have pleotropic immunomodulatory effects and their mechanisms of actions need to be elucidated.

## MECHANISMS OF ACTION OF MYCOPLASMAL LIPOPROTEINS

A key family of receptors that play an important role in recognizing, and inducing responses to Mycoplasmal lipoproteins are the Toll-Like Receptors (TLRs) that are key players in inflammatory responses upon infection (Akira et al., 2001; Wang et al., 2017). Two of the most well-studied TLRs are TLR2 and TLR6, both of which are pivotal in recognizing Mycoplasmal Macrophage-Activating Lipopeptide-2 (MALP-2), with the absence of either TLR leading to a lack of recognition and macrophage activation (Takeuchi et al., 2002). Mycoplasmal lipoproteins can also activate myeloid cells through the inflammasome (Chen et al., 2017; Goret et al., 2017). Amongst the Mycoplasma capable of activating inflammasome with their lipoproteins are M. salivarium and M. pneumoniae (Shimizu, 2016; Sugiyama et al., 2016). Through the activation of the inflammasome, the pre-cursors of proinflammatory cytokines, such as IL-1β and IL-18, can be cleaved and activated in a caspase-dependent manner prior to their release (Boyden and Dietrich, 2006).

Non-pathogenic Mycoplasma also express lipoproteins, suggesting that a non-TLR dependent, or inflammasome dependent, mechanism exists by which Mycoplasmal lipoproteins can lead to immune activation (Shimizu, 2016). M. pneumoniae was still able to induce a strong inflammatory response in TLR2 KO mice (Shimizu et al., 2014). Introduction of known autophagy inhibitors to the TLR2 KO mice in turn lead to a significant fall in the inflammatory response, suggesting a role of autophagy in also leading to inflammatory responses (Shimizu et al., 2014).

Another mechanism by which Mycoplasmal lipoproteins are able to lead to immune system activation is through their initial adhesion to epithelia such as the respiratory tract in the case of M. pneumoniae (Jordan et al., 2001). Such cytadherence is mediated through adhesins such as P1(MPN141) as well as other accessory proteins, and treatment of Mycoplasma with proteases or anti-P1 antibodies was seen to lead to a significant decrease in proinflammatory cytokine release (Yang et al., 2002; Hoek et al., 2005). Cytadherence on behalf of M. pneumoniae was seen to generate a proinflammatory response through pathways involving both autophagy/TLR4 as well as inflammasome (Shimizu et al., 2014).

In addition to their roles as inductors of the inflammatory response, TLRs also play an important role in inducing apoptosis in both monocyte cell lines (ex. HL-60, HEK293, THP-1) and lymphocyte cell lines (ex. MOLT-4) upon binding of Mycoplasmal lipoproteins (Aliprantis et al., 1999; Into et al., 2002a; Into et al., 2004). The two main TLRs seen to play a role in this apoptotic pathway are TLR-2 and TLR-6, which through their activation allow induction of NF-kB (Aliprantis et al., 1999; Wu et al., 2008) and the release of Nitric Oxide from macrophages (Muhlradt and Frisch, 1994; Into et al., 2004). Mycoplasmal induced apoptosis seems to be mediated by activation of the p38 MAPK, the apoptosis signal-regulating kinase (ASK1) and the NF-kB pathways (Into and Shibata, 2005). Mycoplasmas such as M. salivarium and M. fermentans have also been seen to induce apoptosis through the more conventional caspase pathways (Into et al., 2002a,b).

On the other hand, certain mycoplasmal lipoproteins, such as those seen in M. fermentans, can inhibit TNF-α mediated apoptosis (Gerlic et al., 2007). Thus, further research is needed to understand the pleotropic effects of mycoplasmal lipoproteins on apoptosis.

A better understanding of mechanisms of immunomodulation mediated by mycoplasmal lipoproteins may lead to therapies to attenuate the detrimental effects of mycoplasma infections.

#### MYCOPLASMAL LIPOPROTEINS AS THERAPEUTIC TARGETS

One of the most difficult parts of managing a Mycoplasmal infection is the inability to use many antibiotics such as beta-lactams (Bebear and Pereyre, 2005). The development of drugs such as tigecycline that aim to reduce production of chemokines and lipoproteins through targeting of bacterial ribosomes is a growing discipline within the field of Mycoplasma research (Saliba et al., 2009; Salvatore et al., 2009). Studies of tigecycline in a murine model have been able to confirm a significant drop of cytokine and chemokine production following infection with M. pneumoniae (Salvatore et al., 2009). Many studies that focus on development of a possible vaccine against Mycoplasma use inactivated Mycoplasma that can then be introduced to patients (Wenzel et al., 1977; David et al., 2010). Atalla et al. (2015) were able to show that inactivation of Mycoplasma using a photoactivatable alkylating agent (INA), with preservation of the surface lipoproteins, might be a possible mechanism by which Mycoplasma can be inactivated for vaccine development (Atalla et al., 2015). Through a better understanding of bacterial lipoproteins and their interaction with the immune system, researchers are now aiming to develop

#### REFERENCES

Akira, S., Takeda, K., and Kaisho, T. (2001). Toll-like receptors: critical proteins linking innate and acquired immunity. Nat. Immunol. 2, 675–680. doi: 10.1038/ 90609

vaccines that rely on the introduction of bacterial lipoproteins into patients to generate antibody responses (Leng et al., 2015). For example, the above has been achieved with the AvgC lipoprotein from M. agalactiae and shows promise for a vaccine against M. agalactiae infection (Santona et al., 2002). Such an approach to vaccine development is also currently being studied using bacterial lipoproteins, such as the OspA from Spirochetes (Izac et al., 2017; Zhao et al., 2017). An innovative approach for treatment of Mycoplasmal infections would involve direct targeting of lipoproteins; however, such an approach remains to be established given the structure of mycoplasmal lipoproteins and their diverse immunomodulatory mechanisms. Importantly, the polypeptide Aspidasept <sup>R</sup> which neutralizes endotoxin in vivo was previously shown to inhibit cytokines induced by the Mycoplasmal lipoproteins (Martinez de Tejada et al., 2015). Thus, a better understanding of Mycoplasmal lipoprotein epitopes, may promote progress in the field of antibodies, polypeptides or vaccines against Mycoplasma.

## CONCLUSION

Further studies are needed to promote understanding of the mechanisms of transcriptional regulation, phase variations and the numerous immunomodulatory effects (both acute and chronic) of Mycoplasmal lipoproteins. Given the decreasing efficacy of antibiotics against mycoplasmas, there is need for novel immunotherapies. Novel strategies such as use of inactivating agents that preserve the mycoplasmal surface lipoproteins, may advance vaccine development (Atalla et al., 2015). Novel approaches such as use of polypeptides that attenuate proinflammatory effects of mycoplasmal lipoproteins need further investigation as novel therapeutic approaches for mycoplasmal infections (Martinez de Tejada et al., 2015). Ultimately, a better understanding of mycoplasmal lipoproteins can set the basis for the development of vaccines or antibodies against mycoplasmal lipoproteins and perhaps shed light on the pathogenesis of other vector based pathogens such as Spirochetes.

#### AUTHOR CONTRIBUTIONS

AC, NG, VY, JP, NM, and TK wrote and edited the manuscript. TK conceptualized the manuscript.

## FUNDING

This work was supported by NIH grants NIH K08AI08272, NIH/NCATS Grant #UL1TR000124.

Aliprantis, A. O., Yang, R. B., Mark, M. R., Suggett, S., Devaux, B., Radolf, J. D., et al. (1999). Cell activation and apoptosis by bacterial lipoproteins through toll-like receptor-2. Science 285, 736–739. doi: 10.1126/science.285.5428.736

Arai, S., Munakata, T., and Kuwano, K. (1983). Mycoplasma interaction with lymphocytes and phagocytes: role of hydrogen

peroxide released from M. pneumoniae. Yale J. Biol. Med. 56, 631–638.


sequence: adding value, function and reading frames. Nucleic Acids Res. 28, 3278–3288. doi: 10.1093/nar/28.17.3278


lipoproteins of Mycoplasma fermentans induces apoptotic cell death after NFkappa B activation. Cell. Microbiol. 6, 187–199. doi: 10.1046/j.1462-5822.2003. 00356.x



NF-kappaB pathway via TLR2, MyD88, and IRAK4. Sci. Rep. 7:4349. doi: 10. 1038/s41598-017-04729-y


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Christodoulides, Gupta, Yacoubian, Maithel, Parker and Kelesidis. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Mucosal-Associated Invariant T Cell Interactions with Commensal and Pathogenic Bacteria: Potential Role in Antimicrobial Immunity in the Child

*Liana Ghazarian1 , Sophie Caillat-Zucman1,2\* and Véronique Houdouin1,3*

*<sup>1</sup> INSERM UMR1149, Centre de Recherche sur l'Inflammation, Université Paris Diderot, Paris, France, 2 Laboratoire d'Immunologie, Hôpital Saint Louis, AP-HP, Paris, France, 3Service des Maladies Digestives et Respiratoires de l'Enfant, Hôpital Robert Debré, AP-HP, Paris, France*

Mucosal-associated invariant T (MAIT) cells are unconventional CD3+CD161high T lymphocytes that recognize vitamin B2 (riboflavin) biosynthesis precursor derivatives presented by the MHC-I related protein, MR1. In humans, their T cell receptor is composed of a Vα7.2-Jα33/20/12 chain, combined with a restricted set of Vβ chains. MAIT cells are very abundant in the liver (up to 40% of resident T cells) and in mucosal tissues, such as the lung and gut. In adult peripheral blood, they represent up to 10% of circulating T cells, whereas they are very few in cord blood. This large number of MAIT cells in the adult likely results from their gradual expansion with age following repeated encounters with riboflavin-producing microbes. Upon recognition of MR1 ligands, MAIT cells have the capacity to rapidly eliminate bacterially infected cells through the production of inflammatory cytokines (IFNγ, TNFα, and IL-17) and cytotoxic effector molecules (perforin and granzyme B). Thus, MAIT cells may play a crucial role in antimicrobial defense, in particular at mucosal sites. In addition, MAIT cells have been implicated in diseases of non-microbial etiology, including autoimmunity and other inflammatory diseases. Although their participation in various clinical settings has received increased attention in adults, data in children are scarce. Due to their innate-like characteristics, MAIT cells might be particularly important to control microbial infections in the young age, when long-term protective adaptive immunity is not fully developed. Herein, we review the data showing how MAIT cells may control microbial infections and how they discriminate pathogens from commensals, with a focus on models relevant for childhood infections.

Keywords: mucosal-associated invariant T cells, invariant T cells, innate immunity, antimicrobial defense, riboflavin

## INTRODUCTION

T lymphocytes are mainly categorized into either conventional CD4 or CD8 T cells, or unconventional invariant T cells. In conventional T cells, combinations of T cell receptor (TCR) α and β chains are unlimited and adapted for optimal T cell responses to numerous types of pathogens, whereas those of unconventional T cells are much more limited and suited for innate-like immunity. Together with this limited diversity, unconventional T cells are restricted by non-classical MHC molecules, while conventional T cells recognize classical MHC/peptide complexes. In humans,

#### *Edited by:*

*Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia*

#### *Reviewed by:*

*Johan K. Sandberg, Karolinska Institute (KI), Sweden Mario M. D'Elios, University of Florence, Italy*

*\*Correspondence: Sophie Caillat-Zucman sophie.caillat-zucman@aphp.fr*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 11 October 2017 Accepted: 05 December 2017 Published: 15 December 2017*

#### *Citation:*

*Ghazarian L, Caillat-Zucman S and Houdouin V (2017) Mucosal-Associated Invariant T Cell Interactions with Commensal and Pathogenic Bacteria: Potential Role in Antimicrobial Immunity in the Child. Front. Immunol. 8:1837. doi: 10.3389/fimmu.2017.01837*

mucosal-associated invariant T (MAIT) cells represent the most abundant semi-invariant αβT cell subset (1–3). MAIT cells are preferentially localized in mucosal tissues and react against a newly identified class of microbial-derived antigen precursors presented by the non-classical MHC-I-related molecule, MR1. Upon microbial infection, MAIT cells rapidly produce cytokines and cytotoxic effectors. MAIT cells are protective in experimental models of infection and are decreased in the blood of patients with bacterial infections. Here, we review the rapidly evolving field of the protective role of MAIT cell in infectious diseases, with a particular emphasis on models that may be of special interest in children.

#### MAIN CHARACTERISTICS OF MAIT CELLS

Mucosal-associated invariant T cells represent an abundant proportion of resident T cells in tissues (20–40% in the liver, 1–8% in the colon lamina propria, and 10–20% in the lung and female genital tract) (4–10). They also represent 1–10% of the entire CD3 T cell pool in human peripheral blood (7, 11). This compares with around 0.1% for invariant Natural Killer T (iNKT) cells, another population of unconventional innate-like T cells. In contrast, MAIT cells are 10-fold less abundant in mice than in humans and iNKT are more numerous. MAIT cells express a semi-invariant TCR made of a canonical TCRα chain (Vα7.2-Jα33/20/12 in humans, Vα19-Jα33 in mice) paired with a limited number of TCRβ chains (12–15). The MAIT TCR recognizes the conserved, monomorphic, MHC class I-related molecule, MR1 (10), which binds riboflavin (vitamin B2) biosynthesis precursor derivatives, such as 5-(2-oxopropylideneamino)-6-d-ribitylaminouracil (5-OP-RU) produced by most, but not all, bacteria and yeasts (16, 17). MAIT cell activation requires key genes encoding enzymes that form an early intermediate (5-A-RU) in bacterial riboflavin synthesis. Although 5-A-RU does not bind MR1 or activate MAIT cells directly, it forms potent MAIT-activating antigens *via* non-enzymatic reactions with distinct host- or bacteria-derived small chemical molecules, such as glyoxal and methylglyoxal, derived from other metabolic pathways (16, 17). This represents a unique mechanism for creating T-cell ligands from disparate metabolite building blocks. A wide range of bacteria and fungi, but not mammalian cells or viruses, are able to synthesize riboflavin and hence provide MR1 ligands (7, 11, 17). Thus, only microbes that possess a riboflavin biosynthetic pathway have a direct, MR1-dependent, MAIT-activating capacity. Certain bacteria, including *Enterococcus faecalis*, *Listeria monocytogenes*, and group A *Streptococcus* do not activate MAIT cells, likely due to the lack of an intact riboflavin biosynthetic pathway in these strains (7). As humans do not synthesize riboflavin, the MR1–MAIT axis accordingly represents a sophisticated discriminatory mechanism for targeting microbial antigens while protecting the host.

The vast majority of human MAIT cells are CD8<sup>+</sup>, although some CD4<sup>+</sup> and double-negative CD4<sup>−</sup>CD8<sup>−</sup> MAIT subsets are also detected (2, 14, 18). In addition, MAIT cells express high levels of the C-type lectin CD161 and IL-18 receptor α (IL-18Rα) (7, 11, 19). Recently, they have become easily identifiable in the peripheral blood by MR1 tetramers loaded with the bacterial ligand 5-OP-RU (available from the NIH tetramer facility) (14). MAIT cells also express the CXCR6 and CCR9 chemokine receptors, which are involved in trafficking to peripheral tissues, especially the intestine and liver (4, 10, 20) but do not express CCR7, involved in migration to lymph nodes. Like iNKT cells, MAIT cells express the master promyelocytic leukemia zinc finger transcription factor (PLZF), suggesting a common thymic differentiation program (3, 21). They also express RORγ, Tbet, Helios, and Eomes (22), consistent with their various effector functions.

Upon TCR-dependent recognition of microbial antigens, MAIT cells display immediate effector responses, by secreting inflammatory cytokines (IFNγ, TNF-α, IL-17, and sometimes IL-22) and mediating perforin-dependent cytotoxicity against bacterially infected cells (7, 11, 20, 23, 24) (**Figure 1**). This strongly supports their involvement in antimicrobial defense. Cytokines produced by MAIT cells may not only act directly on infected target cells, but also promote activation of other immune cells and orchestrate adaptive immunity through dendritic cell (DC) maturation (25, 26). Importantly, human MAIT cells can also be activated *in vitro* in a TCR-MR1 independent fashion in response to cytokines such as IL-12, IL-18, IL-15, and/or interferon α/β (27–29). Consequently, MAIT cells can be activated in various non-bacterial inflammatory conditions in which these cytokines are produced, in particular during acute or chronic viral infections such as dengue, influenza virus, HCV, and HIV (28, 30–34). For the same reasons, MAIT cells may participate in non-infectious pathological conditions, such as autoimmune disorders and cancer [for review, see Ref. (35–37)].

Finally, in addition to microbial products derived from vitamin B2 synthesis, other MR1-binding ligands have been identified, including the non-stimulatory folic acid (vitamin B9) derivative 6-formyl-pterin (6-FP) (17), and various activating and nonactivating drugs and drug-like molecules (38). So far, the clinical relevance of these ligands is yet to be elucidated.

#### MAIT CELL DEVELOPMENT

MAIT cells are selected on MR1-expressing CD4+CD8+ thymocytes (39) and exit the thymus with a naïve phenotype before acquiring memory characteristics and expanding in the periphery (4, 18). As recently demonstrated using MR1 tetramers, the intrathymic development of MAIT cells is divided into three stages defined by expression of CD161 and CD27. Immature stage 1 and stage 2 MAIT cells (CD161− in the human) predominate in thymus but represent minor subsets in periphery, where mature stage 3 MAIT cells (CD161high) are largely predominant. In germ-free mice, immature stage 1 MAIT cells are generated in the thymus but mature MAIT cells are absent from the periphery, which suggests that colonization by the commensal microbiota provides a key maturation signal. Indeed, colonization of the gut with even a single type of bacteria, capable of providing a ligand for MR1, is enough to restore the normal development of both thymic and peripheral MAIT cells (10, 18, 40).

At birth, cord blood MAIT cells are naïve and represent a very small proportion of T cells (less than 0.1% of T cells), while

they are predominant and exhibit mature characteristics in adult peripheral blood (4, 18, 20, 40) (**Figure 2**). This indicates that MAIT cell thymopoiesis is complemented by an important postnatal peripheral expansion. Surprisingly, mature tissue-resident MAIT cells are detected in the intestine, lung, and liver (but not in the spleen and mesenteric lymph nodes) of second trimester human fetuses (41). The nature of MR1 ligands present during fetal life remains elusive, as it is believed that the fetus *in utero* is sterile and that colonization with microorganisms starts only after birth. Nevertheless, even in the absence of live microbes in the placenta, maternal gestational commensals may be a source of diffusible metabolites reaching fetal tissues (42). Moreover, recent studies indicate that microbial colonization already occurs *in utero* (43, 44). Therefore, early interactions with the maternal or fetal microbiome may influence MAIT cell development, as suggested for other immune cell subsets (45).

The mechanisms driving postnatal MAIT cell expansion in the human remain unclear. In mice, the presence of B cells is necessary for peripheral expansion of MAIT cells but not for their thymic selection (18). In patients with common variable immunodeficiency, some of whom have undetectable circulating B cells, MAIT cell frequencies are decreased, but there is no association between MAIT cell and B cell frequencies (46). It is likely that even a small number of B cells in the lamina propria are sufficient for driving peripheral MAIT cell expansion, as shown in mice with a transmembrane immunoglobulin-μ mutation, in which B lymphocytes are absent in the peripheral blood, but some immunoglobulin A-producing B lymphocytes are found in the intestine (10). This may suggest that the initial proliferation of MAIT cells occurs close to the intestine where bacterial-derived MR1 ligands are abundant. MAIT cells do not expand in mice lacking MR1 in the periphery, or in mice colonized with bacteria lacking MR1 ligand (7). Furthermore, variable microbe-mediated expansion of peripheral MAIT cells was demonstrated in different mouse models, in particular a tremendous expansion during *Francisella tularensis* infection (24). Taken together, these observations indicate that peripheral MAIT cell expansion is likely dependent on encounters with microbial-derived ligands, although this remains difficult to demonstrate in humans. Few studies showed that MAIT cell frequencies gradually increase with age in the peripheral blood of healthy children (4, 40). Moreover, MAIT cell numbers exhibit very large interindividual variability (over one log range) in the blood of both children and adults, but the relationship with previous infections has never

been documented. It is tempting to speculate that the peripheral development of MAIT cells follows a two-step program, in which early interactions with the commensal microbiota provide a first maturation signal, followed by variable MAIT cell expansion related to encounters with different microbes. Only a careful longitudinal analysis of MAIT cells levels in children of various ages with documented microbial infection history will confirm this hypothesis.

#### ANTIMICROBIAL FUNCTION OF MAIT CELLS

*In vitro*, MAIT cells are activated in the presence of MR1 expressing cells loaded with bacterial preparations (fixed *Escherichia coli* and *Mycobacterium tuberculosis* lysate) or cells experimentally infected with various strains of bacteria and yeasts (*E. coli*, *Salmonella typhimurium*, *M. tuberculosis*, and *Candida albicans*). Such activated MAIT cells produce inflammatory cytokines and cytolytic molecules (7, 11, 23, 27, 47) and can kill infected epithelial cells (22, 23, 27). MAIT cells are also able to inhibit *Mycobacterium bovis* bacillus Calmette–Guérin (BCG) growth in infected macrophages (12, 48), suggesting that they may control microbial burden *in vivo*.

A better understanding of the role of MAIT cells in the control of microbial infections has been obtained through several mouse models, in particular transgenic mice overexpressing MAIT cells and MR1-deficient mice, compared with wild-type mice. In MAIT transgenic mice, activated MAIT cells accumulate at the site of *E. coli* or *Mycobacterium abscessus* infection and promote bacterial clearance, except if mice are MR1-deficient (7). A low dose of *M. bovis* aerosol results in a much stronger infection in MR1-deficient mice compared with control mice, indicating the important role of MAIT cells in the early control of mycobacterial infection in the lung (48). An increase in bacterial load is also observed in MAIT-deficient mice infected with *Klebsiella pneumonia* (49). MAIT cells accumulate in the lung of mice after intranasal inoculation of *S. typhimurium* (50). The contribution of MAIT cells is best demonstrated in a model of pulmonary infection with *F. tularensis* (24, 51) MAIT cell numbers progressively increase in the lung reaching their peak of expansion in the late phase of bacterial clearance. High MAIT cell numbers persist even after bacterial clearance, suggesting that they participate in the long-term control of infection. Interestingly, in MR1-deficient mice, not only bacterial clearance is delayed, but there is also a delay in the recruitment of conventional CD4 and CD8 T lymphocytes into the lung, indicating that MAIT cells also contribute to the establishment of adaptive immune responses. Indeed, *F. tularensis*-infected macrophages activate MAIT cells which produce GM-CSF, driving the differentiation of inflammatory monocytes into monocyte-derived DCs in the lung (51). These results show that MAIT cells are able to influence early activation and recruitment of T cells through DC maturation.

Altogether, these experimental models indicate that MAIT cells accumulate at the site of bacterial infection and are protective in various experimental infection models. However, studies in mice are not always contributive to understand the role of MAIT cells in humans, because of fundamental differences regarding their frequency and repertoire diversity. Classical mouse laboratory strains have very few, oligoclonal, MAIT cells while transgenic mice which have a high amount of monoclonal MAIT cells (52, 53). In contrast, humans exhibit high numbers of oligoclonal MAIT cells.

To date, no selective MAIT-cell deficiency has been reported in the human. Therefore, the contribution of MAIT cells to antimicrobial defense indirectly relies on correlation studies showing modifications of MAIT cell numbers in infected patients compared with healthy controls. MAIT cell frequencies are decreased in the blood of patients with various bacterial infections, including active tuberculosis (TB) (7, 11, 54–57), *Vibrio cholera* (58), and *Helicobacter pylori* (59). MAIT cell frequencies are decreased in cystic fibrosis patients with lung bacterial infections, in particular with *Pseudomonas aeruginosa*, and notably, frequencies are even lower in patients with higher inflammation and correlate with the severity of the lung disease (60). In critically ill patients with severe non-streptococcal bacterial infections, a prolonged MAIT cell depletion is associated with further development of intensive care unit-acquired infections, suggesting that MAIT cells might be protective in such a clinical setting (61). All these studies were conducted in adult patients.

As indicated earlier, MAIT cell activation can occur during various viral infections. Since viruses are unable to directly stimulate MAIT cells *via* MR1, it is likely that such MR1-independent activation is related to cytokines released from other virusinfected cells (34). In HIV infection, MAIT cells show signs of exhaustion and decline in numbers, both in the peripheral blood and gut mucosa (31, 33, 62, 63). This may leave patients particularly vulnerable to opportunistic infections. Moreover, depletion of gut mucosal MAIT cells may contribute to microbial translocation due to a compromised mucosal barrier. Upon antiretroviral treatment (ART), MAIT cells appear to be restored in the gut mucosa but not in the peripheral blood in adult patients. In HIV children, however, peripheral MAIT cells recover after ART, even more so if the treatment is started at a younger age (64). These observations support the hypothesis that the dynamics of MAIT cell peripheral expansion and tissue distribution may vary throughout life.

The case of *M. tuberculosis* infection is discussed here in more details, because it deserves particular interest for children. Indeed, the risk of rapid progression to active TB is higher in children than in adults, but in the absence of reliable biomarkers it remains very difficult to differentiate children at risk to develop active TB from those who will remain healthy and develop a latent TB infection. MAIT cells are decreased in the peripheral blood of adult patients with active TB compared with patients with latent infection and subjects without a history of *M. tuberculosis* exposure (11, 56). In addition to their low frequency, MAIT cells from patients with active TB exhibit high expression of programmed death-1 (PD-1), suggesting that they have been persistently stimulated *in vivo*, and blockade of the PD-1 pathway improves their IFNγ production in response to stimulation with a BCG vaccine (54). MAIT cells from patients with active TB have also impaired functional capacities in response to *M. tuberculosis* compared with those from patients with latent TB and healthy controls (55). Altogether, these data suggest that the degree of peripheral MAIT cell depletion correlates with disease outcome. However, MAIT cell frequencies show a high variability between individuals (healthy controls as well as infected patients), making it unlikely to use them as a clear-cut biomarker of disease outcome, unless longitudinal studies in large cohorts of patients provide convincing results.

It is usually proposed that the reduced MAIT cell numbers in the peripheral blood of infected patients is a consequence of their recruitment to the infected tissues. However, data on MAIT cell accumulation in the tissues remain controversial, owing to the difficulty to perform longitudinal studies in patients. Thus, MAIT cells are detected in the lungs of patients with active TB (7, 11). At contrast, MAIT cell frequencies are reduced in pleural effusions, but increased in ascitic fluids from patients with tuberculous peritonitis, suggesting that MAIT levels may vary depending on the tissues (54). So far, one cannot exclude that a low frequency of MAIT cells in some individuals may, by itself, favor bacterial colonization and promote disease progression. As recently shown, a polymorphism in the human MR1 gene, associated with MR1 expression, is associated with susceptibility to meningeal TB in Vietnamese adult patients (65). It will be crucial to know if such association is observed in other microbial infections in various populations, to determine if impaired MR1 antigen presentation is involved in susceptibility to infection.

## HOW MAIT CELLS DISTINGUISH PATHOGENS FROM COMMENSALS?

Because MAIT cells are activated in the presence of microbialderived MR1 ligands and are able to kill infected cells, their activation must be tightly controlled to avoid inappropriate responses to commensals. This is particularly crucial in mucosae (gut and lung) where MAIT cells are abundant and in close vicinity to the microbiota. Several lines of evidence indicate that MAIT cells can adapt their proliferative and effector responses depending on the amount, nature, and location of microbial ligands, and on the presence of co-stimulatory signals (66–69). Thus, MR1-mediated presentation of microbial ligands may not be sufficient to optimal MAIT cell responses. MR1 transcripts are detected in multiple tissues, but MR1 expression at the cell surface is very low in the absence of infection. MR1 is retained in the endoplasmic reticulum until ligand binding occurs, at which time it is rapidly transported to the cell surface (70). In antigen-presenting cells, uptake of intact bacteria is required for efficient MR1-mediated MAIT cell activation, while stimulation with soluble ligand is inefficient. In addition, the amount of MR1 at the cell surface is differentially regulated in different cell types. Toll-like receptor (TLR) stimulation may modify MR1 expression on antigen-presenting cells and B cells (71, 72). In mice, the administration of the synthetic MR1 ligand 5-OP-RU alone causes MAIT cell activation but does not result in MAIT proliferation, while the addition of TLR agonists causes high levels of activation and proliferation of the MAIT cell pool (50). Altogether, these data may explain why the mere presence of commensal-derived ligands is not sufficient to induce surface MR1 expression and MAIT cell activation in the gut. It is likely that a high infiltration of pathogen bacteria, due to the disruption of the intestinal barrier, together with strong inflammatory signals, is required. Moreover, compared with peripheral blood, MAIT cells from mucosae have increased expression of certain genes, such as *TNF*, *IL23R*, and *CD40L*, that would allow them to respond quickly to bacterial infiltration (69). Finally, active compounds produced by some commensals may maintain a state of suppression of gut MAIT cells, as suggested by the decreased IFNγ production by MAIT cells in response to *Staphylococcus aureus* stimulation if commensal *Lactobacilli* bacteria are present (73). The production of folic acid by *Lactobacillus plantarum* was involved in the maintenance of regulatory T cells (74). Interestingly, the folic acid derivative, 6-FP, is able to bind MR1 but acts as an antagonist ligand for MAIT cell activation (17). Whether *L. plantarum* has an intact riboflavin biosynthetic pathway able to produce activating MR1 ligands that compete with 6-FP remains an open issue.

#### REFERENCES


The recent success of a synbiotic trial associating *L. plantarum* to fructooligosaccharides to prevent sepsis in rural Indian newborns paves the way for such investigation (75).

### CONCLUSION

Our knowledge of MAIT cells and their role in the control of microbial infections has grown substantially in the recent years. In humans, numerous studies were conducted in adults, but studies regarding MAIT cell function in children are still lacking. Whether blood MAIT cell frequency could be used as biomarker for disease outcome requires further investigation in longitudinal cohorts. A better knowledge of MAIT cell interactions with pathogens and cross talk with other immune cells will also be crucial for the development of new therapeutic or vaccine strategies to prevent the development of infectious diseases.

## AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.

## FUNDING

LG was supported by a grant from Agence Nationale de la Recherche (ANR) (PRTS 14-CE15-0005, acronym NEOMAIT).


a pool of type-17 precommitted CD8+ T cells. *Blood* (2012) 119:422–33. doi:10.1182/blood-2011-05-353789


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Ghazarian, Caillat-Zucman and Houdouin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

*Svetlana F. Khaiboullina 1,2,3†, Silvana Levis 4†, Sergey P. Morzunov <sup>5</sup> , Ekaterina V. Martynova2 , Vladimir A. Anokhin6 , Oleg A. Gusev 2,7,Stephen C. St Jeor <sup>3</sup> , Vincent C. Lombardi 1,2\* and Albert A. Rizvanov <sup>2</sup> \**

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Colleen B. Jonsson, University of Tennessee, Knoxville, USA Jan Clement, University Hospitals Leuven, Belgium Alexei B. Shevelev, Emanuel Institute of Biochemical Physics Russian Academy of Sciences, Russia*

#### *\*Correspondence:*

*Vincent C. Lombardi vclombardi@nvcbr.org; Albert A. Rizvanov albert.rizvanov@kpfu.ru*

*† These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 06 January 2017 Accepted: 27 April 2017 Published: 18 May 2017*

#### *Citation:*

*Khaiboullina SF, Levis S, Morzunov SP, Martynova EV, Anokhin VA, Gusev OA, St Jeor SC, Lombardi VC and Rizvanov AA (2017) Serum Cytokine Profiles Differentiating Hemorrhagic Fever with Renal Syndrome and Hantavirus Pulmonary Syndrome. Front. Immunol. 8:567. doi: 10.3389/fimmu.2017.00567*

*1Nevada Center for Biomedical Research, Reno, NV, USA, 2 Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, 3Department of Microbiology and Immunology, University of Nevada School of Medicine, Reno, NV, USA, 4 Instituto Nacional de Enfermedades Virales Humanas "Dr. Julio I. Maiztegui", Pergamino, Argentina, 5Department of Pathology, University of Nevada School of Medicine, Reno, NV, USA, 6Kazan State Medical University, Kazan, Russia, 7Preventive Medicine and Diagnosis Innovation Program, Center for Life Science Technologies, Yokohama, Japan*

Hantavirus infection is an acute zoonosis that clinically manifests in two primary forms, hemorrhagic fever with renal syndrome (HFRS) and hantavirus pulmonary syndrome (HPS). HFRS is endemic in Europe and Russia, where the mild form of the disease is prevalent in the Tatarstan region. HPS is endemic in Argentina, as well as other countries of North and South American. HFRS and HPS are usually acquired via the upper respiratory tract by inhalation of virus-contaminated aerosol. Although the pathogenesis of HFRS and HPS remains largely unknown, postmortem tissue studies have identified endothelial cells as the primary target of infection. Importantly, cell damage due to virus replication, or subsequent tissue repair, has not been documented. Since no single factor has been identified that explains the complexity of HFRS or HPS pathogenesis, it has been suggested that a cytokine storm may play a crucial role in the manifestation of both diseases. In order to identify potential serological markers that distinguish HFRS and HPS, serum samples collected during early and late phases of the disease were analyzed for 48 analytes using multiplex magnetic bead-based assays. Overall, serum cytokine profiles associated with HPS revealed a more pro-inflammatory milieu as compared to HFRS. Furthermore, HPS was strictly characterized by the upregulation of cytokine levels, in contrast to HFRS where cases were distinguished by a dichotomy in serum cytokine levels. The severe form of hantavirus zoonosis, HPS, was characterized by the upregulation of a higher number of cytokines than HFRS (40 vs 21). In general, our analysis indicates that, although HPS and HFRS share many characteristic features, there are distinct cytokine profiles for these diseases. These profiles suggest a strong activation of an innate immune and inflammatory responses are associated with HPS, relative to HFRS, as well as a robust activation of Th1-type immune responses. Finally, the results of our analysis suggest that serum cytokines profiles of HPS and HFRS cases are consistent with the presence of extracellular matrix degradation, increased mononuclear leukocyte proliferation, and transendothelial migration.

Keywords: hemorrhagic fever with renal syndrome, hantavirus pulmonary syndrome, nephropathia epidemica, cytokine profile, blood serum

## INTRODUCTION

Hantaviruses are negative strand RNA viruses carried by rodents, insectivores, and bats. The epidemiology of hantaviruses reflects the distribution of their primary rodent hosts (1) and can be divided into two groups based on clinical manifestation in human: hantavirus pulmonary syndrome (HPS) and hemorrhagic fever with renal syndrome (HFRS). HPS is endemic in America and diagnosed in many countries including Argentina and USA (2, 3). Several hantaviruses, including Andes and Sin Nombre, were linked to HPS (1). In contrast, HFRS is exclusively diagnosed in Europe and Asia (1, 4, 5). In Europe, a mild form of HFRS, also referred as nephropathia epidemic (NE), is clinically distinguished (6, 7). HFRS/NE is diagnosed in many countries in Europe, including Western part of Russia, where endemically active regions include the republic of Tatartsan (8). Puumala virus (PUUV) is primarily identified as the causative agent of HFRS/ NE in Tatarstan (9).

Hantaviruses cause asymptomatic infection in their rodent hosts, whereas in human, infection with Andes, Sin Nombre, and Puumala can result in an acute and sometimes fatal disease. It is believed that humans acquire infection by inhaling viruscontaminated aerosols (10). As a general rule, humans can only become infected after direct contact with infected rodents or their excreta; however, there has been documented cases of Andes virus being spread from person-to-person (11, 12).

Hantavirus pulmonary syndrome is an acute severe disease characterized by pneumonia, cardiovascular failure, and shock (13). As the respiratory symptoms worsen, the disease progresses into the cardiopulmonary phase characterized by respiratory distress. In a few hours, patients become hypotensive and develop tachycardia, which can lead to cardiovascular shock (13). Additionally, some patients will present with hemorrhagic manifestations (14, 15). The convalescent phase can last for months, especially in patients requiring mechanical ventilation. It has been suggested that the rates of HPS fatalities vary with the hantavirus strain. For example, Figueiredo et al. reported that the hantavirus strain Araraquara is associated with a more severe presentation of HPS as compared to the Juquitiba strain (15).

Clinically, HFRS manifests with bleeding disorders and kidney dysfunction (16–18). Later, hemorrhages appear, which vary from small petechia to severe internal bleeding (16, 17). In some severe cases, disseminated intravascular coagulation syndrome can develop, which is considered as one of the primary causes of death in HFRS (19). Kidney pathology is present in all cases, progressing through several stages of kidney dysfunction including oliguria, polyuria, and convalescent (18, 20).

*In vitro* studies and postmortem observations have shown that hantaviruses are not cytopathic (21–24). Therefore, HPS and HFRS pathogenesis cannot be explained by direct tissue damage due to viral replication. Reactions of the organism to hantavirus infection, particularly the control of inflammatory cytokine expression, have been suggested as a key factor in disease pathogenesis. In accordance, an increased tissue infiltration with mononuclear leukocytes is a hallmark of HPS and HFRS (23–25). It has been suggested that the leukocyte infiltration, found in postmortem HPS and HFRS tissues, may be the result of an increased expression of inflammatory cytokines (24, 26). Additionally, high serum levels of inflammatory cytokines have been described for both HPS and HFRS (27–29). Furthermore, it appears that the severity of the disease associates with proinflammatory cytokine expression in patients with hantavirus infection. For example, Saksida et al. demonstrated that the level of serum pro-inflammatory cytokines was higher in HFRS patients infected with Dobrava virus as compared to PUUV-infected patients (30). Additionally, our previous report points out the potential role of Th1-type immune response in the severity of NE (28). PUUV infection presents with a milder form of the disease and lower mortality rate when compared to Dobrava virus infections. Therefore, it could be suggested that inflammatory cytokine expression may determine the severity of clinical presentations and, potentially impact the mortality rates (6, 31).

In the present study, our analyses revealed that distinct cytokine profiles associated with HPS and HFRS. The HPS profile was suggestive of severe inflammatory responses when compared to that of HFRS. Additionally, our data demonstrate that HPS is characterized by the upregulation of Th1-type immune response early during infection. Pronounced upregulation of cytokines that facilitate innate immune response, especially natural killer (NK) cell function, was also observed in HPS cases relative to HFRS. Furthermore, activation of a mixed population of immune effector cells, including mononuclear and segmented leukocyte, is predicted in HPS cases based on the cytokine profile. Finally, the observed HPS and HFRS serum cytokine profiles are consistent with disease pathology characterized by increased mononuclear leukocyte proliferation, transendothelial migration, and degradation of extracellular matrix.

#### MATERIALS AND METHODS

#### Subjects

One set of samples consisted of sera collected in Argentina during HPS outbreaks in Buenos Aires, Santa Fé, Entre Ríos, and Jujuy provinces. Of the 40 HPS serum samples from 30 total subjects (34 ± 3.5; 26 male, 4 female), 7 individual samples were from fatal cases, 29 were collected from those who survived, and no information was available with respect to the outcome of the other 4. Out of the 29 samples from survivors, serum samples from three cases were collected once at the early stage of the disease, and two sets of serum samples were collected from 13 patients, once during the early stage and once at a late stage of the disease.

Sixty-seven HFRS serum samples (38 ± 4.6; 52 male, 15 female) were collected from patients admitted into Republican Clinical Hospital for Infectious Disease, Republic of Tatarstan. Serum from 33 HFRS cases was collected twice (early and convalescent), while a single serum sample was obtained from one patient. Diagnoses of HFRS and HPS were based on clinical presentation and were serologically confirmed by detection of anti-hantavirus antibodies. In some cases, the diagnosis was confirmed by PCR. Serum samples from 55 healthy individuals (37 ± 5.1; 32 males, 13 females) matched by age, gender, and region were collected and served as controls. Samples collected in the Russian Federation were done so under a protocol approved by the Institutional Review Board of the Kazan Federal University and informed consent was obtained from each respective subject according to the guidelines approved under this protocol (Article 20, Federal Law "Protection of Health Right of Citizens of Russian Federation" N323-FZ, 11.21.2011). Sample collection in Argentina was made under a protocol approved by the Institutional Review Board of the Instituto Nacional de Enfermedades Virales Humanas Argentina.

#### Multiplex Analysis

Serum cytokine levels were analyzed using multiplex magnetic bead-based antibody detection kits following the manufacturer's instructions (Bio-Rad, Hercules, CA, USA). In order to survey 48 individual analytes, the Bio-Plex Pro Human Cytokine 27-plex and 21-plex immunoassay kits were used. For each subject, 50 µl of serum was analyzed on a Luminex 200 analyzer (Luminex, Austin, TX, USA) utilizing MasterPlex CT control software and MasterPlex QT analysis software (MiraBio, San Bruno, CA, USA). The median fluorescence intensities were determined using a minimum of 50 beads per analyte and serum concentrations were calculated using standard curves for each analyte generated with standards provided with the Bio-Plex kit. Each serum sample was analyzed in duplicates.

#### Cytokine Network and Functional Analysis

Ingenuity Pathway Analysis (IPA) Software and DAVID 6.7 Software (32) packages were used independently, to examine cytokine-based enrichment of molecular pathways. The list of genes coding significantly altered cytokines (with threshold of ±2-fold in the average value of the cytokine level, *P* value < 0.05) were used and the pathways showing the enrichments with significance of *P* < 0.05 were identified and analyzed. Both platforms provided corroborative results and the same significantly enriched pathways lists.

#### FANTOM5 Analysis

Raw cytokine levels, measured using the Bio-Plex system, were first normalized against the median level of the 48 cytokines in each sample. Per group median normalization was then carried out for two separate groups, Argentina (Andes) and Tartarstan (HFRS). For each cytokine/group, the level of a given cytokine was normalized to the median level across the group (including both controls and cases). In the Argentina and Tartarstan groups, there were also subgroups corresponding to sera taken upon admission to the hospital and sera taken upon discharge. The median fold changes were compared between cases and controls for each group or subgroup. Expression profiles of the upregulated cytokines were manually examined in the FANTOM5 datasets (PMID: 24670764, PMID: 25678556), to determine what primary cell types they are expressed by and whether they are induced upon pathogenic stimuli.

#### Statistical Analysis

Statistical analysis was performed using Statistica and XLSTAT software (StatSoft, Tulsa, OK, USA, and Addinsoft, New York, NY, USA, respectively). Statistical analysis was performed using Student's *t*-test for comparisons between individual experimental groups. Data are presented as mean < SE. Significance was established at a value of *P* < 0.05.

## RESULTS

### Serum Samples

Hantavirus pulmonary syndrome and NE serum samples were collected at different time points during hospitalization. Due to the nature of the disease, two-time point collection was not always feasible. For example, only a single serum sample was collected from each of the seven fatal HPS cases. The majority of HPS samples (12 samples) from Argentina were collected upon admission to the hospital, while a small number of samples (seven samples) were also obtained before discharge. NE and HPS sera were obtained as clinical surplus material; therefore, no patient identifiers were available to match the first and second samples. Information provided was limited to diagnosis and virus strain. HPS sera from Argentina were mostly Lechiguanas strain, with the exception of three samples confirmed by PCR to be the Orán strain. HFRS samples were collected in the Republic of Tatarstan, an area identified as endemic for Puumala hantavirus (8). Based on these data, we partitioned the serum samples into Andes and PUUV infections.

#### Serum Cytokine Profile HPS Cases from Argentina

Two sets of HPS serum samples were collected, early at the time of hospitalization (7.05 ± 0.98 days) and during the convalescent period (42.2 ± 6.9 days). Serum levels of 41 cytokines were upregulated during the early stage of the disease (**Table 1**). The majority of these cytokines, a total of 35, remained upregulated at the late stage disease. For most of the cytokines, serum concentration had declined by the late stage, while levels of some cytokines remained upregulated (CXCL1, LIF, CXCL9, TNFβ) or was further elevated (CXCL10, CCL2, IL-1α, IL-2RA, IL-3, IL-12p40, CCL27, CCL7, bNGF, SCF, TRAIL). When early- and late-stage serum cytokine levels were compared, the only significant differences detected were in the concentrations of CXCL10 and MCSF. Interestingly, higher CXCL10 and lower MCSF levels were found in the late stage serum as compared to that in early time points of the disease.

Serum samples from seven fatal HPS cases were collected upon admission to the hospital. The average date of serum collection from fatal cases was 6.1 ± 1.18, which is close to those collected in early phase non-fatal cases (7.05 ± 0.98 days). Therefore, early phase of non-fatal cases and fatal cases could be compared (**Table 2**). Serum levels of 30 cytokines were upregulated in fatal HPS cases. Although the majority of these cytokines overlapped with those upregulated in non-fatal cases, there were cytokines uniquely elevated in fatal cases. For example, IL-2, IL-2RA, IL-4, IL-7, IL-17A, CCL4, CCL11, CCL27, bFGF, PDGF-BB, TNFα, and VEGF (**Table 1**) were upregulated in non-fatal cases, while IL-6 was significantly upregulated only in fatal cases when compared to healthy controls (**Table 2**). When cytokines upregulated in fatal and non-fatal cases were compared, significant differences in



*\*P-values for controls vs early HPS cases.*

*\*\*P-values for controls vs late HPS cases.*

*# P-values for late vs early HPS cases.*

*a Values differ significantly only in early serum from non-fatal HPS cases and not in fatal cases.*

serum levels of IL-2RA, IL-18, CXCL1, HGF, MCSF, MIF, CXCL9, and SCF were identified (**Table 2**).

#### HFRS Cases from Russia

Serum cytokines of HFRS cases can be divided into two groups, with 22 cytokines being upregulated and 12 cytokines downregulated (**Table 3**). Sixteen cytokines were significantly upregulated in the early stage of HFRS, while 11 were significantly downregulated as compared to controls. At the late stage of the disease, 18 cytokines were upregulated and 12 downregulated. Those that were downregulated during the convalescent phase differed from those in the early stage. For example, compared to controls, the level of CCL5 changed significantly only during the late stage of the disease, while the serum level of bFGF was significantly downregulated only during the early stage of the disease. When upregulated cytokines were compared, nine cytokines (IL-1β, IL-1RA, IL-2, IL-4, IL-6, IL-7, IL-12p75, IFNγ, and TNFα) differed significantly between the early and late stage of the disease. Interestingly, the average serum concentrations of these cytokines were higher in the late stage as compared to the early stage of the disease.

#### HFRS vs HPS Cytokine Profile

While the serum levels of 12 cytokines were lower in early HFRS relative to the controls, no cytokines were observed to be downregulated in HPS cases (**Tables 1** and **3**, respectively). All 12 cytokines downregulated during the early stage of HFRS were upregulated in HPS cases.



*\*P-values for fatal HPS cases vs controls.*

*\*\*P-values for fatal vs non-fatal HPS cases.*

*a Values differ significantly only in fatal HPS cases and not in early non-fatal cases.*

Twenty-two cytokines were upregulated in the sera of earlystage HFRS cases, while 41 cytokines were upregulated in early HPS cases (**Tables 1** and **3**, respectively). Furthermore, only 14 cytokines were upregulated in both HFRS and HPS cases at the early stage (IL-1RA, IL-2, IL-4, IL-7, IL-12p40, IL-12p75, IL-13, IL-16, CXCL9, CXCL10, CXCL12, MIF, SCGFb, and VEGF). As the disease progressed, 17 cytokines were upregulated in HFRS cases, while a much larger group of cytokines, total of 35, was upregulated in HPS cases (**Tables 1** and **3**, respectively). Among these cytokines, 13 (IL-3, IL-4, IL-7, IL-12p40, IL-13, IL-16, CCL4, CXCL10, CXCL12, GMCSF, IFNγ, MIF, and SCGFb) were upregulated in the late stages of both HFRS and HPS. It should be noted that the profiles of overlapping cytokines differed in the late stages of HFRS and HPS as compared to that during the early phases of the disease. Only seven cytokines (IL-13, CXCL10, IL-12p40, IL-16, MIF, SCGFb, and CXCL12) were upregulated in early and late stages of HFRS and HPS, while levels of IL-12p75 and CXCL9 remained elevated only in late-stage HPS samples.

#### FANTOM5 Analysis

FANTOM5 analysis identified CXCL12 as the most upregulated cytokine in the serum of HPS cases. CXCL12 is induced in response to strong pro-inflammatory stimuli such as TNFα and IL-1 (33). CXCL12 is produced by a plethora of cells including endothelial cells, smooth muscle cells, and macrophages; however, platelets are also a major source of this cytokine in circulation (34–38). It has been shown that CXCL12 increases angiogenesis by binding to heparan sulfate proteoglycans on endothelial cells (33). Therefore, it could be suggested that HPS cases are characterized by increased angiogenesis, possibly due to endothelial damage.

Hemorrhagic fever with renal syndrome cases were characterized by a strong expression of MIF, IL-12p40, IL-3, IL-16, CXCL9, CCL27, CXCL1, and HGF. On the contrary, upregulation of MIF, IL-12p40, IL-3, and IL-16 was modest in HPS, while activation of CXCL12 expression was the characteristic feature of HPS. This observation suggests that virus strain-based reactions to hantavirus infection contributes to the profile. Strong upregulation of IL-12p40, CXCL9, IL-16, and CCL27, mainly produced by activated mononuclear leukocytes including lymphocytes, monocytes, and dendritic cells, are suggestive of a role for mononuclear leukocytes in pathogenesis of HFRS.

#### IPA Analysis of Serum Cytokine Profile in HFRS and HPS

Differentially expressed serum cytokines levels for HFRS and HPS cases, as well as fatal and non-fatal HPS cases, were further investigated by functional pathway analysis using two independent pathway enrichment analysis engines. First, the HFRS and



*\*P-values for controls vs early HFRS cases.*

*\*\*P-values controls vs late HFRS cases.*

*# P-values for early and late HFRS cases.*

HPS cytokine profiles of early samples were compared. The most important canonical pathways altered in the early phase of HFRS include cytokine–cytokine receptor interaction chemokine signaling, MAPK signaling, and neurotropin signaling pathways. For early HPS, a larger number of pathways were changed including cytokine–cytokine receptor interaction, RIG-I-like signaling, chemokine signaling, STAT signaling, cytosolic DNA-sensing, regulation of autophagy, TOLL-like receptor signaling, and NOD-like receptor signaling pathways. Additionally, pathways regulating antigen processing and presentation, as well as NK-cell-meditated cytotoxicity, were identified as significant in early-stage HPS cases.

Serum cytokine profile analysis of samples from late phase of HFRS did not reveal significantly altered pathways as compared to healthy controls. Analysis of serum samples from the late phase of HPS revealed fewer pathways as compared to that during the early phase of the disease and included NOD-like receptor signaling, cytokine–cytokine receptor interaction and RIG-I-like receptor signaling. When fatal and non-fatal HPS cases were compared, the cytokine–cytokine receptor interaction and JAK–STAT signaling pathways had significant differences. The serum cytokine profiles suggest that the JAK–STAT signaling pathway was suppressed in fatal cases as compared to non-fatal HPS cases.

#### DISCUSSION

Hemorrhagic fever with renal syndrome and HPS are acute infectious diseases caused by a distinct group of hantaviruses circulating among small rodents. In Tatarstan, HFRS is associated with infection by Arvicolinae-borne hantaviruses and HPS is a disease caused by Sigmodontinae-associated hantaviruses (39). Although clinically HPS and HFRS may overlap (40–42), pneumonia and cardiovascular insufficiency are central to HPS, while renal involvement and dysregulation in blood coagulation being the main sequelae in HFRS. The mortality rates also differ, with HPS having a higher mortality rate than HFRS, 39.3 vs 0.4, respectively (6, 43). Although there are distinct clinical presentations associated with Sigmadontinae- and Arvicolinaeborn hantaviruses, these viruses are non-cytopathic *in vitro* (21, 22). Furthermore, no virus replication-associated cell death has been reported in tissues collected postmortem from HFRS and HPS cases (23, 24), suggesting that a response to infection plays a major role in the pathogenesis of these diseases. Differences in serum cytokine expression are one of the most consistent findings in HPS and HFRS. Many studies have documented a significant upregulation of pro-inflammatory cytokines, leading to the "cytokine storm" hypothesis to explain the pathogenesis of hantavirus-caused diseases.

Our data provide evidence to support the role of cytokines in HPS and HFRS pathogenesis. Striking differences in cytokine profiles were found between HPS and HFRS samples. HPS, the more severe form of hantavirus zoonosis, is characterized by upregulated serum cytokine levels, with no cytokines being downregulated (**Tables 1** and **3**). On the contrary, HFRS cases were characterized by dichotomy in the serum cytokine profile (**Table 3**), with cytokines being both upregulated and downregulated. Additionally, we observed higher number of cytokines upregulated in HPS as compared to HFRS, 41 vs 22, respectively. These data suggest that the severity of the disease could be associated with a "cytokine storm" as the body's response to infection.

Our data support previous observations that chemokines promoting mononuclear leukocyte migration are upregulated in hantavirus-infected patients (28). In this study, upregulation of CXCL9 and CXCL10 were characteristic for both HPS and HFRS throughout the course of the disease. Here, we extend the knowledge of chemokine expression in HPS and HFRS. For example, increased serum levels of IL-16, a chemoattractant acting on activated T lymphocytes, was found in both HPS and HFRS (44). Additionally, we have shown that CXCL12, a potent chemoattractant for mononuclear leukocytes, is upregulated in patients with hantavirus infection (45, 46). Furthermore, there were chemokines uniquely upregulated in HPS cases including CCL2, CXCL1, and CCL7. These cytokines are known to promote tissue migration of monocytes, memory T cells, and dendritic cells (47, 48).

CXCL1, a neutrophil chemoattractant produced by activated macrophages and epithelial cells (49, 50), was upregulated in HPS cases, while being downregulated in HFRS samples. This suggests that although leukocyte tissue infiltration is characteristic to both HPS and HFRS, there is a mixed phenotype of migrating immune effector cells in HPS cases including neutrophils and mononuclear leukocytes. Since neutrophil migration is often associated with tissue damage and vascular leakage, the role of neutrophil migration in HPS pathogenesis requires further investigation.

Early activation of leukocytes is evident in the serum of HPS cases and differs from HFRS. For example, IL-15 is upregulated only in HPS, suggesting activation and proliferation of T cells and NK cells. IL-15 functions through JAK kinase, STAT3, STAT5, and STAT6, which is supported by IPA analysis. Additionally, activation of GCSF was detected only in HPS cases, at both early and late stages of the disease, suggesting sustained proliferation, differentiation, and survival of monocytes and macrophages in HPS. Elevated serum levels of IL-3, SCGFb, and SCF in HPS indicates increased proliferation of bone marrow progenitors, and upregulation of M-CSF and GM-CSF further suggests that myeloid progenitor proliferation is increased in HPS. Also, it should be noted that levels of IFN-γ remained persistently increased in HPS cases throughout the course of the disease, whereas IFN-γ levels were upregulated during late -phase HFRS, consistent with our previous observations (28). IFN-γ is associated with activation of Th1-type immune responses and is produced by activated T helper cells, cytotoxic T lymphocytes (CTL), and NK cells.

Our findings that IL-18 was upregulated in the serum of HPS cases, while being downregulated in the serum from those with HFRS, further corroborate the activation of Th1. IL-18 is a strong activator of Th1-type immunity, and alone or together with IL-12, upregulates the production of IFNγ by T cells and NK cells (51). These data suggest that early activation of NK cells and CTL is characteristic to HPS, while being less pronounced in HFRS. Excessive activation of NK cells and CTL can cause tissue damage, which has been shown to play a role in the pathogenesis of organ damage (52, 53). Recently, Prescott et al. suggested that an adaptive immune response has no influence on hantavirus replication or disease pathogenesis, based on studies using a hamster model of HPS (54). However, clinical studies indicate the potential role of innate and adaptive immune responses in disease pathogenesis (55–58). Our data support the role of adaptive immune responses in the pathogenesis of hantavirus infection. Also, it could be suggested that early strong activation of Th1-type immune response may be associated with severe clinical presentations.

A strong inflammatory response is evident based on serum cytokine profiles for HFRS and HPS. For instance, upregulation of pro-inflammatory cytokines such as IL-1β, IL-6, TNFα, and MIF were found inHFRS and HPS cases. Although the elevation of IL-1β, IL-6, and TNFα in hantavirus cases is well documented, upregulation of serum MIF has not been described previously for either HPS or HFRS cases. MIF is a multifactorial cytokine with a strong pro-inflammatory activity, suggesting a role for MIF in increasing vascular permeability and tissue migration of immune effector cells. Furthermore, increased serum VEGF levels in HFRS and HPS cases also suggest that this pro-inflammatory activity can be promoted by MIF-induced upregulation of metalloproteases, resulting in degradation of the extracellular matrix (59–61). Additionally, MIF can facilitate leukocyte transendothelial migration by upregulation of the adhesion molecules VCAM-1 and ICAM-1 on endothelial cells and monocytes (62, 63). Although HFRS and HPS are characterized by an increased pro-inflammatory serum profile, as described above, the upregulation of IL-18 was unique to HPS. IL-18 has been classified as a cytokine promoting severe inflammatory reactions (64–66). Therefore, we suggest that the clinical presentation of HPS may be due, in part, to the high serum levels of IL-18. It has been demonstrated that IL-18 activation is regulated through the NOD-like receptor regulated pathway (67, 68). IPA analysis of serum cytokine profiles revealed NOD-like pathway involvement only in HPS cases, further suggesting a role for this pathway in HPS pathogenesis.

Our analysis of serum cytokine profiles in HPS and HFRS cases suggests that although HPS and HFRS share many features, there are distinct cytokine profiles differentiating these diseases. They include (1) a severe inflammatory response in HPS cases, where IL-18 may play a central role, (2) a robust and early activation of Th1-type immune response in HPS cases as compared to HFRS, and (3) a strong activation of an innate immune response, especially NK cells in HPS cases. The cytokine profiles are suggestive of degradation of the extracellular matrix, increased mononuclear leukocyte proliferation, and transendothelial migration in both HPS and HFRS.

#### ETHICS STATEMENT

Samples collected in the Russian Federation were done so under a protocol approved by the Institutional Review Board of the Kazan Federal University and informed consent was obtained from each respective subject according to the guidelines approved under this protocol (Article 20, Federal Law "Protection of Health Right of Citizens of Russian Federation" N323-FZ, 11.21.2011). Sample collection in Argentina was made under a protocol approved by the Institutional Review Board of the Instituto Nacional de Enfermedades Virales Humanas Argentina.

#### REFERENCES


## AUTHOR CONTRIBUTIONS

SK: collecting data, data analysis, and writing manuscript. SL: HPS serum sample preparation, Argentina site. SM: HPS serum sample preparation, USA site. EM: HFRS samples preparation and collecting data, Russia site. VA: clinical data analysis and contribution into the discussion. OG: IPA analysis and contribution into discussion. SJ: manuscript editing and study team coordinator. VL: manuscript editing and contribution into discussion. AR: project management and coordination between Argentina, USA, Russia, and Japan sites and intellectual contribution into discussion.

## FUNDING

This study was supported by Russian Science Foundation grant 15-14-00016. The work is performed according to the Russian Government Program of Competitive Growth of Kazan Federal University and subsidy allocated to Kazan Federal University for the state assignment in the sphere of scientific activities. Some of the experiments were conducted with support of Federal Center of Collective Use and Pharmaceutical Research and Education Center, Kazan (Volga Region) Federal University, Kazan, Russia. We thank Dr. Alistair R. Forrest for help with data analysis.


activation in vivo and by IL-12/IL-18 in vitro. *Int Immunol* (2004) 16(1):1–11. doi:10.1093/intimm/dxh002

68. Zaccone P, Phillips J, Conget I, Cooke A, Nicoletti F. IL-18 binding protein fusion construct delays the development of diabetes in adoptive transfer and cyclophosphamide-induced diabetes in NOD mouse. *Clin Immunol* (2005) 115(1):74–9. doi:10.1016/j.clim.2004.11.007

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The handling editor declared a shared affiliation, though no other collaboration, with one of the authors (VA) and states that the process nevertheless met the standards of a fair and objective review.

*Copyright © 2017 Khaiboullina, Levis, Morzunov, Martynova, Anokhin, Gusev, St Jeor, Lombardi and Rizvanov. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Epstein-Barr Virus DNA Enhances Diptericin Expression and Increases Hemocyte Numbers in Drosophila melanogaster via the Immune Deficiency Pathway

Nour Sherri, Noor Salloum, Carine Mouawad, Nathaline Haidar-Ahmad, Margret Shirinian\* and Elias A. Rahal\*

Department of Experimental Pathology, Microbiology, and Immunology, American University of Beirut, Beirut, Lebanon

#### Edited by:

Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia

#### Reviewed by:

Antonio M. Mendes, Instituto de Medicina Molecular (IMM), Portugal Hovakim Zakaryan, Institute of Molecular Biology (NAS RA), Armenia

#### \*Correspondence:

Margret Shirinian ms241@aub.edu.lb Elias A. Rahal er00@aub.edu.lb

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 15 September 2017 Accepted: 24 May 2018 Published: 11 June 2018

#### Citation:

Sherri N, Salloum N, Mouawad C, Haidar-Ahmad N, Shirinian M and Rahal EA (2018) Epstein-Barr Virus DNA Enhances Diptericin Expression and Increases Hemocyte Numbers in Drosophila melanogaster via the Immune Deficiency Pathway. Front. Microbiol. 9:1268. doi: 10.3389/fmicb.2018.01268 Infection with the Epstein-Barr virus (EBV) is associated with several malignancies and autoimmune diseases in humans. The following EBV infection and establishment of latency, recurrences frequently occur resulting in potential viral DNA shedding, which may then trigger the activation of immune pathways. We have previously demonstrated that levels of the pro-inflammatory cytokine IL-17, which is associated with several autoimmune diseases, are increased in response to EBV DNA injection in mice. Whether other pro-inflammatory pathways are induced in EBV DNA pathobiology remains to be investigated. The complexity of mammalian immune systems presents a challenge to studying differential activities of their intricate immune pathways in response to a particular immune stimulus. In this study, we used Drosophila melanogaster to identify innate humoral and cellular immune pathways that are activated in response to EBV DNA. Injection of wild-type adult flies with EBV DNA induced the immune deficiency (IMD) pathway resulting in enhanced expression of the antimicrobial peptide diptericin. Furthermore, EBV DNA increased the number of hemocytes in flies. Conditional silencing of the IMD pathway decreased diptericin expression in addition to curbing of hemocyte proliferation in response to challenge with EBV DNA. Comparatively, upon injecting mice with EBV DNA, we detected enhanced expression of tumor necrosis factor-α (TNFα); this enhancement is rather comparable to IMD pathway activation in flies. This study hence indicates that D. melanogaster could possibly be utilized to identify immune mediators that may also play a role in the response to EBV DNA in higher systems.

Keywords: Epstein-Barr virus (EBV), human herpesvirus 4, Drosophila melanogaster, diptericin, immune deficiency (IMD) pathway, tumor necrosis factor-α (TNF-α), deoxyribonucleic acid (DNA), proinflammatory

#### INTRODUCTION

The Epstein-Bar virus (EBV) is a human pathogen that belongs to the herpes family of viruses. Like other herpes viruses, EBV establishes latency in the host after primary infection with potential future reactivation of viral replication. In addition to causing infectious mononucleosis, this virus is associated with several types of malignancies such as Burkitt and Hodgkin lymphomas, nasopharyngeal carcinoma, and post-transplant lymphoproliferative

disorders (PTLD). Moreover, this virus is associated with multiple autoimmune diseases including systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), and multiple sclerosis (MS) (Lünemann et al., 2007; Draborg et al., 2013). SLE patients were demonstrated to have high EBV viral loads in their peripheral blood mononuclear cells (PBMCs) (Gross et al., 2005; Yu et al., 2005). Similar observations were seen in RA patients whereby high EBV viral loads were detected in the blood, synovial fluid cells, and synovial membranes (Blaschke et al., 2000). Furthermore, various studies suggest that the risk of developing MS increases in EBV-infected individuals (Ascherio et al., 2001; Delorenze et al., 2006). A number of underlying mechanisms that link EBV to autoimmunity have been proposed; these have ranged from molecular mimicry to sequestered antigen release (reviewed in Ascherio and Munger, 2015). Nevertheless, despite these various proposed mechanisms, evidence that conclusively indicates one to mediate autoimmunity by EBV remains lacking; this likely indicates that multiple mechanisms are involved.

We have previously reported that injecting mice with EBV DNA triggers the expression of IL-17, a proinflammatory cytokine associated with autoimmune processes (Rahal et al., 2015). We also observed enhanced levels of other proinflammatory cytokines such as IL-23 and IFNγ. Others have reported that EBV DNA enhances the production of IL-8 from primary human monocytes and IFN-α from human plasmacytoid dendritic cell (Fiola et al., 2010). Owing to the fact that the latency established by EBV in the host is life-long and may result in frequent reactivations and consequent shedding of viral DNA, the proinflammatory pathways triggered by this DNA may contribute to autoimmune processes resulting in triggering or exacerbation of disease. The complexity of mammalian immune systems presents a challenge to comprehensively identify the various immune pathways triggered by EBV DNA and that may serve as potential therapeutic targets in autoimmune diseases. Hence, we attempted to establish Drosophila melanogaster as a relatively less complex system to detect innate immune pathways that are activated in response to EBV DNA. Furthermore, to the best of our knowledge, this is the first study examining the effect of viral DNA itself, rather than an infection, on the immune system in flies. The fly immune system is largely devoid of adaptable immune components and hence it solely relies on innate immune responses. Innate immunity in flies involves humoral as well as cellular responses. The humoral immune response involves three major pathways: the Toll, the immune deficiency (IMD), and the Janus kinase/signal transducers and activators of transcription (JAK-STAT) pathways. The production of about twenty antimicrobial peptides (AMPs) is under the control of the Toll and IMD pathways (Imler and Bulet, 2005). The expression of most of these AMPs such as attacin and cecropin may be triggered by either pathway; on the other hand, the production of drosomycin is primarily reliant on the Toll pathway while that of drosocin and diptericin is dependent on the IMD pathway. Activation of the JAK-STAT pathway has also been shown to respond to immunostimulatory challenges by triggering the expression of the stress-inducible humoral factor Turandot (TotA) (Agaisse et al., 2003). As for the cellular innate immune arm in flies it is reliant on defensive activities performed by the hemocyte compartment which 95% consists of plasmatocytes (Buchon et al., 2014). We report here that injecting EBV DNA into adult flies results in triggering the IMD pathway and in the concomitant enhancement of plasmatocyte proliferation.

## MATERIALS AND METHODS

#### Flies

Flies were raised and crossed at 25◦C; standard Drosophila husbandry procedures were followed. W1118 wild-type flies (Bloomington Drosophila Stock Center #3605, Bloomington, IN, United States), UAS-STAT92e (Ghiglione et al., 2002), UAS-Relish ((Ayyar et al., 2007), and UAS-Toll10b (Maxton-Kuchenmeister et al., 1999) flies were employed. The IMD-RNAi line (Vienna Stock center #9253) was obtained from the Vienna Drosophila Stock Center (Vienna, Austria). In addition, the hemocyte driver Cg25C-GAL4 (Bloomington stock center #7011) was used.

Overexpression of Relish, Toll10b, and STAT92e was attained using the following crosses, respectively: CG25C-Gal4>UAS-Relish, CG25C-Gal4>UAS-Toll10b, and CG25C-Gal4>UAS-STAT92e. Conditional silencing of IMD was performed using the following cross: CG25C-Gal4>UAS-IMD RNAi.

## Fly Treatments

EBV DNA was obtained from Advanced Biotechnologies (Columbia, MD, United States) while bacterial DNA was prepared from an isolate of Staphylococcus epidermidis by phenol precipitation followed by extraction with ethanol.

To assess the effect of EBV DNA on the expression of drosomycin, diptericin, and TotA, 1-day-old flies were injected with various copy numbers of EBV DNA. As non-viral DNA control, fly groups received bacterial DNA in amounts equivalent to the weight of EBV DNA copy numbers administered. As negative controls, some fly groups received no injections while others received sterile water, the DNA solvent. Groups of 1 day-old flies that overexpressed Toll10b, STAT92e, and Relish were included as positive controls for Toll, JAK-STAT, and IMD pathway activation as well. To examine the effect of EBV DNA on older flies, 1-week-old flies were injected with EBV DNA or with sterile water. All injections were administered into the thorax of CO2-anesthetized flies using a Nano-injector (World Precision Instruments, Sarasota, FL, United States) and glass capillary needles. A total volume of 55.2 nl was injected into each fly.

The copy numbers of EBV DNA injected into flies were extrapolated from our previous studies in mice (Rahal et al., 2015). The following formula was used to calculate Drosophila equivalent copy numbers and was modified from the Food and Drug Administration (FDA) recommended formula for dose conversions (14):

Drosophila equivalent dose = Mouse dose × Mouse Km Drosophila Km

Where

$$\text{Km} = \frac{\text{Weight (kg)}}{\text{Body surface area (m}^2)}$$

The body surface area of D. melanogaster was calculated using Mosteller's formula (Mosteller, 1987).

#### Gene Expression Studies in Flies

Ten flies were collected per group per time point for gene expression studies. RNA extraction was subsequently performed using TRIzol (Sigma-Aldrich, St. Louis, MO, United States) according to the manufacturer's specifications. Then, cDNA synthesis was carried out using the QuantiTect <sup>R</sup> Reverse Transcription Kit (QIAGEN, Hilden, Germany). Real-time reverse transcriptase PCR was then performed to analyze relative gene expression levels. Primers used for this purpose were obtained from Thermo Scientific (Ulm, Germany). Previously published primers (**Table 1**) were used to analyze the expression of the drosomycin (Tanji et al., 2007), diptericin (Avadhanula et al., 2009), and TotA (Kallio et al., 2010) genes. Primers used to assess the expression of the IMD gene were selected using the NCBI primer designing tool. Transcription of the house keeping gene RPL32, used as an internal control, was detected also using previously published primers (Hedges and Johnson, 2008). Real-time PCR reactions each consisted of 10 µl and contained 5 µl of SYBR green, 150 pmoles of the forward primer, 150 pmoles of the reverse primer, and 150 ng of cDNA. Samples were analyzed in triplicates. Real time detection was performed in a BioRad CFX96 Real Time System employing a C1000 Thermal Cycler (Munich, Germany). A PCR initial activation step of 95◦C for 5 min was used followed by 40 cycles of 95◦C for 15 and 30 s at the annealing temperature for each primer as published. Relative gene expression normalized to the water-injected group was calculated using the 11Cq method. Samples were analyzed in triplicates.

#### Hemocyte Count

Flies were examined for hemocyte counts as previously described (Ghosh et al., 2015). For this purpose, three female flies per time point from each assessed group were anesthetized with CO2 and then their wings were excised. The flies were then placed


in phosphate buffered saline (PBS) using 10 µl per fly. A fine incision was used to expose the thorax of each fly and bleeding was allowed into the PBS for 20 s. Subsequently, hemocyte counts were determined by examining 10 µl of the bleed using a hemocytometer under a light microscope and employing a 40× magnification. Analysis was performed in triplicates for each time point.

#### Analysis of TNFα Expression in Mice

Experimental procedures conducted on mice were approved by the Institutional Animal Care and Use Committee (IACUC) at the American University of Beirut (AUB). BALB/c mice, 4-6 weeks of age, were obtained from the Animal Care Facility at AUB. Mice were intraperitoneally injected with 144 × 10<sup>3</sup> copies of EBV DNA, 22.3 pg of S. epidermidis DNA (equivalent to the weight of 144 × 10<sup>3</sup> EBV DNA copies) or sterile water. Each injection consisted of 100 µl. Three mice per treatment were sacrificed on day 6 post-injection, their spleens were collected, pooled and then subjected to RNA extraction using TRIzol. The RNA was used for cDNA synthesis using the QuantiTect <sup>R</sup> Reverse Transcription Kit. Real-time reverse transcriptase PCR was then performed to detect relative gene expression levels of TNFα using previously published primers (Peinnequin et al., 2004). Previously published primers (Pindado et al., 2007) were also employed to detect the expression of β-actin, used as an internal control. Real-time PCR reactions were performed and relative gene expression normalized to the water-injected mice was calculated as described for fly gene expression studies described above. Analysis was conducted in triplicates.

#### Statistical Analysis

To analyze statistical significance unpaired t-tests were used to compare means and were performed using Graphpad; p-values less than 0.05 were considered statistically significant.

## RESULTS

#### EBV DNA Enhances the Expression of Diptericin in Flies

To assess the effect of EBV DNA on the humoral arm of innate immunity in D. melanogaster, flies were intrathoracically injected with 70, 140, or 280 copies of EBV DNA. Flies were then analyzed for the expression of drosomycin, diptericin, and TotA as molecular indicators of activation of the Toll, IMD, and JAK-STAT pathways, respectively. Whereas we did not observe an increase in the expression of drosomycin (**Figure 1**) or TotA (**Figure 2**) upon EBV DNA treatment, we detected a 115-fold increase (p = 0.0002) in the transcriptional levels of diptericin (**Figure 3**) in the group injected with 70 copies of EBV DNA on day 1 post-injection. This level then decreased to 5.62 on day 3. No notable changes were observed upon injecting flies with 140 or 280 copies of EBV DNA.

To examine whether the observed effects are specific to EBV DNA itself, flies were injected with an amount of S. epidermidis

DNA equivalent to the weight of the EBV DNA copy numbers employed. Treatment with this non-viral DNA control did not notably alter the expression of the humoral immunity markers assessed.

We then examined the expression of diptericin at 6, 12, 24, 48, 36, and 72 h after injection (**Figure 4A**) and included lower copy numbers of EBV DNA into the analysis. Upon injecting flies with 10 copies of EBV DNA, diptericin levels began to significantly increase after 24 h of injection and continued to elevate reaching a highest detected increase of 16 folds (p = 0.001) 72 h after injection. Flies injected with 35 copies

Indicates p < 0.05.

of EBV DNA also demonstrated significantly increasing levels of diptericin expression after 24 h of injection with levels peaking at 48 h after injection (23-folds, p = 0.0001) then decreasing afterwards. Injection of 70 copies of EBV DNA resulted in a significant increase in the relative gene expression starting after 12 h of injection, peaking at 24 h and decreasing from then onwards. Hence, a time-dependent dose response is seen in the enhancement of diptericin expression by EBV DNA treatment. Injection with 140 or 280 copies did not result in any increase in the levels of diptericin expression indicating that EBV DNA may have a biphasic hormetic effect on diptericin expression in flies whereby low levels are stimulatory but high levels are inhibitory. To examine the effect of EBV DNA in older flies, 1-week-old wild-type flies were injected with 70 copies of EBV DNA; diptericin expression levels were then determined in these flies after 24 h of receiving the injection. Diptericin expression levels were reduced by about 140-folds in the 1-week-old flies upon injection with EBV DNA compared to similarly treated 1-day old flies (**Figure 4B**).

#### Silencing IMD Abrogates EBV DNA-Triggered Expression of Diptericin

To establish that enhanced expression of diptericin in response to EBV DNA is the result of IMD pathway activation rather than an alternate pathway, we knocked down the expression of the IMD mediator, a key component of this pathway. Flies were then injected with 70 copies of EBV DNA and analyzed for expression of diptericin after 24 h of injection (**Figure 5A**). In contrast to wild-type flies, knocking down IMD resulted in inhibition of diptericin expression in response to EBV DNA (p = 0.009). Assessing the efficiency of silencing IMD demonstrated a substantial decrease in the mRNA levels (**Figure 5B**).

water-injected flies on day 1 post-injection. <sup>∗</sup>

## EBV DNA Elevates the Number of Hemocytes in Fly Bleed

We examined the effect of EBV DNA on the cellular arm of fly immunity by injecting adult female flies with 70, 140, or 280 copies of EBV DNA and then assessing the number of hemocytes in the fly bleed on days 1 and 3 post-injection. Treatments with S. epidermidis DNA were included as a non-viral DNA control. We observed a sevenfold increase (p = 0.0009) in the number of circulating hemocytes in flies administered 70 copies of EBV DNA after 1 day of injection. This number had notably dwindled back to normal levels on day 3 after injection in this group. As for flies injected with 140 or 280 copies of EBV DNA or with S. epidermidis DNA no notable elevations in circulating hemocyte numbers were detected (**Figure 6A**). To examine the effect of EBV DNA in older flies, 1-week-old wild-type flies were injected with 70 copies of EBV DNA; hemocytes were then enumerated after 24 h of receiving the injection. The number of hemocytes in the bleed of 1-week-old flies upon injection with EBV DNA was about 30% lower than that in 1-day-old flies that were similarly treated (**Figure 6B**).

## The IMD Pathway Plays a Role in the EBV DNA-Triggered Elevation of Hemocyte Levels

Both circulating hemocyte numbers and diptericin levels were elevated in response to 70 copies of EBV DNA after 24 h of injection but not by 140 or 280 copies of DNA or at later time points. Thus, circulating hemocyte numbers were affected by EBV

DNA in a manner that highly mirrored the observed responses in diptericin expression within EBV DNA-treated fly groups. Consequently, we analyzed whether the IMD pathway underlies this elevation in hemocytes by knocking down the IMD mediator and examining the number of hemocytes in the bleed after 24 h of injecting adult female flies with 70 copies of EBV DNA (**Figure 7**). Knocking down IMD decreased the number of hemocytes 2.5 folds (p = 0.0035) compared to wild-type flies upon injection with 70 copies of EBV DNA. This indicates that this pathway is a key player in the cellular responses seen upon treating flies with EBV DNA.

## EBV DNA Enhances TNFα Expression in Mice

The IMD pathway is often compared to tumor necrosis Factor-α Receptor signaling (TNFR) in mammals (71). Hence, we examined whether EBV DNA triggers TNFα in a mammalian system by injecting mice with 144 × 10<sup>3</sup> copies of EBV DNA and assessing TNFα expression in mouse spleens on day 6 postinjection (**Figure 8**). We had previously detected a peak in mouse IL-17 levels on that day after injection with 144 × 10<sup>3</sup> copies of EBV DNA. Mice injected with EBV DNA showed a 3.5-fold increase in the transcription of TNFα. On the other hand, no significant changes were detected in mice injected with bacterial DNA.

## DISCUSSION

Enhanced production of proinflammatory cytokines triggered by EBV DNA in mammalian systems has been reported by our

FIGURE 8 | Relative gene expression of TNFα in mice injected with EBV DNA. Expression was assessed in splenic tissue from BALB/c mice that were intraperitoneally injected with water, 144 × 10<sup>3</sup> copies of EBV DNA or 22.3 pg of S. epidermidis DNA on day 6 post-injection. <sup>∗</sup> Indicates p < 0.05.

group among others (Fiola et al., 2010; Rahal et al., 2015). This raises the possibility of involvement of other immune modulators in response to EBV DNA. We thus aimed to implement a system that permits relatively simple screening of humoral and cellular responses to viral DNA. Therefore, we used the fruit fly, D. melanogaster, an organism with a versatile library of genetic tools to study such responses. The fruit fly has well-conserved cellular and humoral innate immune pathways in addition to the absence of adaptive immunity; this facilitates the identification of innate pathways, the basis of proinflammatory responses that may be involved in EBV DNA-triggered immune activation in higher mammalian systems.

Upon injecting 70 copies of EBV DNA into flies, the transcriptional level of diptericin, the hallmark of IMD pathway activation, was significantly increased by day 1 post-injection. In a similar manner, hemocyte numbers were significantly elevated in flies upon treatment with EBV DNA. Both diptericin expression and hemocyte proliferation effects had dwindled by day 3 post-injection. On the other hand, injections of 140 or 280 copies of EBV DNA did not result in similar increases in neither diptericin expression levels nor in hemocyte proliferation. This may indicate that EBV DNA results in a biphasic hormetic response in flies whereby lower doses induce a possibly beneficial immunostimulatory response whereas higher doses overwhelm the system and are rather inhibitory. Various agents have been previously described to have such an effect in biological systems (Calabrese and Baldwin, 1999, 2001). The particular underlying molecular etiology of such a response remains to be elucidated. Worth noting is that an injection of 144 × 10<sup>3</sup> copies of EBV DNA in mice, which we had previously reported to induce a prominent elevation of IL-17A levels and demonstrated herein to induce TNFα expression in mice, is equivalent to administration of 140 copies of EBV in flies; whereas this dose appears to be immunostimulatory in mice, a similar observation is not seen in flies. Therefore, whether higher doses of EBV DNA also have a biphasic hormetic immunomodulatory response in mice remains to be investigated. Knocking down IMD abrogated both of the humoral and cellular effects. These results suggest that cross-talk between humoral and cellular pathways may occur in response to EBV DNA injection and that IMD is a key mediator in both types of activation. The IMD pathway has been shown to induce Jun amino-terminal kinase (JNK) dependent expression of the ligands for Drosophila vascular endothelial growth factor (VEGF) and platelet-derived growth factor (PDGF) receptor (PVR), PDGF- and VEGF-related factors 2 and 3 (PVF2 and PVF3) (Bond and Foley, 2009), which in turn induce hemocyte proliferation in Drosophila (Munier et al., 2002). Hence, the increase in diptericin expression levels and hemocyte numbers upon EBV DNA injection is likely initiated by activation of the IMD humoral immune pathway; this subsequently results in cellular responses. The observation that 70 or fewer copies of EBV DNA trigger the IMD pathway but that 140 or 280 copies do not induce such a response likely indicates that higher copy numbers exert an inhibitory effect; the particular nature of this effect was not determined by the current study. Worth noting is that the DNA within a nascent EBV virion upon release from an infected cell is linear and contains unmethylated CpG dinucleotides (Kintner and Sugden, 1981). Hence, any shed DNA in addition to that delivered into a cell upon an infection is in this form. Hence, the DNA we used in the study at hand was also linear and unmethylated.

The transcriptional levels of drosomycin and TotA, which are indicative of Toll and JAK-SAT pathway activation, respectively, were not affected by EBV DNA treatment. This may indicate that EBV DNA does not activate these pathways or that their activation occurs under other conditions than those assessed by the current study. Previous studies have indicated that unmethylated CpG DNA motifs, which are abundant in the EBV genome in a nascent viral particle (Bergbauer et al., 2010), activate innate immunity through TLR9 in mammals (Fiola et al., 2010); hence, D. melanogaster, as implemented for examining the immune effects of EBV DNA in the current study, may be used to assess Toll involvement in the response to EBV DNA in flies under other conditions in further studies. Worth noting is that in D. melanogaster endogenous accumulation of chromosomal DNA also triggers the production of diptericin (Häcker et al., 2000). Although the nature of DNA exerting this effect is not microbial, it may indicate that responses to various types of immunostimulatory DNA occur via activation of the IMD pathway in flies.

The IMD pathway in flies is considered to be comparable to signaling mediated by TNFR in mammals (Govind, 2008). We hence examined whether TNFα expression levels are affected by EBV DNA in mice. We observed elevated expression levels of this proinflammatory mediator. Hence, a congruent response was detected in mice. We have previously reported that IL-17, IL-23, and IFNγ are enhanced in mice administered EBV DNA. The detection of elevated TNFα adds to the complexity of the proinflammatory response that seems to be triggered in mammalian systems in response to EBV DNA.

## CONCLUSION

EBV DNA triggers the IMD pathway in flies and enhances comparable TNFα expression in mice. This rather validates the implementation of D. melanogaster to screen for responses and mediators triggered by viral DNA. Identification of sensors of viral DNA and downstream mediators via the use of the fruit fly may reveal possible therapeutic targets of pro-inflammatory responses in humans.

## AUTHOR CONTRIBUTIONS

NSh participated in the study design, performed the experiments, and contributed to analysis and writing. NSa, CM, and NH-A assisted in experimental procedures. MS and ER designed the study, supervised the work and resultant analysis. MS and ER contributed equally to this work and should both be listed as corresponding authors.

#### FUNDING

fmicb-09-01268 June 7, 2018 Time: 17:38 # 8

ER was funded by the Asmar Research Fund, the Lebanese National Council for Scientific Research (L-CNRS), and by the

## REFERENCES


Medical Practice Plan (MPP) at the American University of Beirut. MS was funded by the Medical Practice Plan (MPP) at the American University of Beirut and the Lebanese National Council for Scientific Research (L-CNRS).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer HZ and handling Editor declared their shared affiliation.

Copyright © 2018 Sherri, Salloum, Mouawad, Haidar-Ahmad, Shirinian and Rahal. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Emergence of CD4+ and CD8+ Polyfunctional T Cell Responses Against Immunodominant Lytic and Latent EBV Antigens in Children With Primary EBV Infection

Janice K. P. Lam<sup>1</sup>† , K. F. Hui<sup>1</sup>† , Raymond J. Ning<sup>1</sup> , X. Q. Xu<sup>1</sup> , K. H. Chan<sup>2</sup> and Alan K. S. Chiang<sup>1</sup> \*

<sup>1</sup> Department of Paediatrics and Adolescent Medicine, Li Ka Shing Faculty of Medicine, Queen Mary Hospital, The University of Hong Kong, Pokfulam, Hong Kong, <sup>2</sup> Department of Microbiology, Li Ka Shing Faculty of Medicine, Queen Mary Hospital, The University of Hong Kong, Pokfulam, Hong Kong

#### Edited by:

Yves Renaudineau, Université de Bretagne Occidentale, France

#### Reviewed by:

Mario M. D'Elios, Università degli Studi di Firenze, Italy Masaaki Miyazawa, Kindai University, Japan Wesley H. Brooks, University of South Florida, United States

> \*Correspondence: Alan K. S. Chiang chiangak@hku.hk

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 29 September 2017 Accepted: 21 February 2018 Published: 07 March 2018

#### Citation:

Lam JKP, Hui KF, Ning RJ, Xu XQ, Chan KH and Chiang AKS (2018) Emergence of CD4+ and CD8+ Polyfunctional T Cell Responses Against Immunodominant Lytic and Latent EBV Antigens in Children With Primary EBV Infection. Front. Microbiol. 9:416. doi: 10.3389/fmicb.2018.00416 Long term carriers were shown to generate robust polyfunctional T cell (PFC) responses against lytic and latent antigens of Epstein-Barr virus (EBV). However, the time of emergence of PFC responses against EBV antigens, pattern of immunodominance and difference between CD4+ and CD8+ T cell responses during various stages of EBV infection are not clearly understood. A longitudinal study was performed to assess the development of antigen-specific PFC responses in children diagnosed to have primary symptomatic (infectious mononucleosis [IM]) and asymptomatic (AS) EBV infection. Evaluation of IFN-γ secreting CD8+ T cell responses upon stimulation by HLA class I-specific peptides of EBV lytic and latent proteins by ELISPOT assay followed by assessment of CD4+ and CD8+ PFC responses upon stimulation by a panel of overlapping EBV peptides for co-expression of IFN-γ, TNF-α, IL-2, perforin and CD107a by flow cytometry were performed. Cytotoxicity of T cells against autologous lymphoblastoid cell lines (LCLs) as well as EBV loads in PBMC and plasma were also determined. Both IM and AS patients had elevated PBMC and plasma viral loads which declined steadily during a 12-month period from the time of diagnosis whilst decrease in the magnitude of CD8+ T cell responses toward EBV lytic peptides in contrast to increase toward latent peptides was shown with no significant difference between those of IM and AS patients. Both lytic and latent antigen-specific CD4+ and CD8+ T cells demonstrated polyfunctionality (defined as greater or equal to three functions) concurrent with enhanced cytotoxicity against autologous LCLs and steady decrease in plasma and PBMC viral loads over time. Immunodominant peptides derived from BZLF1, BRLF1, BMLF1 and EBNA3A-C proteins induced the highest proportion of CD8+ as well as CD4+ PFC responses. Diverse functional subtypes of both CD4+ and CD8+ PFCs were shown to emerge at 6–12 months. In conclusion, EBV antigenspecific CD4+ and CD8+ PFC responses emerge during the first year of primary EBV infection, with greatest responses toward immunodominant epitopes in both lytic and latent proteins, correlating to steady decline in PBMC and plasma viral loads.

Keywords: EBV, polyfunctional T cells, immunodominance, infectious mononucleosis, asymptomatic primary infection

## INTRODUCTION

fmicb-09-00416 March 5, 2018 Time: 15:7 # 2

Epstein-Barr virus (EBV) is a human gamma-1 herpes virus which infects more than 95% of the population (Young and Rickinson, 2004). This virus is strongly associated with a range of lymphoid and epithelial malignancies and autoimmune diseases (Longnecker and Neipel, 2007; Posnett, 2008; Salvetti et al., 2009). EBV preferentially infects naïve B cells at oropharyngeal lymphoid tissues and subsequently establishes a persistent infection in the circulating memory B cells (Babcock et al., 1998; Rickinson et al., 2014). Once EBV establishes a latent infection, it downregulates the expression of most of the viral genes thereby evading the host immune responses (Kurth et al., 2000). Primary EBV infection typically occurs during childhood without apparent clinical symptoms and evolves into a non-symptomatic persistent virus carrier state (Hislop et al., 2007). Delayed primary infection in some individuals leads to the development of a selflimiting disease called infectious mononucleosis (IM) (Krabbe et al., 1981; Sumaya and Ench, 1985; Niedobitek et al., 1997; Rickinson et al., 2014). EBV has transforming capability as manifested by the occurrence of uncontrolled malignant B cell proliferation in immunosuppressed individuals such as solid organ or stem cell transplantation recipients (Murukesan and Mukherjee, 2012).

The understanding of the mechanisms by which the human immune system controls EBV infection is incomplete due to the difficulties in identifying specific EBV peptides across a wide range of HLA alleles and the fact that EBV can express a large number of lytic and latent proteins displaying hierarchy of immunodominance in its life cycle (Pudney et al., 2005). IM, which is characterized by large expansions of CD8+ T cells and NK cells, provides an excellent model for the study of immune responses against EBV (Callan et al., 1996; Rickinson et al., 2014). The current knowledge on EBV-specific T-cell immunity is mostly derived from studies of IM peripheral blood mononuclear cell (PBMC) samples using either MHC-specific tetramers to determine the percentages of EBV-specific T cells or Enzyme-linked immunosorbent spot (ELISPOT) assay to detect the production of interferon-gamma (IFN-γ) by EBV-specific CD8+ T cells (McCutcheon et al., 1997; Schmittel et al., 1997; Helms et al., 2000; Jennes et al., 2002). However, the functional roles of CD4+ and CD8+ T cells in the control of EBV infection are not completely understood.

Polyfunctional T cells (PFC) are T cells which have multiple functions, such as degranulation of cytotoxic proteins and production of multiple cytokines such as IFN-γ, tumor necrosis factor-alpha (TNF-α), interleukin-2 (IL-2) simultaneously. Accumulating evidence suggests that PFC are associated with more effective control of chronic microbial infections including human immunodeficiency virus (HIV), hepatitis C virus (HCV), and cytomegalovirus (CMV) (Harari et al., 2005; Casazza et al., 2006; Ciuffreda et al., 2008; Duvall et al., 2008; Qiu et al., 2012). We and other groups have studied EBV-specific PFC in long term carriers and found that they produced more cytokines per cell than the single functional T cells and may be functionally superior (Smith et al., 2009; Ning et al., 2011). Another study found that frequencies of PFC were higher in HIV non-progressors than in progressors (Betts et al., 2006). Interestingly, a recent study demonstrated that polyfunctional and IFN-γ mono-functional CD4+ T cells are molecularly distinct and the polyfunctional gene signatures in response to Plasmodium falciparum infection and influenza infection were highly conserved (Burel et al., 2017). These findings support the notion that PFC contribute to more robust T-cell immunity in the control of virus infections. However, how PFC arise during primary EBV infection and evolve over time as well as their role in the long term control of EBV from primary infection stage to long term persistence remain unclear.

Here, we conducted a longitudinal study to assess the development and maturation of T cell responses to EBV from acute infection stage to long term persistence in two primary infection cohorts in children, those presenting as IM and those as asymptomatic primary infection (AS). ELISPOT assay was first performed to detect the IFN-γ secreting CD8+ T cell responses upon stimulation by HLA class I-specific peptides of EBV lytic and latent proteins in 18 longitudinally followed IM cases and 12 AS cases. A 9-color flow cytometric assay which simultaneously delineates five parameters: production of IFN-γ, perforin, TNFα and IL-2, and surface mobilization of CD107a (degranulation marker), upon stimulation by overlapping peptide pools of 4 EBV lytic and 5 latent cycle proteins was then performed to further evaluate the EBV-specific CD4+ and CD8+ PFC responses in another 11 IM cases. Corresponding PBMC and plasma viral loads were determined as measurement of viral control. T cell lysis against autologous lymphoblastoid cell line (LCL) in three IM patients was measured to assess the cytotoxic function of the EBV-specific T cells.

## MATERIALS AND METHODS

### Subject Recruitment

Two cohorts of study subjects consisting of 29 children with infectious mononucleosis (IM) and 12 with asymptomatic primary infection (AS) were recruited. Serological screening for EBV was performed to confirm their primary infection state of EBV (Supplementary Table 1). Children with positive viral capsid antigen (VCA)-IgM, VCA-IgG, negative EBNA1 IgG and showed clinical symptoms were identified as IM subjects. For those who showed a serological profile of primary EBV infection with positive or negative VCA-IgM, positive VCA-IgG, negative EBNA1 and low VCA-IgG avidity without symptoms were recruited as AS subjects. As the maturation of VCA-IgG antibody from low to high avidity takes up to 6 months, AS patients were estimated to have been infected by EBV within a period of 6 months. Heparinized peripheral blood samples were collected at the time of first examination and subsequently, at 1, 2–5, and 6–12 months after diagnosis for the longitudinal study. Plasma was isolated and stored in −80◦C until use. Peripheral blood mononuclear cells (PBMC) were isolated by standard Ficoll-Hypaque density gradient method. Collected PBMC were cryopreserved in fetal bovine serum (FBS) (Invitrogen, Carlsbad, CA, United States) with 10% DMSO under liquid nitrogen until use. All patient samples were handled

as potential biohazardous material following the institutional safety procedures. The study protocol was approved by the Institutional Review Board of the University of Hong Kong. Informed written consent was obtained from each participant prior to the study.

#### DNA Extraction, HLA Typing and Quantitative PCR (qPCR)

Qiagen DNeasy blood mini kit (Qiagen) and NucliSENS easyMag instrument (BioMerieux) were used to extract DNA from PBMC and plasma, respectively, in accordance to the manufacturer's instruction. Part of the DNA samples were transferred to The University of Birmingham and the HLA laboratory in Queen Mary Hospital, The University of Hong Kong for HLA typing. The HLA types of all subjects recruited in ELISPOT assay were documented (Supplementary Table 3). The rest of the DNA was used to measure the EBV loads by qPCR with ABI PRISM 7900 sequence detector (Applied Biosystems, United States). EBV loads in PBMC were determined by the amplification of viral DNA polymerase (Pol, BALF5) sequence. Human β2 microglobulin sequence was detected as an internal control. Plasma EBV loads were quantified by expression of the BamH1W repeats in the EBV genome.

## Synthetic EBV Peptides

Two types of synthetic peptides consisting of 24 HLA class I-restricted peptides and 9 overlapping peptide pools were utilized. All HLA class I-restricted peptides were kindly provided by Professor Alan B. Rickinson (The University of Birmingham, Birmingham, United Kingdom) and their sequences and identities are listed in Supplementary Table 2. The HLA class-I restricted peptides were diluted into 2 µg/mL in CTL-Test Medium (Cellular Technology Limited, United States) and 100 µl was added to each well in 96-well plate. Fifteen-mer peptides overlapped by 11 amino acids spanning each of the EBV latent proteins, EBNA1, EBNA3A, EBNA3B, EBNA3C and LMP2 and the lytic proteins, BZLF1, BRLF1, BMLF1 and GP350 were purchased from JPT Peptide Technologies, Berlin, Germany. The lyophilized peptide pools of each EBV protein were reconstituted according to manufacturer's instruction. Final concentration of overlapping peptides was 10 µg/ml per million cells. The HLA class-I restricted peptides were used in the ELISPOT assay and overlapping peptides were used in the flow cytometric analysis.

## ELISPOT Assay

Interferon-gamma (IFN-γ) producing EBV-specific CD8+ T cell responses were quantified in duplicate by ELISPOT assay. Cryopreserved PBMC and HLA class-I restricted peptides were used. The peptide epitopes tested for each individual were listed in Supplementary Table 3. Wells containing phytohemagglutinin (PHA) (1 mg/ml) were used as positive control and wells with only cells were used as a negative control. IFN-γ producing cells were counted by immunospot plate reader (Cellular Technology Limited-ImmunoSpot S5 Macro Analyzer) and expressed in spotforming cells (SFC)/10<sup>6</sup> PBMC. Spots were defined as positive when the SFC/10<sup>6</sup> PBMC was higher than 20. Only positive results were reported in the present study.

## Expansion of T Cells

For better sensitivity of the flow cytometric analysis, the EBV lytic and latent protein-specific T cells were expanded by adding recombinant IL-4 and IL-7 according to a protocol provided by Professor Cliona M Rooney, Baylor College of Medicine, United States (Jones et al., 2010; Gerdemann et al., 2011). In brief, cryopreserved PBMC were thawed and washed with 10% FBS in RPMI 1640. Cells were seeded and incubated with relevant peptide pools, IL-4 and IL-7 at 2 µg/ml, 1000 U/ml and 10 ng/ml (PreproTech) respectively for 5 days at 37◦C, 5% CO<sup>2</sup> in 24-well plate (1 × 10<sup>6</sup> PBMC/ml). PBMC cultured with IL-4 and IL-7 for 5 days were used as negative control.

## Re-stimulation of T Cells

After expansion of T cells for 5 days, PBMC were rested overnight at 37◦C, 5% CO<sup>2</sup> prior to peptide re-stimulation. PBMC were then re-stimulated with corresponding overlapping peptide pools at 2 µg/ml, together with PE-Cy5-conjugated anti-CD107a monoclonal antibody (BD Biosciences Pharmingen, United States), co-stimulatory reagents containing anti-CD28 monoclonal antibody at 1 µg/ml (BD Pharmingen, United States), anti-CD49d monoclonal antibody at 1 µg/ml (BD Pharmingen, United States) and brefeldin A at 10 µg/ml (BD Biosciences Pharmingen, United States) for 6 h at 37◦C, 5% CO2. Cells stimulated with 1 µg/ml staphylococcal enterotoxin B (SEB) were used as a positive control whereas unstimulated PBMC were used as negative control. Both positive and a negative controls were treated with the same costimulatory reagents.

#### Flow Cytometric Analysis

After incubation for 6 h, the treated cells were washed in PBS and stained with Aqua Blue Dye (Invitrogen, United States) for 20 min for dead cell exclusion. APC-Cy7-conjugated anti-CD3 (BD Pharmingen, United States), PE-Texas Redconjugated anti-CD4 (Life Technologies, United States) and Pacific Blue-conjugated anti-CD8 (BD Pharmingen, United States) monoclonal antibody were subsequently added for surface marker staining. The stained cells were washed, fixed and permeabilized by BD FACS fixation/ permeabilization kit (BD Bioscience). Subsequently, PBMC were washed and stained intracellularly with FITC-conjugated anti-IFN-γ (BD Pharmingen, United States), PE-conjugated anti-perforin (Abcam, United States BD-48 clone), PE-Cy7 conjugated anti-TNF-α (BD Pharmingen, United States) and allophycocyanin (APC)-conjugated anti-IL-2 monoclonal antibody (BD Pharmingen, United States) for 30 min. Cells were washed and resuspended in 1% paraformaldehyde prior to flow cytometric analysis. Around 10<sup>6</sup> cells were acquired on FACS LSR-II flow cytometer (BD Biosciences, United States). FACS data was analyzed by FlowJo (Tree Star, United States). The data with lymphocytes and live cells below 20% of the total population were excluded. The combinations of T cell functions were analyzed by the Boolean combination gate in FlowJo software.

T cells with more than or equal to 3 functions were defined as PFCs.

#### Isolation of CD3+ T Cells and Cytolysis Assay

CD3+ T cells, which were used as effector cells in this study, were isolated and purified from PBMC of 3 IM patients by negative selection using Magnetic Beads Sorting Kit (Pan T cell isolation kit, Miltenyi, Germany). The purity of isolated CD3+ T cells was determined by staining with anti-CD3 monoclonal antibody, followed by analysis with BD LSRII flow cytometer. Highly purified CD3+ T cells (>95%) were obtained after sorting. Autologous LCLs (target cells) were stained with 5- and 6-carboxyfluorescein diascetate succinimidyl ester (CFSE) and incubated in 10% FBS/RPMI at 10<sup>6</sup> cells/ml at 37◦C, 5% CO2. The purified and stimulated CD3+ T cells (effector cells) were co-cultured with the target cells at a ratio 10:1 for 4 h at 37◦C, 5% CO2 and stained with propidium iodide (PI) (Life Technologies). Cells were analyzed by flow cytometry. Percentage of PI positive cells within the CFSE positive population was obtained. Cultures with either autologous LCLs or isolated CD3+ T cells alone served as negative controls.

#### Statistical Analysis

Comparisons of EBV loads in PBMC and plasma amongst the longitudinal samples of IM or AS subjects across the study time points were performed using repeated measure ANOVA. Bonferroni's correction was applied. P-value < 0.0001 was regarded as statistically significant. Comparisons between groups of T cell responses across the study time points were performed by Kruskal–Wallis test and Mann–Whitney Test. P-value < 0.05 was regarded as statistically significant. Prism 6 (GraphPad Software, La Jolla, CA, United States) was used for calculations and illustrations.

## RESULTS

## Magnitude of IFN-γ Secreting CD8+ T Cell Responses Toward EBV Lytic Peptides Decreased but Those Toward Latent Peptides Increased Over Time in Both IM and AS Individuals

We measured the magnitude of EBV-specific CD8+ T cell responses toward EBV lytic and latent peptides in both IM and AS individuals by IFN-γ ELISPOT assay. HLA typing of 18 PBMC samples of IM patients (IM1–18) and 12 AS individuals (AS 1–12) were performed (Supplementary Table 3). HLA-class I specific EBV lytic and latent peptides were used to stimulate the PBMC of HLA-class I matched individuals (Supplementary Tables 2, 3). The number of spot-forming cells (SFC) per million PBMC of the IM and AS individuals is shown in **Table 1**. Trend in changes of the EBV antigen-specific T cell response was deduced by comparing the SFC per million PBMC at the earliest time point to that at the longest time point for each study subject (**Table 1**). Difference in the magnitude of the T cell responses less than or equal to 20 SFC per million PBMC was defined as a stable trend whereas that greater than 20 SFC per million PBMC was defined as an increasing or decreasing trend. As indicated in **Table 1**, the magnitude of IFN-γ secreting CD8+ T cell responses toward EBV lytic peptides tended to decrease over time from day 0 to 12 months whereas those toward latent peptides tended to increase over time in the majority of IM and AS subjects. The ELISPOT data of 2 representative cases, IM16 and IM17, were shown in **Figure 1**. PBMC of IM16 (HLA-type: A2, A26, B46 and cw1) collected at 1, 3, 6, and 12 months post-diagnosis were stimulated by 7 peptide epitopes (TLD, VLK, GLC, YVL, SLR, CLG, VQP). Prominent IFN-γ secreting CD8+ T cell responses against an immediate lytic cycle protein BRLF1-derived peptide (YVL) and a latent cycle protein EBNA3A-derived peptide (VQP) were observed (**Figure 1A**). YVL stimulated 793 SFC at 1 month and decreased to 115 SFC at 12 months. In contrast, VQP stimulated 58 SFC at 1 month and increased to 215 SFC at 12 months. IM17 (HLA-type: A1, A11, B4001, B13, Cw7 and Cw10) showed similar trend of T cell responses (**Figure 1B**). Response against ATI, the A11-restricted BRLF1-derived peptide, was found to decrease from 2193 to 433 SFC whereas the response against SSCS, the LMP2A-derived peptide, was found to increase from 15 to 180 SFC from 1 to 12 months (**Figure 1B**). The decreasing trend of IFN-γ secreting CD8+ T cell responses toward lytic peptides and increasing trend of responses toward latent peptides were observed in the majority of the IM and AS subjects (**Table 1**). No significant difference between the responses in IM and AS individuals was detected.

#### PBMC and Plasma Viral Loads in Both IM and AS Individuals Decreased Over Time

High PBMC viral loads in IM patients during acute infection and substantial decrease of viral loads over time were observed in previous studies (Kimura et al., 1999; Balfour et al., 2005; Cheng et al., 2007). We also determined the viral loads in both PBMC and plasma of 29 IM patients and 12 AS subjects by qPCR. The median PBMC viral load in IM patients decreased from 21878 EBV copies/million PBMC at diagnosis (D0) to 209 copies/million PBMC at 12 months after diagnosis (**Figure 2A**). The median plasma viral loads also decreased from 37154 to 5 copies/mL plasma from day 0 to 12 months (**Figure 2A**). Decreases in PBMC and plasma viral loads were also observed in AS individuals (**Figure 2B**). In both IM and AS subjects, the PBMC and plasma viral loads peaked at diagnosis and declined substantially afterward. The viral loads were the lowest at the end of the study period, indicating an establishment of effective viral control. Repeated measure ANOVA with Bonferroni's correction was applied for multiple comparisons. Cases which had a complete set of time points (day 0, day 7 and 1, 3, 6, and 12 months) were evaluated by the program. The results showed that the PBMC (n = 9) and plasma viral loads (n = 7) significantly decreased over time in the IM samples with values [F(5,40) = 12.76, P < 0.0001] and [F(5,30) = 13.23, P < 0.0001], respectively. In addition, the PBMC and plasma viral loads at day 7 and 1, 3, 6, and 12 months were significantly lower than those at day 0 (Supplementary Figure 1).

## Kinetics of Development of CD4+ and CD8+ T Cell Responses to EBV Lytic and Latent Cycle Antigens After Primary Infection

Next, we carried out a detailed characterization of the development of CD4+ and CD8+ T cell responses to several immunodominant EBV lytic and latent antigens in IM patients from primary infection to long term persistence by multi-color flow cytometric assays. Due to the low frequencies of EBVspecific T cells, an expansion of T cells was necessary for the accurate analysis of the subtle changes of functional T cell subsets in the IM patients during different stages of infection (Jennes et al., 2002; Jones et al., 2010). We adopted a T cell expansion protocol provided by Prof. Cliona M Rooney (Baylor College of Medicine, United States) in order to increase the sensitivity of our flow cytometric assay (Gerdemann et al., 2011). Such expansion protocol using IL4 and IL7 had been used in different T cell studies and found to maintain the survival of most of the cytokine secreting T cells in similar magnitude (Vella et al., 1997, 1998; Gerdemann et al., 2011; Shmarov et al., 2016). To confirm the validity of the flow cytometric assay, we stimulated the PBMC with EBV overlapping peptide pools (EBNA1, EBNA-3A, -3B, -3C and BZLF1) in an IM patient and an EBV-seronegative donor. EBV-specific T cells were clearly detected in the IM patient but not in the seronegative donor except for the pre-loaded perforin in a small proportion of T cells (Supplementary Figure 2). In addition, the increasing trend of latent antigen-specific IFN-γ secreting CD8+ T cell responses and decreasing trend of lytic antigen-specific responses detected by ELISPOT could also be replicated by this flow cytometric assay (Supplementary Figure 3).

We proceeded to analyze the EBV-specific T cell responses of 11 longitudinally followed IM patients (IM19-29) toward 4 lytic and 5 latent overlapping EBV peptides. The magnitude of responses was measured as the frequency of responsive T cells which presented at least one function (either IFN-γ, perforin, TNF-α, IL-2, or CD107a) after stimulation by the overlapping peptide pools. No discernible differences in the magnitude and quality of T cell responses were found among the IM patients of different age groups. The overall magnitude of responses of CD4+ T cells to overlapping peptide pools of lytic and latent cycle proteins was similar to that of CD8+ T cells (**Figure 3**). At diagnosis, all EBV lytic peptide pools could stimulate both CD4+ and CD8+ T cell responses, presenting a broad response spectrum. Remarkably, the frequency of EBNA3A, 3B and 3Cspecific CD4+ and CD8+ T cells clearly increased over time and dominated at 1 year after diagnosis. In contrast, responses toward other EBV lytic and latent peptides were relatively stable throughout the period of our study (**Figure 3**).

## Diversity of CD4+ and CD8+ Polyfunctional T Cells (PFC) Responses Increased Over Time

Polyfunctional T cells, which execute multiple functions upon stimulation by a specific antigen, were reported to elicit stronger immune responses (Betts et al., 2006; Almeida et al., 2007; Smith et al., 2009). We examined the evolution of the magnitude and functionality of lytic and latent cycle antigen-specific CD4+ and CD8+ T cells. The functional profiles of CD4+ and CD8+ PFC (with three or more functions) of a representative case, IM 25, are shown in **Figure 4**. Low percentages of lytic and latent antigen-specific CD4+ and CD8+ responsive T cells with 2 or more functions appeared at diagnosis. The CD4+ and CD8+ responsive T cells against the latent antigen EBNA3B increased dramatically over time. Particularly, the frequency of EBNA3B-specific CD8+ PFC increased from approximately 1% at diagnosis to 30% of total antigen-reactive CD8+ T cells at 12 months after diagnosis. In contrast, BZLF1-responsive cells decreased over time and EBNA1- responsive cells remained stable across the time points from day 0 to 12 months (**Figures 4A,B**). At day 0, most of the CD4+ and CD8+ PFC only displayed a restricted number of combinations of functions. Very low percentages of 4-functional CD4+ PFC (IFNγ+/IL-2+/perforin+/TNF-α+) and 3-functional CD8+ PFC (CD107a+/IL-2+/TNF-α+) were detected (**Figures 4A,B**). At later time points, these CD4+ and CD8+ PFC became more highly functional as 4- and 5-functional T cells started to emerge. Interestingly, a dramatic increase in the diversity of functional subsets of both CD4+ and CD8+ T cells was observed (**Figures 4C,D**). Such diverse functional subsets were found in 8 out of the 11 IM cases whereas the remaining 3 cases showed a constant level or slightly decreased variety of combination of functions in the later time points (Supplementary Figure 4).

#### Emergence of CD4+ and CD8+ Polyfunctional T Cell Responses Toward EBV Lytic and Latent Antigens

We summarized the development of EBV-specific CD4+ and CD8+ PFC for all IM patients (**Figure 5**). EBV lytic and latent cycle antigen-specific CD4+ PFC (with three or more functions) existed at diagnosis but the mean frequencies were less than 3% of total responsive CD4+ T cells (**Figure 5A**). The mean frequencies of CD4+ PFC against lytic and latent cycle antigens did not change significantly over time from primary infection to long term persistent stage. The abilities of EBV lytic and latent cycle antigen-specific CD8+ T cells to generate PFC were relatively stronger when compared with CD4+ cells and appeared to change over time (**Figure 5B**). During acute infection, the mean frequencies of EBV-specific CD8+ PFC were less than 3% of the total responsive CD8+ T cells after stimulation. The percentages of CD8+ PFC against early lytic antigen, BMLF1, BRLF1 and BZLF1, increased slightly whereas those of CD8+ PFC against the latent antigens, EBNA3A, 3B and 3C, increased significantly over time. Other subdominant antigens such as EBNA1, LMP2 and GP350 did not stimulate any significant changes in the proportion of PFC. Although the PFC responses were induced by the immunodominant latent antigens derived from EBNA3A-C in most of the IM patients (**Figure 5**), it should be noted that PFC responses could also be induced



Results are expressed as numbers of spot-forming cells (SFC). "–" represents sample not available.

by antigens derived from immediately early (BZLF1 and BRLF1) and early (BMLF1) EBV proteins in some individuals (Refer to **Figures 4**, **5** and Supplementary Figure 4). Taken together, our data suggested that the CD4+ and CD8+ T cells specific to the immunodominant EBV proteins, including BZLF1, BRLF1, BMLF1, EBNA-3A, -3B, and -3C, became highly functional during the long term persistent phase.

different peptides (ATI, SSCS, AVF, IED, and SEN) in duplicate wells. IM17 reacted to only two of the peptides. IED was shown as representative negative control. Wells with cells only (blank) were also shown as negative controls. The numbers of spots were shown above the wells. Positive results in histogram were expressed in the mean ± SEM of numbers of spot-forming cells (SFC) per million PBMC.

## Increased Cytotoxicity of T Cells Against Autologous LCL in IM Patients Over Time

Cytotoxicity assay was carried out to examine whether increased polyfunctionality is accompanied by enhanced cytotoxicity of EBV-specific T cells. T cells were stimulated and expanded by overlapping peptide pools from the immunodominant EBNA3A or 3B protein of each donor. The CD3+ T cells were then purified by negative selection using Miltenyl Magnetic Beads Cell Sorting system. The purity of the T cells was on average greater than 95%, and the viability of T cells after sorting did not change significantly.

The cytotoxicity of T cells against autologous LCL, after stimulation with EBNA3 overlapping peptide pools, was assessed longitudinally for three IM patients (**Figure 6A**). The cytotoxicity of T cells against LCL increased over time. These results were consistent with elevated frequencies of perforin- and CD107a-expressing T cells (**Figure 6B**). Cytotoxic potential was further enhanced by stimulating the T cells with EBNA3 overlapping peptide pools for 5 days prior to setting up the lysis assay, concurrent with increasing polyfunctionality, perforin production, and degranulation (**Figure 6B**).

## DISCUSSION

Identification of polyfunctional CD8+ T cells in the immune control of viral infection had been reported (Betts et al., 2006; Almeida et al., 2007; Smith et al., 2009). However, most of these studies focused on assessing CD8+ T cell responses to human immunodeficiency virus (HIV) and their mechanisms in killing. Much less is known about the role of PFC responses in the control of stable persistent DNA virus such as EBV. Here, we describe a comprehensive study in which the development of EBV-specific CD4+ and CD8+ T cells, as well as their

magnitude and polyfunctionality were monitored from primary infection to long-term persistence in two cohorts of children with infectious mononucleosis (IM) and asymptomatic (AS) primary EBV infection, respectively.

The IFN-γ secreting CD8+ T cells were characterized as the major population of T cells participating in viral control (Callan et al., 1996; Shi and Lutz, 2002). IFN-γ ELISPOT assay became a common quantitative detection method of virus-specific CD8+ cells (Larsson et al., 1999; Yang et al., 2000). A previous study had demonstrated the diagnostic potential of IFN-γ and IL-2 ELISPOT assays in distinguishing active and latent tuberculosis in children infected with Mycobacterium tuberculosis (Chiappini et al., 2012). In the present study, the kinetics of IFN-γ producing CD8+ T cell responses in IM patients and AS subjects was first assessed by ELISPOT assay. The decreasing pattern of IFN-γ producing CD8+ T cell responses toward EBV lytic peptides and increasing pattern of T cell responses toward EBV latent peptides were identified over time in both IM and AS subjects. There was no discernible difference in the trend of evolution of T cell responses in these two study cohorts. However, this widely accepted IFN-γELISPOT assay using HLAtype restricted epitopes to detect EBV antigen-specific T cells could not simultaneously assess the CD4+ cell responses and determine the expression of other functional cytokines (Casazza et al., 2006; Scherrenburg et al., 2008). The difficulties in choosing the specific EBV peptides across a wide range of HLA alleles also restricted the panel of antigen-specific cells to be tested which could result in missing of potentially important information (Supplementary Tables 2, 3).

Accumulating evidence indicated that the "quality" of T cell responses should be assessed by functionality (Seder et al., 2008). The more functions the T cells possess, the more robust the immune protection they can provide (Makedonas et al., 2010). We have also shown that long term carriers could generate robust PFC responses against lytic and latent EBV antigens with the greatest responses toward the immunodominant epitopes of the EBNA3 proteins (Ning et al., 2011). However, the time of emergence of PFC responses against EBV lytic or latent antigens, differences between CD4+ and CD8+ T cell responses, and concurrent EBV loads in PBMC and plasma from acute to chronic stage of infection are not clearly understood. To address these questions, we performed a longitudinal study to assess the development of PFC responses to EBV in 11 IM patients. We assessed five functional parameters: IFN-γ, TNF-α, IL-2, perforin, and CD107a, all of which are typical functions of T cells that were reported in previous studies (Harari et al., 2006; Pantaleo and Harari, 2006; Precopio et al., 2007; Makedonas et al., 2010). T cells with three or more functions after stimulation are defined

as PFC. EBV-specific lytic and latent antigen-derived overlapping peptides were used to stimulate both CD4+ and CD8+ T cell responses. Similar trend of response of IFN-γ specific CD8+ cells was observed as that detected by ELISPOT assay, validating the reliability of data generated by both methods (Supplementary Figure 3). In the IM patients, EBV-specific CD4+ and CD8+ PFC were detectable at diagnosis toward lytic and latent cycle antigens. These CD4+ and CD8+ PFC only displayed a restricted combination of functions, such as the 4- functional CD4+ PFC (IFNγ+/IL-2+/perforin+/TNF-α+) and the 3-functional CD8+ PFC (CD107a+/IL-2+/TNF-α+) (refer to **Figure 4**). At later time points, these CD4+ and CD8+ PFC became more highly functional as 4- and 5-functional T cells started to emerge. The development of PFC responses toward different specific peptides varied among the individuals. In most of the IM patients, the PFC responses were induced by the immunodominant antigens derived from EBNA3A-C (**Figure 5**). Interestingly, in some of the individuals, antigens derived from immediately early (BZLF1 and BRLF1) and early (BMLF1) EBV proteins induced the highest proportion of CD8+ as well as CD4+ PFCs (refer to **Figures 4**, **5** and Supplementary Figure 4). In addition, low frequencies of EBNA1-specific CD4+ T cells were detected at both diagnosis and 6 and 12 months in some IM patients (Supplementary Figure 4), in contrast to the findings of previous studies in which EBNA1-specific CD4+ T cells were not detected until months after the initial diagnosis (Long et al., 2013; Rickinson et al., 2014). Since EBNA1-specific effector T cells might be important in the control of EBV-positive lymphoproliferative diseases (Jones et al., 2010), it is interesting to follow whether the proportion of EBNA1-responsive PFC would increase at longer time points in the future. Diverse combinations of three or more functions of both antigen-specific CD4+ and CD8+ T cells were shown to emerge at 6–12 months, suggesting a potential co-operative relationship between the CD4+ and CD8+ PFC in the long term viral control (**Figures 4C,D** and Supplementary Figure 4). Presence of CD107a+ and perforinreleasing cytotoxic CD4+ PFC was demonstrated in this study (**Figure 4C**). Similarly, a rare population of CD4+ cytotoxic T cells was previously identified in the patients with chronic viral infection of CMV (Casazza et al., 2006). The presence of cytotoxic CD4+ PFC indicates that CD4+ T cells may also play a direct role in the control of persistent EBV infection. The clinical preparation of virus-specific T cells for the control of post-transplant lymphoproliferative disorders was shown to require the presence of CD4+ T cells (Gerdemann et al., 2011; Linnerbauer et al., 2014). In fact, CD4+ PFC was found to be molecularly distinct from IFN-γ mono-functional CD4+ T cells and associate with favorable clinical outcomes in patients with Plasmodium falciparum infection (Burel et al., 2017). Taken together, our data clearly demonstrated the development and

maturation of both CD4+ and CD8+ PFC from early to long term infection stages in IM patients.

and 1 year after diagnosis (12 m). d, CD107a; g, Interferon-gamma; 2, Interleukin-2; p, Perforin; t, Tumor-necrosis factor-alpha.

Decreasing patterns of EBV loads in PBMC and plasma were illustrated in both IM and AS subjects (**Figure 2**), indicating that both study subjects were capable of generating sufficient T-cell immunity for viral control. Moreover, the viral loads per se could not explain the development of IM symptoms in some individuals, although elevated viral loads were observed in both IM and AS individuals during acute EBV infection (Silins et al., 2001). Manifestation of the self-limiting immunopathological conditions arising in IM patients might be related to the inadequate NK cell cytotoxicity toward infected cells during acute infection (Rickinson et al., 2014). In both IM and AS subjects, plasma loads declined dramatically to low or undetected levels, yet, PBMC viral loads showed a slower decline (**Figure 2**). The sharp decrease in plasma loads was probably due to the initial control by NK cell and the early lytic cycle antigenspecific CD8+ T cell responses. In contrast, the slow decline of PBMC was due to the gradual emergence of both lytic and latent antigen-specific polyfunctional CD4+ and CD8+ T cell responses at about 6 months. Indeed, a natural decay of PBMC viral loads in the early stage of infection independent of EBVspecific T-cell immunity was previously proposed (Hadinoto et al., 2008). The viral control thereafter was probably mediated by the development and strengthening of PFC responses toward the immunodominant antigens derived from BZLF1, BRLF1, BMLF1, EBNA3A, 3B and 3C proteins from about 6 to 12 months post-infection. The sustained decline of EBV loads during this period correlated well with the emergence of T cells capable of presenting multiple functions. To further assess

FIGURE 5 | Development of EBV lytic and latent antigen-specific polyfunctional T cells from acute primary infection stage to long-term persistence. (A) CD4+ T cells and (B) CD8+ T cells in 11 IM patients. T cells were expanded by overlapping peptide pools for 5 days and re-stimulated by the corresponding peptide pools for 6 h before flow cytometric assays. Each bar represents the mean ± SEM percentages of polyfunctional T cells (three or more functions after stimulation) out of total EBV-specific T cells upon stimulation with a specific latent overlapping peptide pool.

whether increased PFC responses correlated to viral control in the IM patients, a cytotoxic assay was performed. We stimulated the T cells of three IM patients with EBNA3A or EBNA3B peptide pools and found a steady increase in cytotoxicity of T cells against autologous LCL over time (**Figure 6**). The increasing proportions of EBNA3A- or EBNA3B-specific PFC correlated with the enhanced cytotoxic capacities, supporting that effective viral control was achieved by the capability to generate highly functional T cells.

#### CONCLUSION

EBV antigen-specific CD4+ and CD8+ PFC responses emerge during the first year of primary EBV infection, with greatest responses toward immunodominant epitopes in both lytic (BZLF1, BRLF1 and BMLF1) and latent (EBNA3A-C) proteins, correlating to steady decline in PBMC and plasma viral loads.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of "Informed Consent of Human Subjects of Research, Institutional Review Board of the University of Hong Kong" with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the "Institutional Review Board of the University of Hong Kong."

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

JL, KH, RN, and XX performed the experiments, data analysis, and interpretation. JL, KH, and AC wrote the manuscript. AC coordinated the project and recruited and followed all patients. KC performed the serological assays for all patients.

### FUNDING

This work was supported by the RGC-GRF (# HKU739403M and HKU763407M), CRCG (# 104002548), and Epstein-Barr Virus research (# 20004525) grants.

#### ACKNOWLEDGMENTS

We thank Professor Alan B. Rickinson of University of Birmingham for providing us HLA-class I restricted peptides. We thank Mr. T. F. Chan and Mr. Wilfred H. S. Wong of The University of Hong Kong for their advices on statistical analyses. We also acknowledge the support of all the parents and patients recruited in this study.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.00416/full#supplementary-material


antigen exposure and persistence. J. Immunol. 174, 1037–1045. doi: 10.4049/ jimmunol.174.2.1037


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Lam, Hui, Ning, Xu, Chan and Chiang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# HHV-6A Infection of Endometrial Epithelial Cells Induces Increased Endometrial NK Cell-Mediated Cytotoxicity

Elisabetta Caselli<sup>1</sup> , Daria Bortolotti<sup>1</sup> , Roberto Marci<sup>2</sup> , Antonella Rotola<sup>1</sup> , Valentina Gentili<sup>1</sup> , Irene Soffritti<sup>1</sup> , Maria D'Accolti<sup>1</sup> , Giuseppe Lo Monte<sup>3</sup> , Mariangela Sicolo<sup>1</sup> , Isabel Barao<sup>4</sup> , Dario Di Luca<sup>1</sup> and Roberta Rizzo<sup>1</sup> \*

<sup>1</sup> Section of Microbiology and Medical Genetics, Department of Medical Sciences, University of Ferrara, Ferrara, Italy, <sup>2</sup> School of Medicine, University of Geneva, Geneva, Switzerland, <sup>3</sup> Human Reproduction Centre – Brunico Hospital, Brunico, Italy, <sup>4</sup> School of Medicine, University of Nevada, Reno, NV, United States

Edited by: Yves Renaudineau, University of Western Brittany, France

#### Reviewed by:

Luciana Barros Arruda, Universidade Federal do Rio de Janeiro, Brazil Mario M. D'Elios, University of Florence, Italy

> \*Correspondence: Roberta Rizzo rbr@unife.it

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 20 September 2017 Accepted: 05 December 2017 Published: 15 December 2017

#### Citation:

Caselli E, Bortolotti D, Marci R, Rotola A, Gentili V, Soffritti I, D'Accolti M, Lo Monte G, Sicolo M, Barao I, Di Luca D and Rizzo R (2017) HHV-6A Infection of Endometrial Epithelial Cells Induces Increased Endometrial NK Cell-Mediated Cytotoxicity. Front. Microbiol. 8:2525. doi: 10.3389/fmicb.2017.02525 Background: We have recently reported the presence of Human herpesvirus-6A (HHV-6A) DNA in the 43% of endometrial epithelial cells from primary idiopathic infertile women, with no positivity in fertile women. To investigate the possible effect of HHV-6A infection in endometrial (e)NK cells functions, we examined activating/inhibitory receptors expressed by eNK cells and the corresponding ligands on endometrial cells during HHV-6A infection.

Methods: Endometrial biopsies and uterine flushing samples during the secretory phase were obtained from 20 idiopathic infertile women and twenty fertile women. HHV-6A infection of endometrial epithelial cells was analyzed by Real-Time PCR, immunofluorescence and flow cytometry. eNKs receptors and endometrial ligands expression were evaluated by immunofluorescence and flow cytometry.

Results: We observed the presence of HHV-6A infection (DNA, protein) of endometrial epithelial cells in the 40% of idiopathic infertile women. The eNK from all the subgroups expressed high levels of NKG2D and NKG2A receptors. Functional studies showed that NKG2D activating receptor and FasL are involved in the acquired cytotoxic function of eNK cells during HHV-6A infection of endometrial epithelial cells. In the presence of HHV-6A infection, eNK cells increased expression of CCR2, CXCR3 and CX3CR1 chemokine receptors (p = 0.01) and endometrial epithelial cells up-modulated the corresponding ligands: MCP1 (Monocyte chemotactic protein 1, CCL2), IP-10 (Interferon gammainduced protein 10, CXCL10) and Eotaxin-3 (CCL26).

Conclusion: Our results, for the first time, showed the implication of eNK cells in controlling HHV-6A endometrial infection and clarify the mechanisms that might be implicated in female idiopathic infertility.

Keywords: endometrium, HHV-6, NK cell, infertility, epithelial cell

## INTRODUCTION

fmicb-08-02525 December 14, 2017 Time: 17:41 # 2

Human herpesvirus 6 (HHV-6) is a ubiquitous pathogen of the Betaherpesvirinae subfamily, which primarily infects CD4<sup>+</sup> T cells (Takahashi et al., 1989). Similarly to other herpesviruses, HHV-6 remains in latency into the host, after an initial productive infection (Sandhoff et al., 1991). HHV-6 is a set of two related viruses known as HHV-6A and HHV-6B (Ablashi et al., 2014). Even if these two viruses present a similar genetical sequence, they differ for biological and pathogenic characteristics. HHV-6B causes exanthema subitum in young children (Yamanishi et al., 1988). HHV-6A seems to be involved in other pathologies, such as multiple sclerosis (Soldan et al., 1997) and encephalitis (McCullers et al., 1995). Moreover, we have recently shown the presence of HHV-6A, but not HHV-6B infection in endometrial epithelial cells of a subgroup of idiopathic infertile women (Marci et al., 2016). HHV-6 infection is implicated in immune-suppressive effects: (i) direct infection and induction of apoptosis of CD4+ T lymphocytes (Lusso et al., 1988; Grivel et al., 2003); (ii) lysis of cytotoxic leukocytes (CD8+ T cells, NK cells) (Lusso et al., 1991; Lusso and Gallo, 1995); (iii) block of dendritic cells and macrophages maturation (Kakimoto et al., 2002; Smith et al., 2005); (iv) inability of macrophages and dendritic cells to produce IL-12p70 after interferon gamma induction (Flamand et al., 1995; Smith et al., 2003, 2005); (v) dysregulation of cytokine networks, with increased secretion of IL-10, RANTES, TNF-alpha and IL-1beta (Flamand et al., 1991); (vi) decreased expression of CD14, CD64 and HLA-DR on the surface of monocytes as a mechanism of immune evasion (Janelle and Flamand, 2006).

Natural killer (NK) cells, positive for the surface marker CD56, are the dominant immune cell type at the uterine mucosa during placentation (Siewiera et al., 2013). They accumulate during implantation, where they support invading placental trophoblast cells and the creation of new vessels, essential for blood supply to the fetus.

The human endometrium contains a substantial population of NK cells (eNK cells) which vary in number and in proportion to the total number of endometrial stromal cells during the menstrual cycle. Although present in proliferative endometrium, eNK cells increase in number substantially in the mid-secretory phase and are the major endometrial lymphocyte population in the late secretory phase and the first trimester of pregnancy. eNK cells are CD56bright CD16+ and also express CD9, which is not expressed by peripheral blood NK cells. In contrast to peripheral blood CD56bright CD16– NK cells, eNK cells have abundant cytoplasmic granules containing perforin and granzyme (Bulmer et al., 1991). There is no consensus about the origin of eNK cells. Mature peripheral blood NK cells or immature precursors may migrate into the endometrium from the blood possibly in response to chemokines produced by cells within the endometrium at specific stages of the menstrual cycle and pregnancy, and be modified by other factors within the endometrium. For example, production of CXCL-12 by extravillous trophoblast (EVT) cells may attract NK cells into the decidua in pregnancy (Wu et al., 2005); interleukin (IL)-15, produced by secretory endometrium and decidua, has a selective chemoattractant effect on peripheral blood CD16– NK cells (Kitaya et al., 2007); and transforming growth factor beta 1 (TGF-1) has been suggested as modifying peripheral blood NK cells to eNK cells (Keskin et al., 2007). An alternative suggestion is that eNK cells are derived from haematopoietic precursor cells within the endometrium (Lynch et al., 2007).

The presence of eNK cells in close proximity to the invading extravillous trophoblast cells suggests that they may play a role in this process. eNK cells produce many different cytokines and growth factors (for example, IL-1, IL-2, IL-4, IL-6, IL-8, IL-10, tumor necrosis factor alpha, granulocyte-macrophage colony stimulating factor, TGF-1, leukemia inhibitory factor and interferon gamma) (Jokhi et al., 1997). eNK cells are also an important source of angiogenic growth factors. Production of angiogenin, angiopoietin (Ang)-1, Ang-2, vascular endothelial growth factor (VEGF)-A, VEGF-C, placental growth factor, keratinocyte growth factor, fibroblast growth factor and plateletderived growth factor-BB by eNK cells from secretory phase endometrium and early pregnancy decidua has been reported (Li et al., 2001; Lash et al., 2006). eNK cells secrete also matrix metalloproteinases (MMP)-1, MMP-2, MMP-7, MMP-9, MMP-10, tissue inhibitor of metalloproteinases (TIMP)-1, TIMP-2, TIMP-3, urokinase plasminogen activator (uPA) and uPA receptor (Naruse et al., 2009a,b).

The observation of modified percentages of eNK cells in the endometrium of women with reproductive failure (such as infertility, RM and pre-eclampsia) has suggested they may play a role in the pathogenesis. Several studies have shown increased levels of eNK cells in the pre-pregnancy endometrium of women with recurrent miscarriage (Clifford et al., 1999; Tuckerman et al., 2007). There are studies showing an increased number of eNK cells in decidual tissue (Stallmach et al., 1999; Bachmayer et al., 2006) and others showing a reduction in decidual eNK cell numbers in women with pre-eclampsia (Eide et al., 2006; Williams et al., 2009). The presence of a unique type of NK cells in the endometrium is intriguing and the possibility that a microbial infection might affect their functions should be taken into account.

In this article, we provide the first evidence that eNK cells functions are affected by HHV-6A infection of the endometrial epithelial cells and this modification might influence pregnancy outcome.

## MATERIALS AND METHODS

#### Clinical Samples

Endometrial speciments were obtained from patients admitted for tubal patency assessment by Hystero-sono contrast sonography at secretory stage of the menstrual cycle. We selected women with these characteristics: 21–38 years old, regular menstrual cycle (24–35 days), body mass index (BMI) ranging between 18 and 26 Kg/m2, FSH (day 2–3 of the menstrual cycle) <10 mUI/mL, 17-β-Estradiol <50 pg/ml (day 2–3 of the menstrual cycle), normal karyotype. Women that presented endometritis, endometriosis, tubal factor,

ovulatory dysfunction, anatomical uterine pathologies and recurrent miscarriage were excluded. Endometrial samples were maintained in HEPES-buffered Dulbecco modified Eagle medium/ Hams F-12 (DMEM/F-12; Invitrogen, Carlsbad, CA, United States) with 1% antibiotic- antimycotic solution (final concentrations: 100 µg/ml penicillin G sodium, 100 µg/ml streptomycin sulfate, 0.25 µg/ml amphotericin B; Invitrogen), and 5% newborn calf serum (NCS; CSL Ltd., Parkville, VIC, Australia), stored at 4◦C, and processed within 2 h. Uterine flushing was performed with a 14-gauge Foley three-way balloon catheter (Eschmann) inflating an appropriate (5 mL) amount of sterile physiologic saline solution.

#### Ethics Statement

This study was approved by the "Ferrara Ethics Committee" and we collected written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

## Preparation of Endometrial Epithelial and Stromal Cells

The endometrium was prepared as previously described (Marci et al., 2016). Cell dissociation was performed in Ca2+ and Mg2+ free phosphate buffered saline (PBS, pH 7.4) additionated with 300 µg/ml collagenase type III (Worthington Biochemical Corporation, Freehold, NJ, United States) and 40 µg/ml deoxyribonuclease type I (Roche Diagnostics, Mannheim, Germany) in a shaking incubator (Bioline 4700; Edwards Instrument Company, Narellan, NSW, Australia) rotating at 150 rpm at 37◦C. Every 15 min, the digested tissues were omogenized vigorously and dissociation was checked microscopically. After 45 min, the digests were filtered using a 40-µm sieve (Becton Dickinson Labware, Franklin Lakes, NJ, United States) to obtain a single cells suspension without debris. The digestion process was stopped by the addition of HEPES-buffered DMEM/F-12 containing 5%FCS. We isolated the different endometrial cellular components (mononuclear, stromal and epithelial cells) by centrifugation for 8–10 min at 390 × g on Ficoll-Paque (Pharmacia Biotechnology, Uppsala, Sweden). Endometrial cells were collected from the Ficoll-Paquemedium interface using BerEP4-coated magnetic Dynabeads (Dynal Biotech, Oslo, Norway) positive selection system. The sorted epithelial cells were recovered using a magnetic particle collector (Dynal Biotech) and washed in HEPES-buffered DMEM/F-12/1%FCS. The collected cells were finally seeded on culture plates coated with basement membrane extract (BME) (Matrigel <sup>R</sup> , Collaborative Biomedical Products, Bedford, MA, United States). The fraction containing mononuclear and stromal cells were collected and seeded on 100 mm plastic tissue culture dishes. After 12 h, the supernatant cells were collected from the culture, to recover non-adherent mononuclear cells. Purity of epithelial and stromal components was morphologically evaluated by light microscopy and assessed by cytokeratin-18 (CK18) and vimentin staining for epithelial and stromal cells respectively. Mononuclear cells were CD45pos stained. The purity reached for each cell population was routinely over 98%.

## DNA Analysis

DNA extraction and analysis were performed as previously described (Caselli et al., 2012). PCR and real time quantitative (qPCR) specific for the U94 gene were used to determine HHV-6 DNA presence and load. Samples in which 1 µg of cell DNA harbored more than 100 copies of viral DNA, were considered positive. Human RNase P or beta-actin housekeeping genes were used as a control. All clinical samples were randomly and blindly investigated. Furthermore, when enough material to repeat the analysis was present, the analysis was repeated again in a randomized and blinded fashion at a distant time from the first analyses. HHV-6A or B identification was performed as reported previously (Caselli et al., 2012), by restriction enzyme digestion of the U31 nested PCR amplification product and visualization of the digestion products on ethidium bromide stained agarose gel after electrophoresis migration.

#### RNA Analysis

RNA cell extraction was performed using the RNeasy kit (Qiagen, Hilden, Germany). Extracted RNA did not contain contaminant DNA, as assured by DNase treatment and control β-actin PCR without retrotranscription reverse transcription (Caselli et al., 2012). RNA reverse transcription was performed by the RT2 First strand kit (Qiagen, Hilden, Germany) for analysis of virus transcripts. cDNA aliquots corresponding to 200 ng RNA were used for virus transcription analysis, performed by qPCR detecting the expression of U94 gene, as previously reported (Caselli et al., 2012).

## Immunofluorescence Assay for HHV-6 Detection

HHV-6 antigen expression was analyzed by immunofluorescence with a mouse monoclonal antibodies (mAb) that recognized the glycoprotein gp116 (late antigen) of HHV-6 A and B (ABI, Columbia, MD, United States), as previously described (Marci et al., 2016). Epithelial and stromal cells were respectively stained with mouse anti-Cytokeratin-18, a heterotetramer cytoskeleton protein (CK18-FITC) and rabbit anti-vimentin-PE moAbs (Abcam, Cambridge, United Kingdom), respectively. eNK cells were stained for CD56-PE (BD, Milan, Italy). Epithelial cells were stained for IP-10-FITC (Interferon gamma-induced protein 10/CXCL10), Eotaxin-3-PE (CCL26) and MCP-1-PE (Monocyte chemo-attractant protein 1/CCL2) (eBiosciences, Waltham, MA, United States).

#### NK Cell Purification

Endometrial NK cells were separated from endometrial leukocyte samples using the negative magnetic cell separation (MACS) system (Miltenyi Biotech, Gladbach, Germany) (Marci et al., 2016). The analysis of purified cell fraction by flow cytometry with CD3-PerCp-Cy5.5, CD56-FITC moAbs (e-Bioscience, Frankfurt, DE), demonstrated that the NK cell content was >90% (data not shown). Freshly purified NK cells were cultured for 24 h in presence of suboptimal doses of IL-12 (1 ng/ml).

#### HHV-6A Cell Infection

fmicb-08-02525 December 14, 2017 Time: 17:41 # 4

Primary endometrial cells and KLE endometrial epithelial cell line (ATCC CRL1622) were cultivated in DMEM F12 medium (ATCC 30-2006) in presence of L-glutamine, 1% penicillinstreptomycine and 10% of FCS at 37◦C with the 5% of CO2. The cells were inoculated with HHV-6A (strain U1102) cell-free virus inocula (Marci et al., 2016) and infected with 100 genome equivalents per 1 cell. HHV-6A UV-inactivated viral preparations were used as controls. Infected cells were co-cultured with eNK cells to perform cytotoxicity experiments.

#### Cytotoxicity Assays

Target cells (1 × 10<sup>6</sup> cells) were labeled with 7-AAD/CFSE Cell-Mediated Cytotoxicity Assay Kit (Cayman Chemicals, Ann Arbor, MI, United States), that employs CFSE to label target cells within the mixed cell population and 7-AAD to label dead cells. We added eNK effector cells to labeled target cells at various effector:target ratios in replicate in 200 ml DMEM/F12 containing 5% FCS. Microtiter plates were centrifuged at 1200 rpm for 5 min and incubated at 37◦C. After 4 or 18 h of culture, the cells were analyzed by flow cytometry.

To block activating receptor engagement or FASL/FAS pathway, CFSE labeled target cells were incubated with soluble NKG2D-Fc IgG1 chimeric protein (0.2 mg/ml) (R&D Systems, Italy) and/or anti-FASL mAb (10 mg/ml; R&D Systems, Italy) or an isotype match control (1.0 mg/ml; mouse IgG1) for 20 min on ice then co-cultured with eNK effector cells. eNK cell cytotoxicity against K562 (ATCC CCL243), classical target cell, was used as control.

#### ELISA sFASL, sTRAIL, MCP1, IP10, Eotaxin3

Levels of sTRAIL, sFasL, MCP1, IP10 and Eotaxin3 were assessed in duplicates in cell culture supernatants and uterine flushing samples using commercial enzyme-linked immunosorbent assays (ELISAs) (TRAIL, MCP1, IP10, Eotaxin3: R&D Systems, Amersham, United Kingdom; sFasL: Sigma–Aldrich, St. Louis, MO, United States).

#### Flow Cytometry

Leukocytes were defined as CD45pos, and the different cell subtypes were defined as CD3pos (T cells), CD19pos (B cells) CD14pos (monocytes) and Cd56pos (NK cells). Cell viability was assessed by propidium iodide staining. Anti-isotype controls (Exbio, Praha, CZ) were performed.

1 × 10<sup>5</sup> eNK cells were labeled with fluorophore-conjugated antibodies: CD3-PE-Cy7, CD16-PE, CD56-APC, NKG2D-PE, NKG2C-PE, NKG2A-PE, NKp30-PE, NKp44-PE, NKp46-PE, KIR2DL2/3-PE, KIR2DL1-PE (BD, Italy), KIR2DL4-PE (R&D Systems, Italy); CXCR1, CXCR2, CXCR3, CX3CR1, CXCR4, CCR1, CCR2, CCR3, CCR5, and CCR7 (R&D Systems, Italy). eNK cells were gated as CD56pos CD3neg. eNK cells activation status was determined by CD107a staining (R&D Systems, Italy), as previously reported (Rizzo et al., 2016).

5 × 10<sup>5</sup> epithelial cells were stained specific Ab HLA-I (HLA-A,-B,-C)-PE (BD Biosciences, Italy), HLA-E (clone MEM-E/08, Exbio, Praha, CZ) or HLA-DR (BD Biosciences, Italy) and matched isotype controls.

The NKG2D-ligands were detected on epithelial cells by binding of NKG2D-Fc chimera (R&D Systems, Italy) and indirect labeling with the secondary Ab FITC-coupled mouse anti-human IgG1 (Abcam, Cambridge, United Kingdom).

Data were analyzed using FACS CantoII flow cytometer (BD, Milan, Italy) and FlowJo LLC analysis software (Ashland, OR, United States). Ten thousand events were acquired.

#### Statistical Analysis

Data were analyzed by Student's t-test and Fisher exact test [Stat View software (SAS Institute Inc.)]. Statistical significance was assumed for p < 0.05 (two tailed).

## RESULTS

#### HHV-6 in Clinical Specimens

We enrolled 20 idiopathic infertile women and 20 fertile women with at least one previous successful pregnancy. As reported in **Table 1**, the two cohorts presented no significant differences.

The endometrial biopsies were analyzed for the presence of HHV-6 infection. As reported in **Table 2**, we found HHV-6B DNA in the peripheral blood mononuclear cells of the 24 and 26% of idiopathic infertile women and control women, respectively (p = 0.86; Fisher exact test). These results confirmed the previously reported frequency of the HHV-6B virus in a 25–30% of peripheral blood samples (Caselli and Di Luca, 2007). On the contrary, HHV-6A DNA was not revealed in the peripheral blood of all the subjects. Forty percent (8/20) women with idiopathic infertility were positive for HHV-6 DNA in their endometrial epithelial cells, while fertile women did not present HHV-6A viral DNA in their endometrial epithelial cells (p = 4.5 × 10−<sup>6</sup> ; Fisher exact test). HHV-6B DNA was not

TABLE 1 | Women cohorts: demographical and clinical parameters.


<sup>∗</sup>Student's t-test. ∗∗Fisher exact test.

TABLE 2 | HHV-6 DNA results in peripheral blood mononuclear cells (PBMC) and endometrial biopsies.


<sup>∗</sup>Fisher exact test.

present in all the endometrial biopsies, as previously reported (Marci et al., 2016).

The average viral load in endometrial epithelial cells (vimentinneg CK18pos; Supplementary Figure S1A) from HHV6- A positive infertile women was 500.000 copies/ug of cellular DNA (range 700.000–240.000 copies/µg DNA), corresponding to about 4 copies of viral DNA per diploid cell (**Figure 1A**). Endometrial epithelial cells were HHV-6A positive as shown by staining for gp116 late viral antigen (**Figure 1B**). Stromal cells (Vimentinpos CK18neg; Supplementary Figure S1B) and leukocytes (CD45pos; Supplementary Figure S1C) were negative for HHV-6 DNA (**Table 2**), confirming a localized HHV-6A infection into endometrial epithelium, as previously reported (Marci et al., 2016).

#### Endometrial Epithelial Cells Are Permissive to HHV-6A Infection

Primary endometrial epithelial cells, purified from endometrial biopsies from fertile women negative for HHV-6A infection (Vimentinneg and CK18pos, Supplementary Figure S1), and the KLE endometrial epithelial cell line were infected with HHV-6A cell-free inoculum with 100 genome equivalents per cell. Mock infection with UV-inactivated virus was used as control. We evaluated viral replication 1, 3, 7, and 14 days post-infection (d.p.i.) by analyzing virus DNA, transcription and antigen expression. As shown in **Figure 2A**, HHV-6A DNA was present at all times p.i. Virus load, as evaluated by real-time quantitative PCR in primary endometrial epithelial cells and KLE cell line, respectively, corresponded to 7.0 × 10<sup>4</sup> and 8.1 × 10<sup>4</sup> genome copies per µg of total cellular DNA at 1 d.p.i., 5.4 × 10<sup>4</sup> and 6.2 × 10<sup>4</sup> at 3 d.p.i. and 3.2 × 10<sup>4</sup> and 4.5 × 10<sup>4</sup> at 14 d.p.i., when the experiment was discontinued. Virus U94 mRNA, an immediate early gene, was analyzed to establish viral transcription (Pantry and Medveczky, 2017). We observed the persistence of U94 mRNA in all time points (**Figure 2B**). The late HHV-6 gp116 antigen expression confirmed that both primary and KLE endometrial epithelial cells were permissive to HHV-6A replication. gp116 virus antigen expression, indicative of productive infection, peaked at 7 d.p.i., (**Figure 2C**). Control mock-infected endometrial epithelial cells gave no staining (**Figure 2D**).

We previously showed that HHV-6A positive women have an increased cytolytic function of eNK cells toward HHV-6A infected cells (Marci et al., 2016), suggesting an implication of viral infections in the modulation of eNK cell functions. Therefore, to test the ability of eNK cells in controlling HHV-6A infection, we analyzed their cytotoxicity against HHV-6A-infected endometrial epithelial cells, by 7-AAD/CFSE Cell-Mediated Cytotoxicity Assay Kit.

The co-culture of eNK cells with HHV-6A infected endometrial for 4 h did not result in an efficient killing (**Figure 3B**). However, after 18 h of co-culture, eNK cells killed HHV-6A infected endometrial epithelial cells (**Figure 3D**). With eNK cells from HHV-6A positive women at the effector:target ratio of 50 we observed up to 75% of killing, while only 40% of killing was obtained with eNK cells from HHV-6A negative women (p < 0.001; Student's t-test). The absence of killing of uninfected endometrial epithelial cells, supports the specificity of the eNK cytotoxic function against HHV-6A infection (**Figures 3A,C**). Taken together, these results suggest that in the presence of HHV-6A infection eNK cells from HHV-6A positive women become more cytotoxic against infected endometrial epithelial cells. To exclude the presence of external factors that could modify the eNK cell cytotoxicity in HHV-6A positive women, we tested the killing of K562 cells, a classical NK cell targets. In agreement with previous studies (Kopcow et al., 2005), we observed a low cytotoxicity of eNK cells in the presence

of K562 cells in both HHV-6A positive and negative women settings (Supplementary Figures S2A,B). These results suggest a peculiar modification of eNK cells in the presence of HHV-6A-infection, that seems to induce the loss of their tolerance toward endometrial epithelial cells. This condition is enhanced in HHV-6A positive women.

## NKG2D Receptor Modulates eNK Cell Responsiveness to HHV-6A-Infected Endometrial Epithelial Cells

NK cells cytotoxicity is controlled by several NKRs. To evaluate their possible implication in eNK cell cytotoxicity against HHV-6A-infected endometrial epithelial cells, we analyzed the expression of activating/inhibitory receptors in eNK cells.

eNK cells, gated as CD3neg and CD56pos, were purified from endometrial biopsies from fertile and infertile women. Flow cytometry analysis showed that eNK cells express NKG2D, but express low levels of the activating receptors NKG2C/CD94 and CD16 (**Figure 4A**). Even though the activating receptors NKp44, and NKp30 are expressed in all eNK cells (**Figure 4A**), there is a decreased expression in infertile women positive for HHV-6A infection (**Figures 4A,B**) (p < 0.001; Student's t-test). This difference in expression pattern makes eNK cells from infertile women positive for HHV-6A infection quite different from eNK cells commonly present in endometrial tissues (ref J Immunol, **Figure 4**), and from decidual NK cells, which express relatively high levels of NKp30, NKp44, and NKG2D (Hanna et al., 2006).

Expression of inhibitory receptors showed that all eNK cells express high levels of NKG2A/CD94 receptor and low expression of KIRs (Killer immunoglobulin like receptors) (**Figures 4C,D**).

These results demonstrate a possible role of HHV-6A infection in changing in eNK cell receptor repertoire, decreasing NKp30 and NKp44 and maintaining NKG2D and NKG2A expression.

The expression of inhibitory and activating receptors is implicated in the immunoregulation of eNK cells functions to control the uterine environment and embryo implantation. A modification of receptors ligands caused by HHV-6A infection could undermine eNK cells peculiar functions.

We analyzed activation status of eNK cells, measuring CD107a expression, that correlates with both cytokine secretion and NK cell-mediated lysis of target cells (Alter et al., 2004). We observed a slight but not significant increase in the percentage of eNK positive for CD107a expression in HHV-6A positive infertile women (p = 0.08; Student's t-test) (**Figure 4E**).

These results suggest that contact of eNK cells with HHV-6A infected cells is detrimental to recognize differences in eNK

for-48 h with HHV-6A. eNK cell cytotoxicity was determined by 7-AAD/CFSE Cell-Mediated Cytotoxicity Assay Kit after (A,B) 4 or (C,D) 18 h of contact at different E/T ratios. Each data point is calculated as the mean lysis ± SD from at least five independent experiments done in replicate tissue culture wells. <sup>∗</sup>Significant p-value; Student's t-test.

cells status. For this, we selected to analyze the effect of HHV-6A infection on ligands for the two NK receptors that maintained a high expression in eNK cells: NKG2D and NKG2A.

Because the identities of the cellular ligands of the activating receptor NKG2D is largely unknown, we stained the endometrial epithelial cells with Ig fusion proteins composed of the extracellular portions of NKG2D fused to human IgG1. NKG2D ligands were down-modulated following infection (**Figures 5A,B**). These results suggest the implication of NKG2D activating receptor in the cytotoxic function of eNK cells toward HHV-6A infected endometrial epithelial cells.

We then analyzed the effect of HHV-6A infection and the expression level of HLA-E molecule, the ligand of inhibitory receptor NKG2A. As shown in **Figures 5A,B**, we observed the expression of both the non-classical HLA-E and the classical HLA-A,-B,-C molecules at the surface of endometrial epithelial cells. The infection with HHV-6A reduced HLA-E expression, while no effect was observed for classical HLA-A, -B, -C. This downregulation of HLA-E expression by HHV-6A excluded any effect of NKG2A inhibitory receptor in controlling eNK cell responsiveness to HHV-6A-infected endometrial epithelial cells. HLA-II antigens are not expressed by endometrial epithelial cells, while they are up-modulated during HHV-6A infection, similar to previously reported the effect on thyrocytes (Caselli et al., 2012) (**Figures 5A,B**).

To further confirm that the cytotoxic function of eNK cells is associated to the NKG2D activating receptor, we next investigated lytic capacities of eNK cells from HHV-6A positive women blocking specific receptor/ligand interactions with a Fc-chimeric protein. The addition of NKG2D Fc-chimeric protein decreased eNK cell cytotoxicity resulting in a 20% lysis of infected cells compared to a 60% lysis of HHV-6A-infected cells, that we observed without the block of receptor/ligand interactions (**Figure 5C**) (p < 0.001, Student's t-test). The reduced eNK cytotoxicity in NKG2D ligation blocking experiments sustains a role for NKG2D receptor in eNK cell cytotoxicity.

When we looked at NKG2D and NKG2A ligands expression by endometrial cells from infertile HHV-6A positive and negative women and fertile women, we observed a slight reduction of HLA-E and a significant decrease in NKG2D ligands expression in HHV-6A positive infertile women (**Figure 5D**) (p < 0.001, Student's t-test).

FIGURE 4 | eNK cell receptor repertoire expression. eNK cells purified from endometrial biopsies were stained for surface expression of the indicated (A) activating and (C) inhibitory receptors and (E) CD107a activation marker using fluorochrome-conjugated antibodies and analyzed by flow cytometry. (B,D) Representative FACS dot plots gated on CD56pos CD3neg dNK cells are reported. One representative dot plot out of five independent experiments is shown. <sup>∗</sup>Significant p-value obtained by Student's t-test.

FIGURE 5 | Functional analysis dNK cells specific receptors. (A,B) HHV-6A-infection modulates the expression of NKR ligands on endometrial epithelial cells. The binding of human NKG2D-Fc chimera was used to evaluate the cell surface expression of specific receptor ligands. The expression of HLA-E and HLA-A, -B, -C and HLA-DR molecules was evaluated using specific mAb. (C) Endometrial epithelial cells uninfected (gray lines) or infected (black lines) were incubated with soluble NKG2D-Fc fusion protein at the concentration of 1 mg/ml and eNK cell cytotoxicity was analyzed by 7-AAD/CFSE Cell- Mediated Cytotoxicity Assay Kit after 18 h of co-culture. (D) The expression of NKG2D ligands, NKG2A ligands (HLA-E and HLA-A, -B, -C) and HLA-DR expression on the surface of endometrial epithelial cells from fertile women and HHV-6A positive and negative infertile women. Data sets represent mean lysis ± SD from five independent experiments done in replicate. <sup>∗</sup>Significant p-value; Student's t-test.

## Cytotoxicity of eNK Cells Is Dependent of FasL Killing Pathway

NK cells cytotoxicity is mediated by soluble mediators or by induction of death receptor-ligand pathways such as TRAIL (tumor necrosis factor-related apoptosis-inducing ligand) and Fas ligand (FasL). To evaluate the mechanisms involved in eNK cell killing of HHV-6A-infected endometrial epithelial cells, we analyzed these death receptor-ligand pathways. After 18 h of coculture, FasL (p < 0.001; Student's t-test), and not TRAIL, was up-modulated on the surface of eNK cells (**Figures 6A,B**). The blockade of FasL with a neutralizing antibody to FasL reduced eNK cell cytotoxicity against HHV-6A-infected endometrial

HHV-6A negative infertile women. <sup>∗</sup>Significant p-value; Student's t-test.

epithelial cells, after 18 h of co-culture (**Figure 6C**). The blockade of NKG2D recognition of its ligands reduced the expression of FasL on eNK cells (p < 0.03; Student's t-test) and the co-blockade of FasL and NKG2D ligands abolished eNK cells killing of HHV-6A-infected endometrial epithelial cells (p < 0.02; Student's t-test). These data suggest that NKG2D ligands recognition and mechanisms dependent of the death receptor-ligand Fas/FasL pathway are at the basis of eNK cell killing of HHV-6A infected endometrial epithelial cells.

When we looked at FasL expression by endometrial cells from infertile HHV-6A positive and negative women and fertile women, we observed increased levels of FasL secretion in the uterine flushing samples of infertile HHV-6A positive women (**Figure 6D**) (p < 0.001, Student's t-test).

#### HHV-6A Infection Modulates the Repertoire of Chemokine Receptors and Their Ligands

The repertoire of ligand to chemokines receptors plays a critical role in eNK cells homing to infected cells. To evaluate the mechanisms implicated in eNK cell cytotoxicity against HHV-6A infected endometrial epithelial cells, we analyzed the expression of chemokine receptors and ligands on eNK cells in the presence or in the absence of HHV-6A infection.

Flow cytometry analysis showed that eNK cells express very low or undetectable levels of CXCR1, CXCR2, CXCR3, CXCR4, CCR1, CCR2, CCR3, CCR5, CCR7 and CX3CR1 (data not shown). We observed a slight increase CXCR3, CX3CR1 and CCR2 receptors in eNK cells from infertile women positive for HHV-6A infection (**Figures 7A,B**) (p = 0.02; p = 0.012; p = 0.01, respectively; Student's t-test).

Meanwhile, we observed an induction of MCP1, IP10 and eotaxin-3 expression on the surface of endometrial epithelial cells 7 d.p.i. (**Figure 7C**). These molecules are the ligands for CCR2, CXCR3 and CX3CR1, respectively, that are upmodulated on eNK cells from HHV-6A positive women (**Figures 7A,B**).

When we analyzed uterine flushing samples for the expression of MCP1, IP10 and eotaxin-3, we observed increased levels in HHV-6A positive infertile women in comparison to HHV-6A negative infertile women and fertile women, with IP10 reaching a significant difference (p = 0.031; Student's t-test) (**Figure 7D**).

Since eNK cells are in close contact with endometrial epithelial cells, we used an in vitro culture model of endometrial epithelial cells to test the ability of eNK cells to get in contact with HHV-6A infected endometrial epithelial cells. Endometrial epithelial cells were infected (72 h) or not and co-cultured with eNK cells from HHV-6A positive infertile women for 2 h. As shown in **Figure 8** eNK cells (CD56pos) were able to establish contact with HHV-6A infected endometrial epithelial cells but not with uninfected cells. On the contrary, eNK cells from HHV-6A negative infertile and fertile women were not able to interact with HHV-6A infected endometrial epithelial cells.

Together these results evidence the ability of eNK cells to contact HHV-6A-infected endometrial cells, demonstrating the

reported. One representative dot plot out of five independent experiments is shown. <sup>∗</sup>Significant p-value obtained by Student's t-test. (C) Chemokine receptor ligand expression. MCP1 (Monocyte chemotactic protein 1, CCL2), IP-10 (Interferon gamma-induced protein 10, CXCL10) and Eotaxin-3 (CCL26) expression was evaluated in endometrial epithelial cells 7 d.p.i. White arrows indicate positive staining. Images were taken in fluorescence (Nikon Eclipse TE2000S) equipped with a digital camera. Original magnification 20×. (D) Chemokine receptor ligand expression. MCP1 (Monocyte chemotactic protein 1, CCL2), IP-10 (Interferon gamma-induced protein 10, CXCL10) and Eotaxin-3 (CCL26) expression was evaluated in uterine flushing samples from HHV-6A positive and negative infertile women and fertile women. <sup>∗</sup>Significant p-value obtained by Student's t-test.

FIGURE 8 | Endometrial epithelial cells were infected for 72 h with HHV-6A. eNK cell were co-cultured for 4 h at a E/T ratio of 10. Endometrial epithelial cells were stained with CK18 (green), eNK cells with CD56 (red) and HHV-6A with gp116 (green). Images were taken in bright field left panels) or fluorescence (right panels) (Nikon Eclipse TE2000S) equipped with a digital camera. Original magnification 20× and 100×.

role of eNK cells in controlling HHV-6A infection in the endometrial tissue.

### DISCUSSION

fmicb-08-02525 December 14, 2017 Time: 17:41 # 11

Our study demonstrates, for the first time, the critical role of eNK cells in counteracting HHV-6A infection in endometrial tissues through cytotoxicity. We showed phenotypic and cellular changes in eNK cells that allow the recognition and killing of HHV-6A-infected cells in a FasL-dependent manner.

We demonstrate that under HHV-6A infectious conditions, a significant fraction of eNK cells rapidly dampened down their Nkp30 and Nkp44 expression level, maintaining NKG2D and NKG2A expression.

We further show that control of HHV-6A infection involves NKG2D receptor and that endometrial epithelial cells decrease NKG2D ligands upon HHV-6A infection. This is true also in vivo, where endometrial epithelial cells from HHV-6A positive infertile women decreased the expression of NKG2D ligand. These findings suggest that also HHV-6A, as previously reported for HHV-6B (Schmiedel et al., 2016), modifies the expression of activating NKG2D receptor ligands. It should be noted that we obtained discrepant results with a decrease of NKG2DL expression by endometrial cells and the acquisition of cytotoxic effector functions through NKG2D receptor in eNK cells. We hypothesize a selective decrease in NKG2DL with the maintenance of high affinity ligands. Alternatively, co-engagement of other activating receptors could be enough even if there is less NKG2D ligands. Otherwise, the recruitment of NKG2DL into cholesterol-enriched membrane lipid microdomains may promote their shedding, maintaining their functional activity (Waldhauer et al., 2008; Aguera-Gonzalez et al., 2011). The identification of receptor-ligand interactions sustaining eNK cellular cytotoxicity might help to identify potential therapeutic targets that could limit HHV-6A spreading in endometrial tissues. In fact, it has been previously suggested that NKG2D receptor expressed by a subset of virusspecific lymphocytes might behave as a prototypic costimulatory receptor with a role in the control of viral infection in HLA class IIpos NKG2DLpos cells, as we observed in endometrial epithelial cells (Saez-Borderias et al., 2006).

The engagement of NKG2D might trigger cytokine production (i.e., IFN-type 1, IFN-gamma, TNF-alpha) that control cytotoxicity of eNK cells by FasL-Fas pathway (Nguyen et al., 2002). We observed the up-regulation of FasL on the surface of eNK cells and in the uterine flushing samples of HHV-6A positive infertile women, suggesting that eNK cell killing of HHV-6A infected endometrial epithelial cells proceeds through NKG2D engagement and subsequent mechanisms dependent on the death receptor-ligand Fas/FasL pathway. Interestingly, both eNK cells from fertile and infertile HHV-6 negative women presented an increase in specific lysis when co-cultured with HHV-6A infected endometrial cells. However, eNK cells from fertile and infertile HHV-6 negative women present a lower lysis that is similar to that obtained when they are co-cultured with K562 target cells (Supplementary Figure S2). For this, we concluded that the lysis obtained with eNK cells from fertile or infertile HHV-6 negative women is due to the natural cytotoxicity of eNK cells in the presence of target cells. On the contrary, eNK cells from HHV-6A positive infertile women showed a significantly higher lysis of HHV-6A infected endometrial cells, that overcomes the natural cytotoxicity of eNK cells.

Similarly, HLA-E molecules, the ligand of the inhibitory NKG2A receptor, are down-modulated during HHV-6A infection and slightly reduced in endometrial epithelial cells from HHV-6A positive infertile women, thus promoting an activatory profile. On the contrary, classical HLA class I molecules maintain their level of expression during HHV-6A infection. The difference between HLA-E and classical HLA-I molecules might reside in their intrinsic functional differences, where HLA-E has an immune-regulatory function while classical HLA-I antigens are totally involved in antigen presentation (Rizzo et al., 2017). HHV-6A might interfere with HLA-E surface expression impairing protein translocation from endoplasmic reticulum. Endometrial epithelial cells acquired a de novo expression of HLA-II DR molecules, as previously reported for thyrocytes (Caselli et al., 2012). This "APC-like" phenotype during HHV-6A infection, might be involved in viral clearance by immune cells. The expression of HLA-II antigens and chemokine receptor ligands could facilitate eNK cells homing to HHV-6A infected endometrial cells, as we demonstrated in in vitro culture model of endometrial epithelial cells. Since no evidence of HLA-II up-modulation in endometrial epithelial cells from HHV-6A positive infertile women was observed, further investigations are needed to demonstrate the role of HLA-II molecules expression in HHV-6A antigen presentation.

We demonstrate that during HHV-6A infection, eNK cells seem to become cytotoxic to limit viral infection. We observed phenotipical and functional modifications of both eNK cells and endometrial epithelial cells in HHV-6A positive infertile women samples, suggesting an imprint due to HHV-6A infection on both eNK cell immune-phenotype and receptor repertoire toward a cytotoxic activity. Meanwhile, the down-modulation of NKG2D ligands on endometrial epithelial cells could maintain in vivo eNK cells with a low killer profile, allowing the persistence of a subclinical HHV-6A infection. We demonstrate that NKG2D activating and chemokine receptors are implicated in eNK cell cytotoxicity toward HHV-6A-infected endometrial cells. The ability of eNK cells to "recognize" HHV-6A infected endometrial cells in vitro support the implication of eNK cells in controlling HHV-6A infection and spreading in endometrial environment. To our knowledge, this is the first time evidence for the involvement of eNK cells in controlling HHV-6A endometrial infection. The persistence of activated eNK and of subclinical HHV-6A infection could alter endometrial environment, as demonstrated by the increase in chemokines, mainly IP10, and FasL in uterine flushing samples from HHV-6A positive infertile women. This perturbation of molecular environment might disadvantage embryo implantation and placentation, which require a correct engagement of eNK cells. The presence of activated eNK cells can potentially have serious adverse side effects, as incorrect or insufficient remodeling of the spiral

arteries leading to complications of pregnancy such as preeclampsia, fetal growth restriction and stillbirth. Nowadays, there are treatments for the control of NK cells activation, as prednisolone, intravenous Ig (IVIG), intralipid, and TNF-α– blocking agent. Interesting results have been obtained with IVIG treatment in antiphospholipid syndrome with persistent presence of autoantibodies against beta2 glycoprotein 1, where NK cell expansion (Oliver-Minarro et al., 2009) and Th1 shift (Benagiano et al., 2017) seem to be implicated in recurrent miscarriage. It is necessary to underline that during viral infections it is not inhibition of eNK cells that is needed, but rather the right degree of activation that is of importance.

Future studies, with a large cohort of infertile idiopathic women will be necessary to elucidate the possible role of eNK cell cytotoxicity toward HHV-6A endometrial infected cells, endometrial health status and embryo implantation. Understanding mechanisms that regulate eNK cell activation will clarify their involvement in female idiopathic infertility.

#### AUTHOR CONTRIBUTIONS

EC contributed to the conception of the work and data analysis. DB, AR, VG, IS, MD, MS, and IB contributed to data collection

#### REFERENCES


and analysis. RM and GLM contributed to clinical samples collection. DDL contributed to data interpretation and critical revision of the article. RR contributed to the conception of the work, data acquisition and analysis, and writing the article.

## FUNDING

This work was supported by HHV-6 Foundation grants (PI: EC and RR), by PRIN grant (PI: EC, cod 2015YZB22C), by Merck grant (PI: RR), and by FISM-Fondazione Italiana Sclerosi Multipla grant (PI: RR, cod 2015/R/20), and Young Scientist Liberati grant (PI: GLM).

#### ACKNOWLEDGMENTS

The authors thank Iva Pivanti for her excellent technical assistance, and Linda Sartor for revising the English manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02525/full#supplementary-material


of residual cells or selective recruitment from circulating peripheral blood? J. Reprod. Immunol. 76, 45–53. doi: 10.1016/j.jri.2007.03.010


for human cytomegalovirus. Eur. J. Immunol. 36, 3198–3206. doi: 10.1002/eji. 200636682


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Caselli, Bortolotti, Marci, Rotola, Gentili, Soffritti, D'Accolti, Lo Monte, Sicolo, Barao, Di Luca and Rizzo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

#### *Walter Fierz\**

*labormedizinisches zentrum Dr Risch, Vaduz, Liechtenstein*

Viruses are able to interfere with the immune system by docking to receptors on host cells that are important for proper functioning of the immune system. A well-known example is the human immunodeficiency virus that uses CD4 cell surface molecules to enter host lymphocytes and thereby deleteriously destroying the helper cell population of the immune system. A more complicated mechanism is seen in multiple sclerosis (MS) where human herpes virus-6A (HHV-6A) infects astrocytes by docking to the CD46 surface receptor. Such HHV-6A infection in the brain of MS patients has recently been postulated to enable Epstein–Barr virus (EBV) to transform latently infected B-lymphocytes in brain lesions leading to the well-known phenomenon of oligoclonal immunoglobulin production that is widely used in the diagnosis of MS. The cellular immune response to HHV-6A and EBV is one part of the pathogenic mechanisms in MS. A more subtle pathogenic mechanism can be seen in the downregulation of CD46 on astrocytes by the infecting HHV-6A. Since CD46 is central in regulating the complement system, a lack of CD46 can lead to hyperactivation of the complement system. In fact, activation of the complement system in brain lesions is a well-known pathogenic mechanism in MS. In this review, it is postulated that a similar mechanism is central in the development of age-related macular degeneration (AMD). One of the earliest changes in the retina of AMD patients is the loss of CD46 expression in the retinal pigment epithelial (RPE) cells in the course of geographic atrophy. Furthermore, CD46 deficient mice spontaneously develop dry-type AMD-like changes in their retina. It is also well known that certain genetic polymorphisms in the complement-inhibiting pathways correlate with higher risks of AMD development. The tenet is that HHV-6A infection of the retina leads to downregulation of CD46 and consequently to hyperactivation of the complement system in the eyes of susceptible individuals.

Keywords: human herpes virus-6A, age-related macular degeneration, CD46, complement system proteins, autophagy, parainflammation, inflammaging

## INTRODUCTION

Many microorganisms use a survival strategy based on their interference with the immune system of their hosts. One way to do so is to acquire immune regulatory proteins from the host that subsequently protect them from immune-mediated attacks by the host. Examples are human immunodeficiency virus (HIV) (1) and cytomegalovirus (2) that both incorporate host cell-derived

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Robert Braidwood Sim, University of Leicester, United Kingdom Undurti Narasimha Das, UND Life Sciences LLC, United States*

> *\*Correspondence: Walter Fierz w.f@swissonline.ch*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 18 June 2017 Accepted: 28 September 2017 Published: 17 October 2017*

#### *Citation:*

*Fierz W (2017) Age-Related Macular Degeneration: A Connection between Human Herpes Virus-6A-Induced CD46 Downregulation and Complement Activation? Front. Immunol. 8:1314. doi: 10.3389/fimmu.2017.01314*

complement control proteins like CD55 and CD59 to protect themselves against complement attacks by the host. Similarly, certain *Borrelia burgdorferi* strains arm themselves with the complement regulatory proteins FHL-1/reconectin and Factor H by using complement regulators acquiring surface proteins (3). Another strategy is to use cell surface receptors of host immune cells for infection and thereby directly interfering with immune functions. A well-known example is HIV that infects host T-helper cells using the CD4 receptor (4, 5).

Other pathogenic effects can be seen when viruses infect host cells and thereby change the cell functions without killing the cells in the process. In multiple sclerosis (MS), e.g., human herpes virus-6A (HHV-6A) infects astrocytes in the brain by docking to the CD46 molecules (6–11). One effect of such HHV-6A infection in MS patients has recently been postulated to interfere with Epstein–Barr virus (EBV) in latently infected B-cells in brain lesions (12). Consequently, B-cells would be transformed by EBV and produce clonal immunoglobulins that are common in MS patients and are used as diagnostic markers in the cerebrospinal fluid. In addition, cellular immune responses to HHV-6A and EBV would induce and sustain the inflammatory lesions in MS brains. Furthermore, the infection of astrocytes with HHV-6A also leads to downregulation of the receptor CD46 that was used for entering the cell (8). Since CD46 is important in limiting the activity of the complement system, the downregulation of CD46 leads to hyperactivity of complement (13). In recent years, it has become clear that complement activity in the brain itself is an important factor in the pathogenesis of MS (14).

Based on these observations, it is postulated here that similar HHV-6A/CD46/complement interactions are central in the development of age-related macular degeneration (AMD). In this article, pathogenic mechanisms in AMD as they are known today are summarized and then a link to HHV-6A *via* CD46 is proposed. Finally, the relation of AMD to MS and other diseases where HHV-6A infection plays a pathogenic role is explored.

#### HYPOTHESIS

Age-related macular degeneration, a degenerative disease of the retina, is the leading cause of irreversible central blindness in elderly people [for review, see Ref. (15)]. Although many risk factors are known [for review, see Ref. (16)], the etiology of AMD remains elusive. Based on known pathogenic mechanisms described below, it is proposed that HHV-6A is an etiologic agent for AMD.

#### Inflammation/Parainflammation/ Inflammaging

Inflammation plays an important role in the pathogenesis of AMD [for review, see Ref. (17–20)]; however, the exact inflammatory mechanisms involved remain unclear. Individuals with elevated C-reactive protein, a general systemic marker for inflammation, carry a higher risk of developing AMD (21). Locally in the retina, proinflammatory macrophages (M1) are enriched at the expense of scavenging and anti-inflammatory M2 macrophages (22). A chronic low-grade inflammation, called parainflammation, is generally considered a beneficial response to chronic insults also in AMD (23). A chronic, parainflammation characteristic for aging is called inflammaging [for review, see Ref. (24)]. Similar to age-related diseases in other organs, inflammaging is supposed to manifest itself also in AMD (25, 26).

#### Complement and CD46

A central role in the inflammatory pathogenesis of AMD is accredited to the regulation of the complement system [for review, see Ref. (27, 28)]. The strong genetic risk conferred by a polymorphism of complement factor H (29–33), but also polymorphisms of ARMS2/HTRA1 (34) support this notion. At present, the function of the ARMS2 protein and the biological consequences of the polymorphism are not completely unraveled, but it has recently been found that ARMS2 functions as surface complement regulator and that ARMS2 is involved in complement-mediated clearance of cellular debris (35).

The spectrum of complement activation in the retina of AMD patients ranges from beneficial to detrimental. Therefore, complement regulation plays a key role in the pathogenesis of AMD. Membrane cofactor protein (MCP, CD46) is a well-known regulatory membrane protein that guards cells from complement attack [for review, see Ref. (36)]. CD46 acts as a cofactor for complement factor I, which protects autologous cells against complementmediated injury by cleaving C3b and C4b deposited on the cells surface. An intergenic single-nucleotide polymorphism just 3′ of complement factor I on chromosome 4 is indeed associated with risk of advanced AMD (37).

The importance of the complement regulatory CD46 is demonstrated by the finding that retinal pigment epithelial (RPE) cells lose their CD46 expression very early in the development of geographic atrophy even before any morphological change of RPE (38). The loss of CD46 makes the RPE vulnerable to complement. Furthermore, an additional role of CD46 in RPE seems to lay in the adhesion of the RPE to its basement membrane and Bruch's membrane, thereby safeguarding its integrity (39). The key pathogenic role of CD46 loss in AMD is also demonstrated by an experimental animal model in which *Cd46<sup>−</sup>/<sup>−</sup>* knockout mice develop a dry-type AMD-like phenotype (40).

#### Autophagy and CD46

Autophagy is a hot topic in AMD research (41) and it is likely to play an important role in the pathogenesis of AMD as a highly regulated clearance and recycling mechanism of cytoplasmic contents [for review, see Ref. (42, 43)]. Transforming growth factor β activated kinase 1 (TAK1), a key player in the regulation of autophagy, maintains the normal function of RPE cells (44, 45). Recently, it was discovered that autophagy is triggered, when pathogens are bound to CD46 (46). This may be a cellular adaptation to infection *via* CD46.

#### Chemokines and Chemokine Receptors

Associations of polymorphisms in the gene of chemokine (C-X3-C motif) receptor 1 (*CX3CR1*) with AMD susceptibility have been reported in several studies (47–49). The CX3CR1


polymorphisms result in decreased affinity for its ligand (CX3CL1, fractalkine), which in turn negatively affects microglial and macrophage migration (50). A chemotactic cytokine, RANTES or CCL5, produced by RPE cells also seems to regulate inflammatory cell migration (51).

#### HHV-6A: Regulating Complement, Autophagy, and Chemokines

Human herpes virus-6A uses the membrane protein CD46 as a receptor to enter cells (8, 52, 53). Such infection is followed by downregulation of CD46. Other viruses, like measles virus (CD46), HIV (CD4), and EBV (CD21), also follow similar strategies of receptor downregulation after infection (54). The CD46 downregulation by HHV-6A may functionally impair the protective effect of CD46 against the activation of autologous complement and the consequent cellular damage as shown *in vitro* using measles virus (55). In this way, HHV-6A would interfere with key pathogenic complement mechanisms in AMD when RPE cells are infected *via* CD46.

HHV-6 infection can also impair Toll-like receptor signaling by reducing TAK1 activity as shown in infected dendritic cells (56). When RPE cells are infected with HHV-6A, the essential role of TAK1 for maintaining normal function of RPE cells through regulation of autophagy would be impaired (44, 45).

HHV-6 expresses its own chemokine receptors encoded by the U12, and U51 genes. The open reading frame U12 functionally encodes a calcium-mobilizing receptor for the β-chemokines RANTES, MIP-1α and -1β, and MCP-1 (57, 58), thereby potentially interfering with RANTES regulation of inflammatory cell migration (51). In epithelial cells already secreting RANTES, U51 expression results in specific transcriptional downregulation of the cytokine (59).

Altogether, HHV-6A, infecting RPE cells *via* CD46, would have the potential to interfere on several levels with the parainflammatory mechanisms central to AMD pathogenesis.

## Association of AMD with HHV-6A-Related Diseases

If HHV-6A has an etiologic role in the development of AMD, as hypothesized here, a higher prevalence of AMD would be expected in other diseases where HHV-6A infection is observed. In contrast to HHV-6B, which is the infectious agent of roseola in childhood, no definite clinical picture of acute HHV-6A infection could be established so far (60–62). On the other hand, HHV-6A infection has been associated with several chronic diseases like AIDS, Hashimoto's thyroiditis, and MS and their epidemiological association is summarized in **Table 1**.

#### CONCLUSION

Despite the fact that more and more molecular and genetic mechanisms involved in the pathogenesis of AMD are known today (76–79), the etiologic trigger of the disease has not been identified so far. Center stage is taken by the CD46 on RPE cells with its regulatory role in complement activation, autophagy, and the chemokine/cytokine network. Since CD46 is also the sole cellular receptor for HHV-6A, one is tempted to speculate that HHV-6A might be the trigger for AMD. Supporting evidence comes from the potential of HHV-6A to interfere with inflammatory mechanisms (62, 80). Indirect evidence comes from epidemiological studies that link HHV-6A-related diseases with AMD (**Table 1**).

In order to substantiate the hypothesis, several approaches are possible:


#### REFERENCES


#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

#### ACKNOWLEDGMENTS

The author would like to extend many thanks to Dr. Brigitte Walz for helpful discussions and suggestions in review of this manuscript.

2020 and 2040: a systematic review and meta-analysis. *Lancet Glob Health* (2014) 2:e106–16. doi:10.1016/S2214-109X(13)70145-1


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Fierz. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Host immune Response to influenza A virus infection

*Xiaoyong Chen1 , Shasha Liu2,3, Mohsan Ullah Goraya1 , Mohamed Maarouf 2 ,3, Shile Huang4 and Ji-Long Chen1,2\**

*1Key Laboratory of Fujian-Taiwan Animal Pathogen Biology, College of Animal Sciences, Fujian Agriculture and Forestry University, Fuzhou, China, 2CAS Key Laboratory of Pathogenic Microbiology and Immunology, Institute of Microbiology, Chinese Academy of Sciences (CAS), Beijing, China, 3University of Chinese Academy of Sciences, Beijing, China, 4Department of Biochemistry and Molecular Biology, Louisiana State University Health Sciences Center, Shreveport, LA, United States*

Influenza A viruses (IAVs) are contagious pathogens responsible for severe respiratory infection in humans and animals worldwide. Upon detection of IAV infection, host immune system aims to defend against and clear the viral infection. Innate immune system is comprised of physical barriers (mucus and collectins), various phagocytic cells, group of cytokines, interferons (IFNs), and IFN-stimulated genes, which provide first line of defense against IAV infection. The adaptive immunity is mediated by B cells and T cells, characterized with antigen-specific memory cells, capturing and neutralizing the pathogen. The humoral immune response functions through hemagglutinin-specific circulating antibodies to neutralize IAV. In addition, antibodies can bind to the surface of infected cells and induce antibody-dependent cell-mediated cytotoxicity or complement activation. Although there are neutralizing antibodies against the virus, cellular immunity also plays a crucial role in the fight against IAVs. On the other hand, IAVs have developed multiple strategies to escape from host immune surveillance for successful replication. In this review, we discuss how immune system, especially innate immune system and critical molecules are involved in the antiviral defense against IAVs. In addition, we highlight how IAVs antagonize different immune responses to achieve a successful infection.

#### *Edited by:*

*Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia*

#### *Reviewed by:*

*Jason Kindrachuk, University of Manitoba, Canada Hovakim Zakaryan, Institute of Molecular Biology (NAS RA), Armenia*

> *\*Correspondence: Ji-Long Chen chenjl@im.ac.cn*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 28 July 2017 Accepted: 05 February 2018 Published: 05 March 2018*

#### *Citation:*

*Chen X, Liu S, Goraya MU, Maarouf M, Huang S and Chen J-L (2018) Host Immune Response to Influenza A Virus Infection. Front. Immunol. 9:320. doi: 10.3389/fimmu.2018.00320*

Keywords: influenza A virus, immune response, innate immunity, adaptive immunity, immune evasion

## INTRODUCTION

Influenza viruses belong to the *Orthomyxoviridae* family, which is characterized by a segmented, negative sense, and single-stranded RNA (ssRNA) genome (1). They are categorized into four genera (type A, B, C, and D), among which influenza A virus (IAV) can infect a wide spectrum of animal species (2, 3). The IAV genome is about 13,500 bases long and composed of eight ribonucleoprotein (RNP) units that encode at least 17 distinct proteins, including recently identified NS3, M42, PA-N182, and PA-N155 (4–6). IAVs can be further classified on the basis of the molecular structure and genetic characteristics of hemagglutinin (HA) and neuraminidase (NA) proteins. To date, 16 HA subtypes and 9 NA subtypes have been identified to be circulating in animals and humans (7). Besides, two HA- and NA-like subtypes designating IAV-like viruses (H17N10 and H18N11) have recently been discovered in bats (8). Co-infection with multiple virus strains in individuals can result in re-assortment (antigenic shift) of genes to produce novel subtypes that could give rise to a global influenza outbreak (9). There are approximately 5 million clinical infection cases caused by influenza viruses every year and 250,000–500,000 deaths resulted from the annual epidemics around the globe, particularly in people over 65 years old who account for 90% of all influenza-associated deaths in the USA (10). Moreover, in March 2013, residents in China exhibiting signs of respiratory infection were first reported to be infected by a novel re-assorted IAV of avian-origin, which was isolated from infected patients and identified as H7N9 (11).

There are several important IAV-encoded proteins that have been reported to be associated with the virus pathogenesis and host immune response to the viral infection. It has been revealed that changes at amino acid level in the viral proteins are related to increased disease severity and immune evasion in humans or avian caused by IAVs. For example, HA is the most abundant surface glycoprotein of the virus that has the ability to attach the host cell, causing cellular fusion and viral entry (12). In addition, HA contains epitopes which are key to trigger the production of neutralizing antibodies by B cells. Thus, the epitopes of HA are the dominant determinants that affect viral mutation and recombination mechanisms (13). The high variability of HA allows IAVs to escape from host immune surveillance and results in influenza seasonal epidemics. NA, the second most abundant glycoprotein, cleaves sialic acid (SA) moieties, promotes the release of nascent virions, and facilitates IAV dispersion (14). NA also plays crucial roles in the viral infection and HA-mediated membrane fusion by binding to SA receptors (15, 16). Respiratory epithelial cells constitutively expressed mucin glycoproteins at their surface that include MUC5AC, MUC5B, and MUC1, which play an important role in restricting IAV infection (17–19). For example, these mucin glycoproteins are rich in SA, which act as viral receptor decoys and prevent the viral binding to the target cells (19–21). However, NA can degrade these mucins and thus attenuate their action (21, 22). Moreover, it is shown that amino acid residues 147 and 151 in NA protein are critical for its interaction with SA. For example, D151G mutation in NA is responsible for its binding to avian α2-3 and human α2-6 SA, promoting the association of H3N2 virus with SA receptors. However, this mutation decreases the enzymatic activity of NA needed to detach the HA from its receptors (22, 23).

Nonstructural protein-1 (NS1), a multiple function protein, contains two functional domains (N-terminal RNA-binding domain and C-terminal effector domain). NS1 is a major inhibitor of host innate immune response. For example, it suppresses the production and signaling of type I interferons (IFNs) (24). In addition, NS1 can trigger the apoptosis of human airway epithelial cells *via* a caspase-dependent mechanism during the IAV infection (25). Matrix protein 2 (M2) forms tetrameric proton channels that are responsible to maintain pH across the viral envelope following the viral endocytosis, and help to release the uncoated viral RNP into the cytoplasm and nuclear import to start viral replication. M2 helps to hold the optimum high pH of the trans-Golgi network for HA-induced fusion and prevents premature conformational changes of HA (26). PB1-F2 is encoded by the alternate open reading frame of PB1 gene of IAV. PB1-F2 with N66S mutation binds with mitochondrial antiviral signaling protein (MAVS) and inhibits the initiation of IFNs. The 1918 deadly influenza strain H1N1 and H5N1 with N66S mutation increase the production of pro-inflammatory cytokines and enhance viral replication in the lung (27, 28).

During infection, host innate immunity provides the first line of defense and triggers pro-inflammatory responses (29). Adaptive immunity also plays a critical role in the clearance of viral pathogens during the later stages of infection. Additionally, respiratory mucosal immunity is induced in the related mucosal tissues during the IAV infection and involved in antiviral defense. In spite of several immune mechanisms to neutralize invasive pathogens or restrict viral replication, IAVs still have evolved diverse strategies to evade host immunity and can establish successful infection. Here, we review how host immune system responds to IAV infection and how IAVs evade the host immune surveillance.

### INNATE IMMUNE RESPONSE TO THE IAV INFECTION

#### IAVs Target and Enter Host Cells

Influenza A viruses primarily target and infect airway and alveolar epithelial cells, which contain the SA glycans as receptors, thus causing alveolar epithelial injury and eventually failure of gas exchange (30, 31). Hence, human IAV infection may lead to acute respiratory distress syndrome (ARDS) and even death (32, 33). Various subtypes of IAVs have different abilities to attach human upper respiratory tract (URT). For example, H1N1 adsorbs abundant ciliated epithelial cells and goblet cells, whereas H5N1 hardly attaches to these cells in human URT (30). In contrast, H5N1 infects alveolar macrophages as well as alveolar epithelial cells (34, 35). Additionally, human and avian IAVs could target and infect various cells in the lower respiratory tract (LRT). It has been observed that H1N1 and H3N2 attach more abundantly to human trachea and bronchi and adsorb more cell types than H5N1 (36). Of note, low pathogenic (LP) avian IAVs generally do not cause a severe pneumonia because they bind human submucosal gland cells and their mucus which can restrain and remove these viruses before approaching LRT (37). However, high pathogenic (HP) H5N1 is able to infect type II pneumocytes as these cells possess an active metabolism, therefore providing a possibility to develop severe pneumonia (35). Moreover, it has been shown that H5N1 reduces proliferation of infected endothelial cells and causes excessive production of cytokines, leading to the lung damage (38, 39).

Hemagglutinin protein on the viral envelope can recognize SA receptors on the surface of the host cells, which is the most crucial step in the process of IAV invasion into an organism (40, 41). Influenza viruses have two common cellular receptors: SA α-2,3 galactose (SAα-2,3-Gal) and SA α-2,6 galactose (SAα-2,6-Gal) (42). It has been found that human influenza strains preferentially attach to the SAα-2,6-Gal receptor (43). The 2,6 SA receptors are present on respiratory epithelial cells of the human URT, while 2,3 SA receptors are found on epithelial cells of the birds, pigs, and in the LRT of humans (44). Thus, variation of cell surface receptors contributes as a major barrier to cross-species and zoonotic transmissions of influenza virus. Hydrolysis of HA0 precursor gives rise to a dimer HA1–HA2 linked by disulfide bonds. HA1 binds to cellular receptors and HA2 facilitates the fusion of the virus to cellular membrane (45). It is also found that HA of influenza virus binds to C-type lectin as an alternative of SA (46). Virus attachment to host cell induces endocytosis using the clatherin and clavoline-dependent mechanism. Following the endocytosis, release of viral particle is pH-dependent physiological event that occurs at late lysosome (47). Low pH of endosome opens the M2 proton channel for proton flux into the virus and triggers the uncoating and subsequent release of viral RNPs (48). The free viral RNA is imported to the nucleus by interacting with the cellular importin-α/β for replication (49, 50). Although it is recognized that both virus preference for SA receptor and host restriction factors are critical for IAV binding and entry into host cells, further studies are needed to unravel the complicated mechanisms underlying the interaction of IAVs with human airway cells.

#### Activation of Innate Immune Signaling upon Intracellular Detection of IAV Infection

The innate immune response is the first line of defense against viral infection which is rapid in response, but nonspecific. During the IAV infection, viral conserved components called pathogen associated molecular patterns (PAMPs) are recognized by host pathogen recognition receptors (PRRs), such as retinoic acid-inducible gene-I protein (RIG-I) and toll-like receptor (TLR), leading to activation of innate immune signaling that finally induces the production of various cytokines and antiviral molecules (51, 52). These PAMPs have certain characteristic of viral RNA that are not shared by cellular RNAs, such as regions of double-stranded RNA (dsRNA) or the presence of a 5'-triphosphate group (53, 54).

Pathogen recognition receptors have the ability to distinguish self from non-self molecules within the infected cells. RIG-I is the main receptor to recognize the intracellular ssRNA and transcriptional intermediates of IAVs in the infected host cells (**Figure 1**). Non-self RNA and transcriptional products of IAVs in the cytoplasm are also sensed by melanoma differentiationassociated gene 5 (55). Following the recognition of PAMPs, RIG-I is activated and its caspase activation and recruitment domains (CARDs) are exposed. Then the CARD is modulated by dephosphorylation or ubiquitination by E3 ligases, such as TRIM-containing protein 25 (TRIM25) (56). Thus, CARD-dependent association of RIG-I and MAVS trigger the downstream transduction signaling at the outer mitochondrial membrane (57). Subsequently, the transcription factors, including interferon regulatory factor 3 (IRF3) and IRF7, and nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) are activated, causing the expression of a variety of IFNs and cytokines (**Figure 1**) (58).

Toll-like receptors are critical PRRs that sense the pathogen outside of cell membrane and internally at endosomes and lysosomes (59). TLRs expressed on cell membrane are TLR1, 2, 4, 5, and 6 that recognize PAMPs derived from bacteria, fungi, and protozoa, while TLR3, 7, 8, and 9 are expressed on the surface of endosomes and lysosomes and exclusively recognize nucleic acid PAMPs derived from various viruses, including IAVs (60–62).

TLR3, TLR7, and TLR8 are involved in sensing the IAV components in cytoplasmic endosomes during the virus replication. It is known that TLR3 recognizes dsRNA in endosomes (63). Interestingly, it was shown that TLR3 may recognize recently unidentified RNA structures that are present in phagocytosed cells infected with IAVs (64). In plasmacytoid dendritic cells (pDCs), TLR7 recognizes the ssRNA of the influenza virions that are taken up into the endosomes (65). Then downstream signaling of TLR7 is activated *via* the adaptor protein myeloid differentiation factor 88 in pDCs, which results in the activation of either NF-κB or IRF7 to induce the expression of pro-inflammatory cytokines and type I IFNs, respectively (65). In macrophages and DCs, TLR3 interacts with TIR-domain-containing adapterinducing interferon-β. Such interaction results in activation of the serine–threonine kinases IκKε (IKKε) and TBK1 that phosphorylate IRF3 to regulate the expression of IFN-β (66). In human monocytes and macrophages, TLR8 is stimulated by its ligand ssRNA, leading to the production of IL-12. However, the relationship between TLR8 and IAV infection has not been defined (67).

Moreover, some NOD-like receptors, such as NOD-like receptor family pyrin domain containing 3 (NLRP3, also known as cryopyrin) and NLR apoptosis inhibitory protein 5, have been observed to be activated upon cellular infection with IAV (68). NLRP3 is expressed by number of cells, including DCs, macrophages, neutrophils, monocytes, and human pulmonary epithelial cells (69, 70). Three signals are required for activation of inflamosome to trigger cytokine production. First, NLRP3 is activated through pathogen detection, which induces the expression of the genes encoding pro-IL-1β, pro-IL-18, and pro-caspase-1 (71). Second, the IAV-encoded M2 ion channel is required to trigger NLRP3 inflammasome activation and cleavage of pro-IL-1β and pro-IL-18 (72). Recently, it was revealed that accumulation of IAV PB1-F2 in the lysosome of macrophages acted as the third signal leading to activation of the NLRP3 inflammasome (73).

#### Antiviral Molecules Involved in Innate Immunity against IAV Infection

Activation of specific transcription factors including, NF-κB, IRF3, and IRF7 during IAV infection results in translocation of these factors into the nucleus where they initiate the transcription of genes encoding IFNs and pro-inflammatory cytokines (TNF, IL6, IL1β, etc.). It is well known that type I IFNs, such as IFN-α and IFN-β, and type III IFNs also known as interferon lambdas (IFN-λ1, IFN-λ2, IFN-λ3, IFN-λ4) play important roles in antiviral response in both virus-infected and uninfected cells (74). Infection with IAV induces robust expression of type I and type III IFN genes (75). Following the expression, IFN-α and IFN-β interact with IFN-α/β receptors (IFNAR), while IFN-λs interact with IFNL receptors (IFNLR) in an autocrine or paracrine manner, which activate Janus kinase-signal transducer and activator of transcription (JAK-STAT) signaling pathway. Phosphorylated STAT1 and STAT2 bind with IRF9 to form a complex ISG factor 3 (ISGF3). ISGF3 translocates into nucleus and binds with IFN-stimulated response element, which triggers the transcription of numerous IFN-stimulated genes (ISGs) (**Figure 1**) (76). Previous studies suggest that type I and type III IFNs provide similar defense against IAV infection in wildtype mice (77). It was further shown that IAV infection induced expression of same type of ISGs in epithelial cells of wild type, IFNAR- or IFNLR-deficient mice (78). Only when both IFNα/β and IFN-λ receptors in mice were knocked out, the animals failed to restrict non-pathogenic influenza virus (77). In spite of similar role of IFNα/β and IFN-λs to a certain extent, some notable differences exist. For example, it has been observed that mice infected with influenza virus showed higher pulmonary inflammation and mortality after treatment with IFNα, while IFN-λ remained protective (79, 80).

These ISGs target different steps of IAV life cycle. For example, viral entry into cells can be restricted by several ISGs, including Mx family, interferon-induced transmembrane protein family (IFITMs), cholesterol 25-hydroxylase (CH25H), and TRIM proteins. Mx family is comprised of MxA and MxB in human, Mx1 and Mx2 in mice. Mx proteins are produced by various cells, such as hepatocytes, DCs, endothelial cells, and immune cells (81). Mx proteins were the first ISGs identified to restrict IAV infection of mice (82). Recently, a study showed that MxA could retain incoming viral genome in the cytoplasm of human cell (83). In addition, nuclear Mx1 in mice impedes the process of early transcription of IAV activated by the polymerase in the nucleus (84). It is thought that the sensitivity of IAVs to MxA depends on their nucleocapsid proteins, and usually avian strains of IAVs are more sensitive to MxA than human strains (85, 86). However, role of MxB in humans or Mx2 in mice during IAV infection is poorly understood (87).

Interferon-induced transmembrane protein families are known as new ISGs that restrict early viral entry by altering the cellular membrane properties like cell adhesion, fluidity, and spontaneous curvatures (88, 89). It has been found that IFITM proteins restrict the replication of IAVs by interfering virus–host cell fusion following viral attachment and endocytosis (90). Another antiviral ISG, CH25H is an integral element of cellular membranes and upregulated by IFN signaling. CH25H enzymatic activity converts cholesterol to soluble 25-hydroxycholesterol (25HC), which is involved in antiviral defense against enveloped viruses, including influenza virus through blocking viral fusion. Recently, it was suggested that high concentration of 25HC causes physical changes of cellular membrane properties to prevent viral fusion (91). Moreover, previous studies showed that IFNactivated STAT1 bound to the promoter proximal region of the Ch25h gene to stimulate the production of 25HC that enhanced innate immune response against IAV (92, 93). In addition, TRIM proteins play multiple roles in antiviral immunity. TRIM25, an E3 ubiquitin ligase, is considered to regulate the re-localization of RIG-I to mitochondrion and signal transduction to MAVS for innate immune response against the viral infection (94). TRIM22 blocks IAV genome encapsidation and degrades nucleoprotein of IAV by polyubiquitination (95). TRIM32 binds with influenza PB1 RNA polymerase, reduces the polymerase activity, and thus restricts the viral replication (96).

There are increasing number of ISGs that regulate viral mRNA expression and protein translation. For example, zinc finger antiviral protein (ZAP), oligoadenylate synthase and ribonuclease L (OAS-RNase L), PKR, and ISG15 are involved in the regulation of IAV mRNA levels and protein synthesis. ZAPs inhibit the expression of IAV PB2 and PA proteins by reducing the viral mRNA expression and blocking its translation (97, 98). OAS-RNase L can destroy viral RNA in the cytosol of host cells and finally halt the protein synthesis process and viral replication. Mice with reduced expression of RNase L are more prone to influenza virus infection (99). PKR is expressed by all kind of cells and upregulated by type I and type III IFNs (100). PKR expressed in inactive form is activated by influenza virus infection. PKR is a known anti-IAV factor that binds to viral dsRNA and suppresses viral protein synthesis. Genetically deficient PKR mice are highly susceptible to influenza virus (101). ISG15 is an ubiquitin-like protein and restrict viral replication by interfering with virus release and translation of viral proteins (102).

In addition, many other ISGs are also involved in innate immunity against IAV infection. These include viperin, tetherin, and so on (103, 104). It has been shown that overexpression of viperin (also known as RSAD2) restricts the release of influenza virus by affecting the formation of lipid rafts specific microdomains that are particular budding sites of the virus (105). Tetherin is another potential host antiviral factor. In 2008, it was reported that tetherin inhibited retrovirus release (106). Tetherin appeared to limit cellular export of viral progenies by internalizing and degrading them exported to the surface of infected cells (107). It was also found that tetherin restrained the influenza virus by tethering and degrading newly budded viral particles (108). Recently, it has been known that tetherin was able to suppress the budding of several laboratory oriented and seasonal influenza strains, but unable to restrict pandemic influenza A/Hamburg/4/2009 and wild-type influenza virus particles (103, 104, 108).

#### Cells Involved in Innate Immunity against the IAV Infection

Airway epithelial cells are the first target of IAVs. These cells produce antiviral and chemotactic molecules that initiate immune responses by rapid recruitment of innate effector cells, such as NK cells, monocytes, and neutrophils. All cell types have their own unique mechanisms to interact with virus-infected cells to limit viral replication, and also prime adaptive immune cells for antigen-specific immunity and memory. Tumor necrosis factor alpha (TNF-α) and IL-1 induce endothelial adhesion molecules, which trigger the migration of innate immune cells, such as macrophages, blood borne DCs, and natural killer (NK) cells to the site of infection.

Alveolar macrophages are critical for limiting viral spread. Activated macrophages phagocytose IAV-infected cells and thus limit viral spread and regulate the following adaptive immune response (109). Monocytes derived from bone marrow precursors circulate in bloodstream. During IAV infection, MCP-1 (CCL-2) produced by infected epithelial cells attracts alveolar macrophages and monocytes *via* their CCR2 receptors (110). NK cells are important cytotoxic lymphocytes of innate immune system to eliminate IAV infection. It has been revealed that lysis of IAV-infected cells is mediated by the binding of IAV-HA with the cytotoxicity NKp44 and NKp46 receptors (111). Expression of IAV-HA on the surface of infected cells is recognition signal for NK cells, and thereby NK cells target and lyse the infected cells (112, 113).

Dendritic cells, the specialized antigen-presenting cells, bridge up the innate and adaptive immune responses during the IAV infection. Adaptive immune response begins when naïve and memory T lymphocytes recognize viral antigens presented by DCs. In the naïve steady state, DCs are orchestrated underneath the respiratory tract, including the airway epithelial tissue, lung parenchyma, and the alveolar spaces of the lungs (114, 115), where they constantly monitor for invading pathogens by their dendrites that are extended to airway lumen through the tight junctions of epithelial cells. Upon infection with IAV, the conventional DCs (cDCs) migrate from lungs to lymph nodes through interaction between CCR7 and its ligand CCL19 and CCL21 (116). In the lymph nodes, cDCs present antigens derived from IAV to T lymphocytes (117, 118). The self-infected DCs degrade the viral protein into immune peptides. Immune peptides (epitopes) in the cytosol are exported to the endoplasmatic reticulum, where they bind with major histocompatibility complex (MHC) class I molecule. Following the binding with epitopes, MHC class I is transported to the cell membrane *via* the Golgi complex for recognition by virus-specific CD8<sup>+</sup> cytotoxic T cells (CTL). However, viral proteins degraded in endosomes/ lysosomes are associated with MHC class II molecule. These complexes are presented on the cell membrane for recognition by CD4<sup>+</sup> T helper (Th) cells. This process may lead to B cell proliferation and maturation to antibody producing plasma cells (119). In addition, DCs can exert cytolytic activity and contribute to the formation of bronchus-associated lymphoid tissue (BALT) during the IAV infection (120).

## ADAPTIVE IMMUNITY AGAINST THE IAV INFECTION

T cells and B cells play key roles in adaptive immunity against the IAV infection. T cells are mainly known as CD4<sup>+</sup> T and CD8<sup>+</sup> T cells. CD8<sup>+</sup> T cells differentiate into cytotoxic T lymphocytes (CTLs), which produce cytokines and effector molecules to restrict viral replication and kill virus-infected cells. Therefore, T cells are crucial for the restriction of viral infection. Upon infection with IAV, naïve CD8<sup>+</sup> T cells are activated by DCs migrated from lungs to T-cell zone of the draining lymph nodes, leading to T-cell proliferation and differentiation into CTLs (121, 122). Moreover, type I IFNs, IFN-γ, IL-2, and IL-12 also help CD8<sup>+</sup> T cells to differentiate into CTLs (123, 124). IFN-λs were also shown to enhance the T-cell proliferation during influenza virus vaccination (125). CTLs decrease the expression of CCR7 and upregulate the expression of CXCR3 and CCR4, which enables their migration from lymph nodes to the lungs where they kill IAV-infected cells.

Mechanism by which CTLs function is well understood. Upon targeting the virus-infected cells, CTLs produce cytotoxic granules that contain molecules like perforin and granzymes (e.g., GrA and GrB). Perforin binds target cells to form pores on the cell membrane that promote passive diffusion of granzymes to induce apoptosis. It has also been found that GrA can restrict virus replication *via* cleavage of viral and host cell proteins that are involved in protein synthesis (126, 127). In addition, CTLs have the ability to induce apoptosis by expressing cytokines, such as TNF, FASL, and TRAIL, which recruit death receptors in IAV-infected cells (128). Post-infection virus-specific CTLs and DCs circulate in blood, lymphoid organs, and the site of infection (117, 129). These memory CTL cells are quick in response to secondary IAV infection, and the activation and differentiation process received during first infection affects their proficiency and efficiency during a secondary infection (130). Although neutralizing antibodies protected from second infection with the same serotype of IAV, CTLs are specific for epitopes in conserved IAV proteins, such as NP, M1, and PA. Therefore, the CTL response is heterosubtypic in nature (131).

Studies have shown that IAV-specific CD8<sup>+</sup> T cells can last for 2 years in murine models (132). The cytotoxicity of the memory CD8<sup>+</sup> T cells decreases significantly, which is related to their declined target competence and reduced cytolytic molecule expression (131). Autophagy plays a critical role in the establishment of memory CD8<sup>+</sup> T cells, as Atg7-deficient mice are unable to form CD8<sup>+</sup> T cell memory against IAV infection (133). Notably, IAV-specific memory CD8<sup>+</sup> T cells in the nasal epithelia prevent the spread of the virus from the URT to the lung, thus blocking the development of pulmonary disease (134). Besides, lung-resident memory CD8<sup>+</sup> T cells can defend against heterologous IAV infection, *via* restraining viral replication and facilitating viral elimination (135). Additionally, lung-resident monocytes support to establish lung-resident CD8<sup>+</sup> T cell during IAV infection (136).

CD4<sup>+</sup> T cell is another important type of immune cells that is involved in adaptive immunity against the IAV infection. CD4<sup>+</sup> T cells can also target IAV-infected epithelial cells through MHC class II and induce MHC class II expression in epithelial cells in murine models (137, 138). Multiple co-stimulatory ligands expressed by CD4<sup>+</sup> T cells contribute to B cell activation and antibody production, among which CD40 ligand (CD40L) is noteworthy (139). CD40L has been shown to enhance immune response against the highly mutated HA protein of IAV (140). Similar to CD8<sup>+</sup> T cells, CD4<sup>+</sup> T cells are activated by DCs that migrate from the lung to the draining lymph nodes during the IAV infection (141, 142). CD4<sup>+</sup> T cells differentiate into Th1 cells in response to IAV infection, according to their stimulators, including antigen, co-stimulatory molecules, and cytokines secreted by DCs, epithelial cells, and inflammatory cells (143, 144). Th1 effector CD4<sup>+</sup> T cells express antiviral cytokine, such as IFN-γ, TNF, and IL-2 (145), and activate alveolar macrophages (146). The IL-2 and IFN-γ produced by Th1 cells regulate CD8<sup>+</sup> T-cell differentiation to clear the viral infection (147, 148). CD4<sup>+</sup> T cells are also able to differentiate into Th2, Th17, regulatory T cells (Treg cells), follicular helper T cells, and sometimes as killer cells (149). Th2 cells bind to virus-derived MHC class II-associated peptides by antigen-presenting cells and produce IL-4 and IL-13 to promote B cell responses predominantly (150). It has been observed that Th17 and Treg cells are involved in regulating cellular immunity against IAV infection (151). Although it is known that CD4<sup>+</sup> T cells can direct CD8<sup>+</sup> T cell responses by secreting various cytokines, the precise roles of CD4<sup>+</sup> T cells to facilitate and regulate CD8+ T cell responses to IAV infection remain elusive, because primary CD8<sup>+</sup> T cell response against IAV infection could be initiated independently of CD4<sup>+</sup> T cells in mice (152).

B cells are indispensable for priming the defense against infection with heterosubtypic influenza virus strains. In cooperation with memory T cells, naïve B cells reduce morbidity and promote recovery upon heterosubtypic infection (153). At the same time, non-neutralizing antibodies generated by B cells facilitate viral elimination and accelerate memory CD8<sup>+</sup> T cell expansion after heterosubtypic infection (153). In addition, IAV-specific antibody-dependent cell-mediated cytotoxicity (ADCC) also plays a role in the cross-reaction against diverse HA subtypes (154). Though IgA is important in the protection against IAV infection in the respiratory tract, IgG is the dominant antibody in this process. Additionally, some studies have implied that IgG could inhibit pathogenesis involving influenza, while IgA is more crucial for the inhibition of transmission of IAVs (155).

Till now, the lifespan and response speed of both memory B cells and plasma cells are foremost in the induction of protective antibody response by IAV vaccines. However, in the elderly, the memory B cells are maintained, but the antibody response is not maintained even upon multiple IAV immunizations. This suggests a potential defect with aging in the development of plasma cells (156). A study has shown that autophagy is involved in maintaining memory B cells to counteract IAV infection; Atg7 deficient mice exhibits loss of memory B cells, causing reduced secondary antibody response to IAV infection and displaying severe lung damage (157).

#### RESPIRATORY MUCOSAL IMMUNITY AGAINST THE IAV INFECTION

#### Lymphoid Tissues and Immunoglobins in the Respiratory Tract Involved in Immunity against the IAV Infection

The nasal openings and URT are the main entry sites for IAVs and mucosal immune system also acts as the first line to limit the IAV infection apart from innate immunity. Secretory IgA (s-IgA) and IgM are the major neutralizing antibodies present on mucosa to prevent viral entry. Nasal secretions contain IgA which can neutralize HA and NA of IAVs (129, 158). During primary infection with IAVs, all three major immunoglobulin classes (IgG, IgA, and IgM) are present in mucosal secretion to limit the infection, though IgA and IgM are higher in concentration than IgG (159). It is thought that IgM response is dominant during primary infection, whereas during secondary infection IgG response is dominant for immunoglobin secretion (2, 119, 121). In the URT, mucosal response is induced in the nasopharyngeal-associated lymphoid tissues (NALT) (160–162). When antigens are pinocytosed or phagocytosed by macrophages present on the NALT, they interact with local T and B cells, resulting in development of a large number of IgA Ab-forming cell (IgA-AFC) precursors (163, 164). The primed T and B cells migrate from NALT to the lungs *via* general circulation, where they differentiate into specific IgA-AFC to secrete antiviral antibodies. Thus, NALT appears to be initial inductive site for secretion of s-IgA against IAV infection. In the LRT, mucosal immune responses occur in the BALT (165). BALT is the site for AFC development and production of mucosal s-IgA against IAV infection (166).

### Role of s-IgA Antibody in Defense against the IAV Infection

Secretory IgA is the primary isotype detected at the mucosal surface (167), which contributes to mucosal protection through its distinct ability to remove an agent before it traverses the mucosal barrier and infects the cell (168). By covering the viral surface, s-IgA prevents the influenza virions from adhering to the susceptible cells, and thus inhibits their invading host cells and neutralizes the viruses without complement participation. Investigations have been demonstrated that s-IgA plays vital roles both in protection against homologous IAV infection and in cross-protection against URT infection by the viral variants (169). Generally, parenteral administration of IAV vaccine leads to the generation of serum IgG, but not s-IgA, while s-IgA and IgG are both induced by intranasal administration (168, 170). Further, polymeric s-IgA is involved in defending against influenza in humans. Moreover, the quaternary structure of the polymeric s-IgA seems to play a key role in protecting human URT from influenza, and have more neutralizing capacity against IAVs than dimeric s-IgA (171).

#### ESCAPE OF IAVs FROM HOST IMMUNE SURVEILLANCE

To establish a successful infection, IAVs have evolved multiple strategies to circumvent the host immunity. For example, it is well known that IAV infection triggers robust production of IFNs that induce the expression of numerous antiviral molecules or ISGs. Although IFNs have a strong antiviral activity, they cannot fully control IAV infection due to the virus-mediated suppression of IFNs signaling. The mechanisms by which IAVs escape from host antiviral immune responses are discussed here.

#### The Antagonism of Major IAV Proteins

Hemagglutinin of IAVs has been shown to facilitate IFNAR ubiquitination and degradation, reducing the levels of IFNAR, and thus suppressing the expression of IFN-stimulated antiviral proteins (172). It has been described that two discrete antigenic sites, H9-A and H9-B, may provide a novel mechanism for H9N2 virus to counteract humoral immunity (173). In addition, a study has shown that the escape of H5N1 from vaccine-mediated immunity is caused by the addition of N-glycosylation sites on the globular head of HA (174). In contrast, antibody response against NA of IAV cannot inhibit viral infection, but restrain its diffusion, thus lowering the severity of influenza. IAVs employ NA protein to block the recognition of HA by natural cytotoxicity receptors, NKp46, and NKp44 receptors and evade the NKp46 mediated elimination, leading to minimized clearance of infected cells by NK cells (175).

Nonstructural protein-1 of IAVs is the most important IFNs antagonist protein, acting on multiple targets and suppressing the host IFN response. Viral RNA invading the host cell causes RIG-I ubiquitination by a RING-finger E3 ubiquitin ligase named as TRIM25, which is essential for RIG-I signaling pathway to trigger host antiviral innate immunity (94, 176). However, NS1 protein can inhibit the TRIM25-mediated RIG-I ubiquitination, thereby blocking RIG-I activation (177). Moreover, NS1 has an inhibitory effect on protein kinase RNAactivated (also known as protein kinase R, PKR), but the effect relies on the induced expression of vault RNAs (a kind of small non-coding RNA with approximately 100 bases). They are initially described as fornix RNP complex components (178). Through NS1 protein, influenza virus induces the expression of vault RNA that inhibits the activation of PKR and the production of IFNs and ultimately promotes the replication of the virus. In a recent reverse genetic investigation, it was found that after interfering with NS1, the phosphorylation level of PKR dramatically increased, which was attenuated by forced expression of vault RNAs (179). These data indicate that IAV has evolved a critical mechanism by which NS1-mediated PKR inhibition is mediated by upregulation of the host factor vault RNAs that inactivates PKR and blocks the production of downstream effector molecules of IFNs.

In addition, studies have shown that through the interaction with IκB kinases (IKK) α and β, two important kinases in NF-κB pathway, NS1 protein can block the phosphorylation of these kinases and eventually destroy the NF-κB complex predominating in nucleus as well as the expression of downstream genes (180, 181). Also, through the JAK-STAT pathway, NS1 protein can block IFN-mediated downstream signaling pathway and weaken the antiviral effect mediated by the downstream effector molecules induced by IFNs. Specifically, NS1 acts mainly by lowering the phosphorylation levels of STAT1, STAT2, and STAT3, preventing STAT2 from entering into the nucleus to bind to the DNA sequence of ISGs promoter region, leading to reduced expression of ISGs (182). Importantly, NS1 is not only involved in host innate immunity, but also affects adaptive immunity *via* modulating the maturation and the capacity of DCs to induce T cell responses (183). Evidence also indicates that influenza virus NS1 can bind to cellular double-stranded DNA (dsDNA), counteract the recruitment of RNA polymerase II (Pol II) to DNA, and finally block the transcription of IFNs and ISGs (184).

#### The Antagonism of Other IAV Proteins

Studies have found that PB1-F2 protein has a mitochondrial positioning signal, *via* interacting with MAVS, to counteract RLR-mediated activation of IFN signaling pathway (185). Investigation on the interaction between the virus and host by systematic biology analysis has revealed that PB2 protein, a member of the viral polymerase complex, also plays roles in IFN antagonism (186). Furthermore, PB2 interacts with the MAVS to evade from the host IFN antiviral response, which is similar to the action mode of PB1-F2 protein (187). Recently, viral M2 protein has been found to interfere with the host autophagy (188, 189). These studies have suggested that viral M2 may inhibit the activation of TLR pathway and the generation of IFNs *via* blocking the host autophagy.

### CONCLUSION

It is well known that host immune response to IAV infection comprises multiple intricate processes that coordinate together to play significant roles in the protection of host. Given the high mutation rate of IAVs, it is necessary to have effective vaccination strategies that can induce robust production of specific antibodies and long-lived T cell response to defend against the viral infection. Since host innate immunity is also critical for anti-IAV infection, further efforts are needed to utilize the current knowledge and technology to enhance the host innate immunity for control of the disease. While our understanding of the IAV-host interaction has increased profoundly, extensive studies are required to better understand the dynamics of host immune system upon detection of the evolved IAVs. Bridging these gaps will pave the way

### REFERENCES


not only for designing better vaccines and effective vaccination strategies but also for developing novel antiviral agents.

#### AUTHOR CONTRIBUTIONS

XC, SL, and MG performed systematic literature review and wrote the manuscript. MM and SH revised the manuscript. J-LC organized and provided the frame for the manuscript and critically revised the manuscript. All authors read and approved the final manuscript.

#### FUNDING

This work was supported by National Key Research and Development Program of China (2016YFD0500206, 2016YFD0500205, and 2016YFC1200304), Natural Science Foundation of China (91640101), and National Key National Basic Research Program (973) of China (2015CB910502).


mannose receptor in viral entry. *J Virol* (2010) 84(8):3730–7. doi:10.1128/ JVI.02148-09


interferon at the level of the MAVS adaptor protein. *PLoS Pathog* (2011) 7(6):e1002067. doi:10.1371/journal.ppat.1002067


subvert autophagy and maintain virion stability. *Cell Host Microbe* (2014) 15(2):239–47. doi:10.1016/j.chom.2014.01.006

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer HZ and handling editor declared their shared affiliation.

*Copyright © 2018 Chen, Liu, Goraya, Maarouf, Huang and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

#### *Stéphane Rodriguez1,2, Mikaël Roussel1,2, Karin Tarte1,2 and Patricia Amé-Thomas1,2\**

*1 UMR U1236, INSERM, Université de Rennes 1, Etablissement Français du Sang Bretagne, Equipe labellisée Ligue Contre le Cancer, LabEx IGO, Rennes, France, 2 Centre Hospitalier Universitaire de Rennes, pôle Biologie, Rennes, France*

During the last decades, considerable efforts have been done to decipher mechanisms supported by microorganisms or viruses involved in the development, differentiation, and function of immune cells. Pathogens and their associated secretome as well as the continuous inflammation observed in chronic infection are shaping both innate and adaptive immunity. Secondary lymphoid organs are functional structures ensuring the mounting of adaptive immune response against microorganisms and viruses. Inside these organs, germinal centers (GCs) are the specialized sites where mature B-cell differentiation occurs leading to the release of high-affinity immunoglobulin (Ig)-secreting cells. Different steps are critical to complete B-cell differentiation process, including proliferation, somatic hypermutations in Ig variable genes, affinity-based selection, and class switch recombination. All these steps require intense interactions with cognate CD4<sup>+</sup> helper T cells belonging to follicular helper lineage. Interestingly, pathogens can disturb this subtle machinery affecting the classical adaptive immune response. In this review, we describe how viruses could act directly on GC B cells, either through B-cell infection or by their contribution to B-cell cancer development and maintenance. In addition, we depict the indirect impact of viruses on B-cell response through infection of GC T cells and stromal cells, leading to immune response modulation.

Keywords: germinal center, T follicular helper cells, B cells, fibroblastic reticular cells, follicular dendritic cells, lymphoma, Epstein–Barr virus, human immunodeficiency virus

#### INTRODUCTION

Defense against pathogens involve an immediate innate immune response, going along with the establishment of a long-lasting adaptive immune response. Secondary lymphoid organs (SLOs), such as lymph nodes, spleen, tonsils, or mucosal associated lymphoid tissues, are functional structures widely dispersed in the entire body, and ensuring adaptive immune response initiation (1). Indeed, they are meeting points for antigens and lymphocytes allowing the development of memory B and T cells, as well as the differentiation of specialized antibody-secreting cells. The architecture of these lymphoid structures is highly organized, with well-delimited B- and T-cell zones of maturation delineated by specific stromal cell subsets: fibroblastic reticular cells (FRCs) in T-cell areas and follicular dendritic cells (FDCs) in B-cell zones (2). Mature B cells at their different stages of maturation are predominantly localized in primary or secondary follicles. Different steps are critical to allow complete B-cell differentiation into long-lived plasma cells secreting high-affinity

#### *Edited by:*

*Yves Renaudineau, University of Western Brittany, France*

#### *Reviewed by:*

*Mario M. D'Elios, University of Florence, Italy Adam Cunningham, University of Birmingham, United Kingdom*

#### *\*Correspondence:*

*Patricia Amé-Thomas patricia.ame@univ-rennes1.fr*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 08 August 2017 Accepted: 16 October 2017 Published: 31 October 2017*

#### *Citation:*

*Rodriguez S, Roussel M, Tarte K and Amé-Thomas P (2017) Impact of Chronic Viral Infection on T-Cell Dependent Humoral Immune Response. Front. Immunol. 8:1434. doi: 10.3389/fimmu.2017.01434*

antibodies mediating the humoral immune response. These steps include proliferation, immunoglobulin (Ig) gene modifications, cell selection based on affinity for antigen, and require intense interactions with specific CD4+ helper T cells belonging to follicular helper lineage. Nevertheless, some viruses can directly or indirectly disturb this subtle machinery affecting the T-cell dependent humoral immune response.

#### GERMINAL CENTER (GC) REACTION

B-cell maturation begins in SLOs, where trafficking naive B cells recognized native antigens through their B-cell receptors (BCR), the membrane form of Igs. To continue their maturation after BCR engagement, these activated B cells need a cognate interaction with pre-activated CD4<sup>+</sup> T cells. These CD4<sup>+</sup> T cells present at the T-B border and providing B cell help are named pre-T follicular helper (pre-Tfh) cells and engage immunological synapse, involving various stimulatory molecules, with B cells. After this activation step, part of these B cells differentiates into extra-follicular short-lived plasma cells secreting low affinity antibodies. Alternatively, T cells and B cells enter the follicle by downregulating EBI2 and CCR7 expression (3, 4), allowing the beginning of the GC reaction with maturation of B cells into GC B cells and T cells into mature T follicular helper (Tfh) cells (5). Highly proliferating GC B cells decrease their membrane BCR expression and undergo somatic hypermutations consisting in random point modifications introduced into the antigen binding regions of BCR genes, and allowing BCR affinity modulation. This results in generation of daughter cells harboring BCR with a variable affinity for the stimulating antigen. A selection step follows with the aim of essentially keeping GC B cells with a high-affinity BCR (6). FDCs harbor at their membrane immune complexes containing antigens, which are grabbed by GC B cells with high BCR affinity. Such GC B cells then present antigen-derived peptides to cognate mature GC Tfh cells, which in turn provide them with survival signals and trigger class switch recombination, resulting in the modification of Ig class. Noteworthy, a part of selected GC B cells can be addressed for further expansion and somatic hypermutations (7, 8). Thereafter, GC B cells leave the follicles and differentiate either into circulating memory B cells, or plasmablasts and subsequently long-lived plasma cells secreting high-affinity IgG, IgA, or IgE (**Figure 1**).

#### GC B CELLS

Germinal center B cells are mature B cells with differentiation process in progress, leading to high-affinity memory B cell and antibody-secreting cell production. GCs can be spatially segregated in two distinct compartments: the dark zone and the light zone. In the dark zone, GC B cells termed centroblasts are highly proliferative and undergo somatic hypermutations. By contrast, centrocytes, or GC B cells from the light zone, are non-proliferative cells prone to die unless rescued by the interaction with FDCs and Tfh cells. In addition, centrocytes undergo class switch recombination process (9). Centroblast and centrocyte Ig gene modification processes orchestrated by AID (10) favor genetic instability with increased risk of transforming event acquisition (11). Both of these GC B-cell compartments highly express BCL-6, CXCR5, CD10, and CD21, but can be discriminated otherwise. Centroblasts highly express the chemokine receptor CXCR4 at the cell surface, allowing their localization in the dark zone where CXCL12-producing stromal cells are localized (12, 13). Centrocytes downregulate CXCR4, allowing their migration to the light zone, and display a CD83hi CD40<sup>+</sup> phenotype facilitating cognate interactions with Tfh cells (14). GC B cells are prone to apoptosis and express several proapoptotic molecules such as Fas (CD95).

#### NON-B CELLS INVOLVED IN THE GC RESPONSE

#### T Follicular Helper Cells

T follicular helper cells have been recently described as a *bona fide* lineage of memory CD4<sup>+</sup> helper T cells driven by the transcription factor BCL-6 and specialized in helping the production of high-affinity and class-switched memory B cells and antibody-secreting cells (15–17). Tfh cells are characterized by high expression of program death-1 (PD-1) and the chemokine receptor CXCR5. In association with a low expression of CCR7, their CXCR5hi phenotype allows Tfh cell localization in B-cell follicles. Tfh cells also highly express ICOS, unlike the TBET, GATA3, RORc/RORγt, and Foxp3 transcription factors specific for Th1, Th2, Th17, and regulatory T (Treg) cells, respectively (17, 18). Nevertheless, they have the capacity to synthetize cytokines related to these other helper T-cell lineages, such as IFN-γ, TNF-α, IL-2, IL-4, and IL-17 (19, 20). They also shared the expression of IL-21 with Th17 cells, and human Tfh cells specifically secrete the CXCR5 ligand CXCL13 (19). IL-21 plays a predominant role in the regulation of GC responses and B-cell differentiation (21, 22). Cytokine secretion profile is not uniform at the single cell level. In agreement, the whole Tfh cell population is more heterogeneous than previously assumed and gathers several Tfh cell subsets providing differential help to GC B cells (23). In SLOs, pre-Tfh cells are localized at the T-B border, whereas mature Tfh cells are localized inside the GC, establishing immunological synpases with centrocytes (**Figure 1**).

More recently, a circulating counterpart of these cells has been described in blood (24). Circulating or peripheral Tfh cells are memory CD4<sup>+</sup> helper T cells defined by the expression of CXCR5, but at lower level than in SLOs. They are generally also defined as PD-1<sup>+</sup> CCR7low and can express ICOS (25), but a strict phenotype is not currently consensual. Circulating Tfh cells are functionally defined by the help they can provide to B-cell differentiation into antibody-secreting cells *in vitro*. However, circulating Tfh cells and Tfh cells from SLOs also distinguish clearly by the expression level of BCL-6 that is poorly detectable by flow cytometry in circulating Tfh cells (26).

#### T Follicular Regulatory (Tfr) Cells

T follicular regulatory cells have been first described more than 10 years ago as Foxp3<sup>+</sup> cells within human tonsil GCs (27), but their extended features were further demonstrated in

Figure 1 | Germinal center (GC) reaction. Naive B cells patrol within the B-cell area in order to encounter specific antigens and become activated through their B-cell receptor (BCR) engagement. Then, primed B cells migrate at the T–B border, and meet pre-T follicular helper (pre-Tfh) cells, which are CD4+ T cells that have been previously activated (act T CD4) by processed antigens presented by mature dendritic cells (DC) in T-cell area. This cognate interaction between pre-Tfh cells and primed B cells involved the recognition of processed antigens presented by the primed B cells to pre-Tfh cells. Following this interaction, both cell types downregulate EBI2 and CCR7 and increase BCL-6 expression, a prerequisite to cell migration within follicles, initiation of the GC reaction, and maturation of B cells and pre-Tfh cells in centroblasts and germinal center T follicular helper cells (GC Tfh), respectively. Centroblasts, displaying membrane CXCR4 expression, localize in close contact with CXCL12-expressing reticular cells in the dark zone of GC, proliferate, and undergo somatic hypermutations. This latter process results in generation of centrocytes harboring BCRs with variable affinity for the stimulating antigen, localized in the light zone. A selection step driven by follicular dendritic cells (FDC) occurs, in order to choose centrocytes with a high-affinity BCRs. Centrocytes are non-proliferative cells prone to die unless rescued by their interaction with GC Tfh cells. This interaction involves the presentation by centrocytes of processed Ag in CMH-II to T-cell receptors (TCR) of GC Tfh cells. In addition, centrocytes interact with GC Tfh cells through co-stimulatory molecules, such as CD86/C28, PD-L1/PD-1, CD40/CD40L, and ICOS-ligand (ICOS-L)/ ICOS. This interaction results in B-cell activation of pro-survival pathways and drives centrocytes to undergo class switch recombination. Thereafter, B cells leave follicles and differentiate either into circulating memory B (mem B) cells or long-lived plasma cells secreting high-affinity antibodies (IgG, IgA, or IgE). GC Tfh cells also egress and become circulating Tfh cells. One mechanism involved in the control of humoral response is related to the inhibitory action of specialized cells termed T follicular regulatory T cells (Tfr). Tfr cells are functional regulatory T cells localized in follicles and represent one of the mechanisms controlling the magnitude of the GC response.

mice (28–30). Tfr cells are functional regulatory T cells localized in GCs and represent one of the mechanisms controlling the magnitude of the GC response after immunization. Tfr cells were more likely to control the function and output of an established GC, with extra-follicular Treg cells controlling the initiation of the GC (31). In addition, Tfr cells have been demonstrated to be implicated in the control of humoral autoimmunity in mice (32, 33). Like other Treg cells, they express inhibitory molecules such as GITR or CTLA-4. However, they share CXCR5, PD-1, and ICOS expression with Tfh cells, but do not secrete B-cell helper cytokines IL-21 and IL-4. Finally, they co-express BCL-6 and FOXP3 transcription factors specific of Tfh cell and Treg cell lineages, respectively (18, 28). Recently, it has been demonstrated that they can originate either from thymic Treg cells or from CD4<sup>+</sup> activated T cells specific from the immunizing antigen (34).

#### Stromal Cells

Different stromal cells derived from non-hematopoietic precursors of mesenchymal origin co-exist in SLOs and reside in different areas. Among them, two subsets are of interest in this review: FRCs occupying T-cell areas and FDCs within B-cell zones (35).

Fibroblastic reticular cells are specialized myofibroblasts organized as an intricate tridimensional network allowing structural organization of SLOs (13). FRCs are involved in immune cell recruitment in SLOs through the secretion of CCL19 and CCL21, the two ligands of CCR7. Naive T cells and dendritic cells (DCs) are in constant contact with FRCs within T-cell area and migrate along the FRC network. Furthermore, FRCs promote T-cell (36), DC (37) and B-cell (38, 39) survival.

Follicular dendritic cells are the lymphoid stromal cell subset involved in recruitment of B cells and Tfh cells in follicles through CXCL13 secretion. They are absolutely required for GC maintenance and B-cell retention (40). One of their key functions is the capacity to recycle for long time native antigens captured as complement-coated immune complexes, allowing their presentation for the GC B-cell selection step (41).

### MODULATION OF HUMORAL IMMUNE RESPONSE BY VIRAL INFECTION OF GC B CELLS

The principal agent infecting mature B cells is Epstein–Barr virus (EBV) belonging to *Herpesviridae* family (**Figure 2**). EBV infects the vast majority of humans principally during childhood and persists mostly in latent form throughout life. EBV primary infection is transmitted through saliva exchange before virus entry into epithelial cells. There, virus begins a replication phase through lytic cycles and, therefore, infects B cells through exosome production (42). Once in B cells, linear viral genome circularizes and remains latent as episome within the nucleus. Three types of EBV latency are described, each characterized by the expression of part (type I latency) or all (type III latency) viral proteins, including six EBV nuclear antigens (EBNAs) and three latent membrane proteins (LMP1, LMP2A, and LMP2B). Lytic phase and type III EBV latency are linked to higher immunogenicity and viral replication, while the lack of viral protein expression observed in type I latency is presumed as responsible of the impaired immune recognition allowing virus maintenance. Importantly, GC B cells express a more restricted number of viral proteins, limited to EBNA1, LMP1, and LMP2 (latency II).

Several proteins encoded by EBV genome cross-react with signaling pathways involved in B-cell proliferation and survival. LMP1 is a trans-membrane protein considered as a CD40 mimicry as they share several downstream signaling molecules such as tumor necrosis factor receptor-associated factors, which activate NFκB (43), JAK/STAT, PI3K (44), and MAPK (45) pathways. LMP2A, which is expressed together with LMP1, mimics BCR functions allowing B-cell survival of non-expressing BCR cells isolated from adenoid GC B cells (46). Moreover, this ITAM-containing protein allows the recruitment of the downstream signaling molecules Syk, Cbl, PLCγ2, and p85 (47, 48). These studies performed *in vitro* or with transgenic mice ascribed robust signaling to LMP1 and LMP2, considering that they are able to drive activation and survival in the absence of antigen by hijacking BCR and CD40 signals, and even rescue BCR-defective B cells (43, 49–52). Roughan and Thorley-Lawson showed that *in vivo* EBV-infected cells express LMP1, LMP2, and EBNA1Q-K (the default latency program) but the role of LMP1 and LMP2 is probably less crucial than assumed *in vitro*, perhaps only supplementing physiologic signals to provide a survival advantage in the highly competitive environment of the GC (53). Of note, GC B cells latently infected with EBV undergo *in vivo* intensive proliferation without expansion, due to death or egress process (54).

Epstein–Barr virus persists in circulating memory B cells (55). Controversial studies discussed about direct *in vivo* EBV infection of GC B cells, or GC differentiation access of infected naive B cells. It has been suggested that EBV initially affect naive B cells before experiencing the GC response and preferentially become memory B cells rather than antibody-secreting cells (55). Nevertheless, transgenic mice model of LMP1-expressing B cells revealed that LMP1 induces proliferation and extra-follicular differentiation into antibody-secreting cells and blocks GC engagement (43). *In vitro*, EBV is able to efficiently infect resting naive B cells, leading to continuously proliferating lymphoblastoid cell line establishment. In addition, EBV can also infect GC B cells and rescue them from apoptosis (56). *In vivo*, immunohistochemistry studies revealed that the majority of EBV-infected cells are found within the interfollicular zone of lymph nodes, and only few EBVinfected cells are localized within GCs of healthy carriers of the virus, and infectious mononucleosis patients (53, 57).

Expression of typical GC B-cell markers CD38, CD77, and AID, as well as the activated/memory B-cell marker CD27 in naive B cells transformed *in vitro* by EBV suggest a GC B-cell commitment. Nevertheless, these expressions were lower than in *ex vivo* sorted GC B cells. In addition, EBV transformed GC B cells show a downregulation of CD77, BCL-6, and CD79b, compared to non-transformed GC B cells (58). It has been shown that LMP1 downregulated BCL-6 expression in B-cell lines *in vitro* (49), whereas BCL-6 was proposed to be upregulated *in vivo* in GC B cells in the presence of LMP1 (53). These discrepancies highlight the limitations of lymphoblastoid cell lines as a model of *in vivo* EBV biology.

Mutational status of EBV-infected circulating memory B cells showed that they exhibited similar levels of somatic hypermutations than non-infected memory B cells, evidencing a differentiation step within GCs (59, 60). Most EBV<sup>+</sup> GC B cells belonged to clones of cells harboring somatically mutated V gene rearrangements. Nevertheless, EBV<sup>+</sup> GC B cells proliferate without ongoing somatic hypermutations, in contrast with uninfected B cells in the same microenvironment during infectious mononucleosis (57).

B-cell recruitment and migratory capacity into SLOs may also be influenced by EBV infection. Gene expression analysis of *in vitro* EBV-infected tonsillar B cells demonstrated a CXCR5 downregulation while CCR9 and CXCR3 were upregulated (61). Noteworthy, CCR7, which is important for migration into SLOs, was also downregulated in these B cells. Immunohistochemistry reports showed that CD10<sup>+</sup> cells infected with EBV in tonsils of

Figure 2 | Impact of viruses on the GC reaction. This figure depicts several virus actions on cells involved in the humoral immune response. EBV, Epstein–Barr virus; HIV, human immunodeficiency virus; GC, germinal center; FRC, fibroblastic reticular cells; Tfr, T follicular regulatory cells; Tfh, T follicular helper cells; BnAbs, broadly neutralizing antibodies; Abs, antibodies. 1. HIV stimulates collagen deposition by FRCs, impairing naïve CD4+ T cells to access survival signals. 2. Pre-Tfh cells have been found permissive to HIV infection. 3. Tfr cell frequency is increased during HIV infection. 4. HIV persist in Tfh cells, Tfh cells accumulate but are not effective. 5. FDCs are a source of HIV infection for T cells without being infected. 6. Memory B-cell compartment is decreased in HIV patients because of defective Tfh cell help. 7. Impaired specific humoral immune response and hypergammaglobulinemia in HIV patients. (A) EBV infects naïve B cells. (B) EBV drives B-cell proliferation and the expression of differentiation markers. (C) EBV impairs GC entry. (D) EBV persists in memory B cells. \*FRC network destruction by several viruses inducing an SLO disorganization.

healthy carriers classically expressed CXCR4 and CXCR5 and displayed low CCR7 expression (53). Furthermore, EBI2, which was originally found as strongly upregulated in EBV-infected Burkitt's lymphoma B cells together with CCR7, plays a critical role in the regulation of B cells positioning within the GCs. EBI2 is a GPCR signaling through Gαi, like most chemokine receptors. Its expression is physiologically high in naive B cells, transiently upregulated following BCR stimulation and repressed by BCL-6 during GC B-cell differentiation. Indeed, low CCR7 and EBI2 expression levels are required to guide B cells into follicles where they differentiate into GC B cells (3). B cells *in vitro* infected with EBV show an EBI2 upregulation both in lytic and latent phases (62, 63). Consequently, EBI2 high expression in EBV-infected B cells might favor a decreased GC formation.

To illustrate the impact of EBV proteins during autoimmune diseases, we could focus on two mouse models based on LMP2A and LMP1 expression. Minamitani et al. demonstrated in mice with ectopic B-cell expression of the EBV related protein LMP2A an increased B-cell differentiation in short-lived plasmablasts correlated with an enhanced production of low-affinity antibodies. Moreover, they demonstrated that these plasmablasts are more prone to produce anti ds-DNA or anti-cardiolipin antibodies associated with a glomerulonephritis typical of autoimmune diseases like systemic lupus erythematosus (SLE) (64). Another transgenic mouse model (mCD40-LMP-1) with B cells expressing a chimeric molecule comprising CD40 extracellular and LMP-1 intracellular domains and further crossed with lupus prone mice (B6.Sle1) demonstrated an exacerbation of the disease (65). These mice spontaneously developed GCs, had enlarged lymphoid organs, higher titer of anti-histone antibodies and signs of kidney pathology resulting in an acceleration of the disease. The same mCD40-LMP-1 mice crossed with rheumatoid arthritis mouse model also demonstrated a pathology worsening (66). In human, although EBV and SLE have been correlated, EBV reactivation is likely following SLE flare. However, more longitudinal studies are required to decipher their tight association (67).

Altogether, these reports demonstrate that the impact of EBV on B-cell positioning, proliferation, and survival argues for a disturbed GC reaction and a modified humoral response.

### MODULATION OF HUMORAL IMMUNE RESPONSE BY VIRAL INFECTION OF GC NON-B CELLS

#### T Follicular Cell Infection

Recently, Tfh cells have been described as sources of replication of competent human immunodeficiency virus-1 (HIV-1), serving as a significant cellular reservoir (**Figure 2**). HIV is a lentivirus leading to acquired immunodeficiency syndrome characterized by opportunistic infection and subsequent patient death. HIV is a single-strand RNA virus, which is converted to double-strand DNA by reverse transcription after target cell entry through chemokine receptors. Then, viral genome is integrated to cell DNA through a virally encoded integrase. It has been demonstrated that HIV-1 replication is concentrated within CD4<sup>+</sup> T cells localized in lymph node follicles during the asymptomatic phase of the disease. In accordance, Tfh cells were demonstrated as highly permissive to HIV infection and, therefore, constitute an important virus reservoir (68). CXCR4 was considered as the major route for HIV entry in Tfh cells. Nevertheless, CCR5 expressed by pre-Tfh cells constitutes an additional mechanism of Tfh cell infection (69).

In contrast to the depletion of activated memory CD4<sup>+</sup> T cell associated with disease progression in untreated patients, total and HIV-specific Tfh cell accumulation is observed in chronic infection, associated with a skewing of the B-cell population toward GC B cells. Interestingly, Tfh cell frequency is positively correlated with GC B-cell proportion in chronically HIV-infected patients (70).

T follicular helper cell compartment is a heterogeneous population gathering cells harboring Th1, Th2, or Th17 phenotype depending on their cytokine secretion profile. In simian immunodeficiency virus chronic infection, Tfh cell polarization was demonstrated to be biased toward Th1 phenotype, with a majority of Tfh cells expressing CXCR3 and producing IFN-γ and IL-21 (71). Interestingly, CXCR3<sup>+</sup> Tfh cells comprised a subset of CCR5<sup>+</sup> cells and contained more copies of SIV DNA than their CXCR3 counterpart. Finally, CXCR3<sup>+</sup> Tfh cells could be in part responsible for Tfh cell accumulation and sustained GC reaction due to their excessive IFN-γ secretion (71, 72).

Despite germinal center T follicular helper cells and B-cell accumulation, HIV-1 infected individuals display impaired antigen-specific humoral immunity and poor vaccine response, associated with loss of memory B cell subsets correlated with disease progression (73–75). This impaired humoral immune response could be explained by inadequate B-cell help provided by Tfh cells. Indeed, Tfh and GC B-cell co-cultures resulted in an increased B-cell death and reduced IgG levels when cells were sorted from HIV-infected individuals (76). Impaired Tfh cell function in HIV-1 patients depends on the binding of PD-1 from Tfh cells to PD-L1, which is overexpressed on GC B cells. This interaction results in a decreased ICOS, IL-21, IL-4, and IL-10 expression by Tfh cells.

Noteworthy, in these studies, Tfr cells were not distinguished from Tfh cells and may, therefore, participate to the disturbed Tfh cell function. This question was approached in recent works studying more specifically Tfr cells that have been demonstrated to be increased in chronic HIV infection (77, 78). This expansion was related to an increase of immature DC proportion and IDO secretion favoring Tfr cell differentiation. Tfr cells from HIV-1 patients were shown to express more regulatory molecules, such as CTLA-4, LAG-3, GITR, IL-10, and TGF-β. Moreover, Tfr cell removal from follicular T-cell culture partially restored ICOS expression, IL-4, and IL-21 production, and cell proliferation of Tfh cells. Thus, Tfr cells are another contributor of inefficient GC response in HIV individuals (77).

Hypergammaglobulinemia is classically found in chronic HIV infection. It has been associated with Tfh cell amplification, as a correlation between BCL-6 in Tfh cells and total IgG and IgG1 antibody serum levels have been highlighted (70). Nevertheless, highly potent broadly neutralizing antibodies able to invalidate the majority of globally circulating HIV strains develop only in around 20% of HIV-1 infected patients (79). Broadly neutralizing antibodies against HIV display extensive somatic hypermutations (80), hypothesizing that adequate GC response does not occur in the majority of HIV-infected patients. Generation of such antibodies is one goal for HIV vaccine research. Impairment of antibody production is not restricted to anti-HIV antibodies but also concerns other T-cell-mediated antibody responses, as those elicited by exposure to neo-antigen or recall vaccination (81–83). Interestingly, highly active antiretroviral therapy is able to somehow restore adaptive immune response related to vaccination (84).

Concerning circulating Tfh cells, blood memory PD-1<sup>+</sup> CXCR5+ CD4+ cell frequency is correlated with the development of broadly neutralizing antibodies against HIV (85). In addition, their absolute number is increased in patients with lower viral load and spontaneously controlling disease progression, compared to patients with high uncontrolled viremia (86). Furthermore, higher frequency of circulating Tfh cells has been associated with protective antibody responses in vaccine trials (87). Collectively, these reports hypothesize that circulating Tfh cell frequency might be used as a blood biomarker of good prognosis in HIV-infected patients. Nevertheless, additional studies using a common circulating Tfh phenotype and including a high number of patients should now be conducted.

Altogether, these studies demonstrate an impairment of the adaptive immune response in HIV seropositive patients impacting individual immune protection.

#### SLO Stromal Cells

Lymphoid stromal cells are strongly affected by viral infections and contribute both to the initiation and amplification of antiviral immune response within GCs and to virus-related immunosuppression. FDCs, the lymphoid stromal cell subset involved in the recruitment, selection, and survival of GC B cells, are now considered as a reservoir for HIV. Complement-opsonized HIV is internalized by human FDCs and retained for extensive periods in a non-degradative cycling compartment. In particular, HIV<sup>+</sup> FDCs could be detected in patients under antiretroviral therapy (ART). Infectious virus could then be transmitted to uninfected CD4<sup>+</sup> T cells, in particular Tfh cells, a process that could be involved in ART escape (88). Importantly, HIV pro-viral DNA is never found within FDC genome indicating that FDCs are a source of infection for T cells but are not infected by HIV. Conversely, FDCs could be infected and killed by arboviruses; thus, hampering the capacity of GCs to produce protective antibodies and inducing a transient generalized immunosuppression (89). How this mechanism, recently demonstrated for bluetongue virus in sheep, could be involved in the pathogenesis of hemorrhagic fevers caused by arboviruses in human remains to be elucidated.

Fibroblastic reticular cells are the lymphoid stromal cell subset involved in immune cell recruitment, motility, interaction, and homeostasis within lymph nodes of which they regulate the size and microenvironmental structure (13). Different viral pathogens induce distinct patterns of FRC expansion and activation, yet all induce sustained remodeling that alters responses induced by a subsequent infection (90). Importantly, FRCs are a direct target of several viruses. Ebola, Lassa, and Marburg viruses (91, 92), but also lymphocytic choriomeningitis virus (93) infect and destroy FRC network leading to lymph node disorganization and crippling the immune response to new antigens. Moreover, in chronic infection, in particular HIV, stimulation of the TGF-β receptor on FRCs by TGF-β produced by expanded Tregs leads to overproduction and deposition of collagen, causing lymph node fibrosis and restricting access of naive lymphocytes to FRCdependent survival signals. FRC dysfunction is now considered as a major cause of HIV-related CD4<sup>+</sup> T-cell depletion that in turn deprives FRCs of the lymphotoxin-β receptor signals required for their maintenance and leads to a broad immunosuppression (94). Interestingly, restoration of the FRC network and reconstitution of naive T-cell populations are only optimal when therapy is initiated in the early/acute stage of infection.

To date, very few reports decipher direct and indirect impact of viruses on FRC and FDC biology. Additional studies focusing on stromal cell and virus interactions might improve our knowledge on viral pathogenesis.

#### VIRUS INVOLVEMENT IN DEVELOPMENT AND MAINTENANCE OF GC-DERIVED LYMPHOMAS

Lymphomas constitute a large group of cancer arising from lymphoid or extra-nodal tissues. The nomenclature of these neoplasms regularly evolves, and currently comprises more than 50 distinct clinical, pathological, genetic, and molecular entities (95). For a subgroup of B-cell lymphomas observed in immunecompetent patients and including diffuse large B-cell lymphoma (DLBCL), follicular lymphoma, classical Hodgkin lymphoma (HL), and BL, the cell of origin is located in the GC, where B cells can be infected by EBV. Virus-associated lymphomas also occur in immune-deficient patients. In this case, additionally to the previously cited entities, other lymphomas arising from GC B cells are observed, including the post-transplantation lymphoproliferative disease (PTLD) and the primary central nervous system lymphoma (PCNSL), frequently associated with HIV infection (96). These virus-associated lymphomas were recently reviewed (97, 98). Additional virus-associated lymphomas with a cell of origin outside the GC (e.g., primary effusion lymphoma) are described. The three types of EBV latency previously described are associated with various B-cell lymphomas. Type I latency is associated with BL, whereas type II is found in HL, and finally type III is found in PTLD (99).

Burkitt's lymphoma is an aggressive B-cell lymphoma categorized in endemic, sporadic, and immunodeficiency variants. Translocation of the proto-oncogene *c-MYC* to the Ig heavy or light chain region is the hallmark of this lymphoma. The endemic variant is observed in equatorial Africa where it represents the most common malignancy of childhood, whereas the sporadic variant is seen worldwide with a median age of 30 years. Sporadic BL cases account for 1–2% of all lymphomas in Western Europe and USA. EBV is detected in almost all cases of endemic variant contrasting with the 30% of incidence in the sporadic cases and 25% in immune-deficient patients. EBV is not associated with all BL cases, suggesting that the virus is not directly involved in the pathogenesis but rather acts as a cofactor. However, the role of EBV is not clearly understood. In endemic and immunodeficiency BL, a long-term antigenic stimulation by bacteria, virus, or parasite (in particular malaria) precedes the lymphoma and may induce an exhaustion of T cells and, thus, a defect in response to infected B cells (100, 101). In addition, chronic antigen stimulation in the GC might enhance the likelihood of c-MYC Ig rearrangement (98).

Hodgkin lymphoma is defined by an expansion of Reed– Sternberg cells within a reactive microenvironment. There are classified as nodular lymphocyte predominant HL (NLPHD) and classical HL (cHL). Within cHL, four subtypes are recognized. These subgroups differ clinically and also in frequency of EBV infection ranging from 10 to 40%. Noteworthy, EBV is rarely associated with NLPHD. LMP2 is involved in rescuing Reed– Sternberg cells, which commonly present an aberrant BCR, from apoptosis. LMP1 also participates in the constitutive activation of signaling pathways (e.g., NF-kB, JAK/STAT) in Reed–Sternberg cells (101, 102).

Diffuse large B-cell lymphoma is the most common non-HL, accounting for around 40% of new cases. Although the World Health Organization recognizes DLBCL as a single entity, several subgroups with different outcomes have been described (96). All subtypes are associated with various frequencies of EBV although high rates are associated with immune deficiency lymphomas. Recently, a subtype called "EBV<sup>+</sup> DLBCL not otherwise specified (NOS)" has been recognized in immune-competent patients Rodriguez et al. Virus Impact on GCs

(95). They have a worse prognosis than their EBV counterparts. The role of EBV is not clearly understood in this disease where LMP-1 is detected in almost all cases, but EBNA2 only in a minority of patients (98). PCNSL is an aggressive DLBCL occurring in immune-deficient patients, mainly in HIV infection, and is associated with EBV in all cases. LMP1 and EBNA2 are expressed in tumor cells (97).

Post-transplantation lymphoproliferative disease is a heterogeneous group of lymphomas occurring after allogeneic transplantation of solid organ or hematopoietic stem cell graft. A B-cell phenotype is observed in 85% of cases. In 65% of PTLD, a reactivation of EBV occurs because anti-EBV immunity is impaired by the immunosuppressive therapy following the transplantation (103).

Epstein–Barr virus can also be associated with T-cell and NK-cell lymphomas. Among them, angioimmunoblastic T-cell lymphoma is a rare malignancy primarily involving lymph nodes, and characterized by tumor T cells expressing CXCL13, CD10, PD-1, and BCL-6, such as Tfh cells, their normal counterpart (20, 104). EBV involvement in lymphoma development and maintenance is unclear. Nevertheless, EBV is carried by lymph node B cells and in rare tumor and infiltrating non-tumor T cells (105).

Finally, viruses can also act on the tumor microenvironment (99). As reported previously, HIV infection induces an impairment of the adaptive immune response. In addition, EBVencoded proteins have been demonstrated to increase expression and release of cytokines, such as IL-6 or IL-10, that may impact tumor B-cell growth (106).

Epstein–Barr virus proteins can also induce enhanced expression of vascular endothelial growth factor, IL-8, or hypoxiainducible factor-1α, all contributing to angiogenesis, which is an


important mechanism of tumor growth (107, 108). Moreover, exosomes derived from EBV-infected cells may also contribute to tumor growth by apoptosis induction of CD4<sup>+</sup> EBV-specific T cells, Th1, Th17, and T CD8<sup>+</sup> cells, as well as expansion of Treg cells (99, 109). Interestingly, EBV triggers high PD-L1/CD274 expression on malignant cells in both DLBCL, cHL without amplification of the chromosomal region 9p24.1 that contains the genes encoding both PD-L1 and PD-L2, and PTLD (110). PD-L1 upregulation has been associated with exhaustion of tumor infiltrating T cells and tumor escape suggesting that EBV infection induces PD-L1 expression on lymphocytes in order to promote a tolerogenic immune state. In conclusion, EBV is associated with various B-cell and T-cell lymphomas arising from GC in immune-competent or immune-deficient hosts. So far the pathogenesis is not clearly understood and several mechanisms may be involved. EBV can trigger the tumor cell directly such as interfering with signaling pathways or promoting amplification of oncogenes, or acts indirectly by impairing antitumor response (summarized in **Table 1**).

## CONCLUSION

Like during other maturation stages, B and T cells in GCs can be targeted by virus infection. Accumulating reports demonstrate that GC B-cell and T-cell infections disturb specific adaptive immune responses and vaccination efficiency, and can worsen a pre-established autoimmune disease, or participate to cancer development and maintenance. In addition, recent pieces of evidence elucidate that viruses can also infect GC stromal cells, causing a stroma network disorganization leading to disturbed humoral immunity. Altogether, this underlines that virus infection of all GC cell-components must be take into account to understand impact of viruses on the T-cell dependent humoral immune response. Moreover, a better characterization of the functional consequences of chronic viral infection on various GC cell subsets paves the way for the design of new efficient therapeutic strategies in different pathological context. In particular, PD-L1hi EBV+ HL and DLBCL tumor cells may be suitable therapeutic targets for anti-PD-1/PD-L1 immunotherapy with the aim to unleash host antitumor immune responses. Moreover, the recently identified roles of Tfh cells, FDCs, and FRCs as HIV-1 reservoirs or targets, potentially involved in early and late relapse to ART, could provide new prognosis biomarkers and therapeutic targets, including the use of antifibrotic molecules to revert HIV-1-induced damage to lymphoid stromal cell niches, the purging of the FDC reservoir, or the targeting of infected Tfh cells (111).

## AUTHOR CONTRIBUTIONS

SR, MR, KT, and PA are substantial contributors to the conception of this work.

## ACKNOWLEDGMENTS

SR has been funded by research grants from the Institut National du Cancer (INCa\_6530) and la Région Bretagne SAD 2015.

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Rodriguez, Roussel, Tarte and Amé-Thomas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Herpesviral microRNAs in Cellular Metabolism and Immune Responses

Hyoji Kim, Hisashi Iizasa, Yuichi Kanehiro, Sintayehu Fekadu and Hironori Yoshiyama\*

*Department of Microbiology, Faculty of Medicine, Shimane University, Shimane, Japan*

The microRNAs (miRNAs) function as a key regulator in many biological processes through post-transcriptional suppression of messenger RNAs. Recent advancements have revealed that miRNAs are involved in many biological functions of cells. Not only host cells, but also some viruses encode miRNAs in their genomes. Viral miRNAs regulate cell proliferation, differentiation, apoptosis, and the cell cycle to establish infection and produce viral progeny. Particularly, miRNAs encoded by herpes virus families play integral roles in persistent viral infection either by regulation of metabolic processes or the immune response of host cells. The life-long persistent infection of gamma herpes virus subfamilies, such as Epstein-Barr virus and Kaposi's sarcoma-associated herpesvirus, induces host cells to malignant transformation. The unbalanced metabolic processes and evasion from host immune surveillance by viral miRNAs are induced either by direct targeting of key proteins or indirect regulation of multiple signaling pathways. We provide an overview of the pathogenic roles of viral miRNAs in cellular metabolism and immune responses during herpesvirus infection.

#### Edited by:

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### Reviewed by:

*Keiji Ueda, Osaka University, Japan Laurie Krug, Stony Brook University, United States*

> \*Correspondence: *Hironori Yoshiyama yosiyama@med.shimane-u.ac.jp*

#### Specialty section:

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology*

Received: *18 April 2017* Accepted: *29 June 2017* Published: *18 July 2017*

#### Citation:

*Kim H, Iizasa H, Kanehiro Y, Fekadu S and Yoshiyama H (2017) Herpesviral microRNAs in Cellular Metabolism and Immune Responses. Front. Microbiol. 8:1318. doi: 10.3389/fmicb.2017.01318* Keywords: microRNA, herpesvirus, oncogenesis, immune evasion, cell metabolism

## INTRODUCTION

The microRNAs (miRNAs) are small non-coding RNAs consisting of 19–23 nucleotides. The primary miRNA in the nucleus is cleaved into smaller pre-miRNA consisting of around 70 nucleotides with a hairpin structure. The pre-miRNA is then exported to the cytoplasm, where it is cleaved by Dicer to form mature miRNA. The miRNA is incorporated into the RNA-induced silencing complex (RISC), which contains the essential endonuclease Argonaute 2 (Ago2). The miRNA-RISC interacts with the 3′ untranslated region (UTR) in mRNA. This complex suppresses target gene expression through the translational repression or induction of mRNA deadenylation (Winter et al., 2009). Cellular miRNAs play important parts in the regulation of cellular pathways, of which dysregulation has been linked to many disorders including cancer. Viral miRNAs were reported originally by Pfeffer et al. (2004), and now many DNA viruses are known to contain miRNAs in their genomes. More than 200 viral miRNAs are currently identified mainly in the herpesvirus family (**Table 1**).

The first report of viral miRNA was described in the herpesvirus family, especially in Epstein-Barr virus (EBV) strain B95-8 having a 12-kb deletion in BamHI-A rightward transcripts (BART) region (Pfeffer et al., 2004). Subsequent studies revealed that EBV encodes 44 different mature BART miRNAs and 4 mature BamHI-H rightward open reading frame 1 (BHRF1) miRNAs (Cai et al., 2006; Grundhoff et al., 2006; Zhu et al., 2009) (**Figure 1**). Rhesus lymphocryptovirus encodes miRNAs, which are orthologs of EBV miRNAs and evolutionarily conserved (Cai et al., 2006). Kaposi's sarcoma-associated herpesvirus (KSHV), the etiologic agent of Kaposi's sarcoma, encodes TABLE 1 | The role of herpes virus-encoded miRNAs in immune evasion and cancer metabolism.


*(Continued)*


*(Continued)*


*White or gray backgrounds relate to immune evasion, and black backgrounds relate to cancer metabolism. KSHV, Kaposi's sarcoma-associated herpesvirus; HCMV, human cytomegalovirus; HSV, herpes simplex virus.*

(LMP1) and EBNA2 transcribed in antisense and sense orientations, respectively. EBNA-LP is transcribed from variable numbers of repetitive exons (denoted by black lines between EBNA2 and oriP). EBNA3s and EBNA1 shown by black arrows locate between two oriLyt regions. The highly transcribed non-polyadenylated EBER RNAs are represented in the right top of the diagram as black arrows. TR (represented by the white square) denotes the terminal repeats of EBV DNA. LMP2 locating between EBERs and TR is transcribed in sense. The genomic location of miRNAs encoded by EBV is enlarged and represented linearly. The regions of the BHRF cluster and BART clusters 1 and 2 are expanded to show individual miRNAs. Each miRNA is processed from pre-miRNA precursors and transported from nucleus to cytoplasm. Dicer is a part of the RNase III enzyme and cleaves pre-miRNA into short double-stranded RNA fragments called miRNA. The complex of RNA-induced silencing complex (RISC), Argonaute 2 (Ago2), and miRNA specifically inhibits transcription of the target mRNA.

13 precursor (pre)-miRNAs. These KSHV miRNAs are located in the KSHV latency-associated region (Pfeffer et al., 2005; Grundhoff et al., 2006). By sharing target genes, these KSHV miRNAs function as analogs of cellular oncomirs (Skalsky et al., 2007; Gottwein et al., 2011; Manzano et al., 2013).

This mini-review summarizes the role of viral miRNAs in cellular metabolism and the immune system of hosts to understand the pathologies of viral infection. Although viral miRNAs have also been detected in JC virus and human immunodeficiency virus, we mainly discuss miRNAs from Herpesviridae.

## PATHOLOGICAL ROLES OF EBV MIRNAS

#### EBV-Encoded miRNAs

EBV is associated with many tumors, including lymphoma and epithelial carcinomas (Young and Murray, 2003; Luo et al., 2005). Detection of EBV-encoded small non-coding RNAs (EBERs) by in situ hybridization is used as a diagnostic hallmark of EBV infection in tumor cells. EBERs can be detected in a variety of tumors, such as nasal NK/T-cell lymphoma, post-transplant lymphoma, Burkitt's lymphoma, Hodgkin's disease, diffuse large B-cell lymphoma, nasopharyngeal carcinoma (NPC), and gastric carcinoma (Delecluse et al., 2007). Following primary infection, EBV establishes latent infection. Three latency types (I, II, III) are defined depending on the pattern of expression in viral genes (Middeldorp et al., 2003). BART miRNAs are expressed in all latency types, whereas BHRF1 miRNAs are expressed only in type III latency. BHRF1 transcripts encode four mature BHRF1 miRNAs. BART transcripts have two clusters, cluster 1 and cluster 2, which generate 44 mature BART miRNAs (**Figure 1**) (Pfeffer et al., 2004; Cai et al., 2006; Zhu et al., 2009). EBV B95-8 strain has a 12-kb deletion and lacks most of the BART miRNAs (Baer et al., 1984).

BART miRNAs are expressed more strongly in EBV-associated epithelial cells than in B lymphocytes (Chen et al., 2010). BART transcripts have two TATA-less promoter regions, designated P1 and P2 (Sadler and Raab-Traub, 1995). Although P1 supports substantial activity in epithelial cells and B lymphocytes, P2 exhibits strong activity only in epithelial cells. Several transcription factors, important for cellular metabolism and immunity, regulate BART promoter activity (Chen et al., 2005). P1 activity is negatively regulated by interferon regulatory factor 5 (IRF5) and IRF7, and P2 activity is positively regulated by c-Myc and CCAAT-enhancer-binding protein (C/EBP) family members. Relative expression between positive and negative transcription factors in EBV-infected cells probably controls the expression level of BART miRNAs. BHRF1 miRNAs are generated as part of the Cp- and/or Wp-initiated EBNA transcripts in cells showing latency III infection (Amoroso et al., 2011). The expression of miR-BHRF1-1 absolutely depends on Cp/Wp activity, because it locates at the 5′ UTR of BHRF1 mRNA and overlaps with the EBV replication-activated BHRF1 promoter (Amoroso et al., 2011). However, because miR-BHRF1- 2 and miR-BHRF1-3 locate at the 3′ UTR, they are strongly expressed by an alternative promoter for lytic BHRF1 transcript during lytic replication (Kelly et al., 2009; Xing and Kieff, 2011). The expression of BART miRNAs inversely correlates with the methylation status of the promoter in EBV-infected B lymphocytes. Promoter methylation may be important in regulating the expression of BART and BHRF1 miRNAs (Kim do et al., 2011).

EBV miRNAs are transferred to adjacent cells via exosomes (Pegtel et al., 2010). The secreted viral miRNAs modify the expression of target genes in recipient cells (Haneklaus et al., 2012). Some BART miRNAs show distinctive expression between cells and exosomes. MiR-BART7 is expressed more abundantly in exosomes than in EBV-positive NPC cells (Meckes et al., 2010). However, miR-BART8-5p is expressed less abundantly in exosomes than in lymphoblastoid cell lines (Hoshina et al., 2016). These findings indicate that some EBV miRNAs are selectively packaged and transported into recipient cells via exosomes.

## Immune Evasion by EBV miRNAs

EBV can establish either latent infection or lytic replication (Kenney and Mertz, 2014). In tumor cells, EBV usually maintains latent infection rather than entering into lytic replication (Cohen, 2000). During latent infection, EBV transcribes viral miRNAs to escape from the host immune system by targeting both cellular and viral genes. Lowered expression of viral proteins enables infected cells to escape antigenic recognition by the host immune system. Expression of latent membrane protein 1 (LMP1) and LMP2A is downregulated by BART cluster 1 miRNAs (miR-BART16, 17-5p, and 1-5p) (Lo et al., 2007) and miR-BART22 (Lung et al., 2009), respectively. LMP1 and LMP2A are oncogenic viral proteins that promote EBV-positive malignancies by engaging a number of signal pathways, such as the NFκB, JNK/p38-SAPK, PI3K/Akt, ERK-MAPK, and JAK/STAT pathways, followed by subsequent induction of morphological and phenotypic alterations (Young and Rickinson, 2004). LMP1 and LMP2A alter the host immune system by cooperating with environmental and host genetic factors (Dawson et al., 2012).

Viral miRNAs help maintain latent infection by expressing limited numbers of viral genes that allow EBV to evade host immune surveillance. MiR-BART20-5p suppresses lytic replication by targeting EBV immediate-early genes BZLF1 and BRLF1 (Jung et al., 2014), key regulators of the expression of EBV proteins and the production of progeny virus (Pattle and Farrell, 2006). Downregulation of Dicer by miR-BART6- 5p reduces the expression of BZLF1, BRLF1, and both EBNA2 and LMP1 latent proteins in C666-1, an EBV-positive NPC cell line (Iizasa et al., 2010). Likewise, miR-BART2-5p blocks lytic replication by inhibiting expression of viral DNA polymerase BALF5 (Barth et al., 2008).

EBV miRNAs can also block immune response by reducing cytokines, chemokines, and T-cell stimulatory molecules. MiR-BHRF1-3 modulates host interferon (IFN) response by targeting CXCL11, an IFN-inducible T-cell-attracting chemokine (Xia et al., 2008). Suppression of MHC class I-related chain B (MICB) by miR-BART2-5p protects EBV-infected cells from attack by NK cells and T cells. MICB is also a target gene for other herpesvirus miRNAs, including KSHV-encoded miR-K12-1 and human cytomegalovirus (HCMV)-encoded miR-UL112 (Nachmani et al., 2009). Importin 7 (IPO7), a receptor for importing transcription factors into the nucleus, has important function for innate immunity, because loss of IPO7 in macrophages inhibits interleukin (IL)-6 secretion (Yang et al., 2009). IPO7 is a putative target for miR-BART1-3p and miR-BART3 (Dolken et al., 2010). MiR-BART15 also reduces IL-1β production from the inflammasome by targeting NLRP3 (NLR family, pyrin domain containing 3; also known as cryopyrin) at the same 3′ UTR region that host miR-223 recognizes (Haneklaus et al., 2012). These findings suggest that EBV miRNAs inhibit immune response by multiple mechanisms.

#### EBV miRNAs in Cancer Metabolism

Compared with normal cells, cancer cells increase metabolic autonomy, which uses nutrients and promotes metabolic processes of the host cells to support proliferation. Recent studies showed that several EBV miRNAs regulate cellular metabolic processes. MiR-BART1 is one of the metabolic regulators strongly expressed in NPC. The expression levels of phosphoglycerate dehydrogenase (PHGDH) and EH domaincontaining protein 1 (EHD1) were significantly upregulated by miR-BART1 expression in an NPC cell line, CNE1 (Ye et al., 2013). PHGDH is important for the synthesis of serine and glycine, a central metabolite for a variety of biosynthetic pathways, by removing 3-phosphoglycerate during glycolysis (DeBerardinis, 2011). EHD1 regulates endosomal transport of plasma membrane proteins such as transferrin receptor and ß1 integrin (Jovic et al., 2007).

Wan et al. reported that 9 EBV miRNAs (miR-BART1-5p, 3, 4, 5, 6, 7, 8, 10, and 18-3p) are highly expressed in NPC tissues. Genome pathway analysis indicated that upregulated EBV miRNAs mainly target transforming growth factor β (TGF-β) and Wnt signaling pathways (Wan et al., 2015), which are involved in many biological processes during oncogenesis, including reprogramming of tumor cell bioenergetics (Sherwood, 2015) and the bioenergetic shift toward catabolism (Guido et al., 2012). Previous studies showed that both pathways are dysregulated in NPC (Zeng et al., 2007; Chen et al., 2009). Consistent with this result, the TGF-β signaling pathway is suppressed by upregulation of miR-BHRF1, because miR-BHRF1 targets the small ubiquitin-like modifier-regulated component SMAD3 and the transcription co-regulators JUN and FOS (Callegari et al., 2014). Recently, target genes for EBV miRNAs were identified by high-throughput sequencing of RNA isolated by the methods of crosslinking immunoprecipitation (HITS-CLIP) and photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP). These EBV miRNA targets include many components of the Wnt signaling pathway (Riley et al., 2012; Skalsky et al., 2012).

The PI3K/AKT/mTOR pathway is a critical regulator in cell survival, growth, protein synthesis, and glucose metabolism (Yap et al., 2008). Moreover, the AKT pathway promotes the expression of genes involved in glycolysis and lipid genesis (Wullschleger et al., 2006). In NPC cells, miR-BART1 significantly reduces phosphatase and tensin homolog (PTEN) expression while increasing the phosphorylation level of pAKT, pFAK, p130Cas, pShc, and pERK1/2 (Cai et al., 2015). PTEN inhibits the PI3K/AKT pathway by dephosphorylating PIP3 and increasing PIP2, resulting in a reduction of membrane recruitment of AKTs (Rafalski and Brunet, 2011). Thus, miR-BART1 activates migration, invasion, and metastasis of NPC cells via suppression of PTEN (Cai et al., 2015).

MiR-BART7 is expressed strongly in NPC cells and promotes cell proliferation, migration, and invasion. Pathway analysis indicates that the expression level of various genes is altered by the expression of miR-BART7. These target genes belong to the signaling pathway of calcium and the immune system, ionotropic glutamate receptor, ATP-binding cassette transporters, nuclear receptors in lipid metabolism and toxicity, the TGF-ß-signaling pathway, and metabolism of lipids and lipoproteins (Chan et al., 2012). Therefore, the aberrant expression of EBV miRNAs may contribute to metabolic abnormality and oncogenesis in EBVinfected cells by unbalancing various signaling pathways.

## MIRNAS IN OTHER HERPESVIRUSES

## Herpesvirus-Encoded miRNAs

Herpesviruses other than EBV and KSHV also encode miRNAs. Herpes simplex virus (HSV) has two serotypes, HSV-1 and HSV-2, which infect oral or genital mucosa. The latencyassociated transcript functioning as a primary miRNA precursor is exclusively expressed during latent infection (Wagner et al., 1988; Umbach et al., 2008). Since the first report in 2006, 27 mature miRNA sequences have been identified in the HSV-1 genome (Cui et al., 2006; Jurak et al., 2010). Similarly, HSV-2 encodes 24 mature miRNAs (Umbach et al., 2010). Several miRNAs are conserved between HSV-1 and HSV-2, especially in their seed regions. These viral miRNAs have analogous functions in immune evasion and virus propagation (Jurak et al., 2010; Umbach et al., 2010).

Unlike other herpesviruses, HCMV miRNAs are not clustered in latent transcripts, but are distributed throughout the viral genome (Buck et al., 2007). Currently, 26 mature HCMV miRNA sequences are uploaded on miRBase (http://www.mirbase.org). HCMV miRNAs target multiple genes related to immune response, cell cycle control, and vesicle trafficking (Hook et al., 2014).

Varicella-zoster virus (VZV) is a pathogenic human virus that causes chicken pox and shingles. Unlike other herpesviruses examined, although many small RNA sequencing studies have been performed, VZV miRNAs have not been identified yet (Umbach et al., 2009).

Human herpesvirus 6 (HHV-6), with its two variants HHV-6A and HHV-6B, is a ubiquitous pathogen in general human populations. Both are very closely related, with nearly 90% homology at the genomic level. Deep sequencing of small RNA species identified a small non-coding RNA with the characteristics of a viral miRNA from cells harboring HHV-6A. Growth analyses of mutant viruses revealed that miR-U86 directly impacts lytic replication by targeting HHV-6A immediate-early gene U86 (Nukui et al., 2015). HHV-6B encodes four pre-miRNAs expressed from direct repeat regions located at either side of the genome. HHV-6B miR-Ro6-2 is a seed ortholog of host miR-582-5p, which targets SMAD3 to downregulate TGFβ. HHV-6B miRNAs also have the potential to regulate viral replication (Tuddenham et al., 2012).

HHV-7 is a ubiquitous T-lymphotropic virus infecting most humans. As with VZV, HHV-7-encoded miRNAs have not yet been identified (Louten et al., 2015).

#### Herpes Viral miRNAs in Immune Evasion

EBV miR-BART2-5p and KSHV miR-K12-7 target the 3′ UTR of MICB at different locations (Nachmani et al., 2009). HCMV miR-UL112 and cellular miR-376a synergistically downregulate MICB expression and subsequently help the virus evade innate immune recognition (Nachmani et al., 2010). Recognition of HCMV-infected cells by cytotoxic T lymphocytes is impaired due to the reduced expression of aminopeptidase ERAP1 by HCMV miR-US4-1 (Kim et al., 2011). In human fibroblast cells, HCMV clinical strain-specific miR-UL148D was shown to block the human chemokine RANTES, which attracts immune cells during inflammation and the immune response (Kim et al., 2012).

HSV-1 miR-H8 targets the glycosylphosphatidylinositol gene, which results in reduced expression of several immunemodulating proteins, viral expansion, and viral evasion from natural killer cell elimination (Enk et al., 2016). HSV miRNAs also target viral genes to maintain latency and suppress immune function. HSV-1 miR-H6 and miR-H2 reduce infected cell polypeptide 4 (ICP4) and ICP0, respectively (Umbach et al., 2008; Duan et al., 2012). Both miR-H3 and miR-H4 target ICP34.5 mRNA (Umbach et al., 2008). HSV-2 miRNAs also contribute to latency and immune evasion similarly to HSV-1 miRNAs because of the close homology of these two viruses.

In KSHV-infected primary effusion lymphoma cells, KSHV miR-K1 inhibits viral lytic replication by targeting the 3′ UTR of IκBα protein, an inhibitor of the NFκB complexes. Enhanced NFκB activity evades host immune system and promotes cell survival (Lei et al., 2010). The KSHV miRNA cluster also represses a network of targets associated with STAT3 and suppresses STAT3 activation upon IL-6 treatment. KSHV miR-K6-5 targets the 3′ UTR of PKCδ, a Ser/Thr kinase that phosphorylates and activates STAT3. KSHV miR-K921 also targets a second Ser/Tr kinase, IRAK1. Repression of BIRC5, a transcriptional target of STAT3, by KSHV miR-K12-5 promotes KSHV infection. These multiple KSHV miRNAs that repress STAT3 can weaken the innate immune responses to type-I interferons and inhibit the induction of antiviral genes, such as IRF1, IFITM1, and ISG15 (Ramalingam and Ziegelbauer, 2017).

#### Herpes Viral miRNAs in Cancer Metabolism

Several core cellular metabolic pathways are significantly altered by herpesvirus infection and the expression of viral miRNAs (Sanchez and Lagunoff, 2015). HSV-1 miR-H4-5p directly targets cyclin-dependent kinase inhibitor 2A (p16) mRNA in neuroblastoma cell lines. Suppression of miR-H4-5p inhibits cell proliferation, invasion, and progression of the cell cycle via the p16-mediated PI3K-AKT signaling pathway (Zhao et al., 2015). Likewise, HCMV miR-UL112-3p modulates the TLR/IRAK1/NFκB signaling pathway by targeting Toll-like receptor 2 mRNA (Landais et al., 2015). NFκB signaling is known to upregulate the expression of Glut3 in p53-deficient cells (Kawauchi et al., 2008), suggesting that inhibition of NFκB signaling by miR-UL112-3p and miR-US5-1 might suppress aerobic glycolysis (Hancock et al., 2017).

KSHV has been shown to alter host cell energy metabolism by concurrent regulation of two independent pathways. First, KSHV miRNAs stabilize and activate transcription factor HIF1α, a master regulator of cell metabolism, by targeting hypoxia-inducible factor prolyl hydroxylase, EGLN2. Second, downregulation of the mitochondrial heat shock protein A9 (HSPA9) by KSHV miRNAs reduces mitochondrial copy numbers and enhances anaerobic glycolysis (Warburg effect). Downregulation of EGLN2 and HSPA9 allows cell proliferation in a low oxygen condition (Yogev et al., 2014). KSHV miR-K12-11 and miR-K12-3 prevent lytic reactivation by reducing the expression of cellular transcription factors MYB, C/EBPα, and Ets-1, which are reported as activators of the RTA promoter (Plaisance-Bonstaff et al., 2014).

Murine gammaherpesvirus 68 (MHV-68), a natural pathogen of wild rodents, encodes for 14 pre-miRNAs. All MHV-68 miRNAs are located downstream of viral tRNA-like elements and transcribed by RNA polymerase III. Recent research showed that an MHV-68 mutant lacking the expression of all miRNAs results in a higher viral genomic load in the spleen. This report shows that MHV-68 miRNAs contribute to the maintenance of latency in vivo (Steer et al., 2016).

The generation and analysis of mutant viruses revealed that MHV-68 miRNAs are dispensable for short-term virus replication but are important for the establishment of lifelong infection in memory B cells. Furthermore, a lack of miRNA expression results in the complete attenuation of lethal disease in a virus-induced pneumonia model, demonstrating a key role for the viral miRNAs in pathogenesis (Feldman et al., 2014).

Similar to EBERs in EBV, MHV-68 encodes non-coding RNAs called TMERs (tRNA-miRNA-encoded RNAs), which are highly expressed in latently infected cells. TMERs harbor a predicted tRNA-like element and two downstream pre-miRNA hairpins and are processed by tRNase Z instead of Drosha, similar to cellular non-coding tRNAs, to generate mature miRNAs. Analysis of individual TMER mutant viruses has shown TMER4 to be a key mediator of virus dissemination. Interestingly, TMER4 miRNA seed sequence mutants do not compromise TMER4 function. These results demonstrate a crucial miRNAindependent function of TMER4 in hematogenous dissemination and the establishment of peripheral latency (Feldman et al., 2016).

#### SUMMARY AND CONCLUSIONS

Viral miRNAs play important roles in cancer development and progression by modulating immune response and metabolic circuits. Although further study is necessary to understand the pathogenic significance of viral miRNAs, viral miRNAs can be applied for the diagnosis of cancer, identification of drug targets, and therapeutic use.

## AUTHOR CONTRIBUTIONS

HK wrote the manuscript. SF and YK assisted in creating the table. HI and HY edited the paper and contributed financial assistance.

#### FUNDING

This study was supported by KAKENHI (Grant-in-Aid for Scientific Research) from the Ministry of Education, Culture, Sports, Science, and Technology (HI: 26460465, and HY: 16H05843), and a Health Labor Sciences Research Grant from the Ministry of Health Labor and Welfare, Japan (17fk0310105h0001).

#### ACKNOWLEDGMENTS

The authors thank Ms. Sayuri Hamada for her support in the preparation of the figure.

#### REFERENCES


human chemokine RANTES during infection. PLoS Pathog. 8:e1002577. doi: 10.1371/journal.ppat.1002577


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Kim, Iizasa, Kanehiro, Fekadu and Yoshiyama. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# HHV-6A/6B Infection of NK Cells Modulates the Expression of miRNAs and Transcription Factors Potentially Associated to Impaired NK Activity

Roberta Rizzo, Irene Soffritti, Maria D'Accolti, Daria Bortolotti, Dario Di Luca and Elisabetta Caselli\*

Section of Microbiology and Medical Genetics, Department of Medical Sciences, University of Ferrara, Ferrara, Italy

Natural killer (NK) cells have a critical role in controlling virus infections, and viruses have evolved several mechanisms to escape NK cell functions. In particular, Human herpesvirus 6 (HHV-6) is associated with diseases characterized by immune dysregulation and has been reported to infect NK cells. We recently found that HHV-6 in vitro infection of human thyroid follicular epithelial cells and T-lymphocytes modulates several miRNAs associated with alterations in immune response. Since miRNAs are key regulators of many immune pathways, including NK cell functions, we aimed to study the impact of HHV-6A and -6B in vitro infection on the intracellular mediators correlated to NK cell function. To this purpose, a human NK cell line (NK-92) was infected in vitro with HHV-6A or 6B and analyzed for alterations in the expression of miRNAs and transcription factors. The results showed that both viruses establish lytic replication in NK-92 cells, as shown by the presence of viral DNA, expression of lytic transcripts and antigens, and by the induction of an evident cytopathic effect. Notably, both viruses, although with species-specific differences, induced significant modifications in miRNA expression of miRNAs known for their role in NK cell development, maturation and effector functions (miR-146, miR-155, miR-181, miR-223), and on at least 13 miRNAs with recognized role in inflammation and autoimmunity. Also the expression of transcription factors was significantly modified by HHV-6A/6B infection, with an early increase of ATF3, JUN and FOXA2 by both species, whereas HHV-6A specifically induced a 15-fold decrease of POU2AF1, and HHV-6B an increase of FOXO1 and a decrease of ESR1. Overall, our data show that HHV-6A and -6B infections have a remarkable effect on the expression of miRNAs and transcription factors, which might be important in the induction of NK cell function impairment, virus escape strategies and related pathologies.

Keywords: HHV-6, miRNA, transcription factors, natural killer cells, virus infection

## INTRODUCTION

Natural killer (NK) cells belong to the innate immune system and are essential effector cells in the control of virus infections (Cooper et al., 2001). Their activity during antiviral response is crucial to control initial virus replication, by killing virus-infected cells prior to the development of adaptive/specific immunity, but they also act as essential regulators of the adaptive immunity

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Birgit Strobl, Veterinärmedizinische Universität Wien, Austria Masaaki Miyazawa, Kindai University, Japan

> \*Correspondence: Elisabetta Caselli csb@unife.it

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 31 July 2017 Accepted: 19 October 2017 Published: 31 October 2017

#### Citation:

Rizzo R, Soffritti I, D'Accolti M, Bortolotti D, Di Luca D and Caselli E (2017) HHV-6A/6B Infection of NK Cells Modulates the Expression of miRNAs and Transcription Factors Potentially Associated to Impaired NK Activity. Front. Microbiol. 8:2143. doi: 10.3389/fmicb.2017.02143

(Vivier et al., 2008). Their relevance is supported by the observation that individuals with defects in the NK cell component of the innate immunity are more susceptible to virus infection (Orange, 2002), including herpesviruses (Fleisher et al., 1982; Biron et al., 1989), and consequently more prone to develop symptomatic disease following infection. On the other hand, the importance of NK cell activity during virus infections is reflected by the many mechanisms acquired by viruses to evade NK cell-mediated immune responses (Vossen et al., 2002; Arens, 2012). In particular, human herpesvirus 6A and 6B (HHV-6A and 6B), as all viruses belonging to the Herpesviridae family, have developed several mechanisms to control and inactivate the immune response in order to establish a lifelong infection in their hosts.

HHV-6A and 6B are members of the Roseolovirus group of the β herpesvirinae subfamily and, although they share high sequence homology, are classified as distinct species. In fact, they show important differences in biologic properties, epidemiology, and disease association (Ablashi et al., 2014). HHV-6B infects humans in early childhood and is responsible of Exanthem subitum (Yamanishi et al., 1988), while primary infection with HHV-6A still has to be clearly identified. Both HHV-6A and - 6B establish a latent infection in the host following resolution of primary infection. Reactivations in the adult have been associated to the development of multiple symptomatic diseases often characterized by immune dysregulation (multiple sclerosis, Sjögren's syndrome, autoimmune thyroiditis, and others) (Caselli and Di Luca, 2007). Both viruses are considered lymphotropic, showing an elective tropism for CD4+ T-lymphocytes and being able to infect several different cell types of the immune system, including NK cells (Lusso et al., 1993; Caselli and Di Luca, 2007).

Interestingly, in vivo and in vitro evidences indicate that HHV-6A/6B interfere with the immune system of the infected host in several ways (Lusso, 2006; Dagna et al., 2013). They can modulate surface antigens important for T-cell activation, such as human leukocyte antigen (HLA) class I molecule expression in dendritic cells (Hirata et al., 2001); they also can affect cytokine and chemokine productions, including selective suppression of IL-12, affecting the generation of effective cellular immune responses (Smith et al., 2003; Dagna et al., 2013). Furthermore, we recently observed that HHV-6A infection induces the expression of the tolerogenic nonclassical class I HLA-G molecule in primary human mesothelial cells, leading to impairment of NK cell recognition and killing of infected cells (Caselli et al., 2015). With reference to the NK cell component of the immune response, HHV-6A was reported to establish a productive infection in CD3-negative NK cell clones, leading to the de novo expression of CD4 on the NK cell surface (Lusso et al., 1993), and HHV-6B was recently shown to induce down-modulation of the activating NKG2D ligand in infected cells (Schmiedel et al., 2016).

Notably, it has been recently reported that NK cells may be directly involved in the onset and progression of autoimmune diseases, through their potential autoreactivity or through their interaction with the other immune cells (Schleinitz et al., 2010; Poggi and Zocchi, 2014), thus supporting the hypothesis of a correlation between HHV-6A/6B infection, NK cell function and autoimmunity.

On the other hand, miRNAs are known to play an essential role in fine-tuning host immune homeostasis and responses, as miRNA-mediated regulation of gene expression has a profound impact on immune cell development, function, and response to invading pathogens. Interestingly, we recently observed that HHV-6A/6B infection of human thyrocytes and T-lymphocytes profoundly remodulates the expression of cellular miRNAs, inducing specific miRNAs associated to autoimmunity in vivo (Caselli et al., 2017), and of transcription factors (unpublished observations).

To study the effects of HHV-6A and -6B on NK cell functions, we analyzed the effect of in vitro infection of NK cells on the expression of miRNAs. We also investigated the expression of transcription factors in infected NK cells, in the attempt to further clarify the details of intracellular alterations induced by these viruses with relevance on the immune function.

## MATERIALS AND METHODS

### Cells and Viruses

The human NK-92 natural killer cell line (ATCC <sup>R</sup> CRL-2407TM) was used for all infection experiments (Gong et al., 1994). Cells were expanded in Alpha Minimum Essential Medium supplemented with 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 0.1 mM 2-mercaptoethanol, and 20% fetal bovine serum (FBS) (complete medium).

Cell-free virus inocula were obtained as described previously (Caselli et al., 2006; Caruso et al., 2009), and quantified by quantitative real-time PCR (qPCR) (Caselli et al., 2017). All the experiments were performed by using the same virus inoculum, containing 10<sup>10</sup> genome copies/ml, corresponding to about 10<sup>9</sup> infecting particles/ml, as previously described (Caselli et al., 2015).

All experiments involving virus production and infection were performed under standard BLS-2 biosafety level.

## NK Cell Infection

NK-92 cells were seeded at optimal density and after 24 h HHV-6A or 6B were added at a 100:1 multiplicity of infection (MOI, virus genomes:cell ratio). Virus adsorption was carried out in a 2% FBS medium for 3 h, then the excess virus was eliminated by centrifugation and washing in PBS, and cells were finally seeded at 0.5 × 10<sup>6</sup> cells/ml in complete medium with high FBS concentration. Control cells were treated with the same procedure but uninfected. At 1, 2, 3, and 6 days post-infection (d.p.i.) aliquots of cultures were collected and analyzed for virus DNA presence, virus transcription and antigen expression, as well as for expression of miRNAs and transcription factors. Evaluation of cell viability was performed by cell counting after Trypan Blue exclusion test.

#### qPCR Analyses of Virus DNA Presence

Total DNA was extracted from the infected or uninfected cells by a commercial kit (Exgene Cell SV kit, GeneAll Biotechnology,

Korea), and quantified by spectrophotometric reading at 260 and 280 nm. Virus DNA presence was verified and quantified as previously described by a qPCR detecting the conserved U94 gene of HHV-6A and 6B with equivalent efficiency (Caselli et al., 2017).

#### Microarray and RT-qPCR Analyses

RNA was extracted from cells by the miRNeasy kit (Qiagen, Hilden, Germany), allowing the extraction of total RNA including miRNA fraction, as previously described (Caselli et al., 2017). Extracted RNA were devoid of contaminant DNA, as assured by DNase treatment and control β-actin PCR without reverse transcription (Caselli et al., 2012). RNA reverse transcription was performed by the RT2 First strand Kit (Qiagen, Hilden, Germany) for analyses of virus transcripts and human transcription factors, whereas for miRNA analyses the miScript RT kit was used (Qiagen, Hilden, Germany). cDNA aliquots corresponding to 200 ng RNA were used for virus transcription analysis, performed by qPCR detecting the expression of U94, U42 and U22 genes, as previously reported (Caselli et al., 2012). The expression of transcription factors was analyzed by the 'Human Transcription Factor' microarray, detecting and quantifying simultaneously 83 different human transcription factors (Qiagen, Hilden, Germany), using 500 ng cDNA as the template.

miRNA analyses were performed on 100 ng of specifically reverse transcribed cDNA, by the 'Human Inflammatory Response & Autoimmunity' Microarray (Qiagen, Hilden, Germany), quantifying simultaneously 84 different miRNAs involved in the immune response, and by 20 individual assays chosen to detect and quantify miRNAs specifically involved in the NK cell functions. Namely, the following individual miRNAs were analyzed: miR155∗\_1, miR155\_2, miR146a\_1, miR16- 1 <sup>∗</sup>\_1, miR16\_2, miR181a∗\_1, miR181a\_2, miR181b\_1, miR181b-3p\_1, miR10a∗\_1, miR10a\_2, miR150\_1, miR150- 3p\_1, miR27a\_1, miR27a∗\_1, miR27b∗\_1, miR27b\_2, miR223\_1, miR223∗\_1, miR378∗\_1; miRTC\_1, and SNORD61\_11 were used as internal controls (all Qiagen, Hilden, Germany).

All qPCR amplification results obtained by transcription factors microarray, miRNA microarray and individual assays, were analyzed and normalized by a specific Qiagen software, to obtain comparable values between control and infected cells at each time post-infection.

#### Immunofluorescence Analysis

Immunofluorescence for HHV-6A/6B antigen expression was performed with mouse monoclonal antibodies (mAb) directed against glycoprotein gp116 (late antigen) of HHV-6 A and B (ABI, Columbia, MD, United States), as previously described (Caselli et al., 2006). Briefly, aliquots of infected or uninfected NK-92 cells were collected by centrifugation 10 min at 1000 × g, counted, spotted on a glass slide (50,000 cell in 10 µl), dried at room temperature and subsequently stained as described elsewhere (Caselli et al., 2006).

## Statistical Analysis

Statistical analysis of collected data was performed using the Stat View software package (SAS Institute, Inc., Cary, NC, United States). Comparative analysis between individual parameters in infected and control groups was performed by Student's t-test, and p-values ≤ 0.01 were considered significant. For multiple comparisons, the Bonferroni correction was applied, and corrected p-values (pc) ≤ 0.01 were considered significant.

## RESULTS

## HHV-6A and 6B Infect Productively NK-92 Cells

Both viruses established a productive/lytic infection in NK-92 cells, as shown in **Figure 1**, confirming that human NK cells are permissive to viral infection (Lusso et al., 1993), and showing that the NK-92 cell line could be used for all the subsequent analyses.

In fact, virus DNA was present in infected cells at all timepoints post-infection (1 to 6 d.p.i.), as measured by specific U94 qPCR, and the increased genome number, compared to timepoint 0, confirmed that virus replication was actually taking place in infected cells. The analysis of virus transcription confirmed the establishment of productive infection in NK cells, as evidenced by the presence of the immediate-early U42 and late U22 transcripts, together with the U94 transcript (which is detected both during lytic and latent HHV-6 infection), at all time-points post-infection (p.i.) (**Figure 1A**). Establishment of lytic infection was confirmed by IFA results, showing abundant expression of the late gp116 envelope glycoprotein at 6 d.p.i. in infected cells (**Figure 1B**).

Infected NK-92 cells appeared damaged and less able to form clusters compared to uninfected controls, with more pronounced effects in the case of HHV-6A, compared to HHV-6B (**Figure 1B**). The virus-induced damages were confirmed by viable cell counts, showing a slight decrease in HHV-6A and 6B infected cells at 1 d.p.i. (−14.5 ± 3.8% compared to control uninfected cells), but a more evident reduction at later time-points, with HHV-6A infection causing a 66.7 ± 9.8% cell loss and HHV-6B a 39.8 ± 6.5% cell loss, at 6 d.p.i., compared to controls.

## miRNA Expression Modulation by HHV-6A and 6B Infection

miRNA expression in infected cells was first studied by a microarray assay simultaneously detecting 84 different miRNAs associated to the development of inflammation and autoimmunity. The results, summarized in Supplementary Table S1, show that both HHV-6A and 6B induce evident alterations in the expression of cellular miRNAs. In particular, 23 miRNAs resulted significantly modulated (p<sup>c</sup> ≤ 0.01) compared to controls, in at least one time-point p.i. With the aim to highlight only the most prominent effects caused by virus infection, we focused our attention on those miRNAs showing at least fourfold changes compared to controls, arbitrarily chosen as a cut-off value. Thirteen miRNAs were modulated more than fourfold by virus infections, with clear early and late effects and differences between the two viruses (**Figure 2**). In particular, both HHV-6A and 6B induced an early up-regulation of miR-301a and miR-548e (1 d.p.i.), an increase of miR-101 and a decrease of miR-let-7c

and miR-340 at 3 d.p.i., and a down-regulation of miR-23 at late time-points p.i. (6 d.p.i.). Other effects were specific for each virus species. Namely, HHV-6A specifically induced an early upregulation of miR-590 (1 d.p.i.), miR-15a and miR-21 (3 d.p.i.), a sustained up-regulation of miR-29b, miR-101 (3 and 6 d.p.i.), miR-301a and miR-548e (1 and 6 d.p.i.) and a late up-regulation of miR-340 and miR-381 (6 d.p.i.) By contrast, HHV-6B infection specifically up-modulated the expression of miR-301b (2 and 3 d.p.i.) and miR-548e (1 and 3 d.p.i.), whereas it down-regulated miR-590 (2 and 3 d.p.i.) and miR-15a (6 d.p.i.).

Since microarray assays did not include important miRNAs known for their role in NK cell development and function, we analyzed by individual assays the following miRNAs: miR10, miR16, miR27, miR146, miR150, miR155, miR181, miR223, miR378, chosen as they have already been reported to be associated with maturation and activation of NK cells (Liu et al., 2012; Beaulieu et al., 2013; Sullivan et al., 2013).

The results showed that four additional miRNAs were altered by HHV-6A and 6B infection (**Figure 3**), as both viruses, although to a different extent, induced a significant decrease (up to 12-fold compared to controls; p ≤ 0.001) of miR-155 and miR-181, which are respectively involved in cytotoxicity and maturation of NK cells, and a concomitant significant increase (up to 14-fold; p ≤ 0.001) in the expression of miR-146 and miR-223, which are associated to NK cell effector functions, including IFNγ and TNFα production.

## Transcription Factors Modulation by HHV-6A/6B Infection

We analyzed the impact of HHV-6A and -6B infection on the expression of human transcription factors, with the simultaneous analysis of 83 different transcription factors by microarray.

The results showed that HHV-6 infection strongly modulates the expression of transcription factors (**Figure 4**). In fact, more than 30 transcription factors were significantly up or downmodulated compared to controls, at different time-points p.i (p<sup>c</sup> < 0.01). In particular, the remodulation of transcription factors was evident at late time-points p.i., whereas at early time-points p.i. viral infection had lower impact, with clear differences between the two species. In fact, while HHV-6A was essentially down-modulating a few transcription factors

at 1 d.p.i. (DR1, HNF4A, POU2AF1, PPARγ), HHV-6B had an essentially up-regulating effect, increasing the expression of FOXA2, FOXG1, GATA2, HNF1A, and decreasing only the expression of MYOD1. By contrast, at 6 d.p.i. both HHV-6A and HHV-6B infections resulted in the up-modulation of several transcription factors, perhaps related to the evident CPE and cell lysis induced by viruses in the infected NK cells.

Both viruses induced the up-regulation of ATF3, CEBPA, CEBPB, JUN and FOXA2 factors, and the down-modulation of PPARγ factor. However, other factors were modulated by only one virus. For example, EGR1 and FOS were up-regulated only by HHV-6A (at 2, 3, 6 d.p.i.), which also specifically induced the decrease of POU2AF1 expression (at 1, 2 d.p.i.) and NFYB (at 2, 3, 6 d.p.i.). Instead, HHV-6B induced an increase in FOXO1 (2, 3 and 6 d.p.i.) and CREB1 (6 d.p.i.) expression and a down-regulation of ESR1 at 3 d.p.i. Other transcription factors displayed a biphasic modulation, as they were differently regulated by viruses at early and late time-points post-infection: the GATA family, HNF1A and HNF4A.

#### DISCUSSION

In their continuous interaction with the host immune system, herpesviruses have developed a large array of strategies to escape the host defense mechanisms. In particular, although different for many biological and pathological aspects, both HHV-6A and 6B display a strong tropism toward lymphocytes, causing important modifications and cytopathic effects in infected cells. Both virus species infect preferentially CD4+ T cells, and HHV-6A productively infects also different types of cytotoxic effectors, including CD8+ T cells, γδT cells, and NK cells (Lusso et al., 1993; Caselli and Di Luca, 2007). By contrast, HHV-6B tropism seems to be more restricted compared to HHV-6A, infecting poorly or at all cytotoxic effector cells (Lusso et al., 1991), but it was recently shown to down-modulate the activating NK cell ligands in SupT1 T cells, impairing the recognition of virus-infected cells by NK cells (Schmiedel et al., 2016).

Although the importance of the interplay between immune cells and HHV-6A/6B has been recognized as a key factor in viral immune evasion strategies and in the development of associated pathologies, few mechanisms have been elucidated. Our study shows for the first time that HHV-6A and HHV-6B productively infect NK cells and induce evident modifications in the expression of two sets of intracellular mediators of effector functions: miRNAs and transcription factors. In particular, microarray data showed that HHV-6A and 6B manipulate the expression of cellular miRNAs in NK cells, inducing both common and species-specific effects.

The results show that the two viruses may induce a substantial alteration in NK cell functions, as most of the modulated

miRNAs are involved in important immune functions. In fact, both HHV-6A and 6B down-modulate miR-let-7, which is highly expressed in NK cells, CD4+ and CD8+ T cells. Interestingly, this miRNA was reported to be down-modulated also by murine Cytomegalovirus (Beaulieu et al., 2013) and HIV-1 infection (Swaminathan et al., 2012), where it was associated with an increase of anti-inflammatory IL-10, thus providing viruses with an important immune escape mechanism. Similarly, decrease in miR-340 expression might protect viruses from the immune response, as miR-340 directly targets the Th2 effector proinflammatory cytokine IL-4 (Podshivalova and Salomon, 2013; Kim et al., 2016). Studies have in fact reported that IL-4 can enhance viral virulence in diverse models, including HSV-1 eye disease, possibly by suppressing cytotoxic lymphocyte response (Ghiasi et al., 1999; Jackson et al., 2001; Kerr et al., 2004). On the contrary, the increase of miR-301a (up-regulated by both viruses) has been reported to block the IRF1 innate immune response against Japanese encephalitis virus and might therefore favor HHV-6 pathogenesis by inhibiting IFNβ production (Hazra et al., 2017). Interestingly, miR-301 was also up-regulated in T cells in the central nervous system of animals with experimental autoimmune encephalomyelitis (Mycko et al., 2012), an animal model for the process of autoimmune demyelination occurring during multiple sclerosis, a disease with a possible association with HHV-6 infection (Leibovitch and Jacobson, 2014). The up-regulated miR-101 and miR-381 are involved in the polarization and activation of the cells of the innate immune compartment, regulating the inflammatory response (Zhu et al., 2010; Essandoh et al., 2016; Wen et al., 2016); the increased miR-548 may represent a mechanism facilitating viral pathogenesis as it negatively correlates with IFNγR1 levels (Xing et al., 2014). miR-21 and miR-590, up-regulated by HHV-6A, are involved in differentiation and activation of T cells, particularly during autoimmunity (Sousa et al., 2017), with miR-590 also downregulating critical genes of signaling pathways similar in cancer and inflammation (Sheikholeslami et al., 2017). miR-29 and miR-15, both increased by HHV-6A infection, have a specific role in regulating the cytotoxic activity of NK cells, inhibiting the production of IFNγ by directly targeting the 3<sup>0</sup> UTR of its mRNA (Ma et al., 2011; Leong et al., 2014).


FIGURE 4 | Modulation induced by HHV-6A and 6B infection on human transcription factors mRNAs in NK-92 cells. NK-92 cells were uninfected (control) or infected with HHV-6A or 6B cell-free inocula, then analyzed at 1, 2, 3, and 6 d.p.i. for mRNA expression by a microarray detecting 83 transcription factors. Results are expressed as fold-change compared to control values, and represent the mean value of duplicate samples from three independent experiments. Statistically significant differences after Bonferroni correction are marked with asterisks (p<sup>c</sup> ≤ 0.01) and highlighted by colors (yellow ≥ 4-fold up-regulations; blue ≥ 4-fold down-regulations). Exact p<sup>c</sup> values of all significant parameters varied from 0.005 to 0.01.

Interestingly, HHV-6B, but not HHV-6A, infection downmodulated miR-590 and miR-15, suggesting that the two species might impact differently on NK cell functions. Notably, however, both species significantly decreased miR-155, important in the effector functions of NK cells as it stimulates IFNγ production in activated NK cells (Trotta et al., 2012), and miR-181, essential for the correct maturation of NK cells (Cichocki et al., 2011). Both viruses induced an increase of miR-146, which negatively regulates NK activity by inhibiting cytotoxicity, IFNγ and TNFα production (Xu et al., 2016), and of miR-223, which inhibits the production of granzyme B by murine NK cells (Fehniger et al., 2010; Leong et al., 2014). Intriguingly, miR-223 has been associated to the pathogenesis of autoimmune thyroiditis, another disease which has been correlated to HHV-6 infection (Caselli et al., 2012). Interestingly, the modifications of miRNA expression induced by HHV-6A and 6B infections in NK cells were not superimposable to those observed in T cells (Caselli et al., 2017), demonstrating that, as well as species-specific, the effects induced by the two viruses are also cell type-specific, as they impact differently on the expression of miRNAs in different immune cell types.

Our analyses showed also that HHV-6A and 6B modulate the expression of several transcription factors in infected NK cells, with effects shared by both viruses, or specifically induced by only one of them, underlining again the different impact of the two viruses on NK cells. In fact, both viruses induced the up-regulation of ATF3, CEBPA, JUN and FOXA2, and down-regulation of PPARγ factor. Whereas HHV-6A induced an increase in EGR1 and FOS expression and a decrease of POU2AF1 expression, HHV-6B induced an increase in FOXO1 and a down-regulation of ESR1 expression. Although many of the modulated factors are not yet associated to specific functions in NK cells biology, some of them have already been described to have a role in this context, or in closely related immune aspects.

Interestingly, ATF3, upregulated by both viruses, was reported to regulate negatively NK cell functions in MCMV infected mice, by modulating IFNγ expression (Rosenberger et al., 2008). On the other hand, although not yet studied in NK cells, several evidences currently associate PPARγ to Th lymphocyte differentiation, B lymphocyte effector functions and cytokine expression (da Rocha Junior et al., 2013). POU2AF1, down-modulated by HHV-6A, was recently reported to induce upregulation of host defense genes, including IL-6, in airway epithelium (Zhou et al., 2016). FOXO1 and ESR1, respectively increased and decreased by HHV-6B, are important regulators of the immune response, being FOXO1 a negative regulator of NK cell maturation and functions (Deng et al., 2015), whereas ESR1 has been associated to regulation of inflammatory pathways of innate immune cells (Kovats, 2015). Other factors, including JUN, FOS, CEBPA, are involved in several biological processes as regulators of cell cycle progression, hematopoietic cell differentiation and apoptosis (Foletta et al., 1998; Wisdom et al., 1999; Ponti et al., 2002; Healy et al., 2013), and might be studied in detail in the NK cell context.

Overall, our data show for the first time that infection by HHV-6A and 6B profoundly impacts the intracellular environment of infected NK cells, likely inducing biological effects helping the viruses to escape the innate immune response. Although the two different HHV-6 species induce many common effects, our data also show species-specific effects on miRNAs and transcription factors expression. The differences might possibly result in a different biological impact of the two viruses, potentially associated to specific pathological conditions.

## CONCLUSION

HHV-6A and 6B induce significant alterations on the expression of several miRNAs and transcription factors in infected NK cells. These alterations might lead to important modifications in NK cell ability to control HHV-6 infections, enabling immune evasion and facilitating viral replication cycle, and impacting on NK cell involvement in inflammation and autoimmune reactions.

Functional studies should be conducted to investigate the role of the factors altered by HHV-6 infection, and to evaluate their possible implication in pathological alterations and disease progression.

## AUTHOR CONTRIBUTIONS

RR contributed to the conception of the work and data analysis. IS, MD, and DB contributed to data collection and analysis. DDL contributed to data interpretation and critical revision of the article. EC contributed to the conception of the work, data acquisition and analysis, and writing the article.

## FUNDING

This work was supported by HHV-6 Foundation grants (PI: EC and RR), by PRIN grant (PI: EC, cod 2015YZB22C), by Merck grant (PI: RR), and by FISM-Fondazione Italiana Sclerosi Multipla grant (PI: RR, cod 2015/R/20).

## ACKNOWLEDGMENT

We thank Iva Pivanti for her excellent technical assistance, and Linda Sartor for revising the English manuscript.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02143/full#supplementary-material

## REFERENCES

fmicb-08-02143 October 28, 2017 Time: 16:21 # 9


type 1 ocular replication and eye disease. J. Infect. Dis. 179, 1086–1093. doi: 10.1086/314736


Eur. J. Immunol. 32, 3358–3365. doi: 10.1002/1521-4141(200212)32:12<3358:: AID-IMMU3358>3.0.CO;2-Q


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Rizzo, Soffritti, D'Accolti, Bortolotti, Di Luca and Caselli. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dynamics of Viral and Host Immune Cell MicroRNA Expression during Acute Infectious Mononucleosis

Vandana Kaul<sup>1</sup>† , Kenneth I. Weinberg<sup>2</sup> , Scott D. Boyd<sup>3</sup> , Daniel Bernstein<sup>4</sup> , Carlos O. Esquivel<sup>1</sup> , Olivia M. Martinez1,5 \* † and Sheri M. Krams1,5 \* †

<sup>1</sup> Division of Abdominal Transplantation, Department of Surgery, Stanford University, Stanford, CA, United States, <sup>2</sup> Division of Stem Cell Transplantation, Department of Pediatrics, Stanford University, Stanford, CA, United States, <sup>3</sup> Department of Pathology, Stanford University, Stanford, CA, United States, <sup>4</sup> Division of Cardiology, Department of Pediatrics, Stanford University, Stanford, CA, United States, <sup>5</sup> Stanford Immunology, Stanford University School of Medicine, Stanford, CA, United States

#### Edited by:

Yves Renaudineau, University of Western Brittany, France

#### Reviewed by:

Mario M. D'Elios, University of Florence, Italy Haitao Guo, University of North Carolina at Chapel Hill, United States

#### \*Correspondence:

Sheri M. Krams smkrams@stanford.edu Olivia M. Martinez omm@stanford.edu †Co-senior authors

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 04 October 2017 Accepted: 21 December 2017 Published: 15 January 2018

#### Citation:

Kaul V, Weinberg KI, Boyd SD, Bernstein D, Esquivel CO, Martinez OM and Krams SM (2018) Dynamics of Viral and Host Immune Cell MicroRNA Expression during Acute Infectious Mononucleosis. Front. Microbiol. 8:2666. doi: 10.3389/fmicb.2017.02666 Epstein–Barr virus (EBV) is the etiological agent of acute infectious mononucleosis (IM). Since acute IM is a self-resolving disease with most patients regaining health in 1–3 weeks there have been few studies examining molecular signatures in early acute stages of the disease. MicroRNAs (miRNAs) have been shown, however, to influence immune cell function and consequently the generation of antibody responses in IM. In this study, we performed a comprehensive analysis of differentially expressed miRNAs in early stage uncomplicated acute IM. miRNAs were profiled from patient peripheral blood obtained at the time of IM diagnosis and at subsequent time points, and pathway analysis performed to identify important immune and cell signaling pathways. We identified 215 differentially regulated miRNAs at the most acute stage of infection when the patients initially sought medical help. The number of differentially expressed miRNAs decreased to 148 and 68 at 1 and 2 months post-primary infection, with no significantly changed miRNAs identified at 7 months post-infection. Interferon signaling, T and B cell signaling and antigen presentation were the top pathways influenced by the miRNAs associated with IM. Thus, a dynamic and regulated expression profile of miRNA accompanies the early acute immune response, and resolution of infection, in IM.

Keywords: RNA, acute infectious mononucleosis, Epstein–Barr virus, immunology

#### INTRODUCTION

Epstein–Barr virus is a gamma-herpes family virus that infects over 95% of humans globally and is the etiological agent of many acute and chronic infections as well as human malignancies (Khan and Hashim, 2014). While EBV infection is widespread, the outcome of infection can depend upon the age of the host. Pediatric cases are mainly asymptomatic and the infection subsides due to a vigorous host T cell response with the virus subsequently transitioning to latency in a subset

**Abbreviations:** EBV: Epstein–Barr virus; FDR, false discovery rate; IPA, ingenuity pathway analysis; miRNA, microRNA; PBMC, peripheral blood mononuclear cells.

of memory B cells (Jayasooriya et al., 2015; Gantt et al., 2016). However, in adolescents, primary infection can manifest as glandular fever, lymphadenopathy, and a sore throat, termed infectious mononucleosis (IM) (Balfour et al., 2013). IM is associated with a transient proliferation of EBV-infected B cells followed by a considerable expansion of EBV-specific T cells (Sixbey et al., 1984; Kurth et al., 2000; Chaganti et al., 2009; Hadinoto et al., 2009; Balfour et al., 2013). IM is typically selflimiting and resolves over a period of weeks to months with the virus persisting into a latent phase in infected B cells (Hochberg et al., 2004).

MicroRNAs are small 18–25 nt long single-stranded RNA molecules which regulate gene expression at the post-transcriptional level. miRNA target specific mRNAs by complementary base-pairing, leading to cleavage or repression of translation, and can act as major regulators of key cellular processes. We (Harris et al., 2010) and others (Forte et al., 2012; Mansouri et al., 2014) have shown that EBV infection can alter the expression of host miRNAs and thus modulate various arms of the cellular machinery including cell differentiation and death, proliferation, and the immune response. Along these lines, miR-194 is suppressed by EBV and participates in regulation of IL-10 expression in EBV lymphomas (Harris et al., 2010) while two miRNA families, the let-7 family and the miR-200 family, as well as miR-143-3p, act as tumor-suppressor miRNAs and are significantly downregulated in EBV-infected gastric carcinoma cells (Marquitz et al., 2014). Other studies have found that miR-155 and miR-21 were induced in B cells by EBV infection and potentially play a role in viral tumorigenesis (Linnstaedt et al., 2010; Rosato et al., 2012; Anastasiadou et al., 2015). Furthermore, p53-targeted tumor suppressor, miR-34a, is strongly induced by EBV infection in many EBV and Kaposi's sarcoma-associated herpesvirus-infected lymphoma cell lines and plays a role in B cell transformation (Forte et al., 2012). In addition, miRNAs have been identified that play a critical role in the T cell response to pathogens. For example, miR-155 is essential for the CD8<sup>+</sup> effector T cell response to mCMV (Dudda et al., 2013).

In addition to host miRNAs, EBV-expressed miRNAs have also been shown to modulate the function of T and B cells during an immune response. In a study by Albanese et al. (2016), EBV-encoded miRNAs were found to inhibit CD8<sup>+</sup> T cellmediated viral surveillance at multiple levels. The viral miRNAs downregulate TAP-complex and HLA allotype that present TAPdependent epitopes, reduce IL-12 secretion by infected B cells to diminish EBV-specific CD8 T cells and repress EBNA1, a viral protein responsible for latency (Albanese et al., 2016). EBV encoded miRNAs miR-BART1, miR-BART2, and miR-BHRF1- 2 have also been shown to target the host adaptive immune response by repressing IL-12 secretion resulting in reduced differentiation of CD4<sup>+</sup> T cells into the Th1 subset and inhibiting lysosomal enzymes involved in MHC class-II peptide processing (Tagawa et al., 2016), while miR-BART16 suppresses Type I IFN signaling (Hooykaas et al., 2017). In the current study, we report the differential expression of host and viral miRNAs in acute IM caused by primary EBV infection. Our findings elucidate the miRNA signature associated with acute IM and demonstrate the dynamic features of this response with the miRNA expression profile reverting to that of healthy individuals within a few months of IM diagnosis in subjects without complications.

## MATERIALS AND METHODS

## Patient Samples

Peripheral blood samples were obtained from symptomatic patients (aged 18–24 years) who visited Vaden Student Health Center at Stanford University and were diagnosed with clinical IM. Samples were obtained upon diagnosis (time point 0) and then at, 1, 2, and 7 months. Blood samples were processed using Ficoll-Paque PLUS (Amersham Pharmacia Biotech, Piscataway, NJ, United States) then PBMC frozen and maintained in liquid nitrogen until use. PBMCs were isolated from blood derived from healthy age-matched individuals. This study was carried out in accordance with the recommendations of Institutional Review Board at Stanford University with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

## RNA Isolation and miRNA Microarray

Patient (n = 16) and healthy control (n = 3) PBMCs were suspended and lysed in TRIzol Reagent (Life Technologies, Foster City, CA, United States) for isolation of total RNA. Glycogen (20 µg) was used for RNA precipitation. Isolated RNA was DNase treated (Qiagen) and clean-up was done on RNeasy spin columns (Qiagen) using a modified protocol which replaced RW1 buffer with RWT buffer for washing. RNA was suspended in RNase free water and stored at −80◦C. RNA quantification was done using Nanodrop 2000 Spectrophotometer (Thermo Fisher scientific, Wilmington, DE, United States). Approximately 200 ng RNA was used for microarray analysis using Affymetrix GeneChip miRNA 4.0 arrays (Affymetrix, Santa Clara, CA, United States) through the PAN (Protein and Nucleic acid) facility at Stanford University. Briefly, 130 ng total RNA was assessed for quality using Bioanalyzer 2100 (Agilent), followed by Poly-A tailing and labeling with biotin using a Affymetrix FlashTag Biotin HSR RNA Labeling kit (Affymetrix, Santa Clara, CA, United States) according to the manufacturer's protocol. The labeled targets are hybridized to the GeneChip <sup>R</sup> miRNA 4.0 arrays per standard miRNA protocol, which includes 16 h (overnight) hybridization at 48◦C at 60 RPM in the Affymetrix GeneChip Hybridization oven 645. The arrays were then washed and stained in the Affymetrix GeneChip Fluidics Station 450. The arrays were scanned using the Affymetrix GeneChip Scanner 3000 7G. The GeneChip <sup>R</sup> miRNA 4.0 arrays contain 30,424 total mature miRNA probe sets including 2578 mature human miRNAs, 2025 pre-miRNA human probes, 1996 Human snoRNA and scaRNA probe sets and miRNAs of 202 other organisms.

## Data Analysis

Raw microarray data were statistically analyzed using Partek Genomics Suite software (Partek Inc.). Normalization was

done suing Robust Multi-array Average (RMA) and log2 transformation of data was performed, followed by one-way ANOVA statistical testing to identify significantly differentially expressed miRNA between different groups (P ≤ 0.05). After running ANOVA, the software performs multi-testing correction using Benjamini–Hochberg Step-Up FDR-controlling procedure for all the expressed miRNAs (default FDR Alpha level 0.05). Venn diagram depicting intersection of differentially expressed miRNAs and Unsupervised Hierarchical Clustering analysis was also performed. Raw Fold change and adjusted p-value data was exported to dot excel file and data selection and graphic visualization was done using Power BI software (Microsoft Inc.). Differentially expressed EBV miRNAs were also identified using the same miRNA microarray dataset using the miRNA filter as "EBV," mRNA expression data from a publicly available GEO dataset (GSE45924) was used to perform miRNA–mRNA integrated analysis. Briefly, SOFT formatted family file (GSE45924-GPL6883\_series\_matrix.piv.txt.fmt) were downloaded from the GEO Database and imported into Partek Genomics Suite for mRNA "Gene Expression" workflow. Illumina HumanRef-8 v3.0 expression beadchip annotation files were uploaded to the Partek Software to obtain Fold Change Values for mRNA differential expression between comparison groups of Acute IM subjects (PBMC) and control subjects (PBMC). Pre-normalized data was log2 transformed, followed by one-way ANOVA statistical testing to identify significantly differentially expressed mRNAs and multiple-testing correction using Benjamini–Hochberg Step-Up FDR-controlling procedure (default FDR Alpha level 0.05). Fold increases of >2, and fold decreases of >2 were considered significant. The differentially expressed mRNA data was exported into a text file and converted into a suitable format for upload to IPA for further analysis.

### Functional and Pathway Enrichment Analysis of Target Genes Controlled by Differentially Expressed miRNAs

Pathway and molecular function enrichment analysis was performed using IPA software (Qiagen, Redwood city, CA, United States). A list of all twofold differentially expressed miRNAs obtained from Partek Genomics Suite software analysis was uploaded into IPA for analysis. miRNA target enrichment was performed using the "miRNA Target Filter" function, generating a list of 181 miRNAs with available targeting information. An mRNA gene expression dataset (GSE45924) obtained from GEO database was added to the "miRNA Target Filter" setting in IPA and "Expression pairing" was performed. An additional filter applied to this list was "Applied filters: confidence = Experimentally Observed OR High (predicted)." Further "Core Analysis" was performed on the target mRNAs based on different experimental time points to determine the affected networks and canonical pathways. Throughout the analysis, a p-value cutoff of 0.05 was applied. The HumanRef-8 v3.0 reference gene list was used to set the population of genes to consider for p-value calculations while performing Core Analysis.

## RESULTS

## Sample QC and Principal Component Analysis

As a means of quality control to determine the biological separation of the groups based on probe intensities, principal component analysis (PCA) was performed using Partek Genomics software. Two separate groups of samples from IM patients at 0 month (at the time of diagnosis) and healthy controls are clearly observed (**Figure 1**). The remaining samples from 1, 2, and 7 months are loosely segregated into groups and fall between these two groups on PC1 with the samples obtained at 7 months grouping closely with those from healthy controls. These findings indicate that the differentially expressed miRNAs are maximal at the time of IM diagnosis and return to the healthy control pattern in subsequent months.

## Differentially Expressed miRNAs

The differential expression of miRNAs were compared between patients at the time of IM diagnosis (0 month), and at 1 and 2 months after diagnosis, compared to age-matched healthy controls **(Figure 2A**). We identified 215 differentially regulated miRNAs at the most acute stage of infection (diagnosis, 0 month). The number of differentially expressed miRNAs decreased to 148 and 68 during the course of 1 and 2 months post-primary infection, with no significantly changed miRNAs identified at 7 months post-infection (data not shown). The 27 top differentially expressed miRNAs at diagnosis, and 1 and 2 months after diagnosis, show that most of the differentially expressed miRNAs are increased (25 increased, 2 decreased) (**Figure 2B**).

Unsupervised hierarchical clustering depicted as a heatmap shows the relative intensity for the 215, 148, and 68 differentially expressed miRNAs at diagnosis (**Figure 3**), 1 month (Supplementary Figure S1) and 2 months (Supplementary Figure S2) as compared to healthy controls, respectively. A fold change of two and FDR adjusted p-value (Benjamini–Hochberg method for multiple corrections) of 0.05 was considered significant in Partek Genomics Suite Software, with black depicting decreased expression levels and blue depicting upregulation. Furthermore, the differentially expressed miRNAs demonstrate a decreasing absolute fold change over time, with the highest fold change generally observed at the time of diagnosis (**Figure 4**). These 41 miRNAs exhibit a dynamic differential expression pattern that trends toward healthy control levels at 2 months post IM diagnosis. Two miRNAs that are highly upregulated, miR-4417 and miR-4485 (data not shown), exhibit a similar trend of decreasing differential expression from diagnosis to 2 months. A compilation of all statistically significant differentially expressed miRNAs at diagnosis and one, and 2 months versus healthy controls is included in Supplementary Table S1. Supplementary Table S2 provides a list of all differentially expressed miRNAs. A Venn diagram depicts the number of miRNAs in common at IM diagnosis and at 1 and 2 months (**Figure 5**). The intersection shows that 56 miRNAs are differentially expressed at the three time points, thus providing important information regarding possible disease

specific signatures. Interestingly, miRNAs uniquely expressed at diagnosis, 1 and 2 months after diagnosis also decrease in number during the course of the infection toward recovery (79 at diagnosis, 15 at 1 month and 9 at 2 months).

#### Pathways Altered by IM

Expression pairing of miRNA data, from our study and gene expression data from GEO set GSE45924 returned 46 miRNAs targeting 17 mRNAs (**Figure 6A**). Comparative analysis of mRNA targets highlighted the most canonical pathways as well as disease and biological functions (**Figures 6B,C**). Furthermore, functional pathway analysis in IPA using the list of differentially expressed miRNAs at 0 month suggests the most impacted networks include antigen presentation, infectious disease, humoral immune response, protein synthesis, cell cycle, and cell death mechanisms. Networks have been shown in addition to their scores and number of mRNA targets (including particular molecules that pass our set filter criterion) (**Table 1**). Further analysis in IPA for top canonical pathways targeted by our list of differentially expressed miRNAs at IM diagnosis and based on the Illumina HumanRef-8 v3.0 gene superset shows candidates including interferon signaling, altered T and B cell signaling, primary immunodeficiency signaling, antigen presentation and communication between innate and adaptive immune cells as the major affected canonical pathways (Supplementary Figure S3).

#### Differentially Expressed EBV miRNAs

Interestingly, when the EBV miRNAs are analyzed using specific filtering criteria in Partek, only ebv-miR-BART-16 and ebv-miR-BART5-3p are uniquely upregulated at IM diagnosis and at 1 and 2 months as compared to healthy controls (ANOVA 0.05, unadjusted p-value) (**Table 2**). It is notable that no differential expression of any EBV miRNAs is observed at 7 months postinfection, which is expected due to resolution of the symptoms in the acute phase and establishment of EBV latency. Thus, of the more than 40 known EBV miRNAs only two, miR-BART5-3p and miR-BART-16 are expressed during acute IM (Skinner et al., 2017).

### DISCUSSION

This study provides the first miRNA expression profile of sequential, peripheral blood samples from college-age patients with acute IM. Our results clearly indicate that primary EBV disease is accompanied by marked expression changes of multiple host and viral miRNA, and that the number of differentially expressed miRNAs decreases substantially after the initial diagnosis. It is noteworthy that there were no differentially expressed miRNAs observed at 7 months as compared to healthy controls, indicating that there is no identifiable miRNA signature that can distinguish between normal college students and those that have resolved IM. Since none of the patients studied had any complications after IM, it remains possible that expression of the differentially expressed host miRNAs would remain dysregulated if IM were not completely resolved.

We determined that 43 of the host miRNAs, including miR-197-3p, miR-491-5p, miR-151a-5p, miR-652-3p, and miR-744-5p, decreased from the levels found at diagnosis over the subsequent 2 months. While the specific function of these miRNA in IM is unknown, other reports have linked these miRNA to disease processes. Recently, miR-197 was found to act synergistically with EBV-BART6-3p to reduce the expression of IL-6R, thereby compromising host immune defense in EBV-positive Burkitt Lymphoma (Zhang et al., 2017). miR-625-3p has been shown to be a potential biomarker for a number of cancerous conditions including malignant pleural mesothelioma (Kirschner et al., 2012). miR-625-3p targets p38

TABLE 1 | List of top five networks with their respective scores obtained from ingenuity pathway analysis (IPA): Data are indicative of networks regulated by mRNA targets of microRNAs (miRNAs) differentially expressed at the time of infectious mononucleosis (IM) diagnosis.


<sup>∗</sup>Focus Molecules (in bold)- molecules that passed the filters and cut-offs that are members of this network.

FIGURE 6 | mRNA targets and associated pathways of the differentially expressed miRNAs. (A) Venn diagram and table comparison of mRNA targets of differentially expressed miRNAs at Diagnosis, 1 and 2 months. (B) Comparison of canonical pathways, (C) Comparison of Diseases and Bio Functions; mRNA targets were obtained from miRNA–mRNA expression pairing of differentially expressed miRNAs and GEO dataset GSM1119621–GSM1119636 in ingenuity pathway analysis (IPA).

TABLE 2 | Differentially expressed EBV miRNAs in patients at diagnosis, 1 and 2 months versus healthy controls, ANOVA 0.05, unadjusted p-value.


MAPK activator MAP2K6 and induces oxaliplatin resistance by abrogating MAP2K6-p38-regulated apoptosis and cell cycle control networks (Rasmussen et al., 2016). A recent study also shows that miR-625-3p is upregulated in CD8<sup>+</sup> T cells during early immune reconstitution following allogeneic stem cell transplantation (Verma et al., 2017). An interesting study with highly virulent H5N2 strain of mouse-adapted avian influenza virus found miR-151a-p as one of the significantly upregulated miRNAs, whose inhibition resulted in a 70% reduction in mortality of inoculated mice (Choi et al., 2014).

The majority of studies to date examining the impact of EBV infection or pathogenesis on miRNA expression have focused on the miRNA profile within the virally infected cell and the implications for tumorigenesis. In contrast, we analyzed the impact of EBV infection on peripheral lymphoid cells that typically participate in the immune response to EBV. A single previous study did examine the cellular miRNAs from preadolescent children with IM in China over the first 14 days after primary infection (Gao et al., 2015). They determined the host miRNA expression in B cells and CD8<sup>+</sup> T cells and EBV miRNAs in plasma and B cells. Interestingly, some miRNAs such as miR-155 and miR-142 were increased in B cells and decreased in CD8<sup>+</sup> T cells, thus, changes would not be reflected in unfractionated PBMC. Our analysis of miRNAs in PBMC demonstrated a different set of host miRNAs as differentially expressed. These distinct patterns are most likely attributed to different methodologies, since our study took an unbiased approach and performed microarray analysis using Affymetrix GeneChip miRNA 4.0 arrays (over 30,000 mature miRNAs) on unfractionated PBMC, whereas the Gao group specifically examined a small subset of 84 miRNAs in sorted B and T cells via PCR. Further, the ages of the subjects were different and there could be other differences including environmental factors and variant strains of EBV. EBV expresses more miRNAs than other characterized human viruses, with at least 44 EBV miRNAs identified to date (Skinner et al., 2017). We determined that ebvmiR-BART16 and ebv-miR-BART5-3p were increased in PBMC as compared to healthy controls. The specific function of these viral miRNA in IM is unknown, but it is interesting that a recent report indicates that BART16 suppresses IFN signaling (Hooykaas et al., 2017). Notably, the EBV miRNA profile we identified in acute IM is distinct from the EBV miRNA profile identified in plasma samples of patients with chronic active EBV infection (Kawano et al., 2013).

In addition to EBV-induced miRNA regulation in infected B cells, others have focused on understanding the EBV pathogenesis and host responses by transcriptomic (mRNA microarray, Illumina) analysis on PBMCs of human subjects in a prospective study of naturally acquired primary EBV infection (Dunmire et al., 2014). The authors found a distinct

gene expression profile in acute but not latent EBV infection. Moreover, upon gene expression profiling in sorted CD8<sup>+</sup> T cells, NK cells, monocytes, and B cells, it was found that CD8<sup>+</sup> T cells and monocytes showed upregulation of key gene groups (Type I interferon regulated genes (IRGs), type II IRGs, and cell cycle/metabolism genes.) during acute infection. This pathway analysis agrees with the pathways predicted to be altered by our miRNA profile during acute IM.

## CONCLUSION

We have defined the host immune cell and EBV miRNA profile elicited by primary EBV infection during acute IM. This profile is indicative of a response that is accompanied by complete resolution of disease. Thus, it may be useful in predicting which individuals that present with IM will have an uncomplicated course and ultimately may shed insight into the host immune response to acute EBV infection.

## AUTHOR CONTRIBUTIONS

VK designed and performed experiments, analyzed the data, and wrote the manuscript. KW obtained the samples and contributed to development of study design. SB contributed to development of study design and critically revised the manuscript. DB contributed to development of study design. CE contributed to development of study design. OM contributed to study design,

#### REFERENCES


provided critical review of the experimental design, wrote, and provided critical review of the manuscript. SK designed the study, analyzed the data, wrote, and critically reviewed the manuscript. All authors reviewed and approved the final manuscript.

## FUNDING

VK was supported by fellowship from the Transplant and Tissue Engineering Center of Excellence at Lucile Packard Children's Hospital. Funding was also obtained from the Lucile Packard Children's Hospital Heart Center Research Fund (DB, KW, and SB). OM was partially funded by AI115313, SK was partially funded by AI119686, KW, SB, DB, CE, OM, SK were partially supported by AI104342.

#### ACKNOWLEDGMENT

The authors would like to acknowledge Natalia V. Kosovilka from Stanford PAN facility for help with microarray experiments.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02666/full#supplementary-material

MiR-34a is growth promoting in EBV-infected B cells. J. Virol. 86, 6889–6898. doi: 10.1128/JVI.07056-11


Epstein-Barr virus infection. J. Infect. Dis. 208, 771–779. doi: 10.1093/infdis/ jit222


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Kaul, Weinberg, Boyd, Bernstein, Esquivel, Martinez and Krams. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Regulation of the Interferon Response by lncRNAs in HCV Infection

#### Saba Valadkhan<sup>1</sup> \* and Puri Fortes<sup>2</sup> \*

<sup>1</sup> Department of Molecular Biology and Microbiology, Case Western Reserve University School of Medicine, Cleveland, OH, United States, <sup>2</sup> Center for Applied Medical Research, Department of Gene Therapy and Hepatology, Navarra Institute for Health Research (IdiSNA), University of Navarra, Pamplona, Spain

#### Edited by:

Wesley H. Brooks, University of South Florida, United States

#### Reviewed by:

Isabel Chillón, European Molecular Biology Laboratory, France Hiroyuki Oshiumi, Kumamoto University, Japan

#### \*Correspondence:

Saba Valadkhan saba.valadkhan@case.edu Puri Fortes pfortes@unav.es

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 22 September 2017 Accepted: 26 January 2018 Published: 16 February 2018

#### Citation:

Valadkhan S and Fortes P (2018) Regulation of the Interferon Response by lncRNAs in HCV Infection. Front. Microbiol. 9:181. doi: 10.3389/fmicb.2018.00181 The interferon (IFN) response is a critical component of the innate immunity antiviral pathways in mammalians. IFN signaling results in increased expression of cellular factors that block key steps in the viral replication cycle. Many IFN-induced antiviral factors act through decreasing viral entry, replication, transcription, translation, packaging and release. However, these effects are also deleterious for the viability of the cell, which necessitates a tight control over the magnitude and duration of the IFN response. This is partially achieved through the IFN-mediated activation of negative regulatory factors that help in termination of the IFN response and return to a normal homeostatic state. Such built-in negative regulatory mechanisms are frequently hijacked by viruses such as the Hepatitis C virus (HCV) to increase viral replication and productive infections. We and others have shown that long non-coding RNAs (lncRNAs) play prominent roles in regulation of the IFN response. Activation of the IFN cascade alters the expression of a large number of lncRNAs, many of which are directly induced by the JAK/STAT pathway and thus, resemble the well-studied protein-coding interferon-stimulated genes (ISGs). While only a handful of IFN- and virally induced lncRNAs have been characterized, recent studies have identified several lncRNAs that act as positive or negative regulators of expression of ISGs during the IFN response. A number of such regulatory lncRNAs have multiple ISG targets, while others act on a single neighboring ISG. Another group of studied lncRNAs act further upstream and regulate the expression of IFN genes or factors that sense the presence of viral genome or replication products. The large number of unstudied IFN- and virally induced lncRNAs makes it highly likely that future studies will reveal a much greater share for this class of transcripts in regulation of the antiviral response. In addition to their physiological roles, the expression of such lncRNAs is frequently modulated by virally encoded factors to interfere with the antiviral response and promote viral replication, thus making them ideal targets for therapeutic intervention.

Keywords: type I IFN, lncRNAs, HCV, proviral, antiviral, IFN response

## LONG NON-CODING RNAs AS A NEW CLASS OF REGULATORY MOLECULES

High throughput transcriptome studies in the last decade have led to the discovery of 1000s of novel RNA molecules that do not appear to code for functional peptides. These transcripts, named the long non-coding RNAs (lncRNAs), can be found in both eukaryotes and prokaryotes. However, they are especially abundant in more complex eukaryotes including animals and plants (Rinn and Chang, 2012; Morris and Mattick, 2014). Interestingly, their abundance seems to correlate with the level of complexity of the organism, for example while protein-coding sequences constitute < 2% of the human genome, a significantly larger fraction of our genomes is coding for lncRNAs (ENCODE Project Consortium, 2012; Clark et al., 2013). lncRNAs have diverse modes of biogenesis and mechanism of action and constitute a heterogeneous group of transcripts. Many of them are transcribed by RNA polymerase II and are spliced and polyadenylated, although on average they contain fewer introns compared to protein-coding genes (Carninci et al., 2005; Derrien et al., 2012; Niazi and Valadkhan, 2012). However, several studied lncRNAs are RNA polymerase I and III transcripts and a significant fraction of lncRNAs are not polyadenylated (Carninci et al., 2005; Derrien et al., 2012; Djebali et al., 2012). Similarly, they show a wide heterogeneity in size and can range from a couple of hundred to tens of thousands of nucleotides in length. An arbitrary lower length limit of 200 nucleotides which has been proposed for this class of RNAs should not be applied too strictly, as it mainly serves as a general guideline for distinguishing lncRNAs from the small non-coding RNA such as small nuclear and nucleolar RNAs, microRNAs, etc. (Clark and Mattick, 2011; Rinn and Chang, 2012; Mattick and Rinn, 2015).

Despite intense research in recent years, the sheer number and diversity of lncRNAs have limited our understanding of the scope of function of lncRNAs in higher eukaryotes. Evidence emerging from existing research indicate that the expression of many lncRNAs is strongly dependent on the cell type and cellular state and is tightly controlled by various cellular signals (Rinn and Chang, 2012; Amaral et al., 2013). Thus, it is likely that many lncRNAs remain to be described and annotated as more cellular states are probed with high throughput transcriptomic studies. Even among the annotated lncRNAs, the vast majority remain unstudied. Finally, many alternatively processed unstudied isoforms of protein-coding RNAs do not have a significant protein-coding capacity and thus, can potentially affect the cellular function as non-coding RNAs (Carninci et al., 2005; Djebali et al., 2012; ENCODE Project Consortium, 2012). Therefore, it is likely that future research will elucidate a much larger share for lncRNAs in cellular function than currently assumed.

Nonetheless, data from the small number of lncRNAs currently studied indicate their involvement in diverse aspects of cellular function (Wapinski and Chang, 2011; Moran et al., 2012; Rinn and Chang, 2012; Amaral et al., 2013; Ulitsky and Bartel, 2013; Yang et al., 2014). Emerging data suggest that a major functional mechanism of lncRNAs involves regulation of nuclear events, including transcriptional regulation and control of the epigenetic state of chromatin (Rinn and Chang, 2012; Amaral et al., 2013; Rinn, 2014). Accordingly, localization of a lncRNA to the nuclear compartment and association with chromatin modifying complexes or transcription factors serve as potential clues into the function of a lncRNA.

Another potential indicator of the function of a lncRNA is the genomic locus from which it originates (**Figure 1**). Genes coding for lncRNAs frequently overlap protein-coding genes or other lncRNA genes in the sense or antisense orientation with the lncRNA initiating from a different promoter. In many cases, such promoters are located within or near the 3<sup>0</sup> UTR of protein-coding genes (Carninci et al., 2005; Derrien et al., 2012; Djebali et al., 2012). Further, many lncRNAs result from selective stabilization of a certain region of a protein-coding gene, such as the 3<sup>0</sup> UTR (Mercer et al., 2011). Such overlapping genes may affect the biogenesis and/or function of the other genes in the locus via several potential mechanisms, ranging from epigenetic regulation of the activity of the locus, to transcriptional interference, to masking or competing with functionally critical elements of the other transcripts arising from the overlapped locus through base-pairing (Valadkhan and Nilsen, 2010). Over 10% of human genes originate from the so-called bidirectional promoters, and in many cases at least one of the two promotersharing genes is a lncRNA (**Figure 1**) (Adachi and Lieber, 2002; Wakano et al., 2012; Uesaka et al., 2014). Data from studied examples has shown that in such loci, one member of the pair can regulate the expression of the other transcript which originates from the same promoter (Wei et al., 2011; Uesaka et al., 2014). Another subclass of lncRNAs, those originating from promoters in enhancer loci, are thought to be crucial for the function of the enhancers from which they are transcribed (**Figure 1**) (Lam et al., 2014). Finally, many lncRNAs are located in vicinity of other genes without overlapping them or sharing architectural elements with them (**Figure 1**). Such intergenic vicinal lncRNAs can potentially affect the expression of their nearby genes via transcriptional interference or epigenetic regulation (Valadkhan and Nilsen, 2010; Rinn and Chang, 2012; Mattick and Rinn, 2015). It should, however, be mentioned that many functional lncRNAs are located far away from their target genes and thus, locus proximity or even overlap is not necessarily a requirement for a regulatory relationship. On the other hand, some lncRNAs may function in cis to regulate the expression of genes located far in the genome but in the same nuclear territory (Hacisuleyman et al., 2014).

As mentioned above, current evidence point to the involvement of lncRNAs in virtually every aspect of cellular function. Several pioneering studies in recent years have provided evidence for the critical role of lncRNAs in regulation of diverse aspects of the immune response, including both innate and adaptive immunity (Fitzgerald and Caffrey, 2014; Heward and Lindsay, 2014; Imamura and Akimitsu, 2014; Stachurska et al., 2014; Marques-Rocha et al., 2015; Satpathy and Chang, 2015; Sigdel et al., 2015; Yu A.D. et al., 2015; Yu Q. et al., 2015). While the main focus of this review is on the host-derived lncRNAs involved in the interferon arm of the antiviral response, recent research has revealed that a number of virally coded lncRNAs also affect the IFN response (summarized in Valadkhan

and Gunawardane, 2015). The following sections include a discussion of the host-derived lncRNAs induced by and/or affecting the interferon response, followed by a more focused overview of the role of lncRNAs in hepatitis C infection as a well-studied example of a viral infection. Comprehensive reviews on the role of lncRNAs in development and function of the immune response are included elsewhere (Valadkhan and Gunawardane, 2015; Carpenter, 2016; Atianand et al., 2017).

#### THE INTERFERON ARM OF THE ANTIVIRAL RESPONSE

A crucial and nearly ubiquitous component of the innate immune response against viruses and many other microbial pathogens is mediated through the interferon (IFN) signaling cascade. IFNs are traditionally divided into three major groups (I, II, and III). While binding distinct receptors, type I (IFN-α, -β, -κ, -ε, and -ω) and type II (IFN-γ) IFNs show extensive overlap in their downstream signaling cascades and regulated genes (Hertzog et al., 2011; Pollard et al., 2013; Rusinova et al., 2013; Bolen et al., 2014). Similarly, type III IFNs (IFNλ1, IFN-λ2, IFN-λ3, also called IL-29, IFN-λ4, IL-28A, and IL-28B, respectively), which are predominantly expressed in plasmacytoid dendritic cells in addition to a number of other cell types, bind a distinct receptor but show downstream overlap with type I IFNs (Meyer, 2009; Pollard et al., 2013; Bolen et al., 2014). The IFN response is initiated through the induction of expression of IFNs via activation of a class of cellular factors that act as the initial sensors of pathogenassociated molecular patterns. Among them, RNA sensors such as retinoic acid inducible gene I (RIG-I), melanoma differentiation-associated gene-5 (MDA5) and membrane-bound Toll-like receptors (TLR 3, 7, or 8) are particularly relevant to antiviral response against RNA viruses, although cells also contain several DNA sensors (**Figure 2**) (Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014). Activation of the sensor molecules, in turn, leads to signal transduction cascades and ultimately induction of the expression of genes with specific inhibitory functions against microorganisms.

In the case of the type I IFN response, recognition of pathogen-associated patterns by RIG-I and MDA5 leads to activation of the IFN-β promoter stimulator 1 (IPS-1), while TLR3 and TLR7/TLR8 induce signaling via TRIF and MyD88, respectively (Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014). The activation of these pathways, in turn, will result in phosphorylation and dimerization of transcription factors including the interferon regulatory factor 3 (IRF-3), and IRF-7 followed by their nuclear translocation (**Figure 2**). Next, together with NF-κB and ATF-2/c-jun, they nucleate the formation of active transcriptional complexes through interactions with several transcription factors and regulatory proteins such as the transcriptional cofactor (CREB) binding protein (CBP)/p300 (Kawai and Akira, 2006; Hiscott, 2007; Randall and Goodbourn, 2008; Takeuchi and Akira, 2009; Onomoto et al., 2010; Jensen and Thomsen, 2012). This, in turn, results in induction of the expression of antiviral cytokines including type I IFNs (IFN-α and IFN-β). While IFN-β is produced in most cells, IFN-α is predominantly produced in hematopoietic cells such as monocyte/macrophages and dendritic cells, especially the plasmacytoid dendritic cells. IFN-γ is induced in response to different extracellular signals, including interleukins IL2, IL12, and IL18, preferentially, in NK and T cells (Green et al., 2017). These cytokines activate JAK/STAT, NF-κB, JNK, ERK and p38 MAPK pathways, depending on the cell line, and induce transcription of IFN-γ in response to NFAT, NF-κB, STAT and AP1 transcription factors.

IFN-α and -β bind to the IFN-α/β receptor (IFNAR) in an autocrine and paracrine manner, triggering the induction of the Janus kinase/signal transducers and activators of transcription (JAK/STAT) signaling cascade (**Figure 2**) (Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014). The outcome is recruitment, followed by phosphorylation, dimerization and finally nuclear translocation of STAT1 and STAT2. Once in the nucleus, the complex of STAT1 and STAT2 is bound to IRF9/p48, forming the IFN-stimulated gene factor 3 (ISGF3) complex which induces the transcription of hundreds of IFN-stimulated genes (ISGs) (**Figure 2**) (Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014). In a highly analogous manner, IFN-γ, the sole member of the type II IFN family, binds the IFN-γ receptor and activates signaling through the JAK/STAT pathway resulting in STAT1 phosphorylation followed by nuclear translocation (Hu and Ivashkiv, 2009). Once activated, STAT1 binds the gamma-activated sequence (GAS) close to its target genes as a homodimer, triggering the transcriptional induction of STAT1 regulated genes. As mentioned above, the majority of IFNγ-regulated genes are also induced by type I IFNs and thus, type I and type II IFN response show significant overlap in terms of the downstream effects (Hu and Ivashkiv, 2009; Pollard et al., 2013).

The final outcome of the IFN signaling is the induction of ISGs, many of which exert antimicrobial activity via regulation of different steps of cellular gene expression in addition to direct antimicrobial effects. For example, several ISGs including IFIT-1, IFIT-2, IFITM3, ISG15, ISG20, RNase L, PKR, viperin and

BST2/Tetherin are known to have antiviral activity (Randall and Goodbourn, 2008). Thus, activation of the IFN response results in the induction of a cell-intrinsic antimicrobial state in both the infected and the neighboring cells which limits the spread of infectious agents. Further, these potent cytokines also affect both the innate and adaptive immune response through promoting antigen presentation, natural killer cell function, production of antibody in B cells and T cell effector function (Ivashkiv and Donlin, 2014). As mentioned above, most cells are able to launch the type I IFN response, however, cell type and context have a strong effect on the magnitude of the induced response and the specific subsets of effector genes activated (Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014). On the other hand, the IFN response itself is subject to negative feedback regulation by a number of ISGs and other cellular signaling pathways (Yoshimura et al., 2007; Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014; Porritt and Hertzog, 2015).

While the IFN cascade is mainly activated in response to the presence of exogenous viral RNAs, a group of long non-coding cellular RNAs, namely the retrotransposon-derived transcripts, also activate the IFN response (Yu Q. et al., 2015). It has been shown that activation of the Long Interspersed Element-1 (LINE-1) retroelements results in induction of the expression of IFN-β and ISGs (Yu Q. et al., 2015). Prior treatment of cells with IFN leads to suppression of LINE-1 replication, and mutations that inactivate different steps of the IFN signaling pathway cause an increase in LINE-1 replication (Yu Q. et al., 2015). Although the mechanism behind the induction of the IFN response by LINE-1 replication is not known, it is likely that the replication state of LINE elements and other retrotransposons are detected by either the same or similar sensors to those that detect the presence of exogenous viral RNA in the cell. Existing data suggest that the replication of another class of retrotransposons, the Small Interspersed Elements (SINEs) similarly results in activation of the IFN response (Leonova et al., 2013). Interestingly, it has been suggested that LINE-1 elements may contribute to the pathogenesis of autoimmune disorders such as Systemic Lupus Erythematosus (SLE) (Crow, 2010; Nakkuntod et al., 2011), which is known

to involve activation of the IFN response (Bronson et al., 2012; Elkon and Wiedeman, 2012). Additional studies of the connection between autoimmunity and the replication state of cellular repeat elements will likely provide novel insight into the mechanistic basis of development of autoimmune diseases.

### lncRNAs AS CRUCIAL REGULATORY FACTORS IN THE IFN RESPONSE

A number of high throughput transcriptomic analyses in both human and mouse have revealed the presence of strong changes in the expression of lncRNAs during the IFN response (Peng et al., 2010; Carnero et al., 2014; Josset et al., 2014; Kambara et al., 2014, 2015; Barriocanal et al., 2015). In the mouse studies, which were performed in both mouse lung tissue and cultured mouse embryonic fibroblasts, a significant fraction of the differentially expressed genes belonged to lncRNAs, which were either up or down-regulated after interferon stimulation and infection with respiratory viruses (Peng et al., 2010; Josset et al., 2014). Interestingly, promoter analysis and expression correlation studies raised the possibility that a significant fraction of the induced lncRNAs may be novel ISGs that were directly induced through the IFN signaling cascade (Peng et al., 2010; Josset et al., 2014).

The results obtained in the murine models were closely analogous to those obtain using human cells. In one such study, primary human hepatocytes from five donors of different ages and genders were subjected to high throughput transcriptome analysis before and at three time points of 3, 9, and 24 h after treatment with IFN-α (Kambara et al., 2014). Similar to what had been observed in the mouse, IFN stimulation resulted in both induction and repression of expression of a large number of lncRNAs. Many of the induced lncRNAs were likely direct targets of the JAK-STAT signaling pathway and thus, novel ISGs. While the majority of the induced lncRNAs remained upregulated for the three time points analyzed in this study, a fraction of them showed a shorter duration of induction and were limited to the earlier time points of the study (Kambara et al., 2014).

A third set of high throughput studies in HuH7 human hepatocytes focused on changes in gene expression observed at later time points (after 72 h) of treatment with a high dose of IFN-α2 using both microarray analysis (Carnero et al., 2014) and RNA-seq (Barriocanal et al., 2015). Interestingly, similar to the results of the study by Kambara et al. (2014), the lncRNAs that showed differential expression were almost equally divided between upregulated and downregulated ones in both microarray and RNA-seq-based datasets, while almost all differentially expressed protein coding genes were upregulated. In addition, both sets of studies revealed a significant number of differentially expressed lncRNAs which were either vicinal to or overlapped protein-coding genes that function in the immune response (Carnero et al., 2014; Kambara et al., 2014, 2015; Barriocanal et al., 2015). As mentioned above, in some cases the function of a lncRNA depends on its genomic locus and thus, these studies potentially point to the presence of an IFN-activated regulatory network of lncRNAs which may function in fine-tuning the immune response.

#### Regulation of the Expression of IFNs by lncRNAs

Induction of transcription of IFN genes by external or internal stimuli constitutes a key step in the IFN response and is known to be regulated by several protein-mediated mechanisms (Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014). Interestingly, recent evidence point to the presence of a number of lncRNA-mediated regulatory mechanisms acting on this step. At the IFNG/IFN-γ locus, lncRNA IFNG-AS1 (IFNG-antisense-1), also known as Tmevpg1 (Vigneau et al., 2003; Collier et al., 2012; Collier, 2014) and NeST (Gomez et al., 2013), is located downstream of the IFNG genic region and in human, overlaps this locus. The expression of IFNG-AS1, which is present in CD4+ and CD8+ T cells in addition to natural killer (NK) cells, shows a strong positive correlation with that of IFNG (Vigneau et al., 2003; Collier et al., 2012; Collier, 2014). The expression of IFNG-AS1 contributes to the induction of IFNG expression in CD4+ cells during their differentiation into Th1 polarization (Collier et al., 2014) and in CD8+ T cells after stimulation with PMA (phorbol 12-myristate 13-acetate) and ionomycin (Gomez et al., 2013). Mechanistic studies have shown that IFNG-AS1 acts in trans via interacting with WDR5, a subunit of the MLL/SET1 histone H3 lysine 4 methyltransferase complex, potentially recruiting the complex to the IFNG locus to change its methylation state (Gomez et al., 2013).

A more recent study in human trophectoderm progenitors suggested that knock down of a novel lncRNA, lncRHOXF1, which is expressed in trophectoderm and primitive endoderm cells in human blastocyst-stage embryos, may lead to upregulation of MDA5, RIG-I and IFN-β (Penkala et al., 2016). While the mechanism of the observed gene expression changes is not yet determined, the above data suggest a repressive role for lncRHOXF1 in the IFN response. NEAT1, a well-studied lncRNA with a structural role in paraspeckles, also regulates IFN-β production and RIG-I function, albeit in the opposite direction (Ma et al., 2017). Studies in the human umbilical vein endothelial cells indicated that upon infection with Hantaan virus, transcription of NEAT1 was induced through activation of RIG-I/IRF7 pathway. NEAT1, in turn, mediated the relocation of the splicing factor prolineand glutamine-rich protein (SFPQ) to paraspeckles, thus removing its transcriptional inhibitory effect on the expression of RIG-I and DDX60 leading to IFN-β production (Ma et al., 2017). In this manner, NEAT1 acts as a positive feedback regulator of RIG-I activation during viral infections. It is likely that future studies will reveal additional lncRNAs which regulate the signaling steps upstream of the expression of IFN genes, thus acting as global regulators of the IFN response.

## lncRNAs Governing the Expression of IFN-Stimulated Genes

As mentioned above, high throughput transcriptomic studies have revealed the presence of a large number of lncRNAs that show differential expression in response to IFN stimulation (see **Table 1** for some functionally studied examples). One such RNA, which was identified in primary human hepatocytes by Kambara et al. (2014), originated from a locus downstream of the protein-coding ISG CMPK2 and showed a strong upregulation after IFN stimulation in additional cell types from both human and mouse. The induction of this RNA, similar to protein-coding ISGs, was dependent on the JAK-STAT signaling pathway. Knock down studies on this lncRNA, which was named lncRNA-CMPK2/NRIR (Negative Regulator of the IFN Response), resulted in a strong reduction in HCV replication in IFN-stimulated hepatocytes. Additional analyses indicated that knockdown of NRIR led to upregulation of both basal and IFNstimulated transcription of a number of protein-coding antiviral ISGs, which could be best explained by loss of transcriptional inhibition in knockdown cells. While the mechanism of function of NRIR has not been determined, its nuclear localization together with its ability to affect the transcription of its target genes were consistent with either an epigenetic or transcriptional regulatory mechanism. Together, these data provided evidence for the presence of a lncRNA-mediated negative regulatory mechanism during the IFN response (Kambara et al., 2014). As NRIR and likely many other unstudied lncRNAs that regulate the IFN response are themselves bona-fide ISGs, it is plausible that at least some such lncRNAs may affect their own expression in addition to that of other target genes, thus creating an additional layer of self-regulatory loops. However, to our knowledge this possibility has remained unstudied.

Shortly after the discovery of NRIR, another lncRNA named NRAV (Negative Regulator of Antiviral Response) was described in a study focusing on genes differentially expressed in response to influenza virus H1N1 infection in A549 human alveolar epithelium cell line (Ouyang et al., 2014). NRAV originated from what is likely a bidirectional promoter that also gave rise to the main isoform of the dynein light chain gene DYNLL1 and was both spliced and polyadenylated. The level of NRAV was markedly reduced following infection with influenza virus and a number of other viruses in several cell lines. Importantly, forced overexpression and knock down experiments revealed that the expression level of NRAV directly correlated with the extent of viral reproduction. A microarray study of NRAV-overexpressing cell lines showed the reduction of the level of a significant number of ISGs, and additional experiments indicated that NRAV, similar to NRIR (Kambara et al., 2014), has the ability to partially block induction of the expression of its target ISGs in response to IFN stimulation (Ouyang et al., 2014).

Another lncRNA with a negative regulatory function in the IFN response is EGOT (Eosinophil Granule Ontogeny


Transcript). EGOT is a structured polyadenylated nuclear lncRNA conserved, at least, in all placental mammals (Rose and Stadler, 2011). Interestingly, EGOT genomic locus shows enhancer marks with high histone 3 lysine 4 monomethylation and low trimethylation, indicating that EGOT could be an enhancer RNA (Heintzman et al., 2007). EGOT overlaps an intron of the inositol 1,4,5-trisphosphate receptor 1 (ITRP1) gene in antisense orientation but EGOT depletion does not affect ITRP1 cellular levels (Prior et al., unpublished observation). EGOT was first described as a lncRNA expressed in eosinophils during development and maturation and is thought to function in mature eosinophils in regulating the levels of toxic molecules, such as the major basic protein and the eosinophil derived neurotoxin (Wagner et al., 2007). However, GTEx studies show that the highest levels of EGOT are found in nonhematopoietic tissues such as breast, vagina, pancreas, pituitary and kidney cortex (GTEx Consortium, 2013). A recent study in HuH7 cells indicated that the level of EGOT shows a dramatic increase after infection with HCV and other RNA viruses (see below) (Carnero et al., 2016). EGOT was also induced in response to very high doses of IFN-α, but at much lower levels compared to what is observed after infection with RNA viruses. Knock down of EGOT in HCV infected cells led to an increase in the expression of a subset of ISGs including GBP1, ISG15, Mx1, BST2, ISG56, IFI6, and IFITM1, resulting in reduced viral replication (Carnero et al., 2016). Taken together, these results indicate that although EGOT is not a bona fide ISG itself, it is yet another lncRNA with a negative regulatory impact on ISG induction and thus, the IFN response.

Considering the rather small number of lncRNAs that have been studied in the context of the IFN response, the discovery of the negative regulatory roles of NRIR, NRAV and EGOT on the transcriptional induction of ISGs suggests that lncRNAs may play a prominent role in negative feedback loops controlling the IFN response and possibly other signaling cascades in the immune response. Interestingly, several protein factors involved in negative regulation of IFN response have been identified (Yoshimura et al., 2007; Hertzog and Williams, 2013; Ivashkiv and Donlin, 2014; Schneider et al., 2014; Porritt and Hertzog, 2015). Defining the interaction of the negative regulatory lncRNAs with the positive and negative regulatory proteins in the context of IFN response will yield a unified picture of the feedback mechanisms which mediate the termination of IFN signaling cascade.

While the above described lncRNAs had a negative regulatory impact on the IFN response, a recent study has provided evidence for lncRNA-mediated positive regulation of the IFN cascade. A screen for IRF3-dependent genes in HuS immortalized human hepatocytes led to the identification of an annotated but unstudied lncRNA (LOC100506895/AC011738.4/ENST00000436105.1) that was down-regulated in an IRF3-dependent manner after poly(I:C) treatment (Nishitsuji et al., 2016). This transcript, which was dubbed lncRNA#32 and was later renamed LUARIS (lncRNA upregulator of antiviral response interferon signaling), overlaps introns 21 and 22 and exon 22 of the protein-coding gene HECW1 in antisense orientation (hg38 chr7:43,508,728- 43,522,542). Unlike NRIR, the level of LUARIS was reduced after stimulation with IFN-β. Interestingly, both in the presence and absence of IFN stimulation, knock down and overexpression of LUARIS led to dramatic reduction and upregulation of expression of several ISGs, respectively. Indeed, forced overexpression of LUARIS led to suppression of replication of a number of viral pathogens including HCV, further proving that it acts as a positive regulator of the IFN response (Nishitsuji et al., 2016). Mechanistic studies indicated that the transcriptional stimulatory action of this lncRNA was likely mediated through its binding to the activating transcription factor 2 (ATF2) (Nishitsuji et al., 2016). Why should a positive regulator of the IFN response be downregulated by IFNs? It is likely that the expression level of LUARIS is used to adjust the magnitude of the IFN response through the action of multiple regulatory pathways that control its transcription or stability. Defining the identity of additional signaling pathways that control the expression of this and other similarly acting lncRNAs will provide crucial insights into the complex network of interactions that regulate the antiviral response.

While the lncRNAs described above regulated a number of ISG targets, another IFN-induced lncRNA named BISPR (BST2 IFN-Stimulated Positive Regulator) seems to affect the expression of a single target gene. Independent studies from two groups (Carnero et al., 2014; Kambara et al., 2014, 2015; Barriocanal et al., 2015) identified BISPR as a lncRNA that was induced in response to IFN-α stimulation through the JAK/STAT pathway in multiple cell lines including the THP1 monocytes (Kambara et al., 2015) and HuH7 human hepatocytes (Carnero et al., 2014; Barriocanal et al., 2015). BISPR originated from a bidirectional promoter that also gave rise to BST2/Tetherin, a well-studied protein-coding ISG. RNAi-mediated knockdown of BISPR reduced the IFN-mediated induction of BST2 expression (Barriocanal et al., 2015; Kambara et al., 2015) pointing to the potential presence of a regulatory mechanism for coordinating the expression of BISPR with that of its promoter sharing gene BST2. Interestingly, after stimulation by IFN-α, the increase in cellular level of BISPR preceded the rise in the level of BST2, suggesting that expression of BISPR either induced or facilitated the initiation of transcription of BST2 (Kambara et al., 2015). Confirming these findings, forced overexpression of the spliced BISPR RNA from a transgene led to up-regulation of BST2, indicating that BISPR controls BST2 expression via interactions mediated through the BISPR RNA itself, rather than by impacting the local chromatin environment through its transcription (Kambara et al., 2015). As a number of protein-coding ISGs and immunity-related protein-coding genes originate from bidirectional promoters which also give rise to lncRNAs, it is plausible that a regulatory mechanism similar to the one described above may be present in at least a subset of them. Taken together, despite the small number of studied lncRNAs, existing data points to a critical role for this class of transcripts in regulation of the IFN response and the antiviral activity against human pathogens such as the hepatitis C virus, which is a well-studied disease model for the impact of IFNs

on human pathogens. Future studies are likely to identify many additional lncRNAs that regulate different steps of this key aspect of the innate immune response.

#### INTERACTION OF THE NON-CODING TRANSCRIPTOME AND RNA VIRUSES: THE EXAMPLE OF HEPATITIS C VIRUS

Hepatitis C (HCV) is a deadly virus that affects ∼2% of the world's population (60–170 million people) (Manns et al., 2017). About 500,000 people die every year from HCV-related diseases (Webster et al., 2015; Bukh, 2016). After the initial infection, most patients develop an asymptomatic chronic infection that causes liver damage and therefore, may progress to liver steatosis, fibrosis, cirrhosis and hepatocellular carcinoma (HCC). In the course of 30 years of HCV infection, ∼20% of patients will develop cirrhosis. In fact, HCV infection is nowadays one of the major inducers of liver cirrhosis. As HCV infection induces the expression of several oncogenic factors and liver cirrhosis is a niche for HCC, ∼3–7% of cirrhotic patients will develop liver tumors every year (Llovet and Villanueva, 2016).

In spite of the efforts of many groups, an effective vaccine to prevent HCV infection has not yet been developed. Recently, several inhibitors that target different viral proteins have been approved for the treatment of HCV infection (D'Ambrosio et al., 2017). The efficacy of these treatments grazes a complete sustained viral response, allowing the claim that HCV infection can be cured (Bourlière et al., 2017). A fraction of the small number of patients that resist the treatment and do not show viral clearance have high levels of liver cirrhosis. Some of the patients with HCV infection and HCC show a transient response with a later relapse of the virus, probably because HCC architecture and/or composition impairs drug penetration and serves as a reservoir for infected cells (Prenner et al., 2017). Some authors have observed an increased risk of HCC recurrence in patients that have cleared HCV infection (Llovet and Villanueva, 2016; Reig et al., 2016). In spite of this, there are good reasons for optimism. After sustained viral responses, liver fibrosis can regress, the risk of cirrhosis-related complications, including HCC, is reduced and the overall survival of the patients increases (van der Meer and Berenguer, 2016; Nahon et al., 2017; van der Meer et al., 2017). However, there is still a long way to achieve the goal of HCV cure. Patients cured of chronic HCV infection are at risk for reinfection if exposed. Work toward the development of an effective vaccine to prevent HCV infection must be intensified to cure the disease (Bukh, 2016). Research about HCV infection should not stop and a rigorous epidemiological follow up should be carried out given the high prevalence of HCV infection and the ability of HCV to generate escape mutants resistant to the treatments (Ramirez et al., 2016).

#### HCV Infection

The seven genotypes of HCV share similar characteristics. HCV viral particle is enveloped, small (40–80 nm) and encloses the genome coated by the core protein (Gastaminza et al., 2010; Catanese et al., 2013). The genome is an RNA molecule of 9.6 Kb of length and positive polarity that may function as messenger RNA (mRNA). Multiple coding regions of the viral genome and the 5<sup>0</sup> and 3<sup>0</sup> untranslated regions (UTRs) are highly structured, well-conserved and required for replication and encapsidation (Piñeiro and Martinez-Salas, 2012; Pirakitikulr et al., 2016; Shi et al., 2016). The structure located at the 5<sup>0</sup> end contains an internal ribosome entry site (IRES) that allows viral RNA cap-independent translation. Translation synthesizes a polyprotein that can be cleaved by viral and cellular proteases co- or post-transcriptionally. Cleavage releases three major structural proteins (core, and the two envelope glycoproteins, E1 and E2) and seven non-structural proteins (p7, NS2, NS3, NS4A, NS4B, NS5A, and NS5B). These proteins are required for polyprotein cleavage, cellular antiviral response blockade, viral replication, assembly and release (Carnero and Fortes, 2016).

The majority of HCV virions are transported in the blood embedded into very low or low-density lipoproteins (VLDLs and LDLs) formed by triglycerides, apolipoproteins and cholesterol or phospholipids (Bassendine et al., 2013). This coat may help virions to escape from neutralizing antibodies and aids hepatocyte infection. LDL Receptor (LDLR), claudin-1 (CLDN1) and occludin (OCLN) are some of the several receptors identified to play a role in viral entry into hepatocytes (Ploss et al., 2009; Zeisel et al., 2011). After endocytosis, the virion uncoats in response to the acidic pH of the endosome by fusing the viral envelope with the endosome and releasing the viral genome into the cytoplasm. Viral RNA translation occurs in the rough endoplasmic reticulum (ER). There, a membranous web (MW) is formed by doublemembrane vesicles, the replication complex is assembled and viral replication takes place (Romero-Brey et al., 2012; Meyers et al., 2016). Replication is carried out by NS5B, the viral RNA-dependent RNA polymerase. The positive-stranded RNAs are used as templates to produce negative-strand RNAs which are then used as guides to synthesize large quantities of new viral genomes. The low fidelity of HCV RNA-dependent RNA polymerase produces a highly variable progeny (Geller et al., 2016). Thus, in every infected patient, the incoming viruses generate a collection of descendants with related but non-identical HCV genomes, known as quasi-species (Martell et al., 1992).

The new viral genomes can be used for translation of new viral proteins, replication to produce new viral genomes or packaging into new viral particles. Assembly takes place close to lipid droplets (LDs) bound to ER membranes. NS5A molecules from the replication complex bind simultaneously to the newly synthesized RNA and core protein, helping the interaction between core protein and viral RNA which will lead to formation of a nucleocapsid bound to a LD (Masaki et al., 2008). The envelope is acquired after budding through the ER and the particle is released bound to LDL or VLDLs through the secretory pathway (Lindenbach and Rice, 2003; Bayer et al., 2016; Syed et al., 2017). Thus, inhibition of the synthesis of lipid components blocks viral assembly (Bassendine et al., 2013).

#### HCV and the Antiviral IFN Response HCV-Infection and Protein Coding Genes That Regulate the IFN Response

Successful replication requires inactivation of the antiviral response, which is initiated soon after infection (Loo et al., 2006; Saito et al., 2008). RIG-I canonical sensor may recognize the incoming viral genome. Non-canonical sensors such as protein kinase R (PKR) and the DEAD box helicase DDX3X also recognize the 50UTR and the 30UTR of HCV genome, respectively (Arnaud et al., 2010, 2011; Li et al., 2013). Later, RIG-I and TLR3 can sense viral dsRNAs produced during replication (Binder et al., 2011; Li et al., 2012). Sensor activation induces NF-κB and IRFs via MAVS and TRIF and the synthesis of several subtypes of IFN (**Figure 2**). While type III IFNλ signaling has a special impact on HCV infection (Ge et al., 2009; Thomas et al., 2009; Boisvert and Shoukry, 2016), type I IFNα has been traditionally used to cure the disease (McHutchison et al., 1998). IFN signaling is a potent activator of the expression of several ISGs that limit HCV replication. They function by reinforcing IFN signaling (STAT1, STAT2, IRF1, 3, 7 and 9, PKR, OAS or RNase L) and blocking every step of the viral cycle: viral entry (IFITM, TRIM, Mx, CH25H), RNA replication, translation and stability (OAS, IFIT, GBP1), assembly and release (tetherin/BST2, viperin) (Carnero and Fortes, 2016). Viperin binds to NS5A in the replication complex and the LDs, blocking NS5A function in viral replication and assembly (Helbig et al., 2005, 2011). BST2 or tetherin impedes the budding of viral particles by attaching virions to the cell surface and allowing their degradation by the lysosomes (Neil et al., 2008; Dafa-Berger et al., 2012). Interestingly, a strong antiviral response is only achieved by the cooperative function of several ISGs and not by their individual actions (Metz et al., 2012).

To combat the antiviral response, HCV has evolved to express viral proteins that block IFN synthesis and signaling and interfere with the functionality of antiviral molecules. NS3-NS4A protease cleaves MAVS and TRIF impeding sensor signaling (Foy et al., 2005; Li et al., 2005; Meylan et al., 2005). In addition, RIG-I signaling may be affected by HCV-mediated induction of autophagy (Ke and Chen, 2011). NS4A/B precursor blocks MHC Class I transport to the cell surface. Finally, core protein blocks IFN signaling by upregulating the expression of SOCS3 or the protein phosphatase PP2Ac, which are negative regulators of STAT1 transcription factor (Bode et al., 2003; Duong et al., 2004; Kawaguchi et al., 2004; Walsh et al., 2006).

The blockade of IFN signaling induced by the virus explains why many patients fail to respond to IFN treatment (Chen et al., 2005; Sarasin-Filipowicz et al., 2008). Surprisingly, many nonresponder patients have increased ISG mRNA levels both in infected and uninfected hepatocytes (Hedegaard et al., 2017). One way to explain why these patients fail to respond to IFN in spite of expressing high levels of ISG mRNAs is that these mRNAs are not efficiently translated into antiviral proteins. Interestingly, it is also possible that some ISG mRNAs are translated into proteins that exert proviral functions in HCV infected cells. Both possibilities may turn out to be correct. DsRNA-activated PKR phosphorylates eukaryotic translation initiation factor eIF2α, leading to inhibition of cap-dependent translation. This does not affect HCV protein translation, which is IRES-mediated and eIF2α-independent (Garaigorta and Chisari, 2009). However, translation of ISG mRNAs is likely to be drastically affected, leading to HCV-infected cells with high levels of ISG mRNAs that are not translated into antiviral proteins. Therefore, PKR is an ISG with a proviral function in HCV infection. Other proviral ISGs include those that function as negative regulators of the antiviral response allowing cells to return to homeostasis after IFN induction and non-canonical sensors such as DDX3X. After viral genome sensing, DDX3X activates IKKα, which induces the expression of lipogenic genes essential for viral assembly (Li et al., 2013; Pène et al., 2015).

A paradigm for an ISG that promotes HCV replication is ISG15 (Broering et al., 2010). ISG15 is an ubiquitin-like moiety cotranslationally attached to proteins by the IFN-induced ISGylation machinery. Therefore, after HCV infection, the IFN response induces ISGylation of new proteins, which are viral proteins and cellular antiviral factors (Durfee et al., 2010). ISGylation may affect the functionality of targeted proteins by modifying their structure and/or stability. IRF3 is stabilized by ISGylation-mediated inhibition of polyubiquitination (Shi et al., 2010). However, RIG-I ISGylation blocks ubiquitination and functionality, leading to decreased levels of IFN and increased replication of HCV (**Figure 2**) (Kim et al., 2008; Broering et al., 2010). Therefore, in HCV infection, ISG15 functions as a proviral ISG. In fact, upregulation of ISG15 is associated with poor response to IFN treatment and poor prognosis in HCV-infected patients (Chen et al., 2011).

The combined action of PKR and the ISGylation and ubiquitination pathways in HCV-infected cells results in decreased cap-dependent protein translation and increased protein modification that leads to protein malfunction and destabilization. Under such a protein-hostile environment it must be challenging to maintain proper cell functionality and virus replication. Therefore, it is conceivable that both cells and viruses have evolved to achieve functionality under these conditions through the expression of functional non-coding RNAs, which are at least partially immune to protein-hostile conditions (Fortes and Morris, 2016).

#### Effects of HCV Infection on the Non-coding Genome

Hepatitis C virus infection causes a deregulation in the expression of the non-coding genome, which may function either to facilitate or to block viral viability. This has been best studied for miRNAs. Several cellular miRNAs have been described to target the HCV genome to help or prevent viral replication or to regulate the expression of cellular factors required for the virus cell cycle or the antiviral response. Generally, virus replication-related factors induce proviral and reduce antiviral miRNAs while the innate immune response plays the opposite effect. The interplay of HCV infection and miRNAs has been reviewed elsewhere (Singaravelu et al., 2014, 2015).

In contrast, the world of infection-induced lncRNAs is widely unexplored and HCV-related lncRNAs are not an exception. The few studies performed clearly indicate that, similar to what has been described for miRNAs, several cellular lncRNAs are

deregulated in response to HCV replication or to the antiviral response induced by viral infection (Barriocanal and Fortes, 2017). The function of some of them has been studied. These lncRNAs affect HCV replication by regulating cell metabolism, proliferation and the antiviral response. HCV could also express viral lncRNAs generated by XRN1 exonuclease-mediated 5<sup>0</sup> to 3 <sup>0</sup> degradation of the viral genome. These subgenomic RNAs have lost the initial sequences of the IRES and therefore they most likely fall into the category of lncRNAs (Moon et al., 2015).

Studies performed so far to identify HCV-deregulated cellular lncRNAs have employed infected cultured cells or liver tissue from infected patients. In the latter case, it is difficult to establish whether the deregulation of the lncRNAs indeed results purely from HCV infection, as the evaluated tissue has also developed liver cirrhosis and/or HCC. Comparison of the level of lncRNAs in these tissues versus healthy liver identifies the lncRNAs that are deregulated by HCV infection, liver cirrhosis or HCC alone or in all possible combinations. In line with this, some of the lncRNAs deregulated in HCV-infected livers have been shown to play a role in the development of liver cirrhosis and/or HCC (Yuan et al., 2014; Fu et al., 2017). This has been recently reviewed (Barriocanal and Fortes, 2017). Other lncRNAs studied in patients with HCV-derived HCC could be bona-fide HCV-induced lncRNAs. This is the case for UCA1, which is also upregulated in liver and serum of patients with HCC. UCA1 serum levels correlate significantly with HCV antibodies and increase in tissue culture cells infected with HCV (Carnero et al., 2016; Kamel et al., 2016; Barriocanal and Fortes, 2017).

Infection with HCV or expression of the core protein in cultured cells has led to the identification of additional HCV-induced oncogenic lncRNAs such as PVT1, CASC15 and HOTAIR (Carnero et al., 2016; Li et al., 2016). The induction of lncRNA HOTAIR by the core protein may in turn lead to increased viral replication by silencing SIRT1 promoter and affecting glucose and lipid metabolism (Li et al., 2016). HCV-induced lncRNAs such as UCA1, PVT1 or CASC15 are upregulated in response to viral replication and not after activation of the antiviral response. Thus, these lncRNAs do not increase when cells are treated with IFN or pathogen associated molecular patterns (PAMPs) such as poly(I:C) or LPS, or in cells infected with other viruses such as HBV, influenza, adenovirus or Semliki Forest Virus (SFV), which replicate in the nucleus or in the cytoplasm, have DNA or RNA viral genomes and lead to acute or chronic infections (Carnero et al., 2016). Instead, it has been proposed that HCV replication may induce specific pathways for the activation of these lncRNAs. For example, viral protein NS5A induces MYC, which in turn activates PVT-1 transcription (Carramusa et al., 2007). Similarly, HCV-induced reactive oxygen species (ROS) is likely to induce the activation of the expression of UCA1 in HCV infected cells. Indeed, ROS inhibits C/EBPα, a negative regulator of UCA1, and stabilizes the UCA1 inducer HIF-1α, leading to UCA1 induction (Miura et al., 2008; Nishina et al., 2008).

On the other hand, several lncRNAs have been described that are upregulated both in HCV-infected cells and in cells treated with IFN or PAMPs such as poly(I:C), or when cells are infected with viruses different than HCV. Therefore, they can be considered to be lncRNAs upregulated by the antiviral response.

As described above, several IFN-induced lncRNAs have been identified after comparing the transcriptome of cells treated with IFN and controls (Carnero et al., 2014; Kambara et al., 2014, 2015; Barriocanal et al., 2015). This is the case for NRIR and BISPR (see above), ISR2 (IFN-stimulated lncRNA2), ISR8 and lncISG15 (Carnero et al., 2014; Barriocanal et al., 2015). Interestingly, the loci of the above lncRNAs are neighboring those of ISGs that play a key role in HCV infection (CMPK2 and Viperin, BST2, GBP1, IRF1 and ISG15, respectively). As detailed above, NRIR, which is a negative regulator of the IFN response, is significantly upregulated in the liver of HCV-infected patients compared to healthy controls (Kambara et al., 2014). Therefore, HCV may use NRIR to increase replication. Similarly, ISR2, ISR8, lncISG15 and BISPR are upregulated in liver and cultured cells infected with HCV compared to controls (Carnero et al., 2014; Barriocanal et al., 2015). Unexpectedly, HCV-induction of ISR2 and ISR8 is higher than induction with other viruses such as influenza, adenovirus or SFV, or mutant versions that allow induction of a strong antiviral response (Carnero et al., 2014). This suggests that HCV may have an additional mechanism distinct from the IFN response for inducing the expression of these lncRNAs. Although their cellular function has not yet been described, guilt-by-association studies predict that ISR8 is an antiviral factor that induces the immune system and the antiviral response and ISR2 regulates the action of antiviral sensors and IFN activation (Carnero et al., 2014). In the case of BISPR,


The genomic position, name and alternative names are indicated. The fold induction has been quantified in cells treated with IFN, poly(I:C) or in cells infected with HCV versus untreated cells.

increased expression by IFN leads to higher levels of BST2 and decreased virion release (Neil et al., 2008; Dafa-Berger et al., 2012; Barriocanal et al., 2015; Kambara et al., 2015). Thus, induction of IFN-induced lncRNAs may have positive effects (in the case of NRIR) or negative effects (in the case of BISPR) on HCV replication.

STAT3 transcription factor, a negative regulator of the IFN pathway, can be induced by IFN, growth factors, stress, several cytokines such as IL6 and by HCV infection (Wang et al., 2011). HCV core and other viral proteins induce ROS and activate STAT3 by several mechanisms (Yoshida et al., 2002; Waris et al., 2005; Machida et al., 2006). STAT3 activation, in turn, favors viral replication by blocking type I IFN pathway and through positive regulation of microtubule dynamics (McCartney et al., 2013). More recently, it has been shown that several lncRNAs are induced by STAT3 activation, including lncIGF2-AS and lnc7SK, which help MW formation by increasing the level of phosphatidylinositol 4-phosphate kinase (Xiong et al., 2015). Therefore, in this manner, both the IFN response and HCV infection benefit viral replication via STAT3-mediated induction of expression of lncRNAs.

Finally, a group of lncRNAs are induced 3–30 fold by IFN or PAMPs such as poly(I:C) and up to 100 times more by HCV infection (Carnero et al., 2016) (**Table 2**). These lncRNAs were identified by transcriptome analysis of cultured liver cells with and without HCV infection and were named CSRs, after HCV Stimulated lncRNAs.

As mentioned above, the highest upregulation of these CSRs has been observed when cells are infected with HCV. However, it has been shown that infection with other viruses also leads to the induction of their expression. DNA viruses such as adenovirus and RNA viruses such as influenza, SFV, HCV or mutant versions of influenza that allow IFN induction, upregulate CSR3, 7 and 31. On the other hand, CSR6 is only induced by adenovirus, CSR20 by influenza virus infections. Similarly, CSR32/EGOT is upregulated in response to RNA viruses (HCV, influenza, SFV) but not DNA viruses (adenovirus or HBV). Existing data suggest that the expression of EGOT (see above) is upregulated through sensing of HCV viral RNA in the cytoplasm. Therefore, increased levels of EGOT are detected shortly after infection even when UV inactivated non-replicative viruses are used (Carnero et al., 2016). Later in the infectious cycle, EGOT levels are increased in response to viral replication. Several sensor molecules are required for EGOT induction, including RIG-I and the noncanonical sensor PKR, which induce transcription through IRF3 and NF-κB, with the latter being required for EGOT transcription. Indeed, there is a good correlation between the levels of TNFα, which induces NF-κB, and cellular level of EGOT in liver tissues derived from HCV-infected patients, suggesting that TNFα could be a major driver of EGOT expression in the liver (Carnero et al., 2016).

As expected from a negative regulator of the IFN response (see above), EGOT depletion leads to decreased levels of viral genome and proteins and viral titers in HCV and SFV-infected cells (Carnero et al., 2016), likely due to increased levels of ISGs, such as GBP1, ISG15, Mx1, BST2, ISG56, IFI6 and IFITM1. Interestingly, some of the ISG targets of EGOT have been described as inhibitors of HCV or SFV entry, replication or release (Landis et al., 1998; Itsui et al., 2009; Raychoudhuri et al., 2011; Wilkins et al., 2013; Amet et al., 2014; Ooi et al., 2015). Kinetic experiments show that EGOT silencing leads first to increased levels of ISGs and then to decreased replication of HCV or SFV genomes. As discussed above, this repressive role of EGOT resembles what has been described for the lncRNAs NRIR and NRAV, negative regulators of the IFN pathway which are induced by IFN or infection (Kambara et al., 2014; Ouyang et al., 2014). Similar to what has been described for EGOT, downregulation of lncCMPK2/NRIR or NRAV lncRNA activates ISG transcription and inhibits HCV or influenza replication, respectively.

## CONCLUSION

In vitro studies on the role of lncRNAs in the IFN response in the absence of pathogens point to the presence of a strong perturbation in the expression of the non-coding transcriptome following IFN stimulation. From the very small number of studies on the role of lncRNAs in the antiviral response, it is evident that changes in the expression of this class of cellular effectors do play a critical role in negative regulation of the downstream steps of the IFN response. Together with proteins that act as negative regulators, this class of lncRNAs create a complex, decentralized regulatory network with overlapping and likely partially redundant functions which allows for extreme fine tuning of the magnitude and duration of the IFN response. However, in the presence of pathogens such as HCV, such negative feedback loops can favor viral replication and in some cases are actively hijacked by the virus to help in viral survival. There are two possible scenarios that can explain this from an evolutionary perspective. One possibility is that the coevolution of the IFN response and pathogens has tipped in favor of viral survival due to the faster rate of viral evolution, with the viruses gaining the ability to exploit the immune response. Alternatively, it is plausible that at least in the case of some chronic infections, such a reduction in the magnitude of the immune response may be beneficial for the organism through limiting the damage incurred on the infected tissue. Future studies, by providing a more complete picture of the interplay of lncRNAs and viral infections, will shed additional light on the highly complex interplay of the antiviral response and pathogens.

## AUTHOR CONTRIBUTIONS

SV and PF wrote and reviewed the manuscript. PF built the table and SV the figures.

## FUNDING

This work was supported by grants from the MINECO/FEDER, EU (SAF2012-40003, SAF2015-70971-R), Instituto de Salud Carlos III (ISCIII)/FEDER (PI16/0845), grant Ortiz

de Landazuri from the Government of Navarra, Fundacio La Marato de TV3 (20132132), and by the project RNAREG [CSD2009-00080], funded by the MINECO

#### REFERENCES


under the program CONSOLIDER INGENIO 2010 to PF and NIH grants 1R01AI120204-01 and 1R21AI127252-01 to SV.



the Toll-like receptor 3 adaptor protein TRIF. Proc. Natl. Acad. Sci. U.S.A. 102, 2992–2997. doi: 10.1073/pnas.0408824102




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Valadkhan and Fortes. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Modulation of Lipid Droplet Metabolism—A Potential Target for Therapeutic Intervention in Flaviviridae Infections

Jingshu Zhang1†, Yun Lan1† and Sumana Sanyal 1, 2 \*

*<sup>1</sup> HKU-Pasteur Research Pole, School of Public Health, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong, China, <sup>2</sup> School of Biomedical Sciences, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong, China*

#### Edited by:

*Wesley H. Brooks, University of South Florida, United States*

#### Reviewed by:

*Abdulsamie Hanano, Atomic Energy Commission of Syria, Syria Kohji Moriishi, University of Yamanashi, Japan Richard Allen White III, Idaho State University, United States*

#### \*Correspondence:

*Sumana Sanyal sanyal@hku.hk These authors have contributed equally to this work.*

*†*

#### Specialty section:

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology*

Received: *28 September 2017* Accepted: *06 November 2017* Published: *28 November 2017*

#### Citation:

*Zhang J, Lan Y and Sanyal S (2017) Modulation of Lipid Droplet Metabolism—A Potential Target for Therapeutic Intervention in Flaviviridae Infections. Front. Microbiol. 8:2286. doi: 10.3389/fmicb.2017.02286* Lipid droplets (LDs) are endoplasmic reticulum (ER)-related dynamic organelles that store and regulate fatty acids and neutral lipids. They play a central role in cellular energy storage, lipid metabolism and cellular homeostasis. It has become evident that viruses have co-evolved in order to exploit host lipid metabolic pathways. This is especially characteristic of the *Flaviviridae* family, including hepatitis C virus (HCV) and several flaviviruses. Devoid of an appropriate lipid biosynthetic machinery of their own, these single-strand positive-sense RNA viruses can induce dramatic changes in host metabolic pathways to establish a favorable environment for viral multiplication and acquire essential components to facilitate their assembly and traffic. Here we have reviewed the current knowledge on the intracellular life cycle of those from the *Flaviviridae* family, with particular emphasis on HCV and dengue virus (DENV), and their association with the biosynthesis and metabolism of LDs, with the aim to identify potential antiviral targets for development of novel therapeutic interventions.

#### Keywords: lipid droplet, lipid metabolism, HCV, flavivirus, dengue

## INTRODUCTION

Cellular homeostasis is maintained by a constant metabolic energy flux. As one of the major energy sources, lipids are synthesized, modified and utilized through various pathways. Lipid droplets (LDs) are ubiquitous and conserved cytoplasmic compartments delineated by a phospholipid monolayer, and serve as energy reservoirs in almost all living organisms. Excess lipids are packaged, stored and distributed in LDs, an organelle which is not only important in lipid storage and metabolism, but protein quality control, pathogenesis, and immune responses (Walther and Farese, 2012).

Since viruses lack the appropriate machinery to conduct their own lipid synthesis, most have evolved mechanisms to hijack host lipid metabolic pathways (including LDs) for completing their intracellular replication cycles. Hepatitis C virus (HCV) has long been demonstrated to do so (Paul et al., 2014). Apart from the cell biology underlying infection, the interplay between viral infection and host lipid metabolic pathways is important not only to elucidate the pathogenicity of this category of viruses but also to assess how they can be targeted as a general means of combating infections.

As a consequence of development of gene editing and mass spectrometry based lipidomics and proteomics technologies, an increasing body of evidence indicates the involvement of host LDs at different steps of the intracellular life cycle of HCV and flaviviruses (Martín-Acebes et al., 2016b). Here, we have

**Abbreviations:** 769662, 6,7-Dihydro-4-hydroxy-3-(2′ -hydroxy[1,1′ -biphenyl]- 4-yl)-6-oxo-thieno[2,3-b]pyridine-5-carbonitrile; AADAC, arylacetamide deacetylase; ACAT, acyl-CoA, cholesterol acyltranserases; ACC, acetyl coenzyme a carboxylase; ADRP, adipose differentiation-related protein; AICAR, aminoimidazole carboxamide ribonucleotide; AMPK, 5′ AMP-activated protein kinase; ApoB100, apolipoprotein B100; ARF, ADP-ribosylation factor; ARF1-COP I, ADP-ribosylation factor-coat protein I; ATGL, adipose triglyceride lipase; AUP1, ancient ubiquitous protein 1; AY9944, trans-1,4-bis(2-Chlorobenzylaminoethyl) cyclohexane dihydrochloride; BA, 2-chloro-5-nitro-N-(pyridyl)benzamide; BAPTA-AM, 1,2-Bis(2-aminophenoxy)ethane-N,N,N′ ,N′ -tetraacetic acid tetrakis(acetoxymethyl ester); BMS-200150, 2-[1-(3,3-diphenylpropyl)-4 piperidinyl]-2,3-dihydro-1H-isoindol-1-one; C75, 4-Methylene-2-octyl-5 oxotetrahydrofuran-3-carboxylic acid; CD2AP, CD2 associated protein; CE, cholesterol ester; CerS, ceramide synthase; CGI58, comparative gene identification-58; CHO, cholesterol; CMA, chaperone-mediated autophagy; COP, coat protein; CPT-1, carnitine palmitoyltransferase 1; D609, tricyclodecan-9 yl-xanthogenate; DAG, diacylyglycerol; DCI, 3,2-trans-enoyl-CoA iosmerase; DDX3X, DEAD box polypeptide 3 X-linked; DENV, dengue virus; DGAT, diacylglycerol acyltransferases; DHCR, dehydrocholesterol reductase; DHF, dengue hemorrhagic fever; DMV, double-membrane vesicle; DNJ, 1 deoxynojirimycin; dsRNA, double-stranded RNA; DSS, dengue shock syndrome; E protein, envelope protein; ER, endoplasmic reticulum; ERAD, ER-associated degradation; ERGIC, ER–Golgi intermediate compartment; FAPP2, Golgiassociated four-phosphate adaptor protein 2; FASN, fatty acid synthase; GBF1, golgi-specific brefeldin A-resistance guanine nucleotide exchange factor 1; GGTase I, geranylgeranyltransferase; GGTI-286, N-4-[2(R)-Amino-3-mercaptopropyl]amino-2-phenylbenzoyl-(L)-leucine methyl ester; GGTI-298, N-4-[2(R)-Amino-3-mercaptopropyl]amino-2-naphthylbenzoyl-(L)-leucine methyl ester trifluoroacetate salt; GW4869, N,N′ -Bis[4-(4,5-dihydro-1Himidazol-2-yl)phenyl]-3,3′ -p-phenylene-bis-acrylamide dihydrochloride; HCV, hepatitis C virus; HMGCoA, 3-hydroxy-3-methylglutaryl CoA; HMGCR, 3-hydroxy-methyglutaryl-Coenzyme A reductase; HMGCS, hydroxymethylglutaryl-CoA synthase; HSL, hormone-sensitive lipase; IKK, IkB kinase; JEV, Japanese encephalitis virus; LD, lipid droplet; LY294002, 2-(4-Morpholinyl)-8-phenyl-4H-1-benzopyran-4-one; MAPK, mitogenactivated protein kinases; MEDICA 16:3,3,14,14-Tetramethylhexadecanedioic acid; MGL, monoglyceride lipase; MS-209, dofequidar fumarate; MTase, methyltransferase; MTOC, microtubule-organizing center; MTP, microsomal triglyceride transfer protein large subunit; MVD, mevalonate (diphospho) decarboxylase; MβCD, methyl-β-cyclodextrin; NIM811, N-methyl-4-isoleucine cyclosporine; NB-DNJ, N-Butyldeoxynojirimycin; NIs, nucleoside/nucleotide analog inhibitors; NNIs, non-nucleoside inhibitors; NS, non-structural; nSMase2, neutral sphingomyelinase 2; OSBP, oxysterol-binding protein; OSC, oxidosqualene cyclase; OSW-1, 3β,16β,17α-trihydroxycholest-5-en-22-one 16-O-(2-O-4-methoxybenzoyl-β-D-xylopyranosyl)-(1→ 3)-(2-O-acetyl-α-larabinopyranoside); PC, phosphatidylcholine; PDMP, D,L-threo-1-phenyl-2 decanoylamino-3-morpholino-1-propanol; PE, phosphatidylethanolamine; PERL, polyunsaturated ER liposomes; PF-429242, 4-[(Diethylamino)methyl]- N-[2-(2-methoxyphenyl)ethyl]-N-(3R)-3-pyrrolidinyl-benzamide; PI3K, phosphatidylinositide 3-kinases; PI4KA, phosphatidylinositol 4-kinases; PI4P, phosphatidylinositol 4-phosphate; PLA2G4A, cytosolic phospholipase A2; PNPLA5, patatin-like phospholipase domain-containing protein 5; PPARα, peroxisome proliferator-activated receptor α; RdRp, RNA-dependent RNA polymerase; S1P, site 1 protease; SAM, S-adenosylmethionine; SCAP, Sterol regulatory element-binding protein cleavage-activating protein; SCPI-1, [N-(4-{[4-(3,4-dichlorophenyl)-1,3-thiazol-2-yl]amino} phenyl)acetamidehydrobromide]; SCP-2, sterol carrier protein 2; SKI-1, subtilisin/kexin-isozyme-1; Smase, sphingomyelinase; SphK, sphingosine kinase; SPP, signal peptide peptidase; SPT, serine palmitoyltransferase; SQS, squalene synthetase; SRB1, scavenger receptor class B member 1; SREBP, sterol regulatory element–binding protein; TAG, triacylglycerol; TBEV, cataloged these interactions and anticipate that this knowledge will be beneficial for identification of host factors as suitable targets for antiviral interventions.

#### LIPID DROPLET—A MULTIFUNCTIONAL ORGANELLE

## Morphology and Composition of LDs

LDs are essentially the emulsion phase of insoluble oil droplets dispersed in aqueous cytoplasm. Compared to other cellular organelles with double-layered membranes, the structure of LDs is rather unique, containing a hydrophobic core and a single layer of amphipathic phospholipids. The neutral lipid core contains predominantly triacylglycerols (TAGs) and cholesterol esters (CEs) (Thiam et al., 2013). Although the composition of the phospholipid monolayer varies in different cell types, phosphatidylcholine (PC) and phosphatidylethanolamine (PE) are the two major phospholipids. The morphology and consumption of LDs are drastically altered by the composition of their phospholipid monolayer (Guo et al., 2008). The surface of the monolayer is decorated with LD-associated proteins, including lipolytic enzymes such as hormone-sensitive lipase (HSL), adipose triglyceride lipase (ATGL), and PAT-domain family (**p**erilipin, **a**dipophilin and **T**IP47) (Tauchi-Sato et al., 2002; Ohsaki et al., 2006; Wilfling et al., 2014a). Despite being present in nearly all cell types across different organisms, LDs are highly heterogeneous and dynamic with varied numbers and sizes (ranging from 100 nm to 100 mm in diameter) in otherwise identical cells. Even within the same cell, LDs expand or shrink in response to cellular signals.

#### Biogenesis of LDs

In eukaryotes, LDs respond to increased cellular fatty acid levels and emerge from the accumulation of neutral lipids in the ER, which harbors enzymes necessary for neutral lipid synthesis in most cell types (Buhman et al., 2001; Pol et al., 2004). First established as an oil-in-water emulsion, the small nascent LDs undergo a series of well-organized processes and grow into larger, mature LDs. The final steps of TAG and CE synthesis are catalyzed by ER-localized diacylglycerol acyltransferases (DGATs) and acyl-CoA:cholesterol acyltranserases (ACATs), respectively. The continuous accumulation of the newly synthesized TAGs and CEs at specific sites at the ER results in separation of two phases, where globules of TAGs arise between the two leaflets of the bilayer and eventually dissociate. DGAT2, which is inserted into one leaflet of the ER membrane, is transported to LDs where it continues to catalyze synthesis of TAGs, hence promoting further growth of LDs (Kassan et al., 2013; Wilfling et al., 2013). This process is thermodynamically enabled by the unique phospholipid monolayer structure of LDs.

tick-borne encephalitis virus; TOFA,5-tetradecyl-oxy-2-furoic acid; U0126, 1,4-Diamino-2,3-dicyano-1,4-bis(2-aminophenylthio)butadiene; U18666A, 3β- (2-Diethylaminoethoxy)androst-5-en-17-one; VAP, vesicle-associated membrane protein-associated protein; VLDL, very low-density lipoprotein; WNV, West Nile virus; YFV, yellow fever virus; ZIKV, Zika virus.

#### The Multifunctionality of LDs

Long been regarded as simple and passive lipid storage compartments, LDs are currently considered highly dynamic and complex. They play a central role in lipid metabolism and are connected to diverse cellular processes like fatty acid trafficking, cellular signaling, protein storage, autophagy, immunity, and virus replication (Singh et al., 2009; Saka and Valdivia, 2012; Rambold et al., 2015; Welte, 2015; Velázquez et al., 2016).

#### LDs as the Central Regulator for Cellular Homeostasis

As metabolically active organelles, LDs regulate the balance between host lipid synthesis and mobilization to maintain cellular homeostasis. Catalyzed by DGAT1 and DGAT2, cellular fatty acids together with diacylyglycerols (DAGs) are converted into TAGs and stored in LDs. TAGs can be further hydrolyzed to generate DAGs or phosphatidic acid (PA) for membrane phospholipid synthesis and free fatty acids (FFAs) for energy production (Pol et al., 2014).

#### LDs as Transient Protein Storage Compartments for Degradation

Due to unique structural features and proximity to the ER, the surface of LDs can also serve as transient storage depots for proteins that are destined for degradation via the ERassociated degradation (ERAD) pathway (Gao and Goodman, 2015). Misfolded proteins in the ER are removed and degraded by the ubiquitin–proteasome system. Current evidence suggests that ubiquitinated apolipoprotein B100 (ApoB100) (Ohsaki et al., 2008) and 3-hydroxy-3-methylglutaryl CoA reductase (HMGCR) (Hartman et al., 2010) are likely degraded on the surface of LDs through proteasomal and autophagic pathways (Ohsaki et al., 2006). HMGCR is one of the rate-limiting enzymes for cholesterol synthesis in mammalian cells. Ubiquitination of HMGCR is mediated by ancient ubiquitous protein 1 (AUP1), a highly conserved monotopic membrane protein localized to both LDs and the ER membrane (Spandl et al., 2011). AUP1 recruits the ubiquitin-conjugating enzyme UBE2G2 to LDs and facilitates its binding with the ER ubiquitin ligases gp78 and Trc8, which subsequently initiates the ubiquitination/degradation of HMGCR resulting in inhibition of cholesterol synthesis (Jo et al., 2013). Apart from providing a molecular link between LDs and the ubiquitination machinery, monoubiquitinated AUP1 was reported to induce LD clustering, a widespread phenomenon observed in multiple cell types across all species (Lohmann et al., 2013). LDs may also provide sequestration platforms for protein storage (Cermelli et al., 2006), such as during the synthesis of eicosanoids, a class of signaling molecules that use LDs as distinct sites for eicosanoid generation (Bozza et al., 2011).

#### Mobilization of Lipids from LDs

Depending on the cell type, starvation and/or physiological conditions, eukaryotic cells mobilize lipids stored in LDs via two major pathways termed lipolysis and lipophagy. In mammalian adipocytes, lipolysis is activated in response to changes in cellular energy and hormone levels. This allows transient docking and activation of three major lipolytic enzymes, ATGL, HSL, and monoglyceride lipase (MGL) which co-ordinate the hydrolysis of TAGs for energy production (Karlsson et al., 1997; Zimmermann et al., 2004; Dugail and Hajduch, 2007; Lass et al., 2011). Perilipins localize to LD surfaces and under basal conditions shield TAGs from cytosolic lipases. During starvation, perilipins are degraded via the chaperone-mediated autophagy (CMA) pathway to facilitate lipolysis by HSL and ATGL (Brasaemle, 2007; Sztalryd and Kimmel, 2014; Kaushik and Cuervo, 2015). Apart from LD-associated proteins, the ADP-ribosylation factorcoat protein I (ARF1-COPI) vesicular trafficking machinery is likely to play an important role in mediating lipolysis by regulating LD composition and targeting ATGL to LDs (Soni et al., 2009; Wilfling et al., 2014b).

The role of autophagy in regulating lipid metabolism has been intensively studied in recent years (Singh et al., 2009; Singh and Cuervo, 2012). Various cell types have been used to demonstrate the process of LD mobilization via the autophagy pathway, such as hypothalamic neurons (Kaushik et al., 2011), glial cells (Martinez-Vicente et al., 2010), and enterocytes (Narabayashi et al., 2015). Autophagy is a conserved cellular process that delivers cytoplasmic contents, including dysfunctional proteins, and excess or damaged organelles to lytic compartments for degradation and recycling. The process can be induced by a number of factors such as ER stress, cellular starvation, and pathogenic infection. Available data support that three distinct types of autophagy can be triggered: macro-, micro- and chaperone-mediated autophagy , amongst which, macroautophagy is the best characterized (Yoshimori, 2004; Mizushima, 2007). Upon activation, cytoplasmic components are first enclosed by a double-layered vesicular structure termed autophagosome, which fuse with lysosomes where internal cargos are degraded (Mizushima, 2007). Multiple factors such as nutrient deprivation, virus infection, and sterol (cholesterol) depletion, can trigger degradation of LDs through the autophagic machinery (Ouimet and Marcel, 2012). LC3II, a structural component of the autophagosomes, and autophagyrelated proteins Atg2, Atg5, and Atg7 are recruited to the surface of LDs to form autophagosomes. LDs are engulfed for lysosomal degradation to release stored lipids, which undergo mitochondrial β-oxidation for energy production. This process is frequently manipulated by flaviviruses to promote their replication (see Usage of LD as an energy reservoir during viral life cycle) (Singh et al., 2009; Heaton and Randall, 2010; Fujimoto and Parton, 2011; Velikkakath et al., 2012). The level and distribution of cellular cholesterol is tightly regulated; excess free cholesterol stored as cholesteryl esters in LDs are hydrolyzed during sterol starvation through autophagy (Cheng et al., 2006; Ouimet and Marcel, 2012). Sterol regulatory element-binding proteins (SREBPs) are the central transcriptional regulators of cholesterol metabolism and lipogenesis. In the presence of high cholesterol content in the cytoplasm, SREBP binds to sterol regulatory element-binding protein cleavage-activating protein (SCAP) and the ER-associated protein Insig. Upon reduction of cellular cholesterol below a threshold, Insig is degraded through the ERAD pathway, the SCAP-SREBP complex is transported to the Golgi, where SREBP undergoes intramembrane proteolysis and translocates to the nucleus. This mature form of SREBP initiates transcription of a series of down-stream genes involved in the biosynthesis of cholesterol (Brown and Goldstein, 1997; Yang et al., 2002).

## THE FLAVIVIRIDAE FAMILY

Viruses of the Flaviviridae family are enveloped single-strand positive-sense RNA viruses, with the nucleocapsids surrounded by two or more types of envelope glycoproteins and lipid bilayers (Lindenbach et al., 2007; Paul and Bartenschlager, 2015). It comprises several different genera including Hepacivirus (e.g., HCV), Flavivirus [e.g., Zika virus (ZIKV), dengue virus (DENV)], Pegivirus, and Pestivirus.

Persistent infection with HCV in humans can develop into serious liver diseases, including fibrosis and liver cirrhosis, which could further progress into hepatocellular carcinoma (Bartenschlager et al., 2013). Medically-relevant flaviviruses, including yellow fever virus (YFV), ZIKV, DENV, West Nile virus (WNV), and Japanese encephalitis virus (JEV), are usually arboviruses (viz., transmitted by arthropods, mainly mosquitoes and ticks) that are responsible for severe mortality in humans and animals worldwide. DENV and YFV infections are known to cause vascular leakage and hemorrhage in some infected patients (Siqueira et al., 2005; Garske et al., 2014; Thanachartwet et al., 2015). JEV and WSN infections on the other hand, tend to cause neurological diseases (Sarkari et al., 2012; Samaan et al., 2016). ZIKV infection is associated with serious birth defects microcephaly in particular—and other neurological disorders (Petersen et al., 2016). Although there has been significant progress in therapeutic interventions for HCV and some other flaviviruses (for example YFV), there is still an urgent need for vaccines and medications against others such as DENV and ZIKV. Additionally, the ever-increasing geographical spread and number of outbreaks caused by these pathogens pose a considerable threat to public health (Gould and Solomon, 2008).

Despite significant differences in transmission, tissue tropism and pathogenesis, viruses of the Flaviviridae family employ similar intracellular replication strategies. After receptormediated endocytosis, the acidic environment in the endosomes triggers fusion between the virion lipid envelope and cellular membranes, followed by viral uncoating. The viral RNA is subsequently released into the cytoplasm and used directly as mRNA for translation of the viral polyprotein. Host and viral proteases cleave the newly synthesized viral polyprotein to generate the structural and non-structural (NS) proteins (Lindenbach et al., 2007). Viral replicase proteins together with other host factors induce massive rearrangements of intracellular membranes to form organelle-like membrane-delineated compartments for efficient RNA replication. At the replication sites, the positive-sense RNA is used as template to generate the negative-sense RNA intermediate, while multiple positive-sense progeny RNAs are produced to be incorporated into nascent virus particles (Paul and Bartenschlager, 2015). Progeny virions are assembled in close proximity to the ER and LDs, and appear to bud into the ER-lumen, followed by transport through the host secretory pathway where they undergo further maturation, and are eventually released from the cell surface (Lindenbach et al., 2007; Paul and Bartenschlager, 2015; **Figure 1**).

## INFLUENCE OF LD METABOLISM ON THE VIRUS LIFE CYCLE

HCV has historically been used for studying interactions between LD metabolism and the viral life cycle. Others from the same family, such as DENV, have recently started receiving more attention in this regard. The magnitude and complexity of these interactions underscore the significance of targeting LD metabolism to control viral infection. As a dynamic cellular lipid storage organelle, LDs participate in the viral life cycle by providing intracellular membrane surface area, lipids, energy, and proteins.

## Contribution of LDs in Virus Replication and Assembly

Upon infection massive intracellular membrane rearrangements are induced by perturbing lipid biosynthetic pathways to spatially segregate the replication and assembly sites (Welsch et al., 2009; Romero-Brey et al., 2012). On the one hand, the two sites need to be separated to avoid competition between the capsid protein and the viral replicase complex at the level of RNA binding. On the other hand, newly synthesized positive-sense progeny RNAs need to be transported from the replication to the assembly sites, where the capsid protein is concentrated. For maximum efficiency in virus assembly the two sites require close proximity to each other (Welsch et al., 2009; Romero-Brey et al., 2012; **Figure 2**).

Association of LDs to Viral Replication Compartments LDs have been reported to associate with virus-induced membrane bound compartments believed to be replication sites. Despite belonging to the same family, HCV and DENV induce morphologically distinctive replication compartments. In the case of HCV infection, the double-membrane vesicles (DMVs) are derived from the ER (Romero-Brey et al., 2012; **Figure 2A**). DMVs are composed of active viral replicase proteins and double-stranded RNA (dsRNA), along with several host components including vesicle-associated membrane proteinassociated protein A (VAP-A) and VAP-B that are crucial for viral RNA replication (Evans et al., 2004; Gao et al., 2004). The highly hydrophobic NS4B of HCV, together with NS5A, are the major viral factors that contribute to DMV formation (Lundin et al., 2006). These virus-induced compartments use cholesterol as a structural component and can be visualized in close proximity to LDs (Romero-Brey et al., 2012; Paul et al., 2013). While DMVs are considered as replication factories of HCV, their association to LDs is still unclear. The interferon-induced antiviral protein viperin, which inhibits HCV RNA replication, localizes to LDs using a similar mechanism as HCV NS5A, indicating the importance of LD-NS5A association during HCV replication (Jiang et al., 2008; Hinson and Cresswell, 2009). LDs release free cholesterol from the esterified form for membrane biogenesis as per the host cellular requirements (Maxfield and Tabas, 2005) and, therefore, may serve as reservoirs for lipids required for

expanding the intracellular membrane surface to form DMVs (see Association of LDs to Viral Replication Compartments). Besides, HCV replication triggers the activation of the cellular SREBP pathway for de novo synthesis of membrane lipids, which, in turn, regulate biogenesis of LDs (see Manipulation of LD Reserves during Viral life Cycle) (Li et al., 2013). Another possibility is that LDs themselves provide a platform for virus assembly and, therefore, require close proximity to the replication sites for efficient recruitment of newly synthesized viral proteins and subsequent virion packaging (see LDs as a Platform for Virion Assembly) (Miyanari et al., 2007).

Unlike HCV, DENV infection induces formation of singlemembrane in-folding into the ER lumen and unstructured convoluted membranes (Welsch et al., 2009; **Figure 2B**). These DENV-induced vesicle-like structures contain viral replicase and dsRNA. Pore-like openings on these structures enable release of newly synthesized viral RNA, facilitating replication and efficient encapsidation (Welsch et al., 2009). Other flaviviruses, such as WNV and tick-borne encephalitis virus (TBEV) share similar features of intracellular membrane rearrangements (Gillespie et al., 2010; Miorin et al., 2013). DENV replication activates the autophagy pathway to mobilize FFAs from LDs and co-opts FA synthase (FASN). FFAs released from LDs are consumed by oxidation in mitochondria to generate ATP, which is required for viral RNA replication (see Usage of LD as an Energy Reservoir during Viral life Cycle) (Heaton and Randall, 2010). Moreover, DENV NS3 recruits FASN to virus replication sites during membrane remodeling in a Rab18-dependent fashion, engaging both LDs and the viral replication complexes in the process (Heaton et al., 2010; Tang et al., 2014). Regardless of the distinct membrane compartmentalization strategies of HCV and DENV both require close juxtaposition of LDs for energy supply and subsequent virion assembly, as reviewed below.

#### LDs as a Platform for Virion Assembly

In the case of HCV infection, after being generated at the ER, the capsid protein localizes to LDs via its domain 2 in a time-dependent manner. They accumulate on discrete regions of LDs before fully covering the surface of LDs (Boulant et al., 2007; Shavinskaya et al., 2007). Host DGAT1 that synthesizes triglycerides stored within LDs, binds to the HCV capsid protein, which in turn acquires access to DGAT1-generated LDs. Viral RNA replication complexes are subsequently recruited to appropriate sites of virus assembly. LD-localized capsid protein provides stability to these structures via interfering with TAG turnover and inducing aggregation of LDs (Boulant et al., 2008; Herker et al., 2010; Harris et al., 2011). Additionally, by replacing LD-localized ADRP, the capsid protein induces imbalance between the minus-end-directed and the plus-enddirected motors, causing movement of LDs on microtubules toward the nucleus so as to enhance interactions between sites of HCV RNA replication and virion assembly (Boulant et al., 2008). The capsid protein recruits viral NS5A, while the Nterminal of NS5A engages viral replication complexes to LDassociated membranes (Boulant et al., 2007; Appel et al., 2008). HCV NS5A also associates with Rab18, a member of the Rab

FIGURE 2 | LDs as platforms for virion assembly in (A) HCV and (B) DENV infection. (A) (1) ADRP-coated LDs are able to interact with microtubules and move toward both plus and minus ends. (2) During HCV infection, viral capsid protein replaces ADRP from LD surface with the assistance of DGAT1. (3) As the consequence of losing ADRP, LD losses the balance of mobility, moving toward MTOC and nucleus. (4) Clustering of LDs at the peripheral of nucleus enables the contact between LDs and the replication complex of HCV. HCV RNA is delivered from ER-bound replication complexes to NS5A, obtaining access to LD surface, followed by nucleocapsid formation (gray-dashed frame and enlarged panel). (5) The nucleocapsid fuses with VLDL to form viral lipoviroparticle in ER. (B) (1) At the ER–Golgi intermediate compartment (ERGIC), ARF1 and its guanine nucleotide exchange factor (GEF) GBF1 together with COPI deliver ATGL and ADRP from ER export sites (ERES) to the surface of LD. DENV subverts this process for the transportation of capsid protein to LD surface. (2) The accumulation of DENV capsid protein on LDs associates with cellular perilipin 3 and intracellular K<sup>+</sup> concentration. (3) Replicated DENV genomes are released through the vesicle pore and then engaged into nucleocapsids that bud through the ER membrane in close proximity. (4) Capsid protein can be released from LDs to the cytosol or other cellular compartments for subsequent viral assembly (gray-dashed frame and enlarged panel). (5) Packed virions accumulate within the lumen of the vesicle packets-containing ER network before transported to Golgi (Boulant et al., 2008; Chatel-Chaix and Bartenschlager, 2014).

GTPase family that plays an essential role in membrane traffic (Salloum et al., 2013). Rab18 localizes directly to the monolayer surface of LDs in response to lipolytic stimulation (Martin et al., 2005), and facilitates association of NS5A and other replicase components with LDs (Salloum et al., 2013; **Figure 2A**). HCV infection increases the expression of apolipoprotein J, which further stabilizes LD-associated capsid protein and NS5A, thereby facilitating virion assembly (Lin et al., 2014). Cellular CD2 associated protein (CD2AP) also regulates HCV assembly by interacting with HCV NS5A while modulating LD biogenesis at the same time (Li, 2017). Dissociation of HCV capsid protein from LDs has no effect on viral RNA replication but decreases production of infectious virions, indicating that LDs either directly provide a platform for HCV assembly or facilitate transport of the capsid protein from RNA translation/replication to the assembly sites (Boulant et al., 2007, 2008; Miyanari et al., 2007). Additionally, during chronic HCV infection, LDs in liver tissues increase in number and size, causing pathological accumulation of liver lipids, also known as hepatic steatosis. The interaction between the HCV capsid protein and LDs is critical for this development. An LD membrane protein, perilipin 3, regulates the capsid-induced steatosis, indicating host LDassociated proteins as an effective preventive measure of HCVinduced pathology (Ferguson et al., 2017).

The DENV capsid protein also interacts with LDs but in a mechanistically distinct manner as compared to HCV. DENV capsid protein accumulates on the surface of LDs via its center domain and the N-terminal disordered region (Samsa et al., 2009; Martins et al., 2012). Additionally, the binding between DENV capsid protein and LDs may also be attributed to the association between capsid protein and LD membrane protein perilipin 3 in a potassium ion-dependent fashion. Changing the concentration of potassium ion concentration regulates the binding and release of capsid protein from LDs. This phenomenon indicates that DENV may manipulate specific intracellular ion concentrations to favor viral replication (Carvalho et al., 2012). HCV may use the same potassium ion-dependent strategy to interact with LDs via its p7 and NS5A proteins (Carvalho et al., 2012). Contrary to DGAT1-dependent trafficking to LDs, the DENV capsid protein utilizes host Golgi-specific brefeldin A-resistance guanine nucleotide exchange factor 1 (GBF1)-ARF-COPI pathway to localize to the surface of LDs (Iglesias et al., 2015; **Figure 2B**). Similar to HCV infection, inhibiting the association between DENV capsid protein and LDs results in attenuated infectious virion production but not viral RNA replication, underscoring the function of LDs as a scaffold for DENV assembly through exposure of the protein cationic surface toward the aqueous environment (Carvalho et al., 2012).

DENV and HCV capsid proteins use distinct mechanisms for LD association. The process by which LDs gain or release viral capsid proteins remain unknown. However, current evidence on the involvement of LDs provides several possible targets for developing antiviral approaches (**Table 1**) (section Targeting LD Metabolism as Antiviral Strategies).

## Usage of LD as an Energy Reservoir during Viral Life Cycle

Replication of the viral genome is an energy-consuming process. In HCV infected cells, cytoplasmic ATP levels decrease dramatically, as a result of active energy consumption. Meanwhile, elevated ATP levels at replication compartments within infected cells have also been reported (Ando et al., 2012). This would involve either incorporation of ATPgenerating machinery into the membrane-associated replication site, or transport of ATP though membrane-to-membrane communication between mitochondria and replication compartments (Ando et al., 2012). The C terminus of Flaviviridae NS3 encodes a DExH/D-box RNA helicase that functions to unwind dsRNA molecules through ATP-hydrolysis (Tai et al., 1996; Dumont et al., 2006). Many of the cellular signaling events activated during viral infection are also regulated by ATP levels (Hardie, 2011). Given the highly reduced and hydrophobic lipids at the core, LDs serve as an efficient storage for energy (Walther and Farese, 2012). FA hydrolysis releases 2.5 times more ATPs per gram compared to glucose, which provides a tremendous reservoir for supplying energy during viral replication. Not surprisingly, many other pathogens also manipulate LD metabolism to acquire fuel for replication.

Energy stored in LDs is released through lipolysis. Mobilization of TAG stores from LDs by lipases produces significant amounts of FFAs that can be used in β-oxidation, generating ATP and other intermediates for the cell. In addition to lipolysis, an alternative route through autophagy, commonly referred to as lipophagy, can also take up and deliver LDs to lytic compartments for lipid hydrolysis (see Mobilization of Lipids from LDs) (Wang, 2016).

A model proposed by Randall and Heaton suggested that DENV infection triggers lipophagy to deplete LDs, releasing FFAs. DENV also induces cellular β-oxidation to consume the FFAs released from lipophagy for energy production. Exogenously supplemented FAs can replace the need for lipophagy during DENV replication, suggesting that flaviviruses manipulate cellular lipid metabolism to create an environment that favors virus replication (Heaton and Randall, 2010). Our own data support this model. AUP1, a monotopic membrane protein localized to both LDs and ER membranes, was identified as a key component in DENV biogenesis. Expression of AUP1 was up-regulated during DENV infection and was found to be necessary for virus-triggered lipophagy to proceed (Zhang et al., 2016). The requirement of lipophagy during other flavivirus infections is still to be investigated.

Virus-induced lipophagy for energy production remains unclear in the context of HCV infection. HCV uses membranes of autophagic vacuoles for viral RNA replication. The induction of autophagosomes is nutrient starvation-independent. An impaired autophagy pathway results in attenuated virion production (Dreux et al., 2009; Sir et al., 2012). Proteomic and lipidomic studies showed an up-regulation of lipogenic enzymes and proteins related to β-oxidation, such as 3,2-trans-enoyl-CoA isomerase (DCI) (Diamond et al., 2010). In line with this study, DCI was reported to be essential for productive HCV infection through regulation of mitochondrial FA oxidation (Rasmussen et al., 2011). Another microarray analysis revealed a down-regulation of genes involved in degradation and oxidation of FAs, and an elevation of genes that control metabolism and transport of FAs (Blackham et al., 2010). Although a direct experimental evidence of lipophagy induced by HCV is still missing, data from several indirect sources strongly suggest the utilization of cellular pathways for β-oxidation of FFAs.

## Manipulation of LD Reserves during Viral Life Cycle

Apart from providing FFAs for β-oxidation during Flaviviridae infection, LDs also function as a reservoir for lipids that are essential for viral replication.

Flaviviridae replication organelles consist of FAs, specific phospholipids, sphingolipids, and cholesterol (Heaton et al., 2010; Perera et al., 2012; Paul et al., 2013; Martín-Acebes et al., 2016a). While DENV obtains FAs by breakdown of LDs via lipophagy (Heaton and Randall, 2010), HCV controls the transcriptional induction of lipid biosynthetic and related genes through SREBP signaling (Olmstead et al., 2012). HCV infection activates the SREBP precursor that localizes to the ER, and triggers its trafficking to the Golgi. Thereafter, the SREBP precursor is proteolytically processed by site 1 protease (S1P) and S2P at Golgi, releasing its N-terminal fragment that is transported into the nucleus and initiates transcription of lipogenic factors such as FASN and 3-hydroxy-3-methylglutaryl CoA (HMGCoA). The 3′ untranslated region of the HCV RNA genome with DEAD


Frontiers in Microbiology | www.frontiersin.org


*(Continued)*


box polypeptide 3 X-linked (DDX3X) further activates IκB kinase (IKK)-α, which translocates to the nucleus and stimulates SREBP transcriptional activity, thus modulating LD biogenesis (Olmstead et al., 2012; Li et al., 2013).

HCV replication organelles use cholesterol as a structural component (Romero-Brey et al., 2012; Paul et al., 2013). Cellular oxysterol-binding protein (OSBP) and phosphatidylinositol 4 kinases (PI4KA) facilitate trafficking of cholesterol to the HCV-rearranged membrane-like structures during replication, highlighting the need for both factors in supporting HCV replication (Wang et al., 2014). OSBPs are speculated to be sterol carriers and might function to transport sterols out of the ER and incorporate them into LDs in a phosphatidylinositol 4 phosphate (PI(4)P)-dependent manner. Sterols and cholesterol are exchanged by OSBP at the ER-Golgi interface (Mesmin et al., 2013). OSBP-related protein 2 that resides on the surface of LDs may also participate in the process of lipid exchange (Hynynen et al., 2009). Notwithstanding its cellular function, the activity of OSBP appears to be dispensable for DENV replication (Hynynen et al., 2009). DENV replication is regulated by endogenous cholesterol production that is controlled by mevalonate (diphospho) decarboxylase (MVD) and exogenous cholesterol uptake (Rothwell et al., 2009). Similarly, WNV also hijacks cellular cholesterol and redistributes it to viral RNA replication compartments (Mackenzie et al., 2007).

Besides cholesterol, sphingomyelin is another essential membrane component of HCV replication organelles. An active role for sphingolipids in HCV RNA replication has been reported. Sphingomyelin enhances binding of the RNA dependent RNA polymerase NS5B to the template RNA and is therefore important for HCV replication (Weng et al., 2010; Hirata et al., 2012). Expression of genes that encode sphingomyelin synthases 1 and 2 is up-regulated upon HCV infection, resulting in enhanced synthesis of sphingomyelin (Hirata et al., 2012). Dynamic pools of sphingomyelin were observed in LDs, with the high affinity sphingomyelin-binding protein ADRP on the surface of LDs (McIntosh et al., 2010). It is likely that LDs participate in the biogenesis of sphingolipids necessary for HCV replication.

In addition to consumption of lipids that are stored in LDs, HCV can also obstruct the turnover of LDs to establish a microenvironment that is more favorable to viral infection. Release of infectious HCV particles relies on secretion of hepatic very low-density lipoprotein (VLDL)—a TAG-rich lipoprotein. For hijacking VLDL secretion, HCV inhibits the function of the putative TAG lipase, arylacetamide deacetylase (AADAC), thus, further impairing TAG lipolysis (Nourbakhsh et al., 2013). Moreover, HCV capsid protein that localizes to LDs through the activity of DGAT1 (Harris et al., 2011), restrains lipolysis of TAG by interacting with ATGL and its activator comparative gene identification-58 (CGI-58) (Camus et al., 2014).

As with LD association, distinct strategies are employed by HCV and DENV for mobilizing lipids within LDs, hence providing insights into LD catabolism and cellular factors as possible targets (**Table 1**).

## TARGETING LD METABOLISM AS ANTIVIRAL STRATEGIES

Although viruses of the Flaviviridae family cause severe human diseases, there are currently no clinically approved drugs available for treatment against them, other than for HCV. Historically, the development of antiviral therapy has largely focused on directly targeting viral components involved in multiple stages of the virus life cycle.

Entry of flaviviruses is mediated by fusion of the viral envelope (E) protein with the host membrane. Blocking virus entry via targeting the viral E protein offers a means to suppress the onset of infection. A few heterocyclic compounds, such as compound 6, NITD-448 and P02, have been identified to directly bind to the hydrophobic pocket of viral E protein and block its conformational change, which is essential for virus-host fusion (Modis et al., 2003, 2004; Zhou et al., 2008; Poh et al., 2009; Wang et al., 2009). Due to the multifunctional nature of the E protein, its inhibitors may potentially block multiple steps in the viral life cycle, including entry and virion assembly/maturation. More importantly, these inhibitors can exert their effect through direct binding to virions without the need to cross the hydrophobic membrane bilayer and be delivered into infected cells. However, due to the complexity and high variability of flaviviral E protein, it is challenging to develop pan-serotype inhibitors (Wang and Shi, 2015).

During replication, the viral genome is translated into a single polyprotein which is cleaved into individual proteins by a viral protease complex. Since polyprotein processing is a prerequisite for viral replication and assembly, these virally encoded proteases are one of the most attractive antiviral targets (Chambers et al., 1990, 1993; Luo et al., 2015). Two HCV NS3/4A serine protease inhibitors, boceprevir and telaprevir, have been approved in combination with PEG-interferon plus ribavirin for treatment of chronic HCV genotype 1 (Ghany et al., 2011). Recent study by Shiryaev and colleagues have identified a group of small molecule antiviral inhibitors that interfering with the productive fold of the NS2B cofactor in the two-component protease, inhibit its cleavage activity and therefore suppress ZIKV infection. The most potent inhibitor NSC157058 was shown to inhibit ZIKV infection in both cultured hfNPCs and mice without significant toxicity (Shiryaev et al., 2017). Despite these advances, resistance to protease inhibitors can occur rapidly, especially for chronic infections such as HCV due to the genetic variability of the virus and high mutation rate (Rong et al., 2010; Wu et al., 2013). Another concern in developing protease-based antiviral therapy is toxicity. Similarities in viral and host cellular serine proteases would presumably create problems in specificity while targeting the virus.

The flaviviral NS3 RNA helicase is located adjacent to the C terminal of the NS3 protease (Luo et al., 2008). The RNA helicase is believed to be required for several different functions such as initiation of RNA synthesis, separating dsRNA structures formed during viral RNA synthesis and as a translocase that eliminates proteins bound to the viral RNA (Sampath and Padmanabhan, 2009). Viruses with a mutated NS3 helicase are unable to replicate properly (Matusan et al., 2001). Several RNA helicase inhibitors have been identified. The antiparasitic drug ivermectin was shown to inhibit WNV, YFV, and DENV at submicromolar levels, and a small molecule inhibitor ST-610 was found to potently and selectively inhibit all four serotypes of DENV in vivo (Mastrangelo et al., 2012; Lim et al., 2013). However, due to a lack of specific binding pockets for RNA and NTPs, molecules that target the RNA helicase via these binding sites might also nonselectively bind to other cellular proteins with helicase/NTPase activities, resulting in significant toxicity (Luo et al., 2015).

The NS5 RNA-dependent RNA polymerase (RdRp) is the most conserved amongst the flavivirus proteins, and is essential for viral RNA synthesis. Since host cells lack these enzymes, the specificity makes them one of the most promising and intensively studied classes of antiviral targets. RdRp can be targeted by nonnucleoside inhibitors (NNIs) and nucleoside/nucleotide analog inhibitors (NIs) (Malet et al., 2008). NNIs directly target the binding pocket of the polymerase and block its conformational change from its inactive to active form (Biswal et al., 2005). Although a number of NNI candidates for HCV are under clinical development, there hasn't been any FDA approved NNIs for flaviviruses yet. The major challenge in the use of NNIs in antiviral therapy is the structural variability of the binding pockets across different serotypes or genotypes as well as the resistant mutation in or near the binding pocket which results in resistance to the NNIs (Sofia et al., 2012). NIs have been widely used in clinics for treatment of hepatitis, HIV and herpesvirus infections (Jordheim et al., 2013; Menéndez-Arias et al., 2014). Compared to other classes of inhibitors, NIs have a higherthreshold for developing resistance, and a relatively broadantiviral spectrum due to the relatively conserved polymerase structure (Delang et al., 2011; Lim et al., 2013). Unlike NNIs which directly bind to RNA polymerase, NIs have to convert into its triphosphate form inside cells by host kinases before exerting their antiviral effects (Stein and Moore, 2001). However, the kinase activity varies significantly in different cell types/hosts, causing variable efficacy of the same NI. Another major issue associated with NIs is the unpredictable toxicity in vitro. Although the toxicity of NIs is often associated with the inhibition of mitochondrial polymerases (Arnold et al., 2012), other mitochondrial perturbations may also attribute to toxicity (Selvaraj et al., 2014).

The N-terminal domain of NS5 contains one methyltransferase (MTase) that catalyzes guanine N-7 and ribose 2′ -O-methylations using S-adenosylmethionine (SAM) as a methyl donor during viral cap formation (Zhou et al., 2007). Non-selective competitive inhibitors, such as S-adenosylhomocysteine and sinefungin bind to SAM binding sites and inhibit its function (Boldescu et al., 2017). Using virtual screening, a group of small compound molecules have been identified with broad-spectrum activity against the MTase proteins of multiple flaviviruses, including DENV2, DENV3, and YFV (Brecher et al., 2015). Apart from the most important antiviral targets such as E protein, NS3 protease and NS5 polymerase, other viral targets such as capsid protein, NS1 and NS4 proteins are also under evaluation. The details of different


 | Comparison of advantages and disadvantages of different antiviral strategies against HCV and flaviviruses.

TABLE

2

Frontiers in Microbiology | www.frontiersin.org

 toxicity *in vivo*

• Unpredictable

viral targets have been reviewed elsewhere (Boldescu et al., 2017; García et al., 2017).

Due to extensive dependence of viruses (replication, assembly, and budding) on host LDs, the interface of virus-host interactions with LDs and/or LD metabolism provides a rich source for potential antiviral interventions (**Table 1**). First, targeting host factors may produce potential broad-spectrum activity against multiple viral infections due to similar intracellular pathways employed by viruses within the same genus or family. Second, given the high replication and mutation rates of viruses, longterm antiviral therapy against chronic infections inevitably selects for the resistant variants which alter the drug target and therefore are less susceptible to the inhibitory effects of the treatment. The resistant mutants eventually become the dominant species and lead to treatment failure and persistent infection. Development of drug-resistance has become a major challenge with direct-acting antivirals when treating chronic infections (Rong et al., 2010). Unlike viral elements, host cellular factors are much less prone to mutation; thus targeting host lipid metabolism provides an attractive approach for long-term treatment of diseases caused by viral infection. However, since LDs play a role in lipid metabolism in-vivo, manipulating a major metabolic pathway may have a more pleiotropic impact on cellular homeostasis (Georgel et al., 2010). Such consequences need to be carefully assessed to hit the right balance between causing host toxicity while preventing viral pathogenesis. Third, several inhibitors targeting host lipid metabolic pathways are well characterized, which can greatly accelerate the process of drug development. Moreover, targeting specific steps of LD biosynthesis, distribution, trafficking, and metabolism which viruses routinely exploit, allows us to design antiviral strategies with an enhanced therapeutic window. For example, triglyceride-synthesizing enzyme DGAT1 has been identified as an important host factor which is required for trafficking of viral capsid protein to LDs, facilitating early steps of viral assembly. Of note, RNAi-mediated silencing of DGAT1 resulted in impaired viral particle production without affecting LD composition (Herker et al., 2010). Currently, novel classes of pharmacological inhibitors targeting DGAT1 have been developed for clinical applications (DeVita and Pinto, 2013). In addition, regulating enzymes in the FA synthesis pathway has been shown to inhibit production of different viruses. C75, a FA synthase inhibitor, displayed a strong inhibitory effect on HCV replication (Yang et al., 2008), DENV production (Samsa et al., 2009), as well as WNV and YFV replication (Martín-Acebes et al., 2011) without causing significant toxicity to host cells. A series of chemical probes (ML-206, ML-219 and ML-220) has been shown to reduce the biogenesis and consumption of LDs without toxicity to mammalian cells (Boxer et al., 2013). These probes may prove to be beneficial in inhibiting virus production. A noteworthy and indirect strategy to interrupt the association between virus and LDs during viral replication and assembly is to target involved viral proteins. During the biosynthesis of the HCV polyproteins, an internal signal sequence between the capsid protein and envelope protein E1 can be preceded by cellular signal peptide peptidase (SPP). This process releases the capsid protein from the ER, followed by its transport to LDs. SPP inhibitor (Z-LL)2-ketone abolishes the cleavage of capsid protein by SPP and thereby inhibits production of infectious HCV (McLauchlan et al., 2002).

Ideally, antiviral treatments should exert their effects as early as possible after infection. This is particularly true for acute flaviviral infections such as DENV. Targeting intracellular host factors, however, is perhaps less effective in preventing the onset of an infection compared to other inhibitors, which block viral entry. The advantages and disadvantages of antiviral strategies against HCV and flaviviruses by targeting viral components and host factors including those involved in LD metabolism are summarized in **Table 2**.

## CONCLUSION

Despite being an immense global health problem, there are no affordable and efficient prophylactic or therapeutic treatments for some pathogenic flaviviruses. It is imperative to have alternative therapeutic strategies of inhibiting specific steps in the intracellular virus life cycle to combat infection. Viruses from the Flaviviridae family often cause perturbations in cellular energy and lipid homeostasis during infection. This has been reported for DENV, WNV, and HCV infection. Therefore, targeting cellular LDs offers possibilities for such interventions, including inhibition of lipid metabolism and disruption of interactions with viral components. Although knowledge on the participation of LDs during infection of HCV and flaviviruses has significantly progressed, comparative studies that aim to determine the shared or specific requirements of LD components for these pathogens are still lacking. In addition, much of the information available is from in-vitro studies, while the in-vivo relevance remains unexplored. Therefore, a more comprehensive understanding of the molecular biology of viruses and their dependence on host LD metabolism is of utmost priority for development of broad-spectrum and specific anti-flaviviral strategies.

## AUTHOR CONTRIBUTIONS

JZ and YL drafted the manuscript and contributed equally to this work. SS supervised, evaluated, and edited the manuscript.

## ACKNOWLEDGMENTS

This work was funded by Research Grants Council (GRF grants 17117914 and 17113915), and partially supported by Health and Medical Research Funds (16150592), theme based research grant from the Research Grants Council (Project No. T11- 705/14N) and research funds from Institut Pasteur (PTR 546). SS is supported by the Croucher Foundation.

#### REFERENCES


the NS5A protein. J. Gen. Virol. 87(Pt 11), 3263–3272. doi: 10.1099/vir.0. 82211-0


usurps lipid droplets for viral particle formation. PLoS Pathog. 5:e1000632. doi: 10.1371/journal.ppat.1000632


enables their connection to the ER for protein targeting. Elife 3:e01607. doi: 10.7554/eLife.01607


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Zhang, Lan and Sanyal. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Impact of Helicobacter pylori Urease upon Platelets and Consequent Contributions to Inflammation

Adriele Scopel-Guerra1†, Deiber Olivera-Severo1, 2†, Fernanda Staniscuaski 1, 3 , Augusto F. Uberti 1, 4, Natália Callai-Silva<sup>1</sup> , Natália Jaeger <sup>5</sup> , Bárbara N. Porto<sup>5</sup> and Celia R. Carlini <sup>6</sup> \*

#### Edited by:

Wesley H. Brooks, University of South Florida, Tampa, United States

#### Reviewed by:

Andrew S. Day, University of Otago, New Zealand Alireza Sadjadi, Tehran University of Medical Sciences, Iran Abdul Sadiq, University of Malakand, Pakistan

> \*Correspondence: Celia R. Carlini celia.carlini@pucrs.br

† These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 20 September 2017 Accepted: 24 November 2017 Published: 12 December 2017

#### Citation:

Scopel-Guerra A, Olivera-Severo D, Staniscuaski F, Uberti AF, Callai-Silva N, Jaeger N, Porto BN and Carlini CR (2017) The Impact of Helicobacter pylori Urease upon Platelets and Consequent Contributions to Inflammation. Front. Microbiol. 8:2447. doi: 10.3389/fmicb.2017.02447 <sup>1</sup> Center of Biotechnology, Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil, <sup>2</sup> Department of Biology, Universidade Regional Integrada do Alto Uruguai e das Missões, São Luiz Gonzaga, Brazil, <sup>3</sup> Department of Molecular Biology and Biotechnology, Institute of Biosciences, Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil, 4 Institute of Biology, Universidade Federal de Pelotas, Pelotas, Brazil, <sup>5</sup> Institute of Biomedical Research, Pontifícia Universidade Católica do Rio Grande do Sul, Porto Alegre, Brazil, <sup>6</sup> Brain Institute (BRAINS-InsCer), Pontifícia Universidade Católica do Rio Grande do Sul, Porto Alegre, Brazil

Gastric infection by Helicobacter pylori is considered a risk factor for gastric and duodenal cancer, and extragastric diseases. Previous data have shown that, in a non-enzymatic way, H. pylori urease (HPU) activates neutrophils to produce ROS and also induces platelet aggregation, requiring ADP secretion modulated by the 12-lipoxygenase pathway, a signaling cascade also triggered by the physiological agonist collagen. Here we investigated further the effects on platelets of recombinant versions of the holoenzyme HPU, and of its two subunits (HpUreA and HpUreB). Although HpUreA had no aggregating activity on platelets, it partially inhibited collagen-induced aggregation. HpUreB induced platelet aggregation in the nanomolar range, and also interfered dose-dependently on both collagen- and ADP-induced platelet aggregation. HPU-induced platelet aggregation was inhibited by antibodies against glycoprotein VI (GPVI), the main collagen receptor in platelets. Flow cytometry analysis revealed exposure of P-selectin in HPU-activated platelets. Anti-glycoprotein IIbIIIa (GPIIbIIIa) antibodies increased the binding of FITC-labeled HPU to activated platelets, whereas anti-GPVI did not. Evaluation of post-transcriptional events in HPU-activated platelets revealed modifications in the pre-mRNA processing of pro-inflammatory proteins, with increased levels of mRNAs encoding IL-1β and CD14. We concluded that HPU activates platelets probably through its HpUreB subunit. Activation of platelets by HPU turns these cells into a pro-inflammatory phenotype. Altogether, our data suggest that H. pylori urease, besides allowing bacterial survival within the gastric mucosa, may have an important, and so far overlooked, role in gastric inflammation mediated by urease-activated neutrophils and platelets.

Keywords: inflammation, mRNA processing, IL-1β, lipoxygenase inhibitors, CD14, GPVI, collagen receptor, platelet aggregation

## INTRODUCTION

Diseases caused by Helicobacter pylori have a great impact on public health, since this bacterium colonizes the gastric mucosa of half of the world's population, with a higher prevalence in the poorer countries (Parkin, 2004). Helicobacter pylori is a major cause of gastric and duodenal pathologies (Ferlay et al., 2013) and it was classified as the first carcinogenic bacterium by the World Health Organization more than 2 decades ago (IARC, 1994). Urease produced by H. pylori enables bacterial colonization of the gastric mucosa by catalyzing the hydrolysis of urea into carbon dioxide and ammonia, thereby causing a local pH increase and alterations of the mucus properties that favor the pathogen's survival (Perrais et al., 2014). Urease-negative strains of H. pylori were unable to infect the gastric mucosa of germfree piglets, ferrets, or mice (Hu and Mobley, 1990; Eaton et al., 1991; Andrutis et al., 1995).

Helicobacter pylori urease (HPU) accounts for ∼10% of total cell protein content (Suzuki et al., 2007). HPU is a large protein, consisting of a dodecameric organization of two subunits (HpUreA, 26.5 kDa; HpUreB, 61.7 kDa; Ha et al., 2001). Structure vs. activity relationships of the non-enzymatic properties of ureases have been so far poorly characterized (Carlini and Ligabue-Braun, 2016). It has been reported that HpUreB interacts with CD74 on gastric epithelial cells inducing IL-8 production (Beswick et al., 2006) and it also binds to Th17 lymphocytes (Zhang et al., 2011). A monopartite nuclear localization signal is present in HpUreA (sequence <sup>21</sup>KKRKEK26), and the protein is able to target the nuclei of COS-7 (Lee et al., 2012) and of AGS gastric epithelial cells, causing alterations of the cellular morphology (Lee et al., 2015). Additionally, H. pylori secreted outer membrane vesicles (OMVs) contain urease-related proteins, including HpUreA and HpUreB (Olofsson et al., 2010). Incubation of AGS gastric epithelial cells with H. pylori OMVs promoted the translocation of HpUreA into the cell cytoplasm and nuclear localization of the protein (Olofsson et al., 2010).

Epidemiological studies have shown that H. pylori infection correlates positively with several extragastric pathologies, such as intestine bowel diseases, cardiovascular and cerebrovascular diseases (Franceschi et al., 2015; Goni and Franceschi, 2016; Kyburz and Muller, 2017). Several hematological diseases such as primary immune thrombocytopenia, iron deficiency anemia, childhood leukemia, and coagulation disorders have been associated with H. pylori infection (Papagiannakis et al., 2013). The role of this pathogen (Christodoulou et al., 2011) and of its virulence factors in these extragastric diseases is still controversial, requiring further studies (Muhammad et al., 2017).

We have previously reported that canatoxin (Carlini and Guimaraes, 1981), an isoform of Canavalia ensiformis urease (Follmer et al., 2001), presents biological properties that are independent of its enzyme activity, including neurotoxicity, activation of blood platelets (Carlini and Guimaraes, 1981; Carlini et al., 1985; Ghazaleh et al., 1997) and in vivo proinflammatory activity (Benjamin et al., 1992; Carlini and Ligabue-Braun, 2016; Olivera-Severo et al., 2017). We have also demonstrated that a recombinant HPU activated platelets through a lipoxygenase-mediated pathway, leading to exocytosis of dense granules and release of adenosine diphosphate (ADP), which then promoted platelet aggregation (Wassermann et al., 2010). Independently of its enzyme activity, HPU displays a potent lipoxygenase-dependent chemotactic effect on neutrophils, both in vivo and in vitro, causing cell migration in levels comparable to those induced by fMLP (Uberti et al., 2013). HPU activated human neutrophils eliciting extracellular ROS production and protected neutrophils as well as cultured gastric epithelial cells against apoptosis, interfering on the levels of mitochondrial proteins regulating this process (Uberti et al., 2013; Olivera-Severo et al., 2017). Recently we reported on the angiogenic potential of HPU, a property that could have implication in the invasion and metastization of gastric tumors (Olivera-Severo et al., 2017).

Platelets are anucleated cells involved, aside of hemostasis and thrombi formation, in physiological processes such as tissue regeneration, angiogenesis, inflammation and immunity (Jurk and Kehrel, 2005; Farndale, 2009; Vieira-de-Abreu et al., 2012). Platelets store distinct compounds in granules that, once released, contribute to the immune response (Vieira-de-Abreu et al., 2012) and they produce IL-1β, a pro-inflammatory cytokine (Lindemann et al., 2001; Morrell et al., 2014). IL-1β is considered a platelet agonist, which acts through an autocrine loop to link thrombosis and immunity, signaling to endothelial cells and promoting neutrophil adhesion (Brown et al., 2013). Platelet activation leads to a sustained synthesis of the pro-IL-1β protein, which accumulates in the platelets' cytosol until cleavage to release the cytokine (Lindemann et al., 2001). Platelets interact with components of the subendothelial matrix, with different immune cells and many pathogens (Clemetson, 2011; Morrell et al., 2014) and participate in cancer metastasis (Jurk and Kehrel, 2005; Farndale, 2009). Platelets signal the innate immune system through the release of microparticles and NET formation (Italiano et al., 2010; Carestia et al., 2016). These various characteristics of platelets enable a large spectrum of possible cell interactions and modulation by different membrane receptors (Cimmino and Golino, 2013; Thomas and Storey, 2015).

Collagen, the main component of the subendothelial matrix, promotes adhesion and activation of platelets. Several platelet collagen receptors are involved in these events (Clemetson and Clemetson, 2001). Among them, glycoprotein VI (GPVI), a member of the immunoglobulin superfamily of receptors (Clemetson et al., 1999; Jandrot-Perrus et al., 2000), plays a crucial role in the responses of platelets to collagen. GPVI participates in platelet adhesion to the subendothelium matrix and triggers a collagen-induced activation, resulting in a thromboxane A2- and ADP-mediated aggregation thus providing a procoagulant surface for thrombin formation (Moroi et al., 1989; Nieswandt et al., 2001; Nieswandt and Watson, 2003). Collagen binding to GPVI receptor involves signaling through the 12-lipoxygenase pathway (Coffey et al., 2004a,b). On the other hand, when platelets are exposed to low collagen doses, activation through GPVI leads to a secretory phenotype accompanied by release of their dense- and alpha granules' contents, but without changing the platelets into their prothrombotic state (Ollivier et al., 2014).

In this study, we aimed to elucidate the mechanism underlying the activation of platelets exposed to an enzymatically active HPU holoenzyme and to its subunits HpUreA and HpUreB. To that aim, we performed aggregation assays and investigated the roles of platelet's glycoprotein VI and IIbIIIa in HPU-induced and subunits-induced responses. Furthermore, we analyzed expression of P-selectin and the processing of pre-mRNAs in platelets treated with these proteins.

## MATERIALS AND METHODS

#### Helicobacter pylori Urease (HPU)

A recombinant Helicobacter pylori urease (HPU) was produced by heterologous expression in Escherichia coli BL21 (DE3)- RIL transformed with a PGEM-T-easy (Promega) plasmid carrying the whole urease operon (kindly provided by Dr. Barbara Zambelli, Universitá di Bologna, Italy). HPU was purified from bacterial extracts according to Olivera-Severo et al. (2017). Protein homogeneity was checked by 0.1% sodium dodecyl sulfate 10% polyacrylamide gel electrophoresis (SDS-PAGE) (Figure S1A). Previous to the experiments, a 0.5 mg protein.mL−<sup>1</sup> solution was dialyzed against 20 mM sodium phosphate 150 mM sodium chloride, pH 7.5 (PBS 7.5), and the buffer from the last dialysis change was used as a negative control in the bioassays.

Fluorescent HPU was prepared by incubation of a 1.0 mg.mL−<sup>1</sup> solution of urease with 0.1% fluorescein isothiocyanate (FITC) in PBS 7.5 for 60 min at 4◦C. The mixture was exhaustively dialyzed against PBS 7.5 and then applied into a Fast-Desalting column (Amersham Biosciences) to remove any unbound FITC (Wassermann et al., 2010).

#### Helicobacter pylori Urease Isolated Subunits

The subunits of Helicobacter pylori urease, HpUreA and HpUreB, were produced in Escherichia coli BL21 (DE3)-RIL transformed with pET101/D-TOPO (Thermo Fisher Scientific) plasmids containing the cDNA encoding each of the subunits (a generous gift from Dr Cesare Montecucco, Universitá di Padova, Italy). Bacteria were grown in Luria broth with the antibiotics chloramphenicol (Sigma Aldrich, USA; 40 µg.mL−<sup>1</sup> ) and ampicillin (Sigma Aldrich, USA; 100 µg.mL−<sup>1</sup> ) for maintenance of the plasmids. The His-tagged subunits HpUreA and HpUreB were purified from bacterial extracts as follows: after cultivation, cells were harvested by centrifugation, suspended in extraction buffer (20 mM sodium phosphate, 500 mM sodium chloride, 5 mM imidazole, pH 7) and lysed using an ultrasonic homogenizer (10 pulses of 30 s) in an ice bath. After centrifugation (20 min, 20,000 × g, 4◦C), the homogenates were submitted to Ni2<sup>+</sup> affinity chromatography on a 5 mL Chelating Sepharose Fast Flow column (GE Healthcare Life Sciences). After removal of the unbound proteins, the column was eluted stepwise with 300 mM imidazole in 20 mM sodium phosphate, 500 mM sodium chloride, pH 7.0 to obtain HpUreA- and HpUreBenriched fractions. For the last step of purification, the fractions from the Ni2+-affinity step were concentrated in Amicon devices with 10 kDa cut-off (Millipore, Belford, MA, USA), applied into a Superdex S200 Hi-load column mounted in a Akta Purifier system (GE Healthcare Life Sciences), equilibrated in PBS 7.5. The protein fractions were analyzed by SDS-PAGE and by Western blots (Figure S1B), pooled and concentrated in Amicon devices (10 kDa cut-off) to a concentration of 0.5 mg.mL−<sup>1</sup> .

## Protein Determination

The protein content of samples was determined by absorbance at 280 nm, or by the Coomassie dye binding method (Bradford, 1976).

### Urease Activity

The urea hydrolyzing activity of HPU was measured by the alkaline nitroprussiate method (Weatherburn, 1967) using ammonium sulfate as reference, to follow the purification of the holoenzyme.

#### Platelet Aggregation

Rabbit platelets were used in the aggregation assays. Animal care and handling followed international guidelines and all protocols were approved by the institutional Ethics Committee (UFRGS process 721.217). Platelet aggregation assays were performed essentially as described by Wassermann et al. (2010) using two different turbidimetric assays. For platelet aggregation assays performed in a plate reader (SpectraMax <sup>R</sup> M3, Molecular Devices), agonists were added to 100 µL aliquots of PRP, without or with inhibitors, to make 150 µL final reaction volume in 96 wells flat bottom plates, with absorbance readings at 650 nm every 11 s during 20 min. For platelet aggregation using a Lummi-aggregometer (Chrono-Log Corp.), after preincubation of PRP (300 µL) for 2 min under continuous stirring at 37◦C, aggregation was triggered by addition of the agonist (maximal volume 30 µL) and the reaction was registered during 10 min. In both assays, platelet's response was quantified as area under the aggregation tracings. Response to the physiological agonist collagen (bovine tendon, 50 µg.mL−<sup>1</sup> ) or ADP (10µM) (final concentrations) was taken as positive control of platelet activation.

In some of the experiments, platelets were pre-incubated at room temperature without stirring with HPU subunits, inhibitors or antibodies before addition of the platelet agonist: eicosanoid synthesis inhibitors esculetin (0.5 and 1 mM) and indomethacin (75–300µM); antibodies against platelet glycoproteins GPVI or IIbIIIa, or anti-platelet cell markers (see the following sections for more details).

## Platelet Isolation

Peripheral human blood of healthy volunteers was obtained in the presence of 0.313% (w/v) sodium citrate. Written informed consent was obtained from the participants of this study. All procedures regarding blood collection and handling were conducted in strict accordance to Brazilian legislation (Law no. 6.638/1979) and were approved by the institutional Ethics Committee (UFRGS process 721.217).

Human platelets were isolated according to Ollivier et al. (2014) with some modifications; briefly, a platelet-rich plasma (PRP) was separated from whole blood by centrifugation at 200 g for 15 min. For flow cytometry assays (see below), the platelets were pelleted from PRP by an additional centrifugation step (800 × g, 10 min), and washed 3 times (800 × g, 10 min) with a modified Tyrode's buffer (3.6 mM citric acid, 0.5 mM glucose, 0.5 mM KCl, 0.1 mM MgCl2, 10.3 mM NaCl, 2 mM CaCl<sup>2</sup> pH 6.5). After the third wash, the platelets were suspended in the reaction buffer (5 mM HEPES, 12 mM NaHCO3, 137 mM NaCl, 2 mM KCl, 2 mM CaCl2, 0.3 mM NaH2PO3, 1 mM MgCl2, 5.5 mM glucose, pH 7.4). For quantitative PCR, washed platelets were immunolabeled as CD61 positive cells and separated in a MidiMACS LS column (Miltenyi Biotec) according to manufacturer's instructions.

#### RNA Extraction and cDNA Synthesis

For RNA extraction, the isolated platelets were suspended in reaction buffer to a final cell concentration of 10<sup>7</sup> .mL−<sup>1</sup> . Platelets were incubated with 50 µg.mL−<sup>1</sup> collagen or 100 nM urease, HpUreA or HpUreB, in reaction buffer, without stirring (to avoid aggregation), at 37◦C, for 30, 90, and 180 min. Total RNA was extracted immediately after the stimuli, using TriZol reagent (Ludwig Biotec), following the manufacturer's instructions. cDNA synthesis from total RNA was performed with the High-Capacity cDNA Reverse Transcription kit (Applied Biosystens/ThermoFischer Scientific), following the manufacturer's instructions, using random primers.

## Pro-Inflammatory Expression Profile Analysis

Real time quantitative PCR was performed to compare CD14, interleukin-1 beta (IL-1β), cyclooxygenase-2 (COX-2), intercellular adhesion molecule 1 (ICAM-1), and inducible nitric oxide synthase (iNOS) genes expression levels in platelets challenged by collagen, the holoenzyme HPU, and by the HpUreA and HpUreB subunits, in three different time intervals at 37◦C, without stirring. The beta actin gene was used (Zsori et al., 2013) to normalize the RNA content of the samples. Gene specific primers were designed to span exon junctions (Table S1). Reactions were carried in an Eco thermocycler (Illumina), using the qPCR-Sybr Green kit (Ludwig Biotec), and following the parameters: 95◦C for 5 min (initial denaturation), 40 cycles at 95◦C for 10 s (denaturation), 60◦C for 15 s (annealing), 72◦C for 15 s (extension). Melting curves were performed at the end of each reaction, with temperatures ranging from 55 to 95◦C (increments of 0.1◦C/s). All cDNA samples were diluted 1:5. Reactions were carried out in technical quadruplicates from two independent biological replicates. Results were analyzed by the 2 <sup>−</sup>11C<sup>T</sup> method (Livak and Schmittgen, 2001).

#### Flow Cytometry Analysis

Platelets (5 µL of PRP) kept at 37◦C without stirring were stimulated with 100 nM or 300 nM HPU, 25 µg.mL−<sup>1</sup> collagen, for 1, 5, and 10 min. Resting platelets (negative controls, no addition) and stimulated platelets in 50 µL of reaction buffer were stained with FITC-labeled anti-CD42 (GP1b) (1:25) (Abcam) and PerCP-labeled anti-CD62P (P-selectin) (1:5) (Abcam) antibodies for 20 min, at room temperature in the dark. After the incubation period, platelets were fixed with 1% paraformaldehyde. Platelets were identified by gating on platelets' size on the basis of forward scatter (FSC) and side scatter (SSC), followed by CD42 expression, a platelet marker. A total of 20,000 events were analyzed for MFI/percentage of CD62P expression. Data were acquired using a FACSCanto II flow cytometer (Becton Dickinson) with BD FACSDiva software and analyzed by Flowjo <sup>R</sup> vX.

In another set of experiments, platelets were previously treated with polyclonal anti-GPVI (1:10) (Santa Cruz Biotech) or monoclonal anti-IIbIIIa (1:10) (Santa Cruz Biotech) for 20 min at room temperature and then stimulated with 100 nM FITCconjugated HPU for 1, 5, and 10 min at 37◦C. Platelets not treated with the antibodies nor exposed to HPU served as negative controls. After the incubation period, cells were fixed with 1% paraformaldehyde. Platelets were identified as described before. A total of 50,000 events were analyzed for percentage of FITC-positive cells. Data were acquired using a FACSCanto II (Beckton Dickinson) with BD FACSDiva software and analyzed by Flowjo <sup>R</sup> vX.

### Statistical Analysis

The statistical significance of the differences between two groups was assessed using the unpaired Student's t-test. For multiple comparisons a two-way analysis of variance (ANOVA) was performed, and the Tukey post hoc test was used to calculate significance. GraphPad Prism6 software (San Diego, CA, USA) was used to perform statistical analysis. Statistically significance was set at p-value ≤ 0.05. Data in graphs represent mean ± standard error of the mean (SEM) of at least three experiments, unless otherwise stated.

## RESULTS

## Effects of HPU Subunits on Platelet Aggregation

Our previous study has shown that HPU induces aggregation of rabbit platelets in nanomolar concentrations (Wassermann et al., 2010). To test if the isolated subunits also induce this effect, HpUreA and HpUreB were tested as platelet agonists. While HpUreB induced platelet aggregation in a dose dependent manner (**Figures 1A,B**), HpUreA had no activity under the same conditions. As reported for HPU, aggregation induced by HpUreB in rabbit platelets also depends on the production of lipoxygenase-derived eicosanoids (**Figures 1C,D**), as indicated by the inhibitory effect of esculetin, which blocks the platelet 12 lipoxygenase, and by the potentiating effect of the cyclooxygenase inhibitor, indomethacin (**Figures 1E,F**) (Wassermann et al., 2010). To investigate whether the isolated subunits could interfere in the aggregation triggered by the physiological agonists collagen or ADP, the platelets were previously incubated without stirring with HpUreA or HpUreB, and then challenged with the agonists (**Figure 2**). Surprisingly, considering the lack of direct effect of HpUreA, both subunits caused a dose-dependent inhibition of platelets' response to the agonists (**Figure 2A**). HpUreB was more effective than HpUreA in inhibiting platelet's response to collagen or ADP (**Figure 2B**). Moreover, HpUreB apparently interfered in the second wave of aggregation, abruptly

FIGURE 1 | Effects of HPU subunits on platelet aggregation. (A,B) HpUreB induces platelet aggregation in a dose dependent manner. (A) Aggregation of rabbit platelets was induced with the indicated final concentrations of HPU (open symbols, results taken from Wassermann et al., 2010) and HpUreB (closed symbols, this work). Aggregation induced by HPU or HpUreB at 1.2–1.4µM was considered 100%. (B) Superimposed tracings of aggregation induced by different HpUreB concentrations as measured in the plate reader. (C–E) HpUreB-induced platelet aggregation depends on lipoxygenase-derived eicosanoid(s). Platelets were pretreated with the inhibitors of eicosanoids synthesis, esculetin (a 12-lipoxygenase inhibitor) (C,D) and indomethacin (a cyclooxygenase inhibitor) (E,F) at room temperature for 5 min without stirring. Aggregation was triggered by addition of 750 nM HpUreB (time zero), and after 2 min at 37◦C stirring was turned on. Platelets response was monitored on SpectraMax M3 plate reader, with readings at 650 nm every 7 s for 20 min. Superimposed individual tracings of typical experiments are shown in (B,D,F). Aggregation responses were quantified as area under the tracings using SotfMax Pro 5.4.1 (A,C,E). Data are expressed as means ± SEM. Statistical significance was determined by ANOVA followed by Tukey-Kramer test. Values of \*\*p < 0.01.

blocking the progression of aggregation in response to released ADP (**Figure 2B**, right panel), an effect not seen in HpUreAtreated platelets. HpUreA and HpUreB also interfered in the HPU-induced platelet aggregation (**Figure 3**). While 1µM HpUreB had a synergistic effect on platelets activated by 300 nM HPU, increasing the aggregation response by 150% (**Figure 3A**), 1µM HpUreA decreased the aggregation response by 75% (**Figure 3B**).

### HPU Interaction with Platelet Membrane Receptors

Activation of human platelets by 100 and 300 nM HPU induced P-selectin exposure in a subpopulation of these cells (**Figure 4A**). This effect occurred immediately after addition of HPU and persisted for at least 10 min. Other signs of activation were also seen, such as increase in the cellular size and membrane complexity (Figure S4). Previously we have hypothesized (see Wassermann et al., 2010) that, in rabbit platelets, HPU may recruit the same signaling pathway triggered by collagen. Considering that anti-GPVI antibodies almost completely blocked HPU-induced aggregation (Figure S2), here we investigated whether HPU interacts with platelets receptors. For that aim pretreatments of human platelets with either anti-GPVI, the main collagen receptor, or anti-GPIIbIIIa, implicated in platelets' activation by fibrinogen and von Willebrand factor, were carried out. Surprisingly, after blockade of GPIIbIIIa by the antibodies, there was an increase in HPU binding to platelets (**Figure 4B**, left panel). We could not observe reactivity of the employed commercial polyclonal anti-GPIIaIIIb antibodies against HpUreB under our experimental conditions (Figure S3). Nonetheless, the signs of platelet activation (increased cell size) could no longer be seen, thereby confirming that HPU-induced

determined by ANOVA followed by Tukey-Kramer test. Values of \*p < 0.05, \*\*p < 0.01.

platelet activation somehow involves binding to the platelet membrane, if not directly to, at least in the vicinity of, this physiologically relevant receptor. On the other hand, the binding of FITC-labeled HPU to platelets was not affected by either mono- or polyclonal antibodies against GPVI (**Figure 4B**, right panel).

## Urease and Its Subunits Modify Pre-mRNA Processing in Platelets

To evaluate the processing of pre-mRNA in platelets, qPCR analysis was performed after platelets treatment with HPU or its subunits. Collagen-activated platelets were used as positive controls. Collagen increased the IL-1β mRNA processing levels (**Figure 5A**), peaking at 30 min of treatment and then gradually decreasing to reach levels below the controls in 3 h-treated platelets. Differently, HPU-treated platelets decreased the process in the short treatment (30 min) and then gradually increased IL-1β pre-mRNA processing to reach, after a 3 h treatment, levels similar to those seen in collagen-treated platelets at 30 min (**Figure 5A**). Exposure of platelets to either HPU's subunits modified the IL-1β mRNA processing, but with different patterns as compared to that produced by the holoenzyme. The timecourse of HpUreA's effect was similar to that seen in collagenexposed platelets. On the other hand, treatment of platelets with HpUreB inhibited IL-1β mRNA processing in all tested times (**Figure 5B**).

Contrasting with the inhibition observed in collagen-treated platelets (**Figure 5C**), the processing of CD14 mRNA was greatly enhanced (about 100-fold) in HPU-treated plateletes (30 min). The levels gradually decreased to reach levels still 4-fold higher than in resting platelets with 3 h of HPU treatment. Similar to the inhibition patterns seen in collagen-stimulated platelets, HpUreA or HpUreB inhibited the processing of CD14 mRNA up to 1.5 h after treatment, returning to control levels after 3 h (**Figure 5D**). Neither HPU nor its subunits interfered on the processing of pre-mRNA of COX-2, ICAM-1, or iNOS under the tested conditions.

#### DISCUSSION

The association of H. pylori infection, a pathogen with a global dispersal, with several extragastric pathologies has not gone unnoticed. Besides its well established role in stomach diseases, gastric and duodenal cancer (Ferlay et al., 2013), epidemiological studies have associated this bacterial infection with cardiovascular and thromboembolic conditions (Manolakis et al., 2007; Papagiannakis et al., 2013) and other extragastric diseases (Franceschi et al., 2015; Goni and Franceschi, 2016; Kyburz and Muller, 2017). The potential contribution of the very abundant bacterial urease to the pathogenesis of these extragastric diseases in H. pylori positive patients has been so far mostly overlooked.

Microbial and plant ureases have many non-enzymatic properties, mostly based on induction of exocytosis and recruitment of eicosanoids pathways, among which are neurotoxicity and pro-inflammatory activity (Carlini and Ligabue-Braun, 2016). One of these non-enzymatic properties of ureases, regardless of their source and quaternary structures, is their ability to induce aggregation of blood platelets (Follmer et al., 2004; Carlini and Ligabue-Braun, 2016). This platelet-aggregating effect of ureases is due to their exocytosis-inducing activity which leads to the release of

ADP from platelets' dense granules, and requires the eicosanoid 12-hydroxy-peroxy-eicosatetraenoic acid (12-HPETE), produced by the platelet 12-lipoxygenase (Carlini et al., 1985; Olivera-Severo et al., 2006a,b; Wassermann et al., 2010). These data suggested that platelet aggregation induced by HPU resembles that of collagen-activated platelets through its GPVI receptor, a response reported to also require 12-HPETE synthesis by the activated platelets (Coffey et al., 2004a,b). Back then, we hypothesized that ureases and collagen may recruit similar or overlapping signaling cascades to exert their actions in platelets (Wassermann et al., 2010).

Here, we aimed to deepen the knowledge on how HPU interacts with platelets. Recombinant versions of HPU, and of its two subunits, HpUreA and HpUreB, were produced and platelet aggregation assays, flow cytometry and quantitative PCR were used to analyze how these proteins affect platelets' physiology.

In Wassermann et al. (2010), we reported that 1µM HPU induced maximal aggregation of rabbit platelets. The response of rabbit platelets to HPU involved the 12-lipoxygenase, as it could be blocked by esculetin, and the eicosanoid 12- HETE, an oxidized derivative of 12-HPETE, was detected in the medium (Wassermann et al., 2010). Here we showed that HpUreB induces rabbit platelet aggregation in the same molar range as that of the holoenzyme. In contrast HpUreA, even in a 10-fold greater concentration had no platelet aggregating activity. This result reinforces the non-enzymatic nature of the platelet-activating property of HPU, since HpUreB has no enzymatic activity (Ha et al., 2001; Suzuki et al., 2007). Platelets' response to HpUreB required production of lipoxygenasederived eicosanoids, as could be deduced from the dosedependent inhibitory effect of esculetin (**Figures 1C,D**). The fact that pretreatment of platelets with the cyclooxygenase inhibitor indomethacin enhanced platelets' reactivity to HpUreB (**Figures 1E,F**) is expected for a lipoxygenase-mediated process, in a condition of increased availability of arachidonic acid, which is the substrate of both enzymes (Wassermann et al., 2010). These

< 0.01, \*\*\*p < 0.001.

data can be interpreted as HpUreB being the HPU's domain active on platelets, and corroborates previous findings showing that this effect does not require the enzyme's ureolytic activity.

Pre-incubation of platelets with either one of the HPU's subunits inhibited, in a dose-dependent manner, the aggregation response to collagen or ADP (**Figure 2**). These data indicate that (1) although not inducing platelet aggregation, HpUreA also interacts with platelets; (2) HPU binds to platelets' membranes through at least two binding sites, one present in each of its subunits; (3) the isolated subunits bind to platelets in the same sites as does HPU itself; and (4) the binding of HPU's subunits partially blocks physiological functions of platelets. While an explanation for these observations is not trivial, the data clearly indicated that the binding sites and/or the type of binding to platelet membrane of the two subunits are quite different, implying that distinct mechanisms of action may underly their effects on platelets. The reason why pre-incubation with either HPU's subunits inhibited the platelet aggregation promoted by collagen or ADP is unclear, considering their effects on HPU-induced aggregation (**Figure 3**). HpUreA inhibited HPUinduced aggregation, which is consistent to its binding to platelets with an antagonist-like behavior. This effect may be related to fact that ureases insert themselves into lipid bilayers thereby affecting membrane physicochemical properties (Piovesan et al., 2014; Micheletto et al., 2016). On the other hand, HpUreB probably interacts specifically with some platelet receptor, as it induces aggregation per se, and it acted synergistically with HPU to amplify the aggregating effect (**Figure 3**).

Activation of human platelets (constitutively positive for CD42, or glycoprotein 1b) by HPU in comparison to collagen was studied by flow cytometry following the expression of Pselectin (CD62P) (**Figure 4**). Platelets are known to modulate their degranulation and release of specific granules according to the stimulus (Jonnalagadda et al., 2012). For instance, collagen-activated platelets, when in low concentrations of the agonist, can degranulate and release their granular contents without a significant exposure of P-selectin (Ollivier et al., 2014). Here we demonstrated that HPU-stimulated human platelets, despite their low expression of P-selectin, show signs of activation, and eventually aggregate under favorable conditions.

Fibrinogen binding to GPIIbIIIa is a requirement for ADPinduced platelet aggregation. We have previously reported that platelet aggregation induced by canatoxin required ADP release and the presence of fibrinogen (Carlini et al., 1985; Follmer et al., 2001). Here we showed that pre-treatment of platelets with antibodies against GPIIbIIIa enhanced the binding of FITClabeled HPU to the cells (**Figure 4B**, left panel). Altogether, these results implicate the involvement of GPIIbIIIa in the ADPdependent platelet-aggregating effect of ureases. Muhammad et al. (2017) reviewed the mechanisms by which H. pylori infection could lead to cardiovascular and thromboembolic diseases, emphasizing that a cross-reactivity between HpUreB and GPIIIa could be the link associating H. pylori infection to immune thrombocytopenia. Bai et al. (2009) produced monoclonal antibodies against a recombinant HpUreB and showed that these antibodies cross-reacted with GPIIIa from normal platelets and partially inhibited aggregation induced by ADP. The binding motif recognized by the monoclonal anti-HpUreB in the platelet GPIIIa was not identified. This cross-reactivity of HpUreB and GPIIIa may be implicated in our observation that platelets pre-treated with anti-GP IIbIIIa bound more FITC-labeled HPU than did unstimulated platelets (**Figure 4B**). The lack of effect of the anti-GPVI antibodies in preventing HPU binding to platelets does not go against our previous hypothesis that HPU (**Figure 4B**, right panel), and now including HpUreB, share with collagen a lipoxygenasemediated pathway of platelet activation (Wassermann et al., 2010). However, binding to GPVI seems not to be the feature that explains the similarity of platelets' response to these agonists.

It is well known that platelets participate in the inflammatory process by modulating the activity of other inflammatory cells (Thomas and Storey, 2015; Koenen, 2016). We have previously demonstrated that HPU displays pro-inflammatory activity in the mouse paw edema model, causing an intense eicosanoid-dependent neutrophil infiltration in the tissues (Uberti et al., 2013). In the same study, HPU showed a non-enzymatic, lipoxygenase-dependent, chemotactic effect on human neutrophils, and induced extracellular production of reactive oxygen species (ROS) by the activated cells (Uberti et al., 2013). Our data reinforced results obtained by other groups showing that purified HPU elicited the production of ROS and inflammatory cytokines by human macrophages and primary monocytes in vitro (Harris and Granger, 1996; Shimoyama et al., 2003), induced transendothelial migration of T cells (Enarsson et al., 2005) and increased the expression of inducible NO synthase (Gobert et al., 2002). All these activities of HPU contribute to tissue inflammation and injury. Here we investigated whether activation of human platelets by HPU leads to a pro-inflammatory phenotype of these cells.

Platelets have reservoirs of pre-mRNA that are processed into the corresponding mRNA upon stimulation (Denis et al., 2005). Treatment of platelets with collagen, HPU, HpUreA, or HpUreB, modified their processing of IL-1β and CD14 pre-mRNAs, each protein producing effects with a distinct kinetics. Collagen and HpUreA had similar time pattern for stimulation of IL-1β premRNA processing, with peaks at 30 min after treatment and then decreasing to levels below control after 3 h (**Figures 5A,B**). The slower but persistent effect of HPU of increasing IL-1β production 3 h after treatment confirms that this protein changes platelets into a pro-inflammatory phenotype. Further studies are necessary to confirm if the pre-mRNA processing changes observed here impact as well the platelets' protein levels of IL-1β and CD-14 after exposure to HPU.

Zhang et al. (2009) reported that platelets express TLR4, CD14 and MyD88, the signaling pathway triggered in response to lipopolysaccharides (LPS), as described in other cell types (Funda et al., 2001). CD14 is a co-receptor of various Toll-like receptors (TRLs) present in hematopoietic and non-hematopoietic cells that recognizes PAMPs (pathogen-associated molecular patters), triggering the innate immune response and participating in inflammation (Zanoni and Granucci, 2013). Distinct from collagen and the subunits HpUreA and HpUreB, that lowered CD14 pre-mRNA processing below control levels, HPU greatly enhanced the processing of CD14 mRNA (**Figure 5**). This result suggests that HPU activates at least part of the signaling pathway triggered by LPS, and the increased expression of CD14 in HPU-activated platelets corroborates their phenotype conversion into pro-inflammatory cells. The absence of CD14 expression in unstimulated platelets, or in platelets stimulated by collagen under normal physiological conditions, in contrast to the increased levels after HPU stimulus, resembles the reaction of intestinal mucosal cells, which only show increased expression levels of CD14 associated to inflammatory processes (Funda et al., 2001).

Similar to cell activation through LPS where a role of CD14 in the induction of TNF-α, IL-1β, IL-6, and IL-8 expression has been identified (Dentener et al., 1993; Schumann et al., 1994), here we hypothesize that CD14 expression may be modulating the IL-1β levels in platelets stimulated with HPU. This does not occur when platelets are stimulated with collagen, HpUreA or HpUreB. It has been described that platelets stimulated by LPS (Brown and McIntyre, 2011) or dengue virus (Hottz et al., 2013) process IL-1β pre-mRNA and that once stimulated, these cells release microparticles loaded with IL-1β by a mechanism not yet completely elucidated (Hottz et al., 2013). If HPU-stimulated platelets also release microparticles containing IL-1β, that could be delivered nearby endothelial cells or immune cells responsive to IL-1β, such as macrophages and lymphocytes, will be a subject of future studies.

The natural existence of any urease's subunits in a free state has never been described thus the relevance of the biological effects described here for HpUreA or HpUreB is uncertain. On the other hand this study provided valuable insights into the structure vs. activity relationships of HPU concerning its effects on platelets. While it became clear that HpUreB carries the major "platelet-active" domain of HPU, the contributions of HpUreA to the platelet activating effect of HPU are less clear. The antagonistic effect of HpUreA against aggregation induced by the physiological agonists ADP and collagen, proved its interaction with platelets' membranes. Moreover only HpUreA increased IL-1β pre-mRNA processing like HPU did. Thus both HPU subunits contribute to the protein's effect on platelets although in different ways. Another evidence of the existence of two platelet binding sites in HPU is the fact that both of its subunits competed with HPU, partially blocking the aggregating effect of the holoenzyme. In alignment with this conclusion, other studies by our group performed with the isolated subunits of the tri-chained Proteus mirabilis urease have identified a "platelet-activating" domain in

#### REFERENCES


one of its small subunits whose sequence is homologous to the C-terminal half of HpUreA (Broll V, unpublished results).

In summary, in this work we have shown that HPU and its subunits affect platelet physiology in ways that may contribute to the pathogenicity of H. pylori by other mechanisms besides enabling bacterial survival in the gastric lumen. The proinflammatory phenotype of HPU-activated platelets implies altered participation of these cells in many physiological processes, possibly contributing to the development of the extragastric diseases associated to H. pylori infection. Altogether our results reinforce the importance of microbial ureases, acting also in non-enzymatic ways, as a virulence factor of pathogenic microorganisms.

#### AUTHOR CONTRIBUTIONS

AS-G and DO-S planned and conducted experiments on the interactions of urease and its subunits with platelets; AS-G, DO-S, and AFU planned and conducted platelet aggregation assays; AS-G, DO-S, AFU, and NC-S produced the recombinant proteins; AS-G and FS planned and conducted mRNA processing experiments; NJ and BP conducted flow cytometry assays; AS-G, DO-S, FS, NJ, and CC have written and revised the manuscript; CC has conceived and coordinated this study.

## FUNDING

This work was supported by Brazilian agencies Coordenação de Pessoal de Nível Superior (CAPES, Edital 63/2010 Toxinologia, proj. 1205/2011; Edital PosDoc-SUS 2909/2010, proj. 054/2010) and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, Edital Universal Proc. No. 475908/2012-0 and No. 446052/2014-1). AS-G was a recipient of a CAPES Ph.D. fellowship.

#### ACKNOWLEDGMENTS

The authors acknowledge Barbara Zambelli, PhD, from Universitá di Bologna, Italy, and Cesare Montecucco, PhD, Universitá di Padova, Italy, for kindling providing the holoprotein (HPU) and HPU's isolated chains constructs.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02447/full#supplementary-material

purpura. Int. J. Hematol. 89, 142–149. doi: 10.1007/s12185-008-0 247-4


binds to CD74 on gastric epithelial cells and induces NF-kappaB activation and interleukin-8 production. Infect. Immun. 74, 1148–1155. doi: 10.1128/IAI.74.2.1148-1155.2006


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Scopel-Guerra, Olivera-Severo, Staniscuaski, Uberti, Callai-Silva, Jaeger, Porto and Carlini. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A New Role for Helicobacter pylori Urease: Contributions to Angiogenesis

Deiber Olivera-Severo1,2† , Augusto F. Uberti1,3† , Miguel S. Marques4,5,6, Marta T. Pinto4,5 , Maria Gomez-Lazaro4,7, Céu Figueiredo4,5,6, Marina Leite4,5 \* † and Célia R. Carlini1,8 \* †

#### Edited by:

Wesley H. Brooks, University of South Florida, United States

#### Reviewed by:

Alejandro Piscoya, Peruvian University of Applied Sciences, Peru Abdul Sadiq, University of Malakand, Pakistan Christian T. K.-H. Stadtlander, Independent Researcher, United States

#### \*Correspondence:

Célia R. Carlini celia.carlini@pucrs.br Marina Leite mleite@ipatimup.pt †These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 31 July 2017 Accepted: 14 September 2017 Published: 27 September 2017

#### Citation:

Olivera-Severo D, Uberti AF, Marques MS, Pinto MT, Gomez-Lazaro M, Figueiredo C, Leite M and Carlini CR (2017) A New Role for Helicobacter pylori Urease: Contributions to Angiogenesis. Front. Microbiol. 8:1883. doi: 10.3389/fmicb.2017.01883 <sup>1</sup> Center of Biotechnology, Universidade Federal Rio Grande do Sul, Porto Alegre, Brazil, <sup>2</sup> Biology Department, Universidade Regional Integrada do Alto Uruguai e das Missões, São Luiz Gonzaga, Brazil, <sup>3</sup> Institute of Biology, Universidade Federal de Pelotas, Pelotas, Brazil, <sup>4</sup> i3S, Instituto de Investigação e Inovação em Saúde, University of Porto, Porto, Portugal, 5 Ipatimup-Institute of Molecular Pathology and Immunology of the University of Porto, Porto, Portugal, <sup>6</sup> Faculty of Medicine of the University of Porto, Porto, Portugal, <sup>7</sup> INEB - Instituto Nacional de Engenharia Biomédica, University of Porto, Porto, Portugal, <sup>8</sup> Brain Institute (BRAINS-InsCer), Pontifícia Universidade Católica do Rio Grande do Sul, Porto Alegre, Brazil

Helicobacter pylori is a pathogen involved in gastric diseases such as ulcers and carcinomas. H. pylori's urease is an important virulence factor produced in large amounts by this bacterium. In previous studies, we have shown that this protein is able to activate several cell types like neutrophils, monocytes, platelets, endothelial cells, and gastric epithelial cells. Angiogenesis is a physiological process implicated in growth, invasion and metastization of tumors. Here, we have analyzed the angiogenic potential of H. pylori urease (HPU) in gastric epithelial cells. No cytotoxicity was observed in AGS, Kato-III, and MKN28 gastric cell lines treated with 300 nM HPU, as evaluated by the 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. As we previously reported in neutrophils, treatment with 300 nM HPU also had an anti-apoptotic effect in gastric epithelial cells leading to a 2.2-fold increase in the levels of Bcl-X<sup>L</sup> after 6 h, and a decrease of 80% in the content of BAD, after 48 h, two mitochondrial proteins involved in regulation of apoptosis. Within 10 min of exposure, HPU is rapidly internalized by gastric epithelial cells. Treatment of the gastric cells with methyl-β-cyclodextrin abolished HPU internalization suggesting a cholesteroldependent process. HPU induces the expression of pro-angiogenic factors and the decrease of expression of anti-angiogenic factors by AGS cells. The angiogenic activity of HPU was analyzed using in vitro and in vivo models. HPU induced formation of tube-like structures by human umbilical vascular endothelial cells in a 9 h experiment. In the chicken embryo chorioallantoic membrane model, HPU induced intense neovascularization after 3 days. In conclusion, our results indicate that besides allowing bacterial colonization of the gastric mucosa, H. pylori's urease triggers processes that initiate pro-angiogenic responses in different cellular models. Thus, this bacterial urease, a major virulence factor, may also play a role in gastric carcinoma development.

Keywords: Helicobacter pylori, urease, endocytosis, vasculogenesis, angiogenesis

## INTRODUCTION

fmicb-08-01883 September 25, 2017 Time: 13:39 # 2

Helicobacter pylori is a Gram-negative bacterium that infects more than half of the world's population and it is the major cause of gastroduodenal diseases, such as gastritis, peptic ulcers, and gastric cancer (IARC, 1994). H. pylori infection was responsible for 770,000 new cases of gastric cancer worldwide in 2012 (Plummer et al., 2016). Gastric cancer is the third most common cause of cancer-related deaths worldwide according to GLOBOCAN 2012 database, and risk factors include smoking, obesity, diet and bacterial infections, most importantly, H. pylori infection (Ferlay et al., 2013).

This bacterium produces large amounts of urease, an important virulence factor involved in a series of processes that allow bacteria to colonize and induce a strong inflammatory response in the gastric epithelium. HPU enzymatic activity causes hydrolysis of urea into ammonia, thereby neutralizing the acid environment of the stomach (Marshall et al., 1990), leading to altered properties of the gastric mucous layer (Perrais et al., 2014). HPU is a large protein, consisting of a dodecameric organization of two subunits (UreA, 26.5 kDa; UreB, 61.7 kDa; Ha et al., 2001). Its smaller subunit UreA was found to localize in the nuclei of cultured gastric epithelial cells, leading to altered morphology of AGS cells (Lee et al., 2012, 2015). Additionally, both HPU subunits, UreA and UreB, were observed inside outer membrane vesicles secreted by H. pylori and vesicles derived from pathogenic bacteria are often involved in toxin delivery to the host (Olofsson et al., 2010). By a mechanism not yet fully understood, HPU is also involved in the dysregulation of gastric epithelial tight junctions (Wroblewski et al., 2009) and independently of its enzyme activity, it promotes activation of neutrophils both in vivo and in vitro conditions (Uberti et al., 2013; Carlini and Ligabue-Braun, 2016).

Angiogenesis, the formation of new blood vessels from the pre-existing vasculature, is essential for tumor growth, invasion, and metastatic dissemination and it plays a central role in the progression of gastric cancer by providing nutrients and oxygen (Hanahan and Weinberg, 2011; De Palma et al., 2017; Macedo et al., 2017). H. pylori infection can induce and modulate the synthesis of angiogenic and invasive factors in gastric cancer cells (Kitadai, 2010), and gastritis H. pylori-positive patients have an increased number of blood vessels in the gastric mucosa compared to H. pylori-negative patients (Yeo et al., 2006). Recently, Liu et al. (2016) showed that H. pylori colonization correlated to the depth of tumor invasion and higher stage metastasis. VEGF is overexpressed in H. pylori-positive patients and gastric cancer tumors (Okines et al., 2011). However, VEGF overexpression is a poor prognosis indicator of increased angiogenesis in gastric cancer, suggesting that different factors may be involved in this process (Pinto et al., 2017).

Here, we seek to investigate whether HPU plays a role in the angiogenesis process induced by H. pylori.

## MATERIALS AND METHODS

## Cell Culture

Human AGS (ATCC <sup>R</sup> CRL-1739TM), AGS-Ecad (Oliveira et al., 2009), Kato-III (ATCC <sup>R</sup> HTB-103TM), MKN28 (JCRB0253TM) were maintained in RPMI 1640 medium with GlutaMAXTM (Invitrogen, Thermo Fisher Scientific, Inc., Waltham, MA, United States) supplemented with 10% fetal bovine serum (FBS, HyCloneTM, GE Healthcare Life Sciences, Logan, UT, United States), 200 IU/mL penicillin G-200 µg/mL streptomycin sulfate (Invitrogen) at 37◦C under 5% CO<sup>2</sup> humidified atmosphere. Human umbilical vein endothelial cells (HUVECs; HUV-EC-C, ATCC <sup>R</sup> CRL1730TM) were maintained in Medium 199 (M199) with Earle's salts, stable glutamine, and 25 mM HEPES buffer (Biowest, Nuaillé, France) supplemented with 10% FBS, 100 IU-100 µg/mL Penicillin G-Streptomycin Sulfate (Gibco), 100 µg/mL Heparin (Sigma–Aldrich, St. Louis, MO, United States), and 30 µg/mL Endothelial Mitogen (ECGS) (BioMedical Technologies, Inc., Stoughton, MA, United States), in gelatin-coated (Sigma–Aldrich, St. Louis, MO, United States) tissue-culture Petri dishes (TPP <sup>R</sup> Plastic Products AG, Trasadingen, Switzerland), at 37◦C under 5% CO<sup>2</sup> humidified atmosphere.

#### Bacterial Strain and Growth Conditions

Bacteria were grown for 48 h at 37◦C under a microaerophilic atmosphere (GENbox microaer; bioMérieux S.A., Marcy l'Etoile, France) in TrypticaseTM Soy Agar with 5% sheep blood (TSAII; Becton, Dickinson and Company, Franklin Lakes, NJ, United States). Experiments were performed with H. pylori strain 26695 (ATCC <sup>R</sup> 700392TM; cag PAI+).

#### HPU Purification

Helicobacter pylori bacteria grown for 48 h in TSAII (40 plates) were harvested in phosphate buffer (20 mM NaH2PO4, pH 7.5) and lysed by ultrasound (Ultrasonic Homogenizer 4710; 10 pulses of 30 s in ice bath). After centrifugation (20 min, 20,000 × g, 4 ◦C), the supernatant was processed according to a previously published protocol to obtain purified HPU (Wassermann et al., 2010). Protein purity was assessed by gel electrophoresis in 10% polyacrylamide gels containing 0.1% sodium dodecyl sulfate (SDS-PAGE).

#### Immunofluorescence and Microscopy

Immunofluorescence studies and analysis by laser scanning confocal microscopy were performed using a PL APO 63× NA 1.40 oil objective (Spectral Confocal Microscope Leica TCS-SP5; Leica Microsystems, Mannheim, Germany). The images were combined and merged using ImageJ (Schneider et al., 2012) with the plugin Bio-Formats (Linkert et al., 2010). For immunocytochemistry, after treatment with 100 nM

**Abbreviations:** b-FGF2, fibroblast growth factor; CAM, chorioallantoic membrane model; EG-VEGF, endocrine gland-derived vascular endothelial growth factor; GFP, green fluorescent protein; HB-EGF, heparin- binding epidermal growth factor-like growth factor; HPU, Helicobacter pylori urease; HUVECs, human umbilical vascular endothelial cells; IGFBP, insulin-like growth factor-binding proteins; IL-1β, interleukin 1β; mβCD, methyl-β-cyclodextrin; MTT, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide; NRG1-B1, neuroglin; PF-4, platelet factor-4; PTX3, pentraxin-3; TSAII, trypticase soy agar with 5% sheep blood; VEGF, vascular endothelial growth factor.

HPU for 5, 15, and 45 min, the AGS cells were fixed with 4% paraformaldehyde (Polysciences, Warrington, PA, United States), washed with phosphate buffered saline (PBS), and permeabilized with ice-cold 100% methanol. Cells were incubated with the primary rabbit polyclonal antibody anti-UreB and mouse monoclonal antibody anti-Early Endosome Antigen 1 (EEA1) (Santa Cruz Biotechnology, Dallas, TX, United States). The secondary antibodies Alexa Fluor 488 goat anti-rabbit IgG and Alexa Fluor 596 goat anti-mouse (Thermo Fisher Scientific) were used to monitor the location of UreB and EEA1. 4<sup>0</sup> ,6-diamidino-2-phenylindole (DAPI) was used to visualize the nuclei (0.1 mg/mL, Molecular Probes, Eugene, OR, United States).

For live cell experiments, AGS cells were previously transfected with Lamp1-GFP (donated by Dr. Allan Levey, Emory University) using Lipofectamine 2000 transfection reagent (Invitrogen) according to manufacturer's recommendations. HPU was covalently tagged with Texas Red (Sulforhodamine 101 acid chloride; Sigma–Aldrich).

Helicobacter pylori urease incubated with Texas Red (0.5 mg/mL) during 1 h, at 4◦C, with continuous stirring. This sample was then exhaustively dialyzed against 20 mM phosphate buffer, pH 7.0, and gel-filtrated using a De-Salting column (Sigma–Aldrich) to remove any excess of free dye.

#### Cholesterol Depletion

Cholesterol was depleted in AGS cells by incubation with 5 mg/mL mβCD (Sigma–Aldrich) in serum-free medium at 37◦C under 5% CO<sup>2</sup> for 60 min (Hutton et al., 2010). This treatment did not affect cell viability as assessed by the trypan blue exclusion test (Strober, 2001).

#### Cell Viability Assay

The viability of the gastric epithelial cells upon HPU treatment was evaluated using the MTT assay 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (M2128) (Sigma–Aldrich), according to manufacturer's instructions. Cells were seeded at a density of 1.5 × 10<sup>5</sup> cells/well in 96-well plates, and then incubated with 300 nM HPU or PBS for 24 h. After incubation, 20 µL of MTT solution (5 mg/mL) was added to each well and incubated for 150 min at 37◦C. The media was carefully removed and 150 µL MTT solvent (4 mM HCl, 0.1% NP40, in isopropanol) was added to the wells. The absorbance was measured at 570 nm in a plate reader (SpectraMax M3, Molecular Devices, Sunnyvale, CA, United States). The experiments were performed in triplicates.

#### Analysis for Apoptosis-Related Proteins by Western Blot

Two different administrations schedules were used for HPU incubation in AGS cells. To investigate an "acute" effect of the protein, two doses of 300 nM HPU were added to the medium at 0 and 2 h, and the experiment was concluded after 6 h of total incubation. To simulate a "chronic" effect, cells received three doses of HPU, with additions of freshly diluted HPU (300 nM) at 0, 6 and 24 h time-points and the experiment was terminated after 48 h. Cell lysates from both groups were prepared, denatured in sample buffer (50 mM Tris-HCl, pH 6.8, 1% SDS, 5% 2-mercaptoethanol, 10% glycerol, 0.001% bromophenol blue) and heated in a boiling water bath for 3 min. Samples (30 µg total protein) were resolved in 10% SDS-PAGE gels and proteins were transferred to polyvinylidene difluoride (PVDF) membranes (Hybond-P, Amersham Pharmacia Biotech). Rainbow markers (Thermo Fisher Scientific, Inc.) were run in parallel to estimate molecular masses. Membranes were blocked with Tween-Tris buffered saline (TBS) (20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 0.1% Tween-20) containing 1% bovine serum albumin (BSA) and probed with rabbit monoclonal antibodies: anti-Bcl-X<sup>L</sup> (Cell Signaling Technology, Danvers, MA, United States, 1:500) and anti-BAD (Cell Signaling Technology, 1:500). Secondary antibodies (antirabbit, 1:20,000) coupled to horseradish peroxidase were from Jackson ImmunoResearch Laboratories, Inc. (West Grove, PA, United States). The protein bands were visualized using a chemiluminescence detection kit (Millipore, Billerica, MA, United States). The levels of protein expression were quantified using the software ImageJ and normalized against β-actin as an endogenous control.

#### Human Angiogenesis-Related Proteins

The Human Angiogenesis Array Kit (Proteome ProfilerTM-ARY007 Array, R&D Systems, Inc., Minneapolis, MN, United States) was used to detect the expression of angiogenesisrelated proteins in cellular extracts from AGS cells treated with 300 nM HPU for 9 h in comparison with PBS-treated cells, as control. The array was performed once according to manufacturer's instructions, using a pool of cellular lysates from three independent experiments, containing 300 µg of total protein.

### Endothelial Cell Capillary-Like Tube Formation Assay

Human umbilical vascular endothelial cells were seeded in 96-wells plates coated with 100 µL of growth factor-reduced MatrigelTM (Corning <sup>R</sup> Inc., Tewksbury, MA, United States) in serum-free RPMI 1640 GlutaMax medium (Gibco) at a density of 6 × 10<sup>4</sup> cells per well in the presence of 300 nM HPU or PBS (control) for a total of 9 h. HUVECs were allowed to stabilize for 3 h in a cell culture incubator at 37◦C with 5% CO<sup>2</sup> humidified atmosphere. The formation of endothelial network was then followed in the center of each well using a Leica DMI 6000 time-lapse microscope (Leica Microsystems, Germany) for 6 h, with a 10x magnification and z-stacks of 2.08 µm were acquired every 30 min. The number of tubes and branching points per microscopic field were automatically quantified using Angiogenesis analyzer plugin (Carpentier, 2012) for ImageJ software (Schneider et al., 2012).

## Chicken Embryo Chorioallantoic Membrane Angiogenesis Assay

The chicken embryo CAM model was used to evaluate angiogenic activity of HPU as previously described (Teresa Pinto et al., 2016).

HPU (50 nM, N = 16; 100 nM, N = 12 and 500 nM, N = 14), 2.78 µM of b-FGF2 (positive control, N = 14) and PBS/vehicle (negative control, N = 16) were tested. Briefly, fertilized chick (Gallus gallus) eggs obtained from commercial sources were incubated horizontally at 37.8◦C in a humidified atmosphere and referred as embryonic day (E). On E3, a square window was opened in the shell after removal of 2–2.5 mL of albumen to allow detachment of the developing CAM. The window was sealed with a transparent adhesive tape and the eggs returned to the incubator. On E10, 10 µL of test solution were placed on the top of growing CAM into a 3 mm silicon ring under sterile conditions. The eggs were re-sealed and returned to the incubator for three more days. After removing the ring, the CAM was excised from the embryos, photographed ex ovo under a stereoscope, at 20x magnification (Olympus, SZX16 coupled with a DP71 camera). The number of new vessels (less than 15 µm diameter) growing radially toward the ring area was counted in a blind fashion manner. Statistical analysis was performed as described below.

#### Statistical Analysis

The statistical significance of the differences between two groups was assessed using the unpaired Student's t-test and for multiple comparisons it was performed a two-way analysis of variance (ANOVA) followed by Dunnett's multi-comparison post hoc test, used to calculate significance. GraphPad Prism6 software (San Diego, CA, United States) was used to perform statistical analysis. Statistically significance was set at p-value ≤ 0.05. Data in graphs represent average ± standard error of the mean (SEM) of at least three experiments, unless otherwise stated.

## RESULTS

### HPU Is Not Cytotoxic to Gastric Epithelial Cells

To address whether purified HPU could affect cell proliferation and viability of gastric epithelial cell lines we performed the MTT cell proliferation assay. The incubation of AGS, Kato-III, and MKN28 cells with 300 nM HPU for 24 h does not interfere with the proliferation rate, nor with the cell viability of gastric cell lines, as indirectly measured in the MTT assay (**Figure 1A**). Further, the levels of Bcl-XL, an anti-apoptotic protein, and the levels of BAD, a pro-apoptotic protein, were evaluated by western blot in AGS cells after incubation with HPU under two experimental conditions. One of them aimed to simulate an "acute" effect of HPU (6 h of treatment), and the other a "chronic" effect (48 h of treatment) (**Figures 1B,C**). HPU induced

a significant decrease in the expression levels of BAD in both "acute" and "chronic" conditions when compared to untreated cells, of 55 and 80%, respectively. The levels of Bcl-XL, on the other hand, increased almost 100% relative to untreated cells, but only in the "acute" model of stimulation.

### HPU Is Internalized by Gastric Epithelial

To explore if HPU is internalized by gastric epithelial cells, we treated AGS cells with HPU and performed confocal immunofluorescence microscopy. After 15 min of incubation, HPU detected with an anti-UreB antibody and an Alexa-488-conjugated secondary antibody was found on the plasma membrane, and some molecules were already in the cytoplasm (**Figure 2A**). After 45 min post-stimulation, HPU was still seen in a specific area of the plasma membrane (**Figure 2B**), suggesting that the internalization process may occur via specific receptors or cellular structures. Next, to gain insight of HPU's internalization route, AGS cells were analyzed at shorter periods of treatment for co-localization of HPU and early endosomes, by confocal immunofluorescence microscopy. As shown in **Figures 2C,D**, during the initial steps of internalization HPU co-localizes with EEA1-positive early endosomes. After 10 min, a co-localization with Lamp1 was observed, suggesting the progression of HPU sorting to late endosomes/lysosomes (**Figures 2E,F**).

In order to investigate the involvement of cholesterol in the binding and endocytosis of HPU, AGS cells were treated with mβCD, a cholesterol scavenger, previously to treatment with HPU. **Figures 2G,H** show that binding of HPU to the cell membrane depends on cholesterol, suggesting that HPU internalization probably occurs through cholesterol-rich plasma membrane structures, such as lipid rafts.

## HPU Induces Angiogenesis

To investigate the involvement of HPU in angiogenesis we performed the endothelial cell tube formation assay. The addition of 300 nM HPU induced the formation of tube-like structure by HUVECs, after 9 h of treatment (**Figure 3A**). Quantification of the total number of tubes and branching points per microscopic field showed a significantly higher number of both structures in HPU-treated HUVECs than in control, the PBS-treated HUVECs (**Figure 3B**).

Further, to evaluate the role of HPU in an in vivo angiogenesis model we used the chick embryo CAM assay. All tested concentrations of HPU (50, 100, and 300 nM) induced a strong angiogenic response comparable to the positive control (b-FGF2) and a significantly higher angiogenic response than the negative control, as measured by the number of new blood vessels formed (**Figures 3C,D**). Together, the data from the in vitro and in vivo angiogenesis assays indicate that HPU can act as a pro-angiogenic factor in H. pylori infections.

## HPU Reshapes the Expression of Angiogenic Factors

To characterize the angiogenic profile of gastric epithelial cells upon treatment with HPU we performed an antibody

FIGURE 2 | Internalization of HPU by gastric epithelial cells occurs through a classic endocytic pathway and is dependent on cholesterol. (A,B) Confocal immunofluorescence microscopy showing internalization of HPU in AGS cells treated with 100 nM HPU. Cells were fixed, permeabilized and processed for the detection of HPU, using an antibody against urease (Ure-B subunit) stained with Alexa-488-conjugated secondary antibody (green), and for the detection of early endosomes, using an antibody for EEA-1 with Alexa-596-conjugated antibody (red). A representative result is shown for cells fixed after 15 min (A) and after 45 min (B) of treatment with HPU, at 37◦C. (C,D) Colocalization (yellow) of HPU (green) and EEA1-positive endosomes (red) in AGS cells treated with HPU for 5 min at 37◦C evaluated by confocal microscopy. (C,D) Show different sections of the same cell agglomerate. Inserts are 100x original magnifications depicting EEA1-positive endosomes (anti-EEA-1 antibody and Alexa-596 secondary antibody) containing HPU (anti-UreB antibody labeled with Alexa-488 secondary antibody)

(Continued)

#### FIGURE 2 | Continued

fmicb-08-01883 September 25, 2017 Time: 13:39 # 6

next to the plasma membrane. (E–G) Endocytosis of HPU involves Lamp1-positive endosomes and is impaired upon cholesterol depletion. (E,F) AGS cells were transfected with a Lamp1-GFP plasmid prior to the addition of TexasRed-labeled HPU at 100 nM. (G,H) Transfected cells were pre-treated with 5 mg/mL mβCD, for 1 h, washed, and then treated with 100 nM TexasRed-labeled HPU. After approximately 1 min of contact of the cells with HPU, the detection of fluorescence by confocal microscope was started and the time point called 0 min. Scale bars 10 µm.

angiogenesis array, using cellular extracts of AGS cells treated with 300 nM HPU and its untreated counterparts, pooled from three independent experiments. The effect of treatment of AGS cells with HPU on the expression of molecules involved in angiogenesis was then analyzed. The increased expression of angiogenin, angiopoietin-1, angiopoietin-2, EG-VEGF, endoglin, HB-EGF, IGFBP-1, IGFBP-2, IGFBP-3, IL-1β, neuroglin-B1 (NRG1-B1) and pentraxin-3 (PTX3), well-known pro-angiogenic factors and the decrease of anti-angiogenic factors platelet factor 4 (PF4), serpin B5, serpin F1 and vasohibin, were detected in the array assay, confirming the participation of HPU in angiogenesisrelated processes (**Figure 4**).

## DISCUSSION

Estimates are that H. pylori infection accounts for 5.5% gastroduodenal cancer worldwide and for more than 60% of all gastric cancer cases (Correa and Piazuelo, 2011). The infection by H. pylori causes damage of gastric tissues in humans, with epithelial cell cytoplasmic vacuolization and disorganization of gastric glands in the mucosa (Kusters et al., 2006). Ulcers due to gastritis in H. pylori infected patients have a typically slow healing process (Brzozowski et al., 1999). Previous studies using different models of endothelial cells in vitro have indicated that angiogenesis, a process that enhances microcirculation and is critical to recover to wounded tissue, is impaired by H. pylori infection (Kim et al., 2004; Tobin et al., 2008). Kim et al. (2004) have demonstrated that treatment of HUVECs with a water extract of H. pylori significantly inhibited capillary tube formation and decreased the expression of VEGF and angiopoietin.

Several damages seen in gastric tissues infected with H. pylori have been ascribed to the vacuolating toxin, VacA. For example, Tobin et al. (2008) reported that a conditioned medium obtained from a VacA-positive H. pylori strain inhibited bovine aortic endothelial cell functions through a VacA-dependent nitric oxide reduction mechanism. This conditioned medium reduced cell proliferation, tube formation and migration, effects that were blocked by the VacA inhibitor 5-nitro-2-(3-phenylpropylamino)benzoic acid. Moreover, no such effects were observed in cells treated with a VacA-negative H. pylori- or Escherichia coli-conditioned media (Tobin et al., 2008). VacA is known to associate with lipid rafts and is internalized by gastric cells (Gauthier et al., 2005). Moreover, VacA is known to reach the cytoplasm and promote vacuolization and affect mitochondrial function in gastric epithelial cells by mechanisms not yet fully characterized (Palframan et al., 2012).

On the other hand, other H. pylori-derived proteins promote angiogenesis and contribute to tumor invasiveness and growth. H. pylori heat shock protein 60 (HSP60) has been reported to play a role in tumor aggressiveness (Li et al., 2014) and to display proangiogenic activity in HUVECs (Lin et al., 2010). Cyclooxygenase-2 has also been implicated in angiogenesis and tumor invasiveness due to H. pylori (Chang et al., 2005). Here we demonstrate that HPU, a crucial virulence factor that allows bacterial survival and colonization of gastric mucosa, also plays a role in processes that underlie tumor growth in the stomach of infected patients.

Searching for a better understanding of how HPU exerts its pro-angiogenic effects, we have monitored the cytotoxicity of the purified urease on three gastric cell lines and studied its internalization by AGS cells. Incubation with 300 nM HPU for 24 h has not interfered in cell proliferation rates or in the cell viability of AGS, Kato-III, and MKN28 cells. This is a relatively high protein concentration considering the dose range (10–500 nM) of HPU which activates other cell types, such as platelets (Wassermann et al., 2010) and neutrophils (Uberti et al., 2013).

HPU's internalization is evident about 5 min following its addition to AGS cells. This internalization seems to be a receptoror structure-specific process since we could observe an increased density of HPU-immunoreactive molecules in a particular region of the plasma membrane after 45 min of incubation, when most of the protein had already been endocytosed (**Figure 2B**). The lack of binding and internalization of HPU after treatment of AGS cells with the cholesterol scavenger mβCD indicates that specific plasma membrane regions, such as cholesterol-rich lipid rafts could be involved in this process. It has been reported that H. pylori takes advantage of cholesterol-rich lipid rafts to induce responses in epithelial cells, such as the activation of NF-κB and expression of IL-8 (Hutton et al., 2010). Particularly relevant to H. pylori pathogenesis is its type IV secretion system, which delivers CagA and VacA into the gastric cell and interacts with lipid rafts (Schraw et al., 2002; Lai et al., 2008).

As shown in **Figures 2C,D**, HPU co-localized with early endosomes, positives for the marker EEA1. Early endosomes are the first endocytic compartment to accept incoming cargo after internalization from the plasma membrane and their primary function is the sorting of internalized molecules to different intracellular locations (Marko et al., 2010).

Lamp-1 is a marker found in either late endosomes or lysosomes, as a glycoprotein present in the cytoplasmic face of these organelles (Cook et al., 2004). **Figure 2F** shows that 10 min after the addition of HPU to AGS cells, the protein could be localized in structures containing Lamp-1, suggesting that the urease may be directed to a degradation route. The intracellular fate of HPU beyond this point was not investigated in this study. On the other hand, as an invasive bacterium, H. pylori induces autophagy in gastric epithelial cells, a process that results in the sequestration of cytosolic components within double-membrane vesicles called autophagosomes (Terebiznik et al., 2009). These vesicles fuse with lysosomes to become autophagolysosomes,

in which the vesicle contents are degraded by lysosomal hydrolases. H. pylori was reported to modulate and subvert this process preventing the formation of autophagolysosomes, thereby allowing bacterial proliferation (Greenfield and Jones, 2013). It has also been described that H. pylori induces the fusion of phagosomes, generating large less-acidic compartments called megasomes, which are necessary for intracellular bacterial survival. The formation of megasomes was shown to be

HPU-dependent (Schwartz and Allen, 2006). As suggested by the HPU-like immunoreactivity localized in Lamp1-positive compartments, the protein itself or fragments of it after lysosomal hydrolysis, may have a role in the regulation of the fusion of lysosomes and autophagosomes.

The strong inflammatory response of gastric epithelial cells to H. pylori infection results in an intense infiltration of the gastric mucosa by polymorphonuclear leukocytes, macrophages, and lymphocytes. The degree of mucosal damage correlates with the degree of neutrophil infiltration (D'Elios et al., 2007). H. pylori was shown to induce apoptosis of gastric epithelial cells both in vivo and in vitro (Cover et al., 2003). Our group has previously shown that HPU activates human neutrophils and delays their apoptosis by altering the expression of pro- and anti-apoptotic proteins (Uberti et al., 2013). Here, we showed that purified HPU tested on three different gastric cancer epithelial cell lines increased their survival and inhibited apoptosis by modulation of the levels of Bcl-XL, an anti-apoptotic protein, and of BAD, a proapoptotic protein. This result goes in line with the hypothesis of a cancer promoting effect of HPU, that could add to or potentiate the action of other tumor-inducing factors produced by the bacterium.

The production of high cytokine levels by gastric cells upon H. pylori infection has been reported (Fiorentino et al., 2013), accompanied by activation of matrix metalloproteinase 10 via EGFR and ERK-mediated pathways (Costa et al., 2016). To evaluate the participation of HPU in tumorigenic processes related to H. pylori, we investigated the role of this protein in angiogenesis by measuring the expression of molecules involved in this phenomenon using an array analysis, and in vitro tubulogenesis and in vivo angiogenesis models.

The capillary-like tube forming assay and the Chicken embryo CAM assay are two well-accepted in vitro and in vivo models in angiogenesis studies (Kim et al., 2004; Teresa Pinto et al., 2016). Platelet factor-4 (PF-4/CXCL4) was reported to inhibit angiogenesis using the CAM assay (Maione et al., 1990). Studies demonstrating down-regulation of VEGF and angiopoietin receptors in HUVECs by H. pylori employed a capillary tube formation assay (Kim et al., 2004). Here, we showed that purified HPU induced angiogenesis in both in vitro and in vivo models. Addition to the CAM of 1–5 picomoles of HPU produced a strong angiogenic effect equivalent to that obtained for 27.8 picomoles of b-FGF2, the positive control (**Figure 3D**). Thus, the angiogenic stimulus provided by HPU is about 10–15-fold more potent than that of b-FGF2.

It has been shown that the array analysis of molecules participating in angiogenesis in AGS cells after treatment with 300 nM HPU led to an increased expression of pro-angiogenic factors such as angiogenin (Miyake et al., 2015), angiopoietins 1 and 2 (Jones et al., 2001; Zhou et al., 2009), EG-VEGF, also known as prokineticin-1 (Goi et al., 2004; Alfaidy et al., 2014), HB-EGF (Ito et al., 2001; Mehta et al., 2008), endoglin (Ollauri-Ibanez et al., 2017), IGFBP 1, 2 and 3, considered as endothelial tumor

markers (Hintz et al., 1991; Lee et al., 2000; Le Jan et al., 2006), IL-1β (Kaluz and Van Meir, 2011; Christofides et al., 2015), NRG1-B1 (Gemberling et al., 2015), and PTX3 (Stallone et al., 2014). The same experiment showed contrasting results regarding the expression of anti-angiogenic factors in HPUtreated AGS cells, revealing decrease in the levels of PF-4 (Maione et al., 1990; Vandercappellen et al., 2011), serpin B5 and serpin F1 (Cher et al., 2003; Law et al., 2006), and vasohibin-1 (Sato, 2013).

## CONCLUSION

The set of results presented here allow us to conclude that H. pylori's urease, a well-recognized virulence factor for its enzymatic property that allows bacterial survival in the stomach mucosa, displays a potent and so far overlooked pro-angiogenic activity on gastric epithelial cells. Altogether with its platelet- and neutrophil-activating properties and pro-inflammatory activity, which are independent of its enzyme nature, HPU has potential to contribute in other ways to the pathogenesis of diseases caused by H. pylori. The data also emphasize the multifunctional character of this protein and throw light on the need of new approaches to better understand the pathologies associated with H. pylori.

## AUTHOR CONTRIBUTIONS

DO-S: planning and conducting the angiogenesis experiments; AU: planning and conducting the cell viability and internalization experiments; MSM: conducting immunofluorescence and cell transfection with GFP-Lamp1 plasmid; MTP: conducting and

#### REFERENCES


analysis of CAM assay; MG-L: conducting immunofluorescence experiments; DO-S, AU, CF, ML, and CC: have written and revised the manuscript; ML (in Portugal) and CC (in Brazil) have coordinated this study.

## FUNDING

This work was supported by FEDER funds through Portugal 2020, NORTE 2020, and COMPETE 2020 programs (NORTE-01-0145-FEDER-000029; POCI-01-0145-FEDER-007274), and by Fundação para a Ciência e a Tecnologia grants (FCT/CNPq-Proc◦ 4.4.1.00 and FCT/CAPES- FCT/4805/28/5/2014/S to ML) and fellowships (SFRH/BD/95631/2013 to MSM; SFRH/BPD/33420/2008 and SFRH/BPD/110065/2015 to ML). The Brazilian agencies Coordenação de Pessoal de Nível Superior (CAPES, Edital 63/2010 Toxinologia, proj.1205/2011; Edital CAPES/FCT proj 386/14) and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, Edital Universal Proc.N◦ 475908/2012-0 and N◦ 446052/2014-1) provided financial support to CC and fellowships (CAPES/PDSE/11164-12-3/2012 to AU; CAPES-FCT/BEX: 8207/14-3/2014 to DO-S).

### ACKNOWLEDGMENT

The authors acknowledge the support of the i3S Scientific Platforms b.IMAGE for laser scanning confocal microscopy and the In vivo CAM Assays Unit.




in inflammation, angiogenesis and cancer. Cytokine Growth Factor Rev. 22, 1–18. doi: 10.1016/j.cytogfr.2010.10.011


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Olivera-Severo, Uberti, Marques, Pinto, Gomez-Lazaro, Figueiredo, Leite and Carlini. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Clinical efficacy, Safety, and immunogenicity of a Live Attenuated Tetravalent Dengue vaccine (CYD-TDv) in Children: A Systematic Review with Meta-analysis

*Moffat Malisheni1,2,3,4\*, Svetlana F. Khaiboullina5,6, Albert A. Rizvanov6 , Noah Takah2,7, Grant Murewanhema2,8 and Matthew Bates9,10*

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Alexei B. Shevelev, Russian Academy of Sciences, Russia Tatiana V. Kuznetsova, National Institute for Health Development, Estonia Lydia Bogomolnaya, Texas A&M University Health Science Center, United States*

#### *\*Correspondence:*

*Moffat Malisheni malishenitasheni@gmail.com*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 28 May 2017 Accepted: 07 July 2017 Published: 04 August 2017*

#### *Citation:*

*Malisheni M, Khaiboullina SF, Rizvanov AA, Takah N, Murewanhema G and Bates M (2017) Clinical Efficacy, Safety, and Immunogenicity of a Live Attenuated Tetravalent Dengue Vaccine (CYD-TDV) in Children: A Systematic Review with Meta-analysis. Front. Immunol. 8:863. doi: 10.3389/fimmu.2017.00863*

*1Ministry of Health, Lusaka, Zambia, 2 Institute of Health and Wellbeing, University of Glasgow, Glasgow, United Kingdom, 3Department of Microbiology and Immunology, National University of Singapore, Singapore, Singapore, 4 Institute of Molecular and Cell Biology, Agency for Science, Technology and Research, Singapore, Singapore, 5Department of Microbiology and Immunology, University of Nevada Reno, Reno, NV, United States, 6Kazan Federal University, Kazan, Russia, 7Ministry of Health, Yaounde, Cameroon, 8College of Health Sciences, University of Zimbabwe, Harare, Zimbabwe, 9University College London Research & Training Programme, University of Zambia, University Teaching Hospital, Lusaka, Zambia, 10HerpeZ, University Teaching Hospital, Lusaka, Zambia*

Background: Dengue hemorrhagic fever is the leading cause of hospitalization and death in children living in Asia and Latin America. There is an urgent need for an effective and safe dengue vaccine to reduce morbidity and mortality in this high-risk population given the lack of dengue specific treatment at present. This review aims to determine the efficacy, safety, and immunogenicity of CYD-TDV vaccine in children.

Methods: This is a systematic review including meta-analysis of randomized controlled clinical trial data from Embase, Medline, the Cochrane Library, Web of Science, and ClinicalTrials.gov. Studies that assessed CYD-TDV vaccine efficacy [(1 − RR)\*100], safety (RR), and immunogenicity (weighted mean difference) in children were included in this study. Random effects model was employed to analyze patient-level data extracted from primary studies.

Results: The overall efficacy of CYD-TDV vaccine was 54% (40–64), while serotypespecific efficacy was 77% (66–85) for DENV4, 75% (65–82) for DENV3, 50% (36–61) for DENV1, and 34% (14–49) for DENV2. 15% (−174–74) vaccine efficacy was obtained for the unknown serotype. Meta-analysis of included studies with longer follow-up time (25 months) revealed that CYD-TDV vaccine significantly increased the risk of injection site reactions (RR = 1.1: 1.04–1.17; *p*-value = 0.001). Immunogenicity (expressed as geometric mean titers) in descending order was 439.7 (331.7–547.7), 323 (247 – 398.7), 144.1 (117.9–170.2), and 105 (88.7–122.8) for DENV3, DENV2, DENV1, and DENV4, respectively.

Conclusion: CYD-TDV vaccine is effective and immunogenic in children overall. Reduced efficacy of CYD-TDV vaccine against DENV2 notoriously known for causing severe dengue infection and dengue outbreaks cause for serious concern. *Post hoc*

**409**

meta-analysis of long-term follow-up data (≥25 months) from children previously vaccinated with CYD-TDV vaccine is needed to make a conclusion regarding CYD-TDV vaccine safety in children. However, CYD-TDV vaccine should be considered for use in regions where DENV2 is not endemic as currently there is no specific treatment for dengue infection.

Keywords: dengue hemorrhagic fever, dengue shock syndrome, CYD-TDV, dengue virus, efficacy, safety, immunogenicity

#### INTRODUCTION

Continuously increasing dengue virus (DENV) related morbidity and mortality poses a serious threat to global public health and has exerted pressure on national health budgets of endemic countries. There are four types of genetically distinct dengue viruses (DENV 1–4) (1), all causing severe dengue infection (2, 3). Brady et al. estimates that four billion people are at risk of acquiring dengue infection worldwide (4) with approximately 284–528 million dengue cases being documented annually (5). Dengue hemorrhagic fever (DHF)/Dengue Shock Syndrome (DSS) comprise 500,000 to one million of these cases leading to over 20,000 fatalities mostly in children (6, 7). The goal of World Health Organization is to reduce dengue related morbidity and mortality by 2020 (8). Despite the availability of vector control programs, dengue infection has continued to rise globally (9) with significant economic burdens and might continue to do so in the future given the ongoing climate change. Introducing an efficacious and safe vaccine in endemic regions has the potential to reduce dengue related hospitalization and death in children due to severe dengue infection (DHF/DSS). The risk of developing DHF/ DSS during secondary infection is increased when an individual is exposed to a dengue serotype that is different from the one previously experienced (10). This occurs due to antibody-dependent enhancement (ADE) which involves low levels of cross-reactive neutralizing antibodies produced during primary dengue infection forming complexes with target cell receptors (3, 11, 12). Consequently enhancing the number of dengue-infected cells and viremia (3, 11, 12). Therefore, a tetravalent dengue vaccine capable of eliciting a balanced immune response against all four dengue serotypes is warranted if complications due to ADE are to be averted (11, 13, 14). CYD-TDV vaccine is the most advanced live attenuated tetravalent dengue vaccine (15). However, comprehensive evidence regarding CYD-TDV vaccine efficacy, safety, and immunogenicity in children exclusively is absent.

A meta-analysis of randomized controlled trials (RCTs) by da Costa et al. has demonstrated that CYD-TDV vaccine is safe and induces a balanced immune response (16). However, safety and immunogenicity were determined for all age groups and no subgroup analysis based on the age of included participants was conducted. The fact that dengue infection is more severe in children compared to adults indicates that the two groups might respond to CYD-TDV vaccine differently with a possibility of adults confounding the true effect of the vaccine in children. The study also assessed vaccine efficacy by combining five primary studies. However, only Sabchareon et al. out of the five combined studies was designed to determine CYD-TDV vaccine efficacy in Thai children (17). The determined vaccine efficacy was 30.2% and was not statistically significant. Two large Phase III clinical studies designed to determine CYD-TDV vaccine efficacy and not included in the meta-analysis by da Costa et al. have been conducted in Asian and Latin American children showing efficacy of 56.5 and 60.8%, respectively (9, 18). However, because these studies were conducted in various age groups from different regions, the findings are not directly comparable. Therefore, to comprehensively address all the aforementioned concerns, we decided to determine the efficacy, safety, and immunogenicity of CYD-TDV vaccine in children by answering the following questions: (i) does CYD-TDV vaccine reduce the incidence of virologically confirmed dengue (VCD) cases in vaccinated compared with unvaccinated children? (ii) Does CYD-TDV vaccine increase the risk of adverse events in vaccinated as compared to the unvaccinated children? (iii) Is there a difference in geometric mean titers (GMTs) between children exposed to CYD-TDV vaccine and those unexposed?

#### METHODS

#### Eligibility Criteria and Definitions

This review was conducted and reported in accordance with the Cochrane and preferred reporting items for systematic reviews and meta-analyses guidelines (19, 20). Population: children were defined as all individuals under the age of 18 years (21). Intervention: CYD-TDV vaccine manufactured by Sanofi Pasteur. Vaccine was reconstituted in 0.4% sodium chloride and 2.5% serum albumin. Comparator: standard of care, placebo, or no intervention. Outcome: the primary end assessing points were CYD-TDV vaccine efficacy in accordance with the "Guidelines for clinical trials of dengue vaccine in endemic areas" (22). Reduction in the incidence of VCD cases per protocol analysis. Safety: AEs—unfavorable medical occurrences that are not treatment related and ARs—those that might be treatment related

**Abbreviations:** ADE, antibody-dependent enhancement; AE, adverse events; AR, adverse reactions; CYD-TDV, live-tetravalent dengue vaccine; DENV, dengue virus; DHF, dengue hemorrhagic fever; DSS, dengue shock syndrome; ELISA, enzyme-linked immunosorbent assay; GMT, geometric mean titre; JE, Japanese encephalitis; PRISMA, preferred reporting items for systematic reviews and meta-analyses; PRNT50, plaque reduction neutralisation test with a 50% plaque reduction threshold; RCT, randomized controlled trial; RR, relative risk; RT-PCR, reverse transcription polymerase chain reaction; SAE, severe adverse events; UAE, unsolicited adverse events; UN, United Nations; VCD, virologically confirmed dengue; WHO, World Health Organization; WMD, weighted mean difference; YF, yellow fever.

(23). Immunogenicity: levels of dengue neutralizing antibodies expressed as GMTs and measured using plaque reduction neutralisation test with a 50% plaque reduction threshold (PRNT50). More information on the PRNT50 test can be obtained in the "guidelines for plaque reduction neutralization testing of human antibodies to dengue viruses" (24). Study design: only RCTs were included in this review. CYD-TDV vaccination interval requirement was that immunizations be conducted at months 0, 6, and 12 (three vaccine regimen). Exclusion criteria: studies which did not assess CYD-TDV vaccine efficacy, safety, or immunogenicity or did not use CYD-TDV vaccine; studies involving participants aged over 17 years; and studies that used non-RCT study designs, non-three vaccine regimen or a three vaccine regimen with a different vaccination interval.

#### Literature Search and Data Extraction

A comprehensive search strategy was developed in collaboration with an experienced medical librarian to identify recently published studies as presented in Appendix I (all referenced appendices are in the Supplementary Material). Embase, Medline, Web of Science, the Cochrane Library, ClinicalTrials. gov, references of included studies, and authors served as sources for published data. Gray literature was not searched because it lacks quality control. The search for published data was initiated on 01/03/2016, concluded on 11/05/2016 and pilot tested according to the method proposed by Long (25) (Appendix II, Figure 1 in Supplementary Material). This was done to ensure that the analysis includes all relevant information and fit to achieve the goal of this study. Corresponding authors of primary studies were contacted to request for numerical data or clarifications in cases where data were incomplete or graphically presented (Appendix II, Figure 2 in Supplementary Material). Information was extracted based on individual patient-level data.

#### Data Items and Summary Measures

Primary end points and their respective summary measures included: overall and serotype-specific CYD-TDV vaccine efficacy (per protocol analysis), CYD-TDV safety [immediate AEs, severe adverse events (SAEs), solicited ARs, solicited injection site reactions (pain, erythema, and swelling), solicited systemic reactions (fever, headache, malaise, myalgia, and asthenia) and unsolicited adverse event (UAEs)]. Relative risk (RR) was the preferred summary measure for CYD-TDV efficacy [(1 − RR)\*100] (17) and safety, whereas immunogenicity (measured as GMTs) was estimated using the weighted mean difference (WMD). Relative risk was defined as the ratio of incidence VCD cases in the CYD-TDV vaccine group divided by the ratio of incidence VCD cases in the unvaccinated group (26). Mean difference was defined as an absolute difference between the GMTs in the intervention and control groups.

#### Risk of Bias and Statistical Analysis

The Cochrane Handbook for Systematic Reviews of Interventions tool (27) was used in this analysis to assess the quality of each of the included studies. The risk of bias was assessed both at the study and the outcome levels. Evidence tables served as the starting point for data synthesis. The tables were reviewed as well as the forest plots to determine the possibility of combining data from studies in a meta-analysis. The *I*2 and *Q* statistics were used to formally check for the presence of heterogeneity and consequently determine whether the effect sizes should be pooled. Heterogeneity was classified as low, medium and high for *I*<sup>2</sup> values corresponding to 25, 50, and 75%, respectively (28). If heterogeneity was either low/medium or reduced to these levels after being resolved, the pooled effect sizes of outcomes were explored. However, if variations in the effect size between pooled studies remained high (*I*<sup>2</sup> ≥ 75%) after efforts to resolve heterogeneity, meta-analysis was not conducted. The heterogeneity was investigated using metaregression and subgroup analysis to explain its possible cause. The Cochrane Collaboration recommends that studies with divergent effect sizes from the rest should be excluded to resolve heterogeneity (27). The influence of individual studies on the overall effect size was formally investigated using meta-influence. Identified studies were removed systematically to reduce heterogeneity across the combined studies. Sensitivity analysis was performed to explore the robustness of the findings using the fixed effects model (Appendices III and IV in Supplementary Material). The random effects model was preferred because the true effect size may not be constant across all the included studies (29) given that they were conducted in different age groups, countries, regions, and ethnic groups. Extracted data were exported from Excel spreadsheet to STATA version 13.0, where all statistical analyses were conducted. Publication bias was explored by employing Harbord's and Egger's tests.

## RESULTS

336 articles were selected for this study. Embase, Medline, Web of Science, the Cochrane Library, and ClinicalTrials.gov yielded 104, 80, 53, 46, and 53 articles, respectively (**Figure 1** below). Duplicate studies were removed using Mendeley reference manager leaving 194 published articles for further analysis. Of these, 174 articles were removed based on titles and abstracts. After an in-depth review, additional eight articles were excluded, including a press release (30), an abstract (31), one paper used a two regimen vaccination protocol (32), two papers used different vaccines, Acambis (33) and PVRV (34), and one meta-analysis of RCTs previously mentioned (16). Also, two studies utilizing a different protocol of the vaccination regimen and vaccine reconstitution protocol were excluded (35, 36). At the end of the selection procedure, nine studies were found eligible for inclusion in meta-analysis (9, 17, 18, 37–42) after one study was excluded (43) because it presented all data graphically.

A summary of the included studies' characteristics is presented in **Table 1**. Additional study activities are presented in Appendix II, Table 1 in Supplementary Material. All three studies that assessed vaccine efficacy used reverse transcription polymerase chain reaction and enzyme-linked immunosorbent assay tests to confirm dengue cases. Six studies did not specify how information regarding CYD-TDV vaccine safety was collected, while four requested parents/guardians of participants to record safety profiles. PRNT50 was utilized

by all included studies to determine immunogenicity. Details regarding the aforesaid are summarized in Appendix II, Table 2 in Supplementary Material. Quality assessment of eligible studies is presented in **Table 2**. Regarding RR and WMD analyses the values of no difference were 1 and 0, respectively. Therefore, 95% confidence intervals (95% CI) that traversed 1 regarding the former and 0 the latter were considered statistically insignificant (alpha > 0.05). This review considered three hypotheses: CYD-TDV vaccination does not reduce the incidence of VCD cases in children; CYD-TDV vaccination increases the risk of ARs in children; and there is no difference in GMT levels between vaccinated and unvaccinated children. Variables used in meta-regression and subgroup analyses included: gender, sample size, randomization ratio, blinding method, placebo type, age group, study location, RCT phase,

and flavivirus (DENV, yellow fever, and Japanese encephalitis) seroprevalence at baseline.

#### CYD-TDV Vaccine Efficacy

The review found a statistically significant pooled overall CYD-TDV vaccine efficacy of 54% (40–64; *p*-value < 0.001). This result implies that the vaccine reduces the risk of acquiring dengue infection in the intervention group relative to the control group by 54%. Serotype-specific efficacy showed that CYD-TDV vaccine was more effective against DENV4 (77%: 66–85; *p*-value < 0.001) and DENV3 (75%: 65–82; *p*-value < 0.001), while it was less effective against DENV1 (50%: 36–61; *p*-value < 0.001), DENV2 (34%: 14–49; *p*-value = 0.002) and unknown DENV serotype (15%: −174–74; *p*-value = 0.79). There was convincing evidence to reject the pre-specified null hypothesis for all but the unknown


#### Table 1 | Study characteristics of the vaccine trials that meet the inclusion criteria.

*RCT, randomized controlled trial; DENV, dengue virus; YF, yellow fever; JE, Japanese encephalitis; FV, flavivirus.*

#### Table 2 | Risk of bias assessment.


serotype, which was not statistically significant. The main findings for CYD-TDV efficacy including evidence for the presence of heterogeneity and publication bias are summarized in **Table 3**.

#### CYD-TDV Vaccine Safety

Generally, solicited injection site reactions (any) and solicited systemic reactions (any, fever, headache, and asthenia) showed an increased but statistically insignificant risk in vaccinated children compared with unvaccinated children. Other than the aforesaid, CYD-TDV vaccine reduced the risk of ARs. However, meta-analysis of included studies with longer follow-up time (25 months) revealed that CYD-TDV vaccination increased the risk of solicited injection site reactions; RR = 1.1 (1.04–1.17; *p*-value = 0.001) and RR = 1.09 (0.97–1.22; *p*-value = 0.145) using the fixed and random effects models, respectively. However, only the fixed model showed a statistically significant risk (Appendix

#### Table 3 | Main CYD-TDV efficacy findings.


*N, sample size; n, number of cases recorded; NA, not applicable (could not be calculated because input data contained 0 values); DENV 1–4, dengue virus serotypes.*

Table 4 | Main CYD-TDV safety findings.


*N, sample size; n, number of cases recorded; NA, not applicable (could not be calculated because few studies were pooled); SAE, severe adverse events; UAE, unsolicited adverse events.*

*a All parameters were determined after resolving heterogeneity by excluding divergent studies.*

IV, Figure 8 in Supplementary Material). Insignificantly increased risk of solicited systemic reactions was also observed using both the random and fixed effects models (Appendix IV, Figures 10 and 11 in Supplementary Material). None of the aforementioned variables used in subgroup analyses were associated with finding of a statistically significant heterogeneity (*I*<sup>2</sup> > 90%; *p*-value < 0.001). However, meta-regression analysis demonstrated that gender explained 95.1% (*p*-value = 0.049) and 96.5% (*p*-value = 0.002) of the variability in the studies that assessed solicited reactions and solicited injection site reactions, respectively (Appendix IV, Figures 2 and 3 in Supplementary Material). Negative gender coefficients entailed that for every unit increase in the proportion of males in the CYD-TDV group, the log RR reduced by 0.45 and 0.91 units for solicited reactions and injection site reactions, respectively. Following stratification of solicited reactions and injection site reactions by gender, heterogeneity in the subgroups of the pooled studies with more males plummeted to 0.0% (*p*-value < 0.41) and 61.4% (*p*-value = 0.035), respectively (**Table 4**). A study by Crevat et al. found that the most frequently reported safety profiles in the intervention group were UAEs (60–70%), solicited ARs (50–55%), solicited systemic reactions (40–50%), and solicited injection site reactions (<20%) (43). There were no immediate AEs or SAEs reported.

#### CYD-TDV Vaccine Induced Immunogenicity

Although it was challenging to determine the overall immunogenicity using the random effects model due to the persistence of variation in the effect sizes even after resolving serotype-specific heterogeneity, the fixed effects model generated 74.28 1/dil (69.90–78.68; *p*-value < 0.001). The combined serotype-specific GMT levels found after resolving heterogeneity in descending order was: DENV3 (439.7 1/dil), DENV2 (323 1/dil), DENV1 (144.1 1/dil), and DENV4 (105.71 1/dil). A different order was detected when the fixed effects model was applied; DENV3 (114.56 1/dil), DENV4 (112.34 1/dil), DENV2 (81.91 1/dil), and DENV1 (40.51 1/dil) (Appendix IV in Supplementary Material). This showed that the estimates of immunogenicity were not robust. Subgroup analysis and the systematic elimination of studies with divergent mean differences revealed that combining studies with different ages resulted in significant heterogeneity, except for DENV1. This was observed even within the same study (Appendix IV, Figures 4 and 5 in Supplementary Material). By contrast, pooling together studies with the same age resulted in low heterogeneity (*I*<sup>2</sup> = 0.0%), except for DENV4 (Appendix IV, Figures 6 and 7 in Supplementary Material). The main immunogenicity findings and respective 95% CIs are summarized in **Table 5**. The highest neutralizing antibody titers in descending order reported by Cravat et al. were as follows: DENV3 (311–387 1/dil), DENV2 (147–213 1/dil), DENV4 (127–160 1/dil), and DENV1 (105–124 1/dil) (43). Tran et al. reported the following GMTs: DENV1—4:129, 216, 169, and 146 1/dil, respectively (39). The study further demonstrated that GMTs in children increased with increasing age: 2–5 years (64.7–143 1/dil), 6–11 years (93.9–185 1/dil) and 12–17 years (135–334 1/dil).

#### DISCUSSION

#### CYD-TDV Vaccine Efficacy

Although our findings show that CYD-TDV vaccine is protective against dengue infection overall, its reduced efficacy against DENV2 is extremely worrying. This is because DENV2 is known to cause severe dengue infection and is twice as likely to result in DHF/DSS compared to other serotypes (6). The Asian DENV2 has also been reported to cause outbreaks of DHF/DSS, highly pathogenic and is gradually replacing the less pathogenic Latin American variant (44). All efficacy studies conducted in Asia demonstrated reduced and statistically insignificant vaccine efficacy against DENV2 compared with the one conducted in Latin America. This might indicate that CYD-TDV vaccine induced antibodies readily neutralize the less pathogenic Latin American than the highly virulent Asian DENV2 variant. It has been reported that DENV2 neutralizing antibodies induced after primary infection with DENV1 demonstrate differential neutralizing activity against the Asian and Latin American DENV2 variants (45).

It is worth noting that the observed 32% CYD-TDV vaccine protection against DENV2 is merely the best estimate. The real vaccine efficacy in the population can be as low as 14% (**Table 3**). Another important point to note is that CYD-TDV vaccine induced DENV2 neutralizing antibodies had the second highest GMT levels and yet provided the least protection. The question is why? Villar et al. concluded that GMT levels elicited by CYD-TDV vaccine do not reflect serotype-specific vaccine efficacy (9). It has also been suggested that dengue antibodies either might not be the immunological correlate of protection or that each dengue serotype has its own protective titer threshold (46). Both findings correspond with our results which show that CYD-TDV vaccine protection was highest for the serotype with the lowest GMT levels and so on (**Tables 3** and **5**). Microevolution due to genetic recombination and natural selection occurring within individual dengue serotypes might explain lower efficacy of the CYD-TDV vaccine against DENV2 (47–49). Wahala et al. have shown that mutations occurring in the E protein (the major target for dengue neutralizing antibodies) have an effect on antibody binding and neutralizing activity (50). Therefore, reduced vaccine efficacy against DENV2 could be due to the fact that the circulating DENV2 virus acquired mutations in the E protein hence becoming antigenically and genetically different from the one included in the CYD-TDV vaccine. Reduced DENV2 vaccine efficacy might as well be as a result of dominant CYD-TDV vaccine induced antibodies that lack or have low serotype-specific neutralizing activity against the circulating DENV2 serotype. This is because high levels of serotype-specific neutralizing antibodies are known to confer protection against subsequent DENV infection (45). Evidence depicting a similar situation can be derived from the seasonal influenza vaccine which was only 23% effective in vaccinees that were infected with a subtype that was different from the one included in the vaccine (51). Although information regarding the unknown DENV serotype is limited, a study by Mustafa et al. has proposed the emergence of a DENV5 serotype (48). By contrast, Hesse argues that the aforesaid is highly unlikely because DENV mutation rate is too low to lead to

#### CYD-TDV Vaccine Safety

the creation of a new serotype (44).

Our overall findings regarding the safety of CYD-TDV vaccine are in agreement with the findings by da Costa et al. (16) in that none of the increased AEs and ARs were statistically significant. Crevat et al. also concluded that CYD-TDV vaccine does not increase the risk of AEs and ARs in children below 2 years (43). However, this study had a small sample size (*N* = 90), short follow-up time (18 months), and it is not clear how allocation concealment was done. All of which might lead to the actual effect of CYD-TDV vaccine being overestimated. Contrary to the aforesaid, revelations from meta-analysis of studies with


*WMD, weighted mean difference expressed as GMTs; GMTs, geometric mean titers (1/dil); DENV 1–4, dengue virus serotypes. a All parameters were determined after resolving heterogeneity by excluding divergent studies.*

longer follow-up time indicate that CYD-TDV increases the risk of injection site reactions significantly thus making it difficult to reject the pre-specified hypothesis (Appendix IV, Figure 8 in Supplementary Material). Similarly, *post hoc* analysis of the data from Capeding et al. (18) has demonstrated that the risk of hospitalization and severe dengue in children aged between 2 and 5 years vaccinated with CYD-TDV was highly significant (RR = 7.45: 1.15–313.8, *p*-value not provided) (52). However, *post hoc* meta-analysis of all studies conducted to assess CYD-TDV vaccine safety in children is required before a comprehensive conclusion can be made.

#### CYD-TDV Vaccine Induced Immunogenicity

Our immunogenicity findings show that GMTs significantly increased in CYD-TDV vaccinated compared to unvaccinated children. Children in Latin America had higher GMT levels compared to those in Asia. Da Costa et al. (16) have reported similar results and evidence from our findings is sufficient to reject the pre-specified null hypothesis. However, there was significant variation in the effect sizes presented in the included studies. Interestingly, reduced heterogeneity was observed when studies with the same age group were combined regardless of study location except for DENV4. By contrast, combining studies with different age groups resulted in significantly increased heterogeneity regardless of study location and, surprisingly, even within the same study. Furthermore, we found that GMT effect sizes increased as the age of study participants increased, which corresponds with the findings by Tran et al. (39). Taken together, our findings clearly demonstrate that age influences CYD-TDV vaccine induced GMT levels. Included studies demonstrated that prior exposure to dengue viruses increased antibody response during subsequent infection. The observed variations in GMT levels among the included studies might be explained by variations in the burden of dengue infection across countries and between the two regions, with Asian countries experiencing a higher burden compared with Latin American countries (53). Although the PRNT is considered the gold standard, discrepancies between laboratories and regions have been reported (54, 55). Furthermore, the scientific community has different views regarding this test. Rainwater-Lovett et al. and Thomas et al. have separately reported that PRNT gives varying results based on the test conditions applied (56, 57). Endy has described as confusing the fact that PRNT does not give information on whether an individual will be protected from subsequent dengue infection using cross-reactive neutralizing antibodies (58). The former concern might be responsible for the observed variations in CYD-TDV vaccine elicited GMT levels across the included studies, while the later may be more applicable to the reduced vaccine efficacy against DENV2 as stated above. To the contrary, Timiryasova et al. have concluded that the PRNT test is fit for purpose and can "detect and measure dengue serotype-specific neutralizing antibodies in human serum samples with acceptable intra-assay and inter-assay precision, accuracy/dilutability, specificity, and with a lower limit of quantitation of 10" (59).

## Strengths and Limitations

This, to the best of our knowledge, is the first systematic review to assess CYD-TDV vaccine efficacy not only in children exclusively but also using primary studies that were designed specifically to determine vaccine efficacy. Our review has demonstrated that children of varying ages respond to CYD-TDV vaccination differently and gender imbalance in the CYD-TDV group introduces heterogeneity when assessing solicited reactions and injection site reactions. Excluding graphically presented data prevented estimation bias. Since all of the included studies were conducted in children located in Asia and Latin America, the findings cannot be generalized to children of all endemic regions.

## Implications for Public Health and Research

Emergence of novel and virulent dengue serotypes has been proposed. To prevent possible pandemics, there is need to strengthen vector control programs, dengue surveillance, diagnostic capabilities and management, and research through capacity building in endemic settings such as Asia, Latin America, and especially Africa, where dengue infection happens to be neglected (60, 61). DENV2 included in the CYD-TDV vaccine needs to be updated to match the currently circulating Asian and Latin American variants. In addition, understanding of dengue neutralizing antibodies by investigating whether they are correlates of protection and what titer thresholds against all four serotypes are optimal for protection is warranted. Therefore, the scientific community needs to quickly come to a consensus regarding the PRNT test. Otherwise, new serological tests capable of being standardized to enable inter laboratory comparability, reproducibility and that can accurately and specifically measure dengue correlates of protection must be developed given that this is crucial for vaccine development. The fact that dengue infection is known to cause DHF/DSS and death in children is an indication that passive immunotherapy using serotype-specific neutralizing monoclonal antibodies that target conserved regions of the E protein might be a viable alternative to vaccination. Passive immunotherapy might significantly benefit younger children who are at higher risk and yet respond poorly to vaccines immunologically. Another reason for considering passive immunotherapy is that higher titers of serotype-specific neutralizing monoclonal antibodies can be administered without enhancing ADE, which is mainly caused by low titer levels of cross-reactive antibodies as previously mentioned. Future vaccine trials should consider employing ≥25 months follow-up time, stratify and provide age specific data to facilitate comprehensive and conclusive analyses. Finally, *post hoc* analysis can reveal vital vaccine-related safety information missed during the duration of the clinical trial. Therefore, the aforementioned analysis should be considered and encouraged where complete clinical data of participants involved in clinical trials are available.

## CONCLUSION

Overall, CYD-TDV vaccine is effective, but less efficacious against DENV2 in children. CYD-TDV vaccine is immunogenic in children with lower GMT levels observed in younger children compared to adolescents. Although the vaccine increased the risk of some safety parameters in vaccinated children insignificantly, meta-analysis of studies with long follow-up time revealed that CYD-TDV vaccine significantly increased the risk of solicited injection site reactions. Therefore, *post hoc* meta-analysis of the long term follow-up data (≥25 months) collected from the children previously vaccinated with CYD-TDV are needed before a comprehensive conclusion regarding CYD-TDV vaccine safety in children can be made. However, given the urgency for a dengue vaccine in endemic regions, CYD-TDV should be considered for use in regions where DENV2 is not endemic as currently there is no specific treatment for dengue infection.

### AUTHOR CONTRIBUTIONS

MM designed the study, performed the statistical analysis, and wrote the final report. SK, NT, and GM independently performed searches, selection, and data extraction of published articles. AR resolved disagreements from searches, selection, and data extraction. MB resolved disagreements from searches, selection, and

#### REFERENCES


data extraction. All authors made contributions and reviewed the final report.

## FUNDING

Sincere gratitude goes to William Spence and Hilda Emengo from the Institute of Health and Wellbeing, University of Glasgow and Healthcare Improvement Scotland, respectively, for agreeing to supervise this work. Further gratitude goes to the Chevening Secretariat (Foreign and Commonwealth Office, United Kingdom) for its financial support that led to the realization of this work. SK and AR were supported by RSF grant 15-14-00016 and Program of Competitive Growth of Kazan Federal University. RA was also supported by state assignment 20.5175.2017/6.7 of the Ministry of Education and Science of Russian Federation.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu.2017.00863/ full#supplementary-material.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Malisheni, Khaiboullina, Rizvanov, Takah, Murewanhema and Bates. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Infectious Agents and Inflammation: The Role of Microbiota in Autoimmune Arthritis

Andrea Picchianti-Diamanti<sup>1</sup> , Maria M. Rosado<sup>2</sup> and Raffaele D'Amelio<sup>1</sup> \*

<sup>1</sup> Department of Clinical and Molecular Medicine, Sant'Andrea University Hospital, Sapienza University of Rome, Rome, Italy, 2 Independent Researcher, Rome, Italy

In higher vertebrates, mucosal sites at the border between the internal and external environments, directly interact with bacteria, viruses, and fungi. Through co-evolution, hosts developed mechanisms of tolerance or ignorance toward some infectious agents, because hosts established "gain of function" interactions with symbiotic bacteria. Indeed, some bacteria assist hosts in different functions, among which are digestion of complex carbohydrates, and absorption and supply of vitamins. There is no doubt that microbiota modulate innate and acquired immune responses starting at birth. However, variations in quality and quantity of bacterial species interfere with the equilibrium between inflammation and tolerance. In fact, correlations between gut bacteria composition and the severity of inflammation were first described for inflammatory bowel diseases and later extended to other pathologies. The genetic background, environmental factors (e.g., stress or smoking), and diet can induce strong changes in the resident bacteria which can expose the intestinal epithelium to a variety of different metabolites, many of which have unknown functions and consequences. In addition, alterations in gut permeability may allow pathogens entry, thereby triggering infection and/or chronic inflammation. In this context, a local event occurring at a mucosal site may be the triggering cause of an autoimmune reaction that eventually involves distant sites or organs. Recently, several studies attributed a pathogenic role to altered oral microbiota in rheumatoid arthritis (RA) and to gut dysbiosis in spondyloarthritis (SpA). There is also growing evidence that different drugs, such as antibiotics and immunosuppressants, can influence and be influenced by the diversity and composition of microbiota in RA and SpA patients. Hence, in complex disorders such RA and SpA, not only the genetic background, gender, and immunologic context of the individual are relevant, but also the history of infections and the structure of the microbial community at mucosal sites should be considered. Here the role of the microbiota and infections in the initiation and progression of chronic arthritis is discussed, as well as how these factors can influence a patient's response to synthetic and biologic immunosuppressive therapy.

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Adriana Gruppi, Decision Research, United States Georg Singer, Medical University of Graz, Austria Yuji Naito, Kyoto Prefectural University of Medicine, Japan

> \*Correspondence: Raffaele D'Amelio raffaele.damelio@uniroma1.it

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 29 July 2017 Accepted: 26 December 2017 Published: 16 January 2018

#### Citation:

Picchianti-Diamanti A, Rosado MM and D'Amelio R (2018) Infectious Agents and Inflammation: The Role of Microbiota in Autoimmune Arthritis. Front. Microbiol. 8:2696. doi: 10.3389/fmicb.2017.02696

Keywords: microbiota, autoimmune arthritis, infections, immunosuppressive therapy, biologics

#### MICROBIOTA AND THE IMMUNE SYSTEM

Host–microbe interactions are the result of a co-evolution process, in which both partners have set borders and have developed mechanisms of tolerance or ignorance. In this context, hosts were able to establish a continuous connection with symbiotic bacteria, representing for the host a "gain of function." Bacteria provide help for different functions, including the digestion of complex carbohydrates as well as the absorption and supply of vitamins (LeBlanc et al., 2013). Worth mentioning is that, in the complex network of interactions of the microbiome, the environment should be taken into account since the outbreak of an infection and/or an inflammatory disease relies on the complex relationships between the microorganisms, the host, and the environment (such as dietary components, pollutants and even drugs) (Kim et al., 2016).

Microbiota are composed of archaea, bacteria, fungi, and viruses forming a complex ecosystem of which bacteria are the most well-characterized member. Any microbial imbalance resulting in a shift (i.e., loss or overgrowth of a species) and/or reduction in microbial diversity is defined as dysbiosis.

The mucosa has a dual function in our body. It works as a physical barrier, promoting pathogen exclusion, and is an absorption site that allows entry of food-derived metabolites and gas exchange. The mucosal immune system is not fully developed when we are born because many of the lymphoid structures, especially in the gut, are formed only upon stimulation through exposure to microorganisms. In fact, mice bred and maintained in germ-free conditions have small Payer's patches and lack isolated lymphoid follicles (ILF) in their bowel. Furthermore, serum levels of IgM natural antibodies and IgA secreting cells in the spleen and intestine are reduced in germ free mice (Rhee et al., 2004). This phenotype is rescued by gut colonization once the animals are transferred from isolators to conventional housing and food. At birth, colonization plays a crucial role in inducing and shaping the development of the secondary lymphoid organs as well as setting-up thresholds of reactivity for both innate and acquired immune responses. Thus, the neonate microbiota intervenes with the fine balance amongst infection, inflammation, and tolerance that may "imprint" the immune system for life.

Recently, Vatanen et al. (2016) followed-up the gut colonization of children from birth to 3 years of age. Children from Finland, Estonia and Russia were selected who shared a human leukocyte antigen (HLA) distribution typical of individuals at risk for autoimmunity (Vatanen et al., 2016). The study showed that, depending on the lipopolysaccharide (LPS) subtype produced by different Bacteroides species, Toll-like receptor-4 (TLR4) could be either activated or inhibited. This had an impact on tolerance to endotoxin, and, consequently, strongly conditioned the genetic predisposition of the infants to type I diabetes (T1D). Thus, the presence of different LPS subtypes in the gut could partially explain the difference in prevalence of early onset autoimmune diseases in Finland and Estonia compared to Russia. In line with the hygiene hypothesis, this work adds another clue to the correlation between exposure or lack of exposure to specific microorganisms in early life and the increase of autoimmune diseases in western countries.

The microbiota is comprised of both pathogenic and nonpathogenic bacteria, however, this distinction is not absolute. A microorganism may behave as commensal or as pathogenic with regards to the dietary components, nutritional milieu, broad-spectrum antibiotic treatment, co-infection or genetic background of its host. Moreover, even though commensals are by definition not supposed to induce immune responses, and lamina propria-derived antigen-presenting cells traffic preferentially to mucosal-associated lymphoid tissue (MALT) in order to restrict systemic immune responses against the microbiota, it is possible to find antibodies specific to commensals in the serum of healthy people, as well as circulating T cells reactive against non-pathogenic bacteria (Macpherson et al., 1996; Ergin et al., 2011). Also, in specific genetic conditions, such as in individuals with defects in interleukin-17 (IL17) signaling pathway, commensal microorganisms become pathogenic through invasion, causing fungal disease (Puel et al., 2011). Another condition in which commensals may be harmful is in mucosal leakage. Changes of permeability at mucosal sites or altered immune functions strongly condition the persistence of antigenic stimulation, not only at the place of microbial entry but also in faraway locations, such as articular sites (Asquith et al., 2014). Moreover, in frail patients, the administration of Saccharomyces cerevisiae as a probiotic has been anecdotally reported to induce sepsis (Montineri et al., 2008; Eren et al., 2014).

It is worth mentioning here that the microbiota itself plays a role in regulating the growth of opportunistic enteric infections, a mechanism termed colonization resistance (Stecher and Hardt, 2011). Often antibiotic administration, by altering the equilibrium of the gut microbiota, leads to overgrowth of Clostridium difficile. This infection represents, nowadays, a major clinical problem because C. difficile is one of the bacterial species with higher antibiotic resistance. Microbiota transplantation from healthy donors, by rescuing microbial equilibrium, has been one of the most successful strategies to treat serious C. difficile infection (Hand, 2016).

Bacterial products, such as the previously mentioned LPS, are sensed by enterocytes through TLRs. Microbial stimulation of enterocytes induce the production of antimicrobial peptides (AMPs) and the fucosylation of small intestinal epithelial cells by Fucosyltransferase 2 (Fut2), both of which can limit specific infections (Pickard et al., 2014). TLRs regulate innate immune responses but are also crucial for epithelial cell homeostasis at mucosal sites. Altered immune function of TLRs, for example due to different polymorphisms, may change binding to bacterial products, secretion of cytokines or chemokines, which consequently interferes with acquired immune cell generation/migration and inflammation. Thus, it can contribute to the persistence of inflammatory diseases such as rheumatoid arthritis (RA) (Rogier et al., 2015).

In addition to bacteria, bacterial-derived products can also control, directly or indirectly, the function of both epithelial and inflammatory cells. For example, two important metabolites are the short-chain fatty acids (SCFAs) and the lipid mediator

prostaglandin E2 (PGE2). Recognition of the SCFAs by innate immune cells is important to modulate inflammation in response not only to intestinal but even to articular damage (Maslowski et al., 2009). Commensals are also able to regulate the inflammatory activity of monocytes, by promoting the release of PGE2 that, in turn, down-modulates the activation of tissuedamaging neutrophils (Grainger et al., 2013).

The gut is the main site for the generation of the two most important T cell populations, the inducible regulatory T cells (iTregs) and CD4IL17-producing cells (Th17), both of which play critical roles in the development and persistence of autoimmune disorders. Colonization of the distal small intestine, by segmented filamentous bacteria (SFB), is fundamental to development of resident lamina propria dendritic cells (DCs) able to release IL6 and IL22 that trigger the loop of Th17-Treg cell in the newborn gut (Ivanov et al., 2009; Maslowski et al., 2009). Although many studies have addressed gut microbiota composition in humans, we still do not know which would be the "best/healthiest" microbial composition for the maintenance of the balance between regulatory (iTregs) and inflammatory (Th17) cells (Hand, 2016). Recent studies revealed that the homeostasis of the colonic Treg compartment relies more on the synergistic effects of different bacterial strains than on a single species of that strain. This observation strongly suggests the importance of the entire bacterial community in shaping the appropriate microenvironment that is able to sustain the production and maintenance of the anti-inflammatory environment (Atarashi et al., 2013).

#### ORAL/GUT DYSBIOSIS AND ARTHRITIS

Rheumatoid arthritis and Spondyloarthritis (SpA) are chronic inflammatory autoimmune diseases potentially leading to quality of life impairment and reduction in life expectancy. RA is characterized by the presence of autoantibodies [rheumatoid factor (RF) and anti-citrullinated protein antibodies (ACPAs)] and is characterized clinically by the emergence of erosive synovitis predominantly involving small joints, such as hands, wrists, and feet. The reported prevalence of the disease is 0.5–1% in the general population (Hayter and Cook, 2012).

Spondyloarthritis is a pleomorphic group of conditions that includes ankylosing spondylitis (AS), psoriatic arthritis (PsA), reactive arthritis (ReA), SpA associated to inflammatory bowel disease (IBD), acute anterior uveitis and axial non-radiographic SpA. The reported prevalence for AS is 2.4/1,000 (Dean et al., 2014). Arthritis is the most frequent extra-intestinal disease manifestation in patients with IBD (range 15–40%); conversely, up to 60% of patients with SpA present subclinical gut inflammation and 10% overt inflammation evolving to Crohn's disease (Mielants et al., 1995; De Vos et al., 1996; Olivieri et al., 2014).

The evidence of a direct role for the gut microbiota in the development of autoimmune arthritis arises from experimental studies on germ-free mice in which a reduced severity and/or incidence of arthritis has been demonstrated (Carding et al., 2015).

However, in man, a direct correlation between gut dysfunction and systemic inflammation has been clearly established only in AS. Some studies by Ciccia et al. (2012) have identified several alterations of the ileum architecture in AS patients that may be the cause for chronic inflammation, both local and systemic. AS patients showed disruption of the basal membrane with compromised epithelial cell permeability, hyperplasia of goblet cells (with increased mucins production) (Ciccia et al., 2012) and activation of Paneth cells, producing high levels of adenosine monophosphate, such as α-defensin 5 (Ciccia et al., 2010) and proinflammatory cytokines, such as IL23 (Ciccia et al., 2009). These characteristics may be the cause or the consequence of dysbiosis, with changes in the diversity of the commensal bacteria (Costello et al., 2015). Increased intestinal permeability, probably determined by genetic factors such as HLA-B27 expression (Rosenbaum and Davey, 2011; Lin et al., 2014; Bowness, 2015), results in continuous antigenic stimulation with activation of effector T cells. These effector cells do not show a clearly defined Th17, Th1, and/or Th9 polarization, due to the over-expression of an IL23 response by Paneth cells (Ciccia et al., 2009). Worth noting is that IL23 can regulate the maturation of autoreactive Th17 cells but can also induce chronic inflammation by stimulating IL17, IL6, IL8, and tumour necrosis factor (TNF)-α production in neutrophils and macrophages. Furthermore, a link between IL23R polymorphisms and susceptibility to psoriasis, PsA and AS have been shown (Liu et al., 2008).

In patients with PsA, subclinical gut inflammation is characterized by the overexpression of IL9 and the generation of Th9, Th17, and Th22 responses. Th9 cells produced in the gut are found to be increased in the peripheral blood and they are able to migrate, enter synovial tissues and become established. In the meantime, gut-activated Paneth cells express both IL9 and IL9 receptors, creating a positive autocrine loop that perpetuates local inflammation (Ciccia et al., 2016).

Solid data suggest that alterations in the oral microbiota can influence both progression and disease outcome in RA patients (Diamanti et al., 2016; **Table 1**). It has been speculated that certain species of bacteria, mainly Porphyromonas gingivalis (PG) which may be present in the oral cavity, and are increased in patients affected by periodontitis, can induce loss of tolerance and lead to citrullination of several peptides and proteins (Wegner et al., 2010; Quirke et al., 2014; Ciccia et al., 2016).

Citrullination is a post-translational modification that converts arginine residues into citrulline leading to relevant changes in the protein structure, charge and function. Porphyromonas gingivalis peptidyl-arginine-deiminases enzyme (PPAD) can catalyze citrullination on different self-proteins (mainly α-enolase and fibrinogen) thus resulting in immune evasion by these neo-epitopes and the production of ACPAs (McInnes and Schett, 2011).

Indeed, it has been reported a direct correlation exists between the serum level of antibodies to PG and ACPAs in RA patients (Mercado et al., 2001; Dissick et al., 2010; Scher et al., 2012; Lange et al., 2016). This observation has been confirmed by the increased prevalence of PG-induced periodontitis in RA patients as compared to healthy individuals. Likewise, other studies



RA, rheumatoid arthritis; SpA, spondyloarthritis; PsA, psoriatic arthritis; ACPAs, anti-citrullinated protein antibodies.

have demonstrated that the presence of periodontitis is a poor prognostic factor in RA patients treated with anti-TNFα agents. The clearing of a periodontal infection can reduce the severity of active RA and lead to a better disease outcome during synthetic and biologic immunosuppressant treatment (Ortiz et al., 2009; Savioli et al., 2012; Kaur et al., 2014).

In line with the above described observations, a case of complete and long-lasting recovery in a patient with early RA after PD treatment has been recently reported, suggesting that, at the onset of disease, periodontitis treatment may avoid the development of a chronic and progressive arthritis (Salemi et al., 2014).

The influence of the intestinal microbiota on RA is still a matter of debate. A lower representation of common commensals such as Bifidobacteria and Bacteroides species (Vaahtovuo et al., 2008) and an increase in Mycoplasma fermentans (Johnson et al., 2000), Escherichia coli (Newkirk et al., 2010) and Proteus mirabilis (Senior et al., 1999) has been reported in early RA patients, whereas another study showed that stool samples from early RA patients had significantly more Lactobacillus communities than healthy subjects (Liu et al., 2013). In patients with long-standing RA a decreased gut microbial diversity seems to correlate with disease duration and autoantibody levels as well as expansion of rare genera, Collinsella, Eggerthella, and Faecalibacterium (Chen et al., 2016). Other authors have reported, in the fecal microbiota of RA patients, an increase of Prevotella copri spp. with concomitant reduction of Bacteroidetes phylum members (Scher et al., 2013). Recently Zhang et al. (2015) described a dysbiosis of both fecal and oral samples from RA patients using metagenomic shotgun sequencing and a metagenome-wide association study; these concordant abnormalities in mouth and gut microflora were restored after immunosuppressive treatment. In particular, Haemophilus spp. was strongly reduced, whereas Lactobacillus salivarius was increased especially in oral microbiota of patients with active RA disease (Zhang et al., 2015; **Table 1**).

The role of the gastrointestinal tract in the pathogenesis of SpA was proposed by Costello et al. (2015) who have evaluated biopsy specimens from 9 early SpA patients finding a higher abundance of five families of bacteria (Lachnospiraceae, Ruminococcaceae, Rikenellaceae, Porphyromonadaceae, and Bacteroidaceae), accompanied by a reduction of Veillonellaceae and Prevotellaceae with respect to healthy controls. Tito et al analyzed the gut microbiota (colon biopsy specimens) of 27 patients with SpA and 15 healthy controls and showed that the intestinal inflammation was associated with the mucosal microbiota composition. Of particular interest, they found that the genus Dialister positively correlated with the SpA activity (Tito et al., 2016). Stebbings et al. (2002) were the first group that reported differences in the fecal samples of AS patients compared to healthy controls by using molecular analysis. They did not find significant differences in the microbiota composition but did find a higher proportion of sulfate-reducing bacteria in AS patients (Stebbings et al., 2002). Fecal samples and blood specimens obtained from children with enthesitis showed less Faecalibacterium prausnitzii spp. and Lachnospiraceae family members and a significant increase in the Bifidobacterium genus (Stoll et al., 2014). Finally, other authors have reported a significant increase in Firmicutes associated with a reduction in members of Actinobacteria phylum and Propionibacteria spp in the skin samples of patients with psoriasis, compared to healthy

controls (Fahlén et al., 2012). The gut microbiota observed in patients with PsA also showed limited diversity compared to healthy controls. In particular, it has been demonstrated that there is a decrease in the bacterial genus Akkermansia, Ruminococcus, and Pseudobutyrivibrio in fecal samples from PsA patients (Scher et al., 2015; **Table 1**).

There is a growing awareness that the microbiota diversity and composition can influence a patient's response to synthetic and biologic immunosuppressive therapy. Immunosuppressive drugs, such as cyclophosphamide and methotrexate (MTX), can induce a diffuse depletion of the gut microbiota associated with a decrease in the commensal anaerobic species in favor of potential pathogens that can damage the gut barrier, altering epithelial cell permeability with consequent bacterial translocation (Viaud et al., 2013; Karin et al., 2014). Different authors have underlined the importance of an intact intestinal microbiota in the response to sulphasalazine (SSA), a synthetic drug frequently used in autoimmune arthritis. They found a decreased efficacy of SSA therapy in patients with ulcerative colitis previously treated with severe antibiotic regimens (Das et al., 1973; Peppercorn and Goldman, 1972). Conversely, the use of a multi-strain probiotic regimen does not seem to interfere with the metabolism of SSA in RA patients (Lee et al., 2010).

By now, data on the impact of microbiota in patients receiving immunosuppressive biologics are limited and only refer to IBD patients. One research group has found a reduction in the bacterial species Firmicutes phylum in patients affected by relapsing Crohn's disease as compared to clinically stable patients with administration of anti-TNFα agents. This modification in gut microbiota was associated with a reduction in F. prausnitzii spp. predicting clinical relapse of the Crohn's disease (Rajca et al., 2014). More recently, Busquets et al. (2015) have demonstrated that adalimumab (fully human anti-TNFα monoclonal antibody) could partially restore a healthy microbiota in a small group of patients affected by Crohn's disease. Treatment resulted in the recovery of Firmicutes, Bacteroides and Actinobacteria diversity species whereas E. coli spp. was decreased (Busquets et al., 2015).

#### INFECTIONS AND ARTHRITIS

In autoimmune arthritis, such as RA and SpA, the hypothesis of an infectious event as the cause for disease induction and relapse has been under consideration for a long time. Arleevskaya et al. (2014) have shown during a 10-year follow-up study that minor infections are more frequent and prolonged in RA patients and their relatives than in families without a history of autoimmunity. This study suggests a role for some predisposing defects of anti-bacterial defense mechanisms, more than for a specific microorganism. However, although by now no causative pathogen has been clearly identified nor a direct correlation with a specific infection has been established, it does not imply its absence. In fact, many pathogens have been indicated as possible triggers for RA disease but heterogeneity of the studies impairs forming any conclusions (Arleevskaya et al., 2016).

The most intriguing and supported idea for a role of infection in triggering RA, is that the microorganism can enter to and be located in the intestine, lungs or mouth; where it can interact with environmental factors such as smoke and drugs and, with a genetic predisposition, lead to mucosal inflammation. This, in turn, results in increased immune activation and spreading of inflammatory mediators, which locally perpetuate the disease. Exacerbated immune activation may also lead to aberrant migration of intestinal macrophages and lymphocytes from inflamed mucosa to the joints (Salmi et al., 1995).

The complex and interwoven link between microbes and joint inflammation is a frequent challenge for clinicians.

A severe increase in the serologic inflammatory parameters (i.e., erythrocyte sedimentation rate and C reactive protein), induced by a concomitant infection in a patient with autoimmune arthritis under immunosuppressive therapy is commonly encountered in the rheumatologic clinical practice. Indeed, the stimulation of systemic inflammation induced by the infectious pathogen can resemble a reactivation of the autoimmune arthritis, leading clinicians to increase the immunosuppressive therapy with a consequent worsening and dissemination of the infection.

The link between microbes and arthritis also has a broad spectrum of clinical expression. In fact, it can range from septic arthritis, in which the pathogen is directly responsible for joint inflammation and damage, to ReA. In the latter, infections by microorganisms, mainly Gram negative, such as Chlamydia trachomatis, Yersinia, Salmonella, Shigella, and Campylobacter (Espinoza and García-Valladares, 2013), find a genetically predisposed background (i.e., HLAB27) eventually resulting in chronic inflammation. However, the mechanism underlying the immune stimulation and consequent migration to distant synovial tissue is poorly understood (Keat et al., 1978; Granfors et al., 1990; Merilahti-Palo et al., 1991). It is worth mentioning that T cell clones specific for enteric bacteria have been found in synovial tissue of patients with ReA (Hermann et al., 1990; Probst et al., 1993).

It can be argued that the early recognition and eradication of the infective focus can lead to a resolution of the inflammatory stimulus and consequent interruption of the chronic immune loop. This can be the case of periodontal treatment in early RA, antibiotic therapy for beta-hemolytic Streptococcus in rheumatic fever and anti-hepatitis C virus (HCV) treatment for mixed cryoglobulinemia.

Recently, Campisi et al. (2016) using a complex and elegant set of experiments in mice, demonstrated that microbial infection at mucosal sites can elicit a break in tolerance by promoting the presentation of self-antigens derived from epithelial cell apoptotic bodies. Apoptosis of infected colonic cells leads to the generation of both anti-pathogen and autoreactive T cells, mainly Th17 cells, with production of auto-antibodies and inflammation (Campisi et al., 2016). These results have important implications for understanding how infections can stimulate the development of autoimmunity in an immunological context different from bystander activation of autoreactive T cells or epitope mimicry. Further studies are needed to establish similar correlations in humans.

The risk of infections in autoimmune patients is higher than in healthy individuals, due to endogenous (dysfunctional immune system) and external factors (i.e., comorbidities and immunosuppressive therapy). The risk has been reported to be approximately double for infections of bone and joints, skin, soft tissues and respiratory tract (Myllykangas-Luosujarvi et al., 1995). Moreover, it is known that infections can induce disease relapses, present a more severe clinical outcome and can be a frequent cause of death in immune-suppressed subjects (Gonzalez et al., 2007).

Immunosuppressive treatment is the main exogenous factor contributing to the increased risk of infections in autoimmune patients. TNFα blocking agents are the first biological immunosuppressive therapy approved for both RA and SpA. Currently five different drugs targeting TNFα are available: infliximab (the first antibody to be approved) is a chimeric monoclonal antibody (mAb) with approximately one third murine and two thirds human sequences; adalimumab, a fully human mAb produced by chinese hamster ovary (CHO) cells; golimumab a fully human mAb; certolizumab pegol, a poly-etyhlene-glycol (PEG)ylated humanized Fab' fragment; and etanercept a fusion protein that consists of the p75 part of TNFR2 and a human IgG1 Fc domain (Salemi et al., 2015; Dieterich et al., 2016). Although these agents have shown a good safety and efficacy profile, up to 40% of the patients may present a primary or secondary drug failure or treatment-related adverse events (Day, 2002).

In the last decade, several research groups have studied in depth the molecular pathways involved in the development of autoimmune arthritis, allowing identification of new therapeutic targets. Among those, the anti-CD20 chimeric mAb (rituximab) (de la Torre et al., 2012); the humanized anti-human IL6 receptor mAb (tocilizumab) (Nishimoto et al., 2007), and the CTLA-4- Ig (abatacept), a fusion protein that works as an inhibitor of T cell activation (Moreland et al., 2006), are currently available for RA patients. More recently, mAb against the p40 subunit of interleukin 12/23 ustekinumab (Gottlieb et al., 2009) and the mAb against the IL17A, secukinumab have been introduced for the treatment of patients with SpA (Patel et al., 2013).

Different authors have evaluated the infection risk induced by corticosteroids (CS) as well as synthetic and biologic immunosuppressant drugs in SpA and RA patients. Studies generally showed that synthetic immunosuppressive therapy (mainly MTX), is relatively safe, whereas data on the possible increase in the number and/or severity of infections in patients taking TNFα blocking agents are contrasting. Doran et al. (2002) showed that the use of CS, older age, extraarticular manifestations of RA, leukopenia, and comorbidities, are strong predictors of infections, whereas the use of synthetic immunosuppressive agents was not associated with increased risk. These results were partially confirmed by Lacaille et al. (2008), who reported an increased risk of infections in RA patients taking CS, but not synthetic immunosuppressive therapy. Data on the infection risk for biologic drugs derives mainly from national registries among which are the British Biologics Register and the RABBIT German Registry. Data from these registries estimated a two-fold increased risk of serious infections which was directly related to the dose of CS, mainly when CS were associated with TNFα inhibitors (Dixon et al., 2006; Zink et al., 2014). Results obtained from meta-analysis studies have also indicated a significant increase in the occurrence of serious infections (40%), in patients receiving anti-TNFα agents, while the data for opportunistic infections were not conclusive (Bongartz et al., 2006; Minozzi et al., 2016).

We have recently described, in a retrospective observational cohort study, that the combination of anti-TNFα with CS appears to be the most "pro-infective" treatment, whereas synthetic immunosuppressant drugs, either alone or associated with CS, were relatively safe in both patients with RA or with SpA (Germano et al., 2014). In contrast with these real-life data, the placebo-controlled trials evaluating TNFα blockers in RA patients, did not find a higher rate of infections in treated patients with respect to the placebo group (Lipsky et al., 2000; Klareskog et al., 2004; Weinblatt et al., 2005).

## CONCLUSION

Patients with autoimmune arthritis are at high risk of infections, due to endogenous (dysfunctional immune system) and external factors (i.e., comorbidities, drugs).

Infections can induce disease relapses and be characterized by a severe clinical outcome in these patients, representing a frequent cause of death. Furthermore, the stimulation of systemic inflammation induced by the infectious pathogen, mimicking a reactivation of the autoimmune arthritis, frequently represents a confounding and dangerous factor for clinicians.

In the last few years the efforts of several researchers have focused on the gut and mouth microflora leading to the identification of specific abnormalities in both its composition and diversity. In susceptible individuals, the presence of PG could be a trigger factor for RA development by inducing the generation of citrullinated proteins via a PPAD enzyme on different self-proteins that results in the production of ACPAs. In SpA patients the increased intestinal permeability, probably induced by genetic factors (e.g., HLA-B27) could lead to a disruption of the basal membrane, hyperplasia of goblet cells and activation of Paneth cells producing high levels of AMPs and IL23. It would result in a continuous antigenic stimulation with activation of effector T cells.

Careful monitoring for infectious foci, especially in the mouth and intestine, should be recommended in order to allow their early recognition and eradication, thus resulting in removal of the inflammatory stimulus, inhibition of the chronic immune loop and improvement of the arthritis.

#### AUTHOR CONTRIBUTIONS

AP-D and MR drafted the manuscript and performed the revision of the literature. RD critically revised the manuscript giving important intellectual contribution.

## REFERENCES



in rheumatoid arthritis, psoriatic arthritis, and ankylosing spondylitis: a systematic review and meta-analysis. Expert Opin. Drug Saf. 15 (Suppl.1), 11–34. doi: 10.1080/14740338.2016.1240783


periodontitis treatment only? Case report and review of the literature. Medicine 93:e195. doi: 10.1097/MD.0000000000000195


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Picchianti-Diamanti, Rosado and D'Amelio. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Intestinal Microbiota Influences Non-intestinal Related Autoimmune Diseases

Maria C. Opazo1,2, Elizabeth M. Ortega-Rocha<sup>3</sup> , Irenice Coronado-Arrázola<sup>4</sup> , Laura C. Bonifaz <sup>5</sup> , Helene Boudin<sup>6</sup> , Michel Neunlist <sup>6</sup> , Susan M. Bueno<sup>4</sup> , Alexis M. Kalergis 4,7 and Claudia A. Riedel 1,2 \*

<sup>1</sup> Laboratorio de Biología Celular y Farmacología, Departamento de Ciencias Biológicas, Facultad de Ciencias Biológicas, Millennium Institute on Immunology and Immunotherapy, Universidad Andres Bello, Santiago, Chile, <sup>2</sup> Facultad de Medicina, Millennium Institute on Immunology and Immunotherapy, Universidad Andres Bello, Santiago, Chile, <sup>3</sup> Laboratorio de Inmunobiología, Facultad de Medicina, Departamento de Biología Celular y Tisular, Universidad Nacional Autónoma de México, Mexico City, Mexico, <sup>4</sup> Departamento de Genética Molecular y Microbiología, Facultad de Ciencias Biológicas, Millennium Institute on Immunology and Immunotherapy, Pontificia Universidad Católica de Chile, Santiago, Chile, <sup>5</sup> Unidad de Investigación Médica en Inmunoquímica Hospital de Especialidades Centro Médico Nacional Siglo XXI, Instituto Mexicano del Seguro Social, Mexico City, Mexico, <sup>6</sup> Institut National de la Santé et de la Recherche Médicale U1235, Institut des Maladies de l'Appareil Digestif, Université de Nantes, Nantes, France, <sup>7</sup> Departamento de Endocrinología, Facultad de Medicina, Pontificia Universidad, Metropolitana, Chile

Edited by: Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Richard Eugene Frye, Phoenix Children's Hospital, United States Matej Oresic, University of Turku, Finland

#### \*Correspondence:

Claudia A. Riedel criedel99@yahoo.com; claudia.riedel@unab.cl

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 09 November 2017 Accepted: 26 February 2018 Published: 12 March 2018

#### Citation:

Opazo MC, Ortega-Rocha EM, Coronado-Arrázola I, Bonifaz LC, Boudin H, Neunlist M, Bueno SM, Kalergis AM and Riedel CA (2018) Intestinal Microbiota Influences Non-intestinal Related Autoimmune Diseases. Front. Microbiol. 9:432. doi: 10.3389/fmicb.2018.00432 The human body is colonized by millions of microorganisms named microbiota that interact with our tissues in a cooperative and non-pathogenic manner. These microorganisms are present in the skin, gut, nasal, oral cavities, and genital tract. In fact, it has been described that the microbiota contributes to balancing the immune system to maintain host homeostasis. The gut is a vital organ where microbiota can influence and determine the function of cells of the immune system and contributes to preserve the wellbeing of the individual. Several articles have emphasized the connection between intestinal autoimmune diseases, such as Crohn's disease with dysbiosis or an imbalance in the microbiota composition in the gut. However, little is known about the role of the microbiota in autoimmune pathologies affecting other tissues than the intestine. This article focuses on what is known about the role that gut microbiota can play in the pathogenesis of non-intestinal autoimmune diseases, such as Grave's diseases, multiple sclerosis, type-1 diabetes, systemic lupus erythematosus, psoriasis, schizophrenia, and autism spectrum disorders. Furthermore, we discuss as to how metabolites derived from bacteria could be used as potential therapies for non-intestinal autoimmune diseases.

#### Keywords: microbiota, autoimmune disease, gut, microbiome, skin, CNS

## INTRODUCTION

Our body is colonized by millions of microorganisms that can survive in extreme environments surpassing difficult conditions, such as low pH or low oxygen (Peterson et al., 2015). The skin, gut, nasal, and oral cavities and genital tract are colonized by hundreds of different types of microorganisms and are known as "normal flora" or microbiota (Peterson et al., 2015). Lederberg defined the microbiota in 2001 as "the ecological community of commensal, symbiotic, and pathogenic microorganisms that share our body space" (Lederberg , 2001). For some authors the concept of microbiota comprehends mainly bacteria and while the concept of microbiota comprehends several different species among them are bacteria, archaea, fungi and viruses (Selber-Hnatiw et al., 2017). Recent scientific advances supported additionally by "omics analyses" have been crucial for the generation of a large amount of data relative to the composition of the microbiota (Almonacid et al., 2017). In fact, scientific progress has allowed the identification of the composition of the microbiota and the identification of specific microorganisms that live in the gut (Ferreira et al., 2017). The analysis of this information has contributed to revealing the complex relationship between the microbiota and the host. Evidence in the literature has shown that alterations in the proportion of these microorganisms can be associated to pathologies affecting humans (Aarts et al., 2017; Almonacid et al., 2017; Ferreira et al., 2017). Along these lines, in the past few years several scientific publications have shown a possible association between microbiota alterations and autoimmune diseases (Alkanani et al., 2015; Ma et al., 2015; Miyake et al., 2015; Breban et al., 2017; Kohling et al., 2017). These pathologies are characterized by an immune response against the body's own tissues causing inflammation and destruction of tissues and/or organs (Nagy et al., 2015). Autoimmune diseases are especially frequent in western countries, affecting majorly women (Davidson and Diamond, 2001). It has been proposed that lifestyle in this "modern era" can be affecting the microbiota composition causing a deregulation of the immune system (Berbers et al., 2017). Evidences in the literature have shown a strong link between microbiota composition and intestinal autoimmune diseases, such as Crohn's disease and inflammatory bowel disease (IBD) (Matsuoka and Kanai, 2015; Nishida et al., 2017; Powell and MacDonald, 2017). However, host gut microbiota seems also capable of influence autoimmune diseases that target tissues other than the intestine, including Type 1 diabetes (De Groot et al., 2017), multiple sclerosis (Hindson, 2017), arthritis (Felix et al., 2017), and psoriasis (Yan et al., 2017). Interestingly diseases like schizophrenia and autism are now considered to also have an inflammatory component suggesting that these ailments could also be associated to changes in intestinal microbiota (Dickerson et al., 2017; Vasquez, 2017; Wu, 2017; Yang et al., 2017; Cox and Weiner, 2018; Kopec et al., 2018). The aim of this review article is to analyze recent information supporting an association between gut microbiota composition and non-intestinal autoimmune diseases.

## THE MICROBIOTA THROUGH HUMAN LIFE

An adult of 70 kg in average has around 39 trillion of bacteria and 30 trillion of human cells (Sender et al., 2016) and at least 20% of the metabolites in the blood are derived from commensal bacteria (Rook et al., 2017). The gut microbiota consists of about 2,000 different bacterial species (Llorente and Schnabl, 2015) and most of them reside at the distal intestine (Kamada et al., 2013b). In general, human gut microbiota is comprised by two main dominants phyla Firmicutes and Bacteroidetes, which are susceptible to alterations due to factors such as age, genetics, diet, environment, and infection (Gill et al., 2006). The neonatal microbiota is highly different compared to adult microbiota (Pickard et al., 2017). Neonatal microbiota is strongly influenced by type of delivery at birth. Thus, a vaginal delivery allows the colonization of the mother's gastrointestinal microorganisms to the neonate, meanwhile in a C-section delivery the infant will present more microorganisms related to mother's skin (Wampach et al., 2017).

Certain data support the notion that bacteria colonization in humans will begin in the gestation at the womb (Stinson et al., 2017). Consistently, several reports have detected the presence of bacterial DNA in the amniotic fluid, umbilical cord, placenta, meconium, and fetal membranes (Khan et al., 2016; Stinson et al., 2017; Tschoeke et al., 2017). This evidence is in contrast with the hypothesis of the sterile womb and that bacteria colonization in humans begins only at birth or at breastfeeding (Funkhouser and Bordenstein, 2013). The neonatal microbiota initially will resemble very much to individual maternal microbiota (Wampach et al., 2017). During the following years, the microbiota will be shaped and changed by nutritional, physiological, and/or pathological events occurring through life (Chu et al., 2016).

Evidence suggests that a proper microbiota homeostasis is required for the maturation of central nervous system (CNS), as well as the immune system during different developmental stages, such as infancy, adolescence, and adulthood (Rook et al., 2017). Certain microorganisms included in the microbiota will contribute to an appropriate development of various human tissues and organs. However, other microorganism could increase the susceptibility to suffering certain pathologies (Berbers et al., 2017). Therefore, the diet is an important factor that can lead to changes in the microbiota composition (Cui et al., 2017). It has been reported that only 24 h are sufficient to alter the composition of the microbiota after a change in the diet of an individual (Wu et al., 2011). For example, high fat diets increase the presence of enterotypes, such as Bacteroides meanwhile a fiber-rich diet increases the amount of the Prevotella genus (Wu et al., 2011). It has been shown that inappropriate changes in the microbiota composition, known as dysbiosis, could cause harmful consequences to the host (La Fata et al., 2017). For example dysbiosis has been reported for patients suffering from type I and type II diabetes, IBD and colorectal cancer (CRC) (Peterson et al., 2015). Because it would be of importance to understand how the microbiota composition can impair the wellness of the host, significant research efforts are currently in place to develop new treatments for these pathologies based on restoring a normal microbiota composition.

## THE INTESTINAL BARRIERS FOR THE MICROBIOTA

The gut has developed different mechanisms to ensure a beneficial intestinal microbiota composition, as well as for regulating microbiota overgrowth and restricting pathogen colonization (Llorente and Schnabl, 2015; Gensollen et al., 2016). It is thought that the gut can produce an intestinal barrier by

secreting mucus and pro-inflammatory molecules that contribute to the establishment of innate and adaptive immunity (Feng and Elson, 2011).

Intestinal epithelial cells are the first constituents of the gut barrier and they cover the intestinal lumen and separate the gut microbiota from the immune system (Farhadi et al., 2003; Feng and Elson, 2011). Epithelial cells are maintained together by tight junctions (TJs), adherents junctions (AJ), and desmosomes (Hartsock and Nelson, 2008). TJs are multiprotein complexes comprised of integral membrane proteins, such as claudins, occludins, and junctional adhesion molecules (Hartsock and Nelson, 2008). The TJs regulate the passing of solutes and fluids through the epithelial cells by passive paracellular diffusion (Choi et al., 2017). As part of the epithelial cell barrier are goblet cells, which secrete glycoproteins to the lumen forming an inner mucus layer that is closer to epithelium and an outer mucus layer that is in contact with bacteria (Hooper and Macpherson, 2010). Additionally, epithelial cells can secrete antimicrobial proteins, such as defensins, cathelicidins, and C-type lectins (Chairatana and Nolan, 2017). These molecules contribute at controlling bacterial growth by either enzymatically degrading their cell wall, disrupting the inner membrane or depriving bacteria from essential heavy metals (Mukherjee and Hooper, 2015). Additionally, enterocytes, enteroendocrine cells, globet cells, and Paneth cells can also produce antimicrobial proteins contributing to the antimicrobial activity (Chairatana and Nolan, 2017).

The three main lymphoid structures of gut immune system that locate at the mucosa are: (1) the Peyer's patches (PP), which is the mucosa-associated lymphoid tissue that can be found in clusters; (2) the lamina propria (LP) located as an isolated lymphoid tissue where effector lymphocytes secrete cytokines and immunoglobulins; and (3) the epithelium layer in which intraepithelial resident lymphocytes can be found (Richards et al., 2016; Shi et al., 2017). The secretion of IgA is considered to be an antimicrobial agent that is accomplished by the help of dendritic cells (DC) from the PP. IgA interacts with bacteria impeding their adhesion to epithelial cells and inhibiting bacterial motility (McGuckin et al., 2011).

The gut microbiota is tightly associated and has constant communication with the mucosal immune system. It is thought that the mucosal immune system limits the invasion of tissues by the normal flora, which entails a high microbial diversity as well as potential pathogens that could have been ingested with the diet (Hooper and Macpherson, 2010). Such a function is in part carried out by DCs located at the mucosal surface where they uptake antigens and prime lymphocytes. DCs can directly sample normal flora and pathogenic bacteria (Kelsall, 2008). Despite all of these barrier mechanisms, bacteria can find ways for trespassing them and go across the epithelial layer triggering bacteria killing mechanisms. Rapidly, trespassing bacteria suffer phagocytosis and elimination by the LP macrophages (Kelsall, 2008).

Activated DCs can promote the differentiation of T cells to regulate immune tolerance (Shi et al., 2017). In fact, there is a high content of T cells at the intestinal mucosa that can be divided in two major subpopulations known as type A or conventional (TCRαβ) located at the PPs and mesenteric lymph nodes and type B or non-conventional (TCRγδ) that can be found almost exclusively at the epithelium (Van Wijk and Cheroutre, 2010). Naïve T cells become activated into effector helper T cells (Th) by mainly differentiating into: (1) Th1 cells that contribute to the elimination of intracellular pathogens; (2) Th2 cells that protect against parasites and mediate allergic reactions; or (3) Th17 cells that contribute to the clearance of foreign pathogens (Geremia et al., 2014). Intestinal DCs can also regulate the differentiation of T cells to regulatory T cells (Tregs). The role of Tregs is to reduce and control the immune response in part by suppressing the activation and proliferation of T helper cells through the secretion of anti-inflammatory cytokines (O'garra and Vieira, 2004). Several studies support the notion that an alteration of the balance between T helper and Treg cells can be closely associated to intestinal autoimmune pathologies (Fasching et al., 2017; **Figure 1**).

## MODULATION OF T CELL DIFFERENTIATION BY THE MICROBIOTA

Microbiota members can regulate the immune response through secretion of metabolites, such as short-chain fatty acids (SCFA). SCFAs are produced in the colon by the microbiota through fermentation of non-digestible carbohydrates including cellulose or inulin among others (Chen et al., 2015). The main products are acetate, propionate and butyrate, which are absorbed by the colon (Rios-Covian et al., 2016). While butyrate acts as an energy source for epithelial cells (Jung et al., 2015) and facilitates tight junction assembly (Peng et al., 2009), acetate and propionate are substrates for gluconeogenesis and lipogenesis in the liver and other peripheral organs (Rios-Covian et al., 2016).

SCFAs can modulate the intestinal immune response (Tremaroli and Backhed, 2012) by regulating T cell differentiation (Cavaglieri et al., 2003), epithelial barrier function, production of antimicrobial peptides, and the secretion of pro-inflammatory mediators (Johnson-Henry et al., 2014). Administration of butyrate in an animal model of colorectal colitis ameliorates the symptoms by increasing the percentage of Tregs and the production of IL-10 and IL-12 in peripheral blood, with an concomitant decrease of RORγt (a Th17 biomarker), IL-17 and IL-23 levels (Zhang et al., 2016). Furthermore, the addition in vitro of butyrate to human peripheral blood mononuclear cells (PBMCs) increased the differentiation of Tregs suggesting for this molecule a regulatory role in Treg/Th17 balance that influences the immune response (Zhang et al., 2016). Administration of SCFAs has been used in animal models of experimental autoimmune encephalomyelitis (EAE). It was observed that the oral administration of SCFAs butyrate, acetate, and propionate could significantly decrease EAE clinical score in mice (Mizuno et al., 2017). In these experiments, propionate showed the higher capacity to protect animals from the development of EAE (Mizuno et al., 2017) Interestingly, while treatment with propionate increased the frequency of Tregs in lymph nodes, treatment with butyrate did so in the spleens (Mizuno et al., 2017). These results reinforce the notion that

the normal flora and pathogenic bacteria. In the mesenteric lymph node DCs can promote differentiation of T cells, to regulate immune tolerance. Once T cells are activated they differentiate to T helper cells (Th), like Th1, Th2, or Th17 cells. Intestinal DCs will also regulate the differentiation of T cells to T regulatory (Treg) cells. Treg cells have the capacity to suppress the activation and proliferation of Th cells by the secretion of anti-inflammatory cytokines.

SCFAs from the intestine can regulate systemic inflammation that is mediated by lymphocytes.

Hashimoto et al. by using Ace2 knockout mice showed that a protein-free diet alters intestinal immunity (Hashimoto et al., 2012). Angiotensin converting enzyme-2 (ACE2) is a key regulatory enzyme of the renin-angiotensin system (RAS) as it catalyzes the conversion of angiotensin II (Ang II) to angiotensin 1–7 (Ang 1–7) the latter can bind the G-coupled protein Mas receptor inducing vasodilatation contrasting the effects of the binding of Ang II to its receptor AT-1 that promotes vasoconstriction and hypertension (Perlot and Penninger, 2013). ACE2 knockout male mice or Ace2−/<sup>y</sup> mice (Ace deficient at the × chromosome) induced with colitis and protein-deprived showed increased infiltration, ulceration, weight loss and higher diarrhea scores, as well as decreased serum levels of tryptophan (Trp) (Hashimoto et al., 2012). Trp is an essential amino acid for mammals and can only be obtained through the diet (Badawy, 2017). These knockout mice supplemented with Trp significantly improved colitis symptoms (Hashimoto et al., 2012). Additional analyses showed that these mice display an altered gut microbiota composition and an increase production of antimicrobial peptides, as compared to Trp-supplemented mice (Hashimoto et al., 2012).

Moreover, it has been observed that the microbiota produces catabolites from Trp or Trp-indole derivatives, such as Indole 3-acetamide, indole-3-acetic acid and indole-3-lactic acid and modulates the mucosal immune response through IL-22, which is produced by innate lymphoid cells 3 (ILC3) (Zelante et al., 2013). Consistently with these findings, Lamas et al. showed that the colitis associated knockout mice for Card9 display a microbiota dysbiosis (Lamas et al., 2016). An adaptor protein involved in the immune response against fungi dysbiosis (Etienne-Mesmin et al., 2017). Authors showed that Card9 knockout mice missed a Trp-metabolizing bacterium. Thus, the consequence is that these mice had low content of indole derivatives, which are important for the production of IL-22 by the ILC3 and T cells at the mucosa. Low levels of IL-22 generate a pro-inflammatory environment (Lamas et al., 2016), because this cytokine has antiinflammatory properties and belongs to the IL-10 cytokine family (Parks et al., 2015). IL-22 participates in host defense against extracellular pathogens by eliciting innate defensive mechanisms that promote the expression of antimicrobial peptides at mucosal surfaces (Rutz et al., 2013) and is also involved in tissue repair by enhancing epithelial cell proliferation (Aujla and Kolls, 2009).

Additionally, IL-22 can influence intestinal epithelial cell glycosylation by inducing the expression of fucosyltransferase 2 (Fut2) that catalyzes the fucosylation of membrane proteins, a post translational modification needed for protection against enteric pathogens, such as S. typhimurium (Okumura and Takeda, 2017). Recent data support the notion that microbiota has immune-modulatory properties, however little is known about the identification of the specific bacteria genera responsible for the phenotype and also few is known at molecular level for which mechanisms these bacteria modulate the inflammatory state of the intestine. So far, it has been described that the presence of polysaccharide A (PSA) in the gut commensal Bacteroides fragilis induces the secretion of IL-10 by Tregs, which in turn decreases inflammation in the gut and in distant tissues, such as the brain (Ochoa-Reparaz et al., 2010; Dasgupta et al., 2014). These findings highlight the role of metabolites produced by the intestinal microbiota to modulate inflammation and their potential use as a therapeutic tool to treat inflammatory and autoimmune diseases.

#### WHAT DO WE KNOW ABOUT THE ROLE OF THE MICROBIOTA IN NON-INTESTINAL AUTOIMMUNE DISEASES?

#### Autoimmune Diseases

Autoimmune diseases are pathologies characterized by an inappropriate immune response against own tissues and molecules that results in tissue-specific or systemic inflammation that leads to organ damage and malfunction (Rose and Bona, 1993; Marmont, 1994). Causes for autoimmune disease are multifactorial and range from genetic predisposition to the exposition of environmental agents, such as infectious agents, xenobiotics, drugs, or stress (Davidson and Diamond, 2001). The disease progresses from initial naive lymphocyte activation to a chronic state characterized by an increase in the number of autoantigens targeted by T cells and antibodies. Activated autoreactive B cells can function as antigen presenting cells for novel peptides and express co-stimulatory molecules. Antigens are processed and presented to naive T cells leading to the activation of additional autoreactive B cells that present new epitopes up to a point in which there is autoreactivity to a large number of autoantigens (Lanzavecchia, 1995; Liang and Mamula, 2000). The production of autoantibodies induces damages to tissues by the formation of immune complexes, cytolysis, or phagocytosis of target self-cells and interfering with proper tissue and cellular functions (O'Garra et al., 1997). Although there are several autoimmune diseases, in this review we will focus on non-intestinal autoimmune disorders, for intestinal autoimmune diseases, please refer to other reports (Gallo et al., 2016; Blander et al., 2017; Passos and Moraes-Filho, 2017).

## Graves's Disease and Hashimoto's Thyroiditis

The Grave's disease (GD) is an autoimmune disease characterized by the targeting of antigens derived from the thyroid gland. In GD there are autoantibodies against the thyroid stimulating hormone receptor (TSHR) (Kristensen, 2016). These autoantibodies activate the TSHR inducing the synthesis and secretion of thyroid hormones by the thyroid gland and causing hyperthyroidism. GD is the most common cause of hyperthyroidism and is more frequently observed in women than in men (Pokhrel and Bhusal, 2017). Shor et al. evaluated the prevalence of gastrointestinal auto antibodies in patients with Hashimoto's thyroiditis and Grave's disease (Shor et al., 2012). These are anti-gliadin antibodies (AGA), tissue transglutaminase (tTG) and anti Saccharomyces cerevisiae antibodies (ASCA). In particular ASCA have proven to be sensitive and highly specific for Crohn's disease. ASCA antibodies were highly prevalent in patients with GD (Shor et al., 2012). Analysis of fecal samples from GD patients showed an increased content of yeast supporting Schor's analyses (Covelli and Ludgate, 2017). It has also been observed the presence of antibodies against Yersinia enterocolitica and to Helicobacter pylori, but these responses vary among patients and are not observed in all the analyzed patients (Kohling et al., 2017). Next-generation sequencing projects intended to analyze and identify bacteria species in patients with GD. Using a TSHR immunized mouse model, it was observed an alteration of immunized animals when compared to controls. In humans, this modification is not fully observed, so far it has been observed in a small number of patients a significant decrease of the Bacteroides genus (Indigo, 2017). These are the first reports associating the gut microbiota and GD; therefore additional work must be accomplished to better understand the influence of the gut microbiota on the development of GD.

Hashimoto's thyroiditis (HT) is an autoimmune disease that is characterized by the infiltration of mononuclear cells in Opazo et al. Microbiota and Non-intestinal Autoimmune Diseases

the thyroid, together with the production of autoantibodies against thyroglobulin and thyroid peroxidase (TPO) (Antonelli et al., 2015). It is thought that environmental factors, such as diets higher in iodide, contribute to the etiology of HT (Rose et al., 2002). Recently, research efforts have focused on the involvement of microbiota in the pathogenesis of autoimmune diseases. The transfer of microbiota from conventional rats to specific pathogen free (SPF) rats increased the susceptibility of the latter to experimental autoimmune thyroiditis (Penhale and Young, 1988), which provides further support for an influence of the microbiota during HT pathogenesis. The use of the probiotic mixture (VSL#3TM) has been successful to reduce the susceptibility to developing autoimmune diseases, such as Type 1 diabetes and colitis by enhancing the production of IL-10 in Peyer's patches and the spleen (Calcinaro et al., 2005; Di Giacinto et al., 2005). Along these lines, it was important to explore whether probiotics could have a positive effect on HT. Contrarily to what was initially thought, the use probiotic strains of Lactobacillus rhamnosus HN001, Bifidobacterium lactis HN019, and L. rhamnosus GG failed to improve the disease outcome in a mouse model for autoimmune thyroiditis (Zhou and Gill, 2005). It has been demonstrated that a dysbiosis state can alter the epithelial barrier permeability leading to a condition known as "leaky gut" (Vaarala et al., 2008). At the histologic level, this is observed as morphological changes in epithelial cells and lymphocyte infiltration (Fritscher-Ravens et al., 2014). Interestingly, a similar observation has been made in patients with HT, in which both the space of two adjacent microvilli and the microvilli thickness are significantly increased. Furthermore, these patients were also evaluated for functional mucosal alterations using a lactulose/manitol test showing an increase in the recovery of lactulose/mannitol, which is consistent with the histological observations (Sasso et al., 2004). These data suggest that the microbiota and the epithelial barrier play an important role of during the pathogenesis of HT.

#### Type I Diabetes

Type 1 diabetes (T1D) is the most prevalent autoimmune disease in young people (<20 years), with a peak at 10–14 years old (Maahs et al., 2010) being more common in boys than in girls (Atkinson et al., 2014). The incidence of T1D is very variable worldwide, being higher in Europeans countries probably due to environmental factors (Xie et al., 2014) T1D is characterized by a T cell-mediated (CD4<sup>+</sup> and CD8+) destruction of β pancreatic cells (Atkinson et al., 2014). There are three major auto antigens associated to T1D, which are insulin, GAD65 (glutamic acid descarboxylase, 65 kDa isoform), and IA2 (Insulin autoantigen 2) (Ounissi-Benkalha and Polychronakos, 2008). The presence of antibodies specific for these antigens has been observed in the serum of T1D patients (Miao et al., 2007). A classical trio of symptoms characterizes T1D, which are polydipsia, polyphagia, and polyuria, accompanied by an overt hypoglycemia (Atkinson et al., 2014). All of them are used as hallmarks for T1D diagnosis in high-risk individuals, such as children and adolescents (Leslie, 2010).

The 60% of patients suffering from T1D has been attributed to a genetic cause (Ounissi-Benkalha and Polychronakos, 2008). The 50% of heritability of T1D is attributed to the human leukocyte antigen (HLA) alleles located in chromosome 6 and the rest is to non-HLA loci (Barrett et al., 2009; Redondo et al., 2017). It has been reported that are at least 40 non-HLA loci such INS, CTLA4, PTPN22, and IL2RA can contribute to disease susceptibility (Barrett et al., 2009). Moreover, T1D can also be triggered by environmental factors, such as cesarean or vaginal birth, diet, early infections in life etc. (Rewers and Ludvigsson, 2016). Experimental evidence has shown that the intestinal microbiota could induce T1D by priming the immune system at an early postnatal period (Endesfelder et al., 2016).

The first evidence linking the gut immune system and T1D derives from animal models. Non-obese diabetic (NOD) mice fed with regular commercial cereal-based chow and mice fed with a 10% casein-based diet presented the highest rates of T1D among the experimental groups (Elliott et al., 1988). These rates (26.9% cereal-based chow vs. 19.1% casein-rich diet) were significantly higher than the expected for this type of mice, which normally develop insulin-dependent diabetes at 200–250 days of age. Authors observed that the incidence of T1D in these mice was due to the 10% of casein in this diet, interestingly this percentage corresponds to the percentage of casein in cow milk (Elliott et al., 1988). Another important observation in children suffering T1D was the presence of antibodies against bovine serum albumin, a protein also contained in cow milk (Savilahti et al., 1988; Karjalainen et al., 1992; Saukkonen et al., 1996). These two observations support the notion that the diet could trigger the development of T1D, due that it contains potential antigens that will prime the immune system (Mejia-Leon and Barca, 2015; Rewers and Ludvigsson, 2016; Virtanen, 2016).

The notion that antigens derived from the diet can prime the immune system, suggests that the immune system is in contact with antigens and the intestinal permeability must be altered. In fact, there is evidence supporting a relationship between T1D and high intestinal permeability (Vaarala, 2008; Li and Atkinson, 2015; Maffeis et al., 2016). A study performed in 46 non-celiac T1D patients showed a significant increase of intestinal permeability as compared to healthy controls (Secondulfo et al., 2004). Authors performed electronic transmission microscopy (TEM) analyses over intestinal biopsies from non-celiac T1D patients. They observed a partial decrease in the microvilli together with morphological alterations at the tight junction domains (Secondulfo et al., 2004). Another study showed that T1D patients have high intestinal permeability measured as the urine levels of lactulose and mannitol 5 h after ingestion and high levels of zonulin in the serum (Sapone et al., 2006). Zonulin is a protein that can regulate intestinal permeability by disassembling tight junctions (Fasano et al., 2000). Studies performed in Biobreeding diabetes-prone (BBdp) rats, widely used as an animal model for studying human T1D (Bortell and Yang, 2012), showed increased intestinal permeability (Meddings et al., 1999). Using this animal model Neu et al. found in the small intestine of these animals a high number of globet cells and high intestinal mucus secretion before the onset of the disease, reflecting an inflammatory response at the intestine (Neu et al., 2005).

Other studies have found that intestinal microbiota could increase the probability to develop T1D. Studies performed in non-obese diabetic (NOD) mice at young age, when these animals are prediabetic, were infected by an oral gavage with wild type C. rodentium showed higher intestinal permeability, developed earlier insulitis (Lee et al., 2010) and showed an increased lymphocytic infiltration at the pancreas Langerhan's islets (In't Veld, 2011). These authors showed that the mutant strain of C. rodentium that lacks the ability to disrupt the intestinal barrier was unable to induce insulitis (Lee et al., 2010). Maffeis et al. observed in Italian T1D affected children an increased intestinal permeability that correlates with alterations in the microbiota composition (Maffeis et al., 2016). Interestingly, the authors found three microbial markers (D. invisus, G. sanguinis, and B. longum) highly represented in T1D affected children as compared to healthy controls (Maffeis et al., 2016). Furthermore, it has been observed that bio-breeding diabetes-resistant (BBDR) rats present more probiotic bacteria, such as Bacterioides, Eubacterium, and Ruminococcus (Roesch et al., 2009). It has also been observed in humans that the Bacteroidaceae family is over-represented in children with T1D together with a decrease of intestinal microbiota dominant species as Bifidobacterium adolescentis and B. pseudocatenulatum (De Goffau et al., 2013). Microbiota composition can be influenced by age, with major changes observed at early ages (Koenig et al., 2011; Arrieta et al., 2014; Rodriguez et al., 2015). Kostic et al. described that the microbiota of geneticallypredisposed infants from 3 to 36 months old, based on HLA genotyping, has reduced alpha diversity and overabundance of Blautia, Rikenellaceae, Ruminococcus, and Streptococcus genera (Kostic et al., 2015). In contrast, Maffeis et al. (2016) showed that D. invisus was completely absent in this samples indicating that there is still contradictory data regarding the microbiota composition in T1D patients.

#### Multiple Sclerosis

Multiple sclerosis (MS) is the most common autoimmune disease that has the CNS as a target (Reich et al., 2018). The immune system in the MS patients reacts against proteins that are found in myelin and neurons inducing axonal damage and neuronal death (Lemus et al., 2018). The MS patient develops several symptoms that lead to chronic disability including cognition impairment, loss of motor control, and sensitivity (Yong et al., 2017). Several reports have shown that MS patients have dysbiosis at the intestinal microbiota and it has been proposed that MS patients could have a specific type of microbiome (Chen et al., 2016; Probstel and Baranzini, 2017; Shahi et al., 2017; Tremlett and Waubant, 2017). It was shown that MS patients have a relatively low abundance of the Bacterioides, Parabacteroides, Prevotella, and Lactobacillus genera and an increased abundance of Akkermansia, Blautia, Ruminococcus, and Bifidobacterium (Jangi et al., 2016; Freedman et al., 2017). Interestingly, Akkermansia is a mucin-degrading microorganism that transforms mucin into SCFAs, suggesting that this microorganism might be trying to compensate the inflammatory state of the MS patient (Derrien et al., 2004, 2011). Cekanaviciute et al. found that samples of gut microbiome from MS patients impaired the differentiation of T cells to CD25+FoxP3<sup>+</sup> Treg cells (Cekanaviciute et al., 2017). These authors found that MS patients have a high presence of the Akkermansia genus, specifically A. calcoaceticus and A. muciniphila (Cekanaviciute et al., 2017). Further, it was observed that PBMCs from healthy donors showed an increased differentiation into effector T cells as compared to Treg cells, when exposed to A. calcoaceticus extracts from MS patients (Cekanaviciute et al., 2017). Once PBMCs are exposed to A. muciniphila extracts from MS patients the differentiation of T cells inclined toward to Th1 cells, suggesting that the microbiota of MS patients has a pro-inflammatory effect (Cekanaviciute et al., 2017).

There are several animal models to study MS, one of them is the experimental autoimmune encephalomyelitis (EAE) (Robinson et al., 2014). EAE has the advantage of reproducing most of the symptoms observed in humans and disease is mediated by T cells, as it is the case for MS patients (Stromnes and Goverman, 2006). The induction of EAE in rodents generates a T cell-mediated response, mainly of Th1 and Th17 types (Kleinewietfeld and Hafler, 2013). Interestingly, the function these cells is influenced by the composition microbiota (Wu and Wu, 2012). It has been shown that in normal conditions commensal microorganisms, such as segmented filamentous bacteria can activate these cells to keep a healthy immune response (Ivanov et al., 2009). Noteworthy, Lee et al. showed that the germ-free mice were more resistant to EAE (Lee et al., 2011). The authors observed that some of these mice did not develop symptoms, while and developed only mild symptoms and had a shorter period of EAE (Lee et al., 2011). Experiments of microbiota transfer from specific pathogen free mice (SPF) mice to germ-free mice enhanced the EAE symptoms in germfree mice, supporting the notion that microbiota can influence the immune response during EAE (Lee et al., 2011).

Other studies have found that the microbiota can alter the ratio of cells that play important roles during autoimmune diseases, such as effector T cells vs. Tregs (Molloy et al., 2012; Kosiewicz et al., 2014). For example, in vitro re-stimulation with MOG35−<sup>55</sup> of T cells from germ-free mice and SPF mice that have been previously immunized with MOG/CFA showed in SPF mice an increased secretion of IFNγ and IL-17, as compared to germ-free mice (Lee et al., 2011). Germ-free mice have increased frequencies of Tregs at day 8th before the onset of EAE and at day 15 at the peak of EAE symptoms (Lee et al., 2011). Recently, it was reported that EAE could be induced in mice that received microbiota from MS patients (Berer et al., 2017). By using next-generation sequencing analyses it was shown that the microbiota composition of MS patients has a low content of the Sutterella genus (Berer et al., 2017). It has been observed that this bacterium plays a beneficial role in patients that suffer from IBD (Morgan and Harris, 2015). There is evidence that Bacteroides can be beneficial for maintaining the homeostasis in models of intra-abdominal sepsis and experimental colitis (Tzianabos et al., 1995; Pagliuca et al., 2016). This seems to be the case for B. fragilis, as its polysaccharide (PSA) induces IL-10 secretion by Tregs (Ochoa-Reparaz et al., 2010). In fact, oral treatment with purified Bacterioides fragilis PSA has been shown to be beneficial because it reduces EAE scores, demyelination, IFNγ, and IL-17 and increases T cells conversion to Tregs and the content of IL-10 (Ochoa-Reparaz et al., 2010). Oral immunization with an oral vaccine of attenuated Salmonella typhimurium expressing the colonization factor antigen I (CFA/I) of Enterotoxigenic Escherichia coli (ETEC) to mice that suffer EAE decreased the EAE scores showing a decreased infiltration of TCD4<sup>+</sup> cells in the spinal cord (Jun et al., 2005). CFA/I plays an important role in the attachment of ETEC to the intestinal epithelia an important step for diarrhea pathogenesis (Li et al., 2009).

The question as to how the microbiota modulates de immune response in the MS and/or EAE remains to be addressed. Rumah et al. showed that bacteria metabolites could modulate the immune response (Rumah et al., 2013). These authors isolated a strain of Costridium perfringens type B bacillus from a woman that suffered MS and showed ovoid lesions at the corpus callosum (Rumah et al., 2013). By using a mouse model it was shown that protoxin (a protein secreted by this strain) caused BBB disruption (Finnie, 1984) and binds to myelin in the CNS inducing oligodedrocyte death, as observed in frozen sections of mouse retina (Rumah et al., 2013). Farrokhi et al. found that serum from MS patients has low levels of the lipid 654, this lipopeptide is secreted by commensal intestinal Bacteroidetes in healthy people (Farrokhi et al., 2013). It has been reported that the lipid 654 could work as a ligand for the human toll-like receptor 2 (TLR2), thus it could contribute at down-regulating innate immunity (Clark et al., 2013). The work of Farrokhi suggests that low levels of lipid 654 could be used as a biomarker of MS, however further research is required to demonstrate that this diagnostic approach would be accurate.

Evidence derived from the EAE model suggests that SCFAs can improve and balance the immune response. Mizuno et al. observed that the oral administration of propionate decreased EAE clinical scores and increased Tregs at the lymph nodes, thus it could favor the suppression effect of Tregs (Mizuno et al., 2017). Additional interesting work has provided evidence that supports the notion that the first symptoms of EAE begins at the intestine (Nouri et al., 2014). Nouri et al. in their article showed that there is an unbalance of Treg/Th17 toward Th17 pro-inflammatory response at the intestine (Nouri et al., 2014) Authors also showed an increased intestinal permeability, altered intestinal morphology across the small intestine characterized by depth crypt, gross mucosal thickness, and high expression of zonulin (Nouri et al., 2014). Summarizing, the evidence available in the literature suggests that intestinal microbiota could potentiate or modulate the immune response of MS or EAE specifically the ration or balance of Treg/Th17.

#### Systemic Lupus Erythematosus

Systemic Lupus Erythematosus (SLE) is a systemic autoimmune disease with unknown etiology that affects predominantly women. SLE is characterized by the presence of hyperactive and aberrant antibody response to nuclear and cytoplasmic antigens (La Paglia et al., 2017). Regarding the influence of microbiota over SLE development, microbiota composition analyses showed that SLE patients display intestinal dysbiosis. A decrease in the Firmicutes with an increase of the Bacteroides phyla was observed (Hevia et al., 2014). These bacteria are the most abundant components of the human microbiota and the same pattern has been observed in patients with Crohn's disease (Qin et al., 2010). Mouse models of lupus showed differences in microbiota composition, as compared to control mice. Females showed a more accelerated development of the disease that was associated with decreased levels of Lactobacillaceae and increased levels of Lachnospiraceae families (Zhang et al., 2014). Treatment with probiotics composed by Lactobacillaceae members has been used as anti-inflammatory therapies given its anti-inflammatory properties (Jirillo et al., 2012). Lopez et al. observed the influence of the microbiota over T cell differentiation in healthy and SLE patients (Lopez et al., 2016). These authors found an increase in lymphocyte activation and differentiation toward Th17 in SLE patients (Lopez et al., 2016). Even though, there is still no clarity about the role of microbiota in the development of SLE, the evidence suggests that the dysbiosis observed in the SLE patients could be related to this disease.

## Psoriasis, Psoriasis Arthritis, and Other Skin Related Autoimmune Pathologies

The literature provides evidence for an association between intestinal microbiota with skin autoimmune diseases (Scher et al., 2015; Eppinga et al., 2016; Forbes et al., 2016; Zakostelska et al., 2016). Intestinal-related autoimmune diseases like Crohn's disease is a known comorbidity of psoriasis, patients with psoriasis have a 2.9-times higher risk of developing Crohn's disease as compared to the general population as well as Crohn's disease patients have a 7-times higher risk of developing psoriasis (Oliveira Mde et al., 2015). There is increasing evidence supporting a role of intestinal dysbiosis as a factor in the pathogenesis of Crohn's disease (Kaur et al., 2011) and this might as well be related to psoriasis pathology.

Another important comorbidity of psoriasis is psoriatic arthritis, which is a type of chronic spondyloarthritis with unknown etiology. A recent study characterized the composition of gut microbiota in patients with psoriatic arthritis, patients with psoriasis and healthy controls. The gut microbiota profile of both groups showed decreased bacterial diversity and a reduced relative abundance of some bacterial taxa as compared to healthy controls, such as Akkermansia, Ruminococcus, and Pseudobutyrivibrio. It is important to highlight that the microbiota profile of psoriatic arthritis resembled that published for patients with IBD, a finding that further supports a possible role for the gut microbiota in this skin disease (Scher et al., 2015). The directionality of this relationship remains poorly understood, but it opens a new and interesting research field.

Other skin diseases related to intestinal dysbiosis are scleroderma and vitiligo. In scleroderma (also known as systemic sclerosis) the majority of patients experience gastrointestinal tract dysfunction that might be related to gut microbiota composition. A study performed by Volkmann et al. (2017) shows a different microbial composition between two different cohorts of patients with systemic sclerosis and healthy controls. Firmicutes showed a significantly increased abundance in systemic sclerosis patients (63.5 and 42.8%) compared to healthy controls (33%). In contrast, Bacteroidetes decreased in one of the cohorts (21.3%) as

compared to healthy controls (63.2%). This study also associated the presence of some genus with disease severity. In both systemic sclerosis cohorts, Clostridium was more abundant in patients with low gastrointestinal symptom severity, Lactobacillus was more abundant in patients with mild constipation and Prevotella was more abundant in patients with moderate to severe gastrointestinal symptom severity. Further research will elucidate the role and molecular pathways in which these genders contribute to disease pathogenesis (Volkmann et al., 2017).

Even though, there is information relative to the influence of intestinal microbiota over skin autoimmune diseases, there are data that make these pathologies more complex. Therefore, an influence of skin microbiota over the skin autoimmune diseases remains to be demonstrated (Sanford and Gallo, 2013; Statnikov et al., 2013; Tett et al., 2017). Briefly scientific work that relates skin microbiota with skin autoimmune diseases will be discussed due to the importance of the skin as a physical barrier that constantly interacts with factors from the environment, such as solar irradiation, grades of humidity or dryness and microorganisms. While a wide range of microbes inhabit the skin, the principal residents microbiota belong to one of these phyla: Actinobacteria, Bacteroidetes, Firmicutes, or Proteobacteria. Interactions between these microorganisms at the skin and the immune system have been suggested to influence tissue integrity and homeostasis (Sanford and Gallo, 2013). Similar to what happens in the gut, skin dysbiosis has been postulated to contribute to the pathology of diverse skin autoimmune diseases (Zeeuwen et al., 2013) like psoriasis, psoriasis arthritis and vitiligo (Gao et al., 2008; Castelino et al., 2014; Ganju et al., 2016). Studies related to chronic plaque psoriasis, the most common form of psoriasis, have shown that the composition of the microbiota both with skin lesions or not differed of healthy skin (Gao et al., 2008; Alekseyenko et al., 2013). Gao et al. showed that Firmicutes were the most abundant and diverse phylum present in the psoriatic patients with skin lesions (relative abundance of 46.2%) as compared to those that did not display lesions (39%) and with healthy subjects (24.4%) (Gao et al., 2008). In contrast, Actinobacteria has lower relative abundance in psoriatic patients with lesions (37.3%) compared to individuals with healthy skins (47.6%) (Gao et al., 2008). The phylum of Proteobacteria was less frequent in psoriatic patients (11.4%) as compared to subjects with healthy skin (21.9%). Alekseyenko et al. described three distinct types of bacteria from their data about the relative abundance of major phyla present in the skin, Actinobacteria, Firmicutes, and Proteobacteria (Alekseyenko et al., 2013). These results are consistent with some of the findings described above. Proteobacteria dominate the cutaneotype 1 they described, while the cutaneotype 2 has higher relative abundance of Actinobacteria and Firmicutes. These authors associated the cutaneotype 2 with psoriasis status in which the higher abundance of Firmicutes coincides with the results by Gao et al. (2008). The differences in the results could be attributed to as to where the sample was obtained from, as it has been demonstrated that skin microbial composition is dependent on the area of the body that is analyzed (Grice and Segre, 2011). Both studies provide evidence that skin microbiota composition is different in patients with psoriasis as compared to healthy controls. Nevertheless, the physiological meaning of these differences has yet to be defined.

The most recent study to date related to psoriasis is a metagenomic analysis that focused on the less explored diversity of the skin microbiota and provided evidence that these less characterized species inhabiting the skin might be relevant in disease pathogenesis. In this study, an increase in members of the genus Staphylococcus was associated to development of psoriasis (Tett et al., 2017).

The skin microbiota is not the only microbial compartment that has been associated to psoriasis pathogenesis. In a murine model of imiquimod-induced psoriasis-like skin inflammation, germ-free, and oral antibiotic-treated mice showed milder skin inflammation as compared to conventional reared mice, along with a decrease in γδTCR and Th17 cells in draining lymph nodes and spleen. These cells are essential components of the IL-23/Th17 axis, which happens to be the main axis in psoriasis pathogenesis. These results suggest that the absence of microbiota or an alteration in their composition caused by antibiotic treatment can decrease the pro-inflammatory T cell response and thus diminish the severity of the imiquimod-induced skin inflammation (Zakostelska et al., 2016). Interestingly, in humans, a link between psoriasis and Crohn's disease has long been acknowledged (Hughes et al., 1983) and discussed extensively in previous articles (Najarian and Gottlieb, 2003). Finally, a recent study characterized for the first time the skin microbial composition of patients with the autoimmune disease vitiligo (Ganju et al., 2016). The data obtained were used to characterize a core skin microbiome for a lesional and non-lesional skin (Ganju et al., 2016). Methylobacterium constitutes a genus exclusive to lesional skin while Anaerococcus, Microbacterium, Streptophyta, and Nocardiode are exclusive to non-lesional skin (Ganju et al., 2016). It remains unclear whether the microbiota differences between groups are a cause or an effect of altered skin physiology of the depigmented patches. Therefore, further research is required to define the contribution of the bacterial component to the pathology of vitiligo (Ganju et al., 2016). Compared to the vast information about the role of microbiota in systemic autoimmune diseases, the contribution of intestinal microbiota to skin disease conditions has not been fully explored and requires further research.

#### AUTOIMMUNITY AND MICROBIOTA IN PSYCHIATRIC DISORDERS

Psychiatric disorders have complex etiologies likely resulting from environmental interactions with genetic factors (Tsuang, 2000; Chaste and Leboyer, 2012). Recently, comorbid immune system dysregulations have emerged as a relevant etiological agent in several psychiatric disorders, raising the concept that autoimmunity might be an important contributing factor (Severance et al., 2016). This notion is supported by occurrence of brain inflammation-induced psychosis and autoimmune encephalitis, such as the anti-NMDA receptor encephalitis. This latter disease is characterized by the presence of autoantibodies against the N-methyl-D-aspartate receptor (NMDAR), a glutamatergic receptor in the brain involved in synaptic transmission (Guasp and Dalmau, 2017). Symptoms include a range of psychotic symptoms early in the course of the disease followed by neurologic deterioration and ultimately protracted cognitive and behavioral deficits (Dalmau et al., 2011). Remarkably, approximately 80% of patients recover with immunotherapy directed to remove the antibodies and antibody-producing plasma cells (Dalmau et al., 2017). Consistently with this notion, various autoantibodies have been detected in subgroups of schizophrenic patients (Pearlman and Najjar, 2014). In particular, meta-analyses have suggested that schizophrenic patients are three times more likely to have high levels of anti-NMDAR antibodies as compared to healthy controls (Pearlman and Najjar, 2014). Recently, Schwarz et al. analyzed fecal microbiota from a small cohort of patients with first-episode psychosis (FEP) (Schwarz et al., 2017). Interestingly, Lactobacillus and Bifidobacterium were increased in FEP patients and correlated with the psychotic symptoms severity (Schwarz et al., 2017). Although up to date this is the only report associating changes in the microbiota with psychotic severity in patients, it was also reported that autistic children show increased amounts of Lactobacillus in their intestinal microbiota (Adams et al., 2011; Schwarz et al., 2017). At family level, Schwarz et al. observed a decreased in the Veillonellaceae family an alteration of the microbiota composition that was also observed in depressive patients (Jiang et al., 2015).

As the largest organ of the immune system, the GI tract serves as an interface between the environment and the host. Therefore, alterations in gut cellular processes may influence immune homeostasis. Data generated to date suggest that increased intestinal permeability and alterations in the gut microbiota composition are strongly associated with psychiatric disorders and could represent potential sources of immune functional impairment (Fiorentino et al., 2016; Esnafoglu et al., 2017; Stevens et al., 2017). Gut microbiota seems to play a critical role in the bidirectional communication between the gastrointestinal tract and the CNS. In addition to regulating gastrointestinal functions, this microbiota-gut-brain axis has been shown to modulate brain functions, such as emotional behavior and stressrelated responsiveness (Diaz Heijtz et al., 2011). The mechanisms implicated in these pathways, although not entirely characterized, involve humoral and/or neural route, such as the vagus nerve (De Palma et al., 2014). An involvement of the vagus nerve in the communication between the gut and the brain was supported by the observation that vagotomy abrogated the decrease of anxietyand depression-related behavior induced by the probiotic L. rhamnosus in mice (Bravo et al., 2011). These observations suggest a close relationship between the different components of the gut-brain axis, several probiotics used in animal models have shown their efficacy with combined and correlated beneficial effects on the GI tract, such as intestinal barrier strengthening, HPA axis activation and behavior, including social interaction, anxiety, and exploratory behavior (Hsiao et al., 2013; Vanhaecke et al., 2017). On the other hand, mice exposed to antibiotics to alter the established microbiota, showed an increased in explorative behavior associated with an enhanced level of brainderived neurotrophic factor (BDNF), an important growth factor for neuronal survival and synaptic plasticity (Bercik et al., 2011). However, another mouse strain exposed to a different antibiotic mixture induced a different behavioral phenotype consisting of an increased depressive-like behavior and altered social interaction, which was associated with reduced level of hippocampal BDNF activity. These data suggest that mouse strains or antimicrobial regimen can differently influence microbiota composition and in turn differently affect behavior (Guida et al., 2017). The use of the probiotic Lactobacillus casei DG in these animals was able to restore the behavioral phenotype and BDNF levels similar to controls (Guida et al., 2017). Interestingly, a decreased serum BDNF levels were observed in patients with major depressive disorders (Pedrotti Moreira et al., 2017) and pharmacological treatment that increases BDNF levels proved to be beneficial for depressive patients (Gupta et al., 2016). Along with the depressive-like behavior, mice exposed to antibiotics showed a marked dysbiosis characterized by a loss of bacterial diversity, an increase in Protebacteria and Actinobacteria and a decrease in Bacteroidetes and Firmicutes. The probiotic treatment was not able to restore this bacterial phenotype, but promoted the increase of Lachnospiraceae, a fiber-degrading and SCFA producer (Guida et al., 2017). In humans, either a decrease or an increase of Lachnospiraceae has been correlated with major depressive disorders (Naseribafrouei et al., 2014; Fung et al., 2017), indicating that there is still a controversy about microbiota composition in this pathology. Such a discrepancy could probably result from the influence of external factors (i.e., diet) between the patients cohorts used in these studies. The causal link of microbiota on GI and brain dysfunction has recently been shown for irritable bowel syndrome (IBS), a common intestinal disorder often accompanied by comorbid anxiety (De Palma et al., 2017). Colonization of germ-free mice with fecal microbiota from IBS patients induced altered GI transit, intestinal barrier dysfunctions, and anxiety-like behavior, suggesting a role for the gut microbiota in the expression of IBS (De Palma et al., 2017). The importance of microbiota has also been highlighted in neurodevelopmental disorders, such as autism spectrum disorder (ASD). In mice, pregnant animals exposition to poly(I:C) triggers a maternal immune activation that can cause in the offspring to atypical social and repetitive behavior reminiscent of ASD-related behavior. The mechanisms leading to behavioral abnormalities in the offspring required a particular type of maternal bacterium, the segmented filamentous bacteria (SFB), which induces intestinal TH17 cells producing IL-17 (Kim et al., 2017). Interestingly, induction of TH17 response by SFB has been shown to modulate gene expression of enteric neurons, which in turn influences microbiome/immune system crosstalk (Yissachar et al., 2017). Besides behavioral abnormalities, offspring from poly(I:C)-treated mice also develop GI symptoms, changes in gut bacterial population and intestinal barrier defects, as reported in some patients with ASD (Horvath and Perman, 2002). Treatment of the offspring with B. fragilis induced a remodeling of the microbiota composition ameliorated GI symptoms and reduced repetitive behavior. These preclinical results stimulated the development of pilot clinical studies using microbiome-driven therapies for ASD. Although not controlled and double-blinded,

a clinical trial based on microbiota transfer of healthy donors in 7–16 years old children with ASD showed improvements in GI symptoms and in social and communication skills that are associated with bacterial community changes (Kang et al., 2017). Similarly, probiotic supplementation with Bifidobacteria and Lactobacillus species in 5–9 years ASD children modified the fecal microbiota composition and improved both GI symptoms and behavior (Shaaban et al., 2017). However, these studies were performed with a small number of patients, and wide-scale randomized controlled trials are needed to critically confirm the efficacy of probiotics in ASD. The use of probiotics or microbiotaderived metabolites that could selectively or simultaneously act on the different components of the microbiota-gut-brain axis raises alternative new treatments for CNS disorders. However, more information is still required to better understand the influence of the microbiota on CNS pathogenesis and to delineate the cellular and molecular mechanisms involved in the communication between microbiota, the gut, and the brain. Also, the implications of intestinal dysbiosis in the etiology and/or pathophysiology of brain diseases remain to be further defined and represent a major challenge for translational research.

## EXTERNAL FACTORS THAT INFLUENCE THE MICROBIOTA

Microbiota is established by many factors that determine the characteristics for each individual, including genetic predisposition, an inheritance from the mother since the fetus is forming until breastfeeding and environmental factors, such as diet, culture, and geographic location (Yatsunenko et al., 2012). However, when there is an imbalance of microorganism's populations present in the gut, known as dysbiosis, many problems at the systemic level are triggered. This situation may cause an incorrect nutrient absorption, favor weight alterations and enhance some immune system diseases (Degruttola et al., 2016).

Diet is one of the important factors affecting the composition of microbiota in humans (Conlon and Bird, 2014; Sonnenburg and Backhed, 2016). The first major change in diet that humans experience is when the consumption of solid food begins. At this stage, microbiota composition and abundance suffer significant modifications, such as a decrease of Bifidobacteria and Enterobacteria (Fallani et al., 2011). This process occurs due to the new substrates that are available in the gut, leading to the proliferation of certain types of microbes (Fallani et al., 2011; Davis et al., 2017; Wampach et al., 2017). Later in life, each time that diet changes this phenomenon takes place, promoting the growth of some microorganisms more than others (Ottman et al., 2012; Odamaki et al., 2016; Sonnenburg and Backhed, 2016). In turn, this evolving microbiota will modulate the development and function of the immune system (Round and Mazmanian, 2009; Conlon and Bird, 2014; Kabat et al., 2014). However, microbiota analyses highlighted that although after a month of receiving a traditional diet, the composition of the microbiota took 1 year to change and be similar to the one found in other people with the same diet (Claesson et al., 2012).

FIGURE 2 | Influence of the gut microbiota in non-intestinal diseases. Gut dysbiosis induced by external factors as diet, infections, or antibiotic overuse lead to an inflammatory response that influence the outcome of several autoimmune diseases as Grave's disease, Hashimoto's Thyroiditis, Multiple Sclerosis, SLE, and type1 diabetes. It has also been observed an important role in skin related autoimmune diseases as Psoriasis. Moreover, the evidences in the literature support that gut microbiota can influence CNS disorders like autism, depression, and schizophrenia.

Human studies have shown that there is a correlation in elderly between the microbiota location and diet, which can influence their health (Claesson et al., 2012). Additionally, analyses of microbiota composition had shown four different dietary groups and found that diet based on "low fat-high fiber" was associated with a more diverse microbiota. A study performed with C57BL/6 mice fed with high-fat diet was aimed at evaluating whether inflammation was due to microbiota changes or cytokines production. Results have shown that microbiota underwent changes 8 weeks after the treatment started, but increased IFNγ and TNF-α cytokines were only detected 16 weeks after administration of a high-fat diet (Guo et al., 2017). These observations underscore the importance of microbiota and corroborate the ability of diet to modulate microbiota composition.

Additionally, evaluation of the feces obtained from three different dietary groups ovo-lacto vegetarian, vegans, and omnivores, showed differences in microbiota composition (Ferrocino et al., 2015). Findings showed that B. fragilis group was diminished on ovo-lacto vegetarian and vegan volunteers, which is associated with low consumption of protein and animal fat (Wu et al., 2011). Also, a significant reduction for the loads for Lactic Acid Bacteria (LAB) was observed for the dietary groups ovo-lacto vegetarian and vegan, which may be due to the absence or low intake of some food such as cheese and yogurt (Zimmer et al., 2012).

Gluten-free diet is worldwide used for patients with celiac disease (CD), which is a chronic enteropathy caused mainly by gluten intolerance. It has been shown that the intestinal microbiota can be modified by a gluten-free diet: two of the groups that are reduced by such a diet are Lactobacillus and Bifidobacterium, which are considered as part of the group of healthy bacteria (De Palma et al., 2009; Lorenzo Pisarello et al., 2015). In contrast, some opportunistic bacteria from Enterobacteriacea family are increased as a result of gluten-free diets (De Palma et al., 2009; Lorenzo Pisarello et al., 2015). One possible explanation for these changes in the microbiota is the low carbohydrate intake due to the restrictions imposed by the gluten-free diet. Healthy bacteria usually use carbohydrates for their metabolism, which are needed for the colonization and fermentation inside the gastrointestinal tract (Sanz, 2010).

A diet that low in carbohydrates accessible for microbiota results in low levels of production of Short Chain Fatty Acids (SCFA) produced by bacteria from the microbiota. Butyrate, propionate, and acetate can modulate immune system by controlling inflammation and promoting an anti-inflammatory environment in the gut (Maslowski et al., 2009; Arpaia et al., 2013). However, when the diet undergoes a detrimental change, the abundance of these beneficial bacteria decreases and therefore SCFA production is reduced. Alteration of SCFA is common in obesity, celiac disease, and Type 2 Diabetes, because it is known that their action ameliorates the disease (Lin et al., 2012; Hong et al., 2016; Lerner et al., 2016).

Consistently with this notion, the Ma-Pi2 diet a diet rich in carbohydrates, whole grains and vegetables was shown to ameliorate dysbiosis and increment the diversity of microbiota



in Type 2 Diabetes patients, as compared to untreated patients. Also, abundance of SCFA producers increased, such as Bacteroides, Dorea, and Faecalibacterium. Furthermore, the implementation of this diet has shown a potential role in the recovery of metabolic control in Type 2 Diabetes (Fallucca et al., 2014; Candela et al., 2016).

#### Overuse of Antibiotic Treatments

Overuse of antibiotics may cause a significant imbalance in microbiota and a disruption of the natural interaction between the microorganisms. One of the most important characteristics of a normal microbiota is the capacity to compete out infectious pathogens (Kamada et al., 2013a). Therefore, microbiota removal by antibiotics may allow the detrimental growth of pathogenic bacteria populations, increasing the probability of an infection. Additionally, antibiotics not only kill pathogens but also beneficial bacteria, eliminating as well the positive effect of the latter. Microbiota modulates the immune response through the molecules it produces, so if beneficial microorganisms decrease, a decrease in the modulation of the immune system can also be observed (Langdon et al., 2016).

In response to a constant exposure of an antibiotic, microorganisms can acquire genetic resistance, leading to an the increment of multi-drug resistance microorganisms (Jernberg et al., 2007), which can become a major public health problem due to the lack of new treatments capable of eliminating these bacteria. In the last years, many cases of infection caused by multi-drug resistant bacteria were reported and this number is expected to increase (Karam et al., 2016; Lee et al., 2016). The effects of overuse of antibiotics can be treated but not reversed. Restoration of the microbiota can take months or even years, but it will not be able to become the same as before (Jernberg et al., 2007, 2010). Importantly, it has also been described that newborns whose mothers have received antibiotics perinatally have a different microbiota composition, as compared to newborns whose mother have not been treated (Fallani et al., 2011). Due to the importance of first microorganisms in the gut of newborn, these changes in microbiota caused by excessive use of antibiotics may have long-term consequences (Langdon et al., 2016).

Interestingly, recent findings have highlighted the natural presence of antibiotic resistance genes in microbiota and their differential occurrence according to diet. Microbiota changes observed in ovo-lacto vegetarian and vegan diets, as compared to omnivores diet, are related to the presence of antibiotic resistance genes. A study performed with 144 volunteers found the presence of 12 antibiotic resistance genes in their microbiota. Among these genes, the occurrence of erm(A) (Erythromycin resistance methylase gene) has been exclusively detected in the feces of vegan subjects, which also show a low abundance of tet(K) (Tetracycline efflux protein), as compared to the genes present in the feces of omnivores. Interestingly, another antibiotic resistance gene with a higher occurrence observed in the feces of omnivores was van(B) (Vancomycin resistance gene; Milanovic et al., 2017).

## CONCLUSIONS

This review article discusses data supporting the influence of the gut microbiota over non-intestinal autoimmune diseases. The central theme of this review is the intestine in which two important actors, microbiota and the immune system are controlling the response to non-intestinal autoimmune diseases (**Figure 2**, **Table 1**). Not much is known about the mechanisms of the interaction between microbiota and the immune system. However, today it is possible to identify certain members of the microbiota that regulate, balance or unbalance the immune response of the host. The current evidence supports the notion that changes or alterations of the microbial species that form part of the intestinal microbiota will affect the balance of Tregs and Th17 cells at the intestine, which could modify the immune response of non-intestinal autoimmune diseases. The experimental evidence suggesting that the cytokines secreted from Treg and Th17 will determine and influence non-intestinal autoimmune responses. It could also be possible that cells of the immune system located at the intestine could to move other organs to establish or modify an autoimmune response. The major message of this review is that the abundant data support the notion that the intestine is a critical organ the appropriate immune balance and for the prevention of non-intestinal autoimmune diseases. The key point is that by modifying the intestinal microbiota of a patient that suffers non-intestinal autoimmune disease it might be possible to improve the outcome of such illness. Interestingly, by modifying the diet it might be possible to improve the intestinal microbiota to promote an antiinflammatory response of a patient suffering from autoimmunity. Thus, the scientific community has paid attention to the potential therapeutical benefits of manipulating the composition of the gut microbiota through oral administration of probiotic or modified organisms expressing selected self-antigens to treat these non-intestinal autoimmune diseases. Work remains to be done in order to fully understand the complex mechanisms of the intestinal microbiota that can impact non-intestinal autoimmune diseases.

## AUTHOR CONTRIBUTIONS

MO, EO-R, and IC-A, have written the first draft of the manuscript. LB, CR, SB, HB, MN, and AK revised and improved the first draft. All authors have seen and agreed on the finally submitted version of the manuscript.

#### FUNDING

MO, IC-A, CR, SB, and AK: Millennium Institute on Immunology and Immunotherapy, IMII P09/16-F. IC-A: CONICYT 63140215. CR: Fondecyt 1161525, Nucleus project UNAB DI-471-15/N. SB: Fondecyt 1170964. AK: Fondecyt 1150862. HB and MN: Region Pays de la Loire (MIBIOGATE) and the Fondation pour la Recherche Medicale.

## REFERENCES


diabetic patients by macrobiotic Ma-Pi 2 diet. Br. J. Nutr. 116, 80–93. doi: 10.1017/S0007114516001045


with beta-cell autoimmunity and those without. Diabetes 62, 1238–1244. doi: 10.2337/db12-0526


causes gut endocannabinoidome changes, hippocampal neuroglial reorganization and depression in mice. Brain Behav. Immun. 67, 230–245. doi: 10.1016/j.bbi.2017.09.001

Guo, X., Li, J., Tang, R., Zhang, G., Zeng, H., Wood, R. J., et al. (2017). High fat diet alters gut microbiota and the expression of paneth cell-antimicrobial peptides preceding changes of circulating inflammatory cytokines. Mediators Inflamm. 2017:9474896. doi: 10.1155/2017/9474896

Gupta, R., Gupta, K., Tripathi, A. K., Bhatia, M. S., and Gupta, L. K. (2016). Effect of mirtazapine treatment on serum levels of brain-derived neurotrophic factor and tumor necrosis factor-alpha in patients of major depressive disorder with severe depression. Pharmacology 97, 184–188. doi: 10.1159/000444220

Hartsock, A., and Nelson, W. J. (2008). Adherens and tight junctions: structure, function and connections to the actin cytoskeleton. Biochim. Biophys. Acta 1778, 660–669. doi: 10.1016/j.bbamem.2007.07.012

Hashimoto, T., Perlot, T., Rehman, A., Trichereau, J., Ishiguro, H., Paolino, M., et al. (2012). ACE2 links amino acid malnutrition to microbial ecology and intestinal inflammation. Nature 487, 477–481. doi: 10.1038/nature11228

Hevia, A., Milani, C., Lopez, P., Cuervo, A., Arboleya, S., Duranti, S., et al. (2014). Intestinal dysbiosis associated with systemic lupus erythematosus. MBio 5, e01548–e01514. doi: 10.1128/mBio.01548-14

Hindson, J. (2017). Multiple sclerosis: a possible link between multiple sclerosis and gut microbiota. Nat. Rev. Neurol. 13:705. doi: 10.1038/nrneurol.2017.142


Hsiao, E. Y., McBride, S. W., Hsien, S., Sharon, G., Hyde, E. R., McCue, T., et al. (2013). Microbiota modulate behavioral and physiological abnormalities associated with neurodevelopmental disorders. Cell 155, 1451–1463. doi: 10.1016/j.cell.2013.11.024

Hughes, S., Williams, S. E., and Turnberg, L. A. (1983). Crohn's disease and psoriasis. N. Engl. J. Med. 308:101.

Indigo, P. (2017). Investigation of Novel Biomarkers and Definition of the Role of the Microbiome in Graves' Orbitopathy. Available online at: http://www.indigoiapp.eu/publishable-summary/ (Accessed September 20, 2017).


Jernberg, C., Lofmark, S., Edlund, C., and Jansson, J. K. (2010). Long-term impacts of antibiotic exposure on the human intestinal microbiota. Microbiology 156, 3216–3223. doi: 10.1099/mic.0.040618-0

Jiang, H., Ling, Z., Zhang, Y., Mao, H., Ma, Z., Yin, Y., et al. (2015). Altered fecal microbiota composition in patients with major depressive disorder. Brain Behav. Immun. 48, 186–194. doi: 10.1016/j.bbi.2015.03.016

Jirillo, E., Jirillo, F., and Magrone, T. (2012). Healthy effects exerted by prebiotics, probiotics, and symbiotics with special reference to their impact on the immune system. Int. J. Vitam. Nutr. Res. 82, 200–208. doi: 10.1024/0300-9831/a000112

Johnson-Henry, K. C., Pinnell, L. J., Waskow, A. M., Irrazabal, T., Martin, A., Hausner, M., et al. (2014). Short-chain fructo-oligosaccharide and inulin modulate inflammatory responses and microbial communities in Caco2-bbe cells and in a mouse model of intestinal injury. J. Nutr. 144, 1725–1733. doi: 10.3945/jn.114.195081


Kamada, N., Chen, G. Y., Inohara, N., and Nunez, G. (2013a). Control of pathogens and pathobionts by the gut microbiota. Nat. Immunol. 14, 685–690. doi: 10.1038/ni.2608

Kamada, N., Seo, S. U., Chen, G. Y., and Nunez, G. (2013b). Role of the gut microbiota in immunity and inflammatory disease. Nat. Rev. Immunol. 13, 321–335. doi: 10.1038/nri3430

Kang, D. W., Adams, J. B., Gregory, A. C., Borody, T., Chittick, L., Fasano, A., et al. (2017). Microbiota Transfer Therapy alters gut ecosystem and improves gastrointestinal and autism symptoms: an open-label study. Microbiome 5:10. doi: 10.1186/s40168-016-0225-7

Karam, G., Chastre, J., Wilcox, M. H., and Vincent, J. L. (2016). Antibiotic strategies in the era of multidrug resistance. Crit. Care 20:136. doi: 10.1186/s13054-016-1320-7

Karjalainen, J., Martin, J. M., Knip, M., Ilonen, J., Robinson, B. H., Savilahti, E., et al. (1992). A bovine albumin peptide as a possible trigger of insulin-dependent diabetes mellitus. N. Engl. J. Med. 327, 302–307. doi: 10.1056/NEJM199207303270502

Kaur, N., Chen, C. C., Luther, J., and Kao, J. Y. (2011). Intestinal dysbiosis in inflammatory bowel disease. Gut Microbes 2, 211–216. doi: 10.4161/gmic.2.4.17863

Kelsall, B. (2008). Recent progress in understanding the phenotype and function of intestinal dendritic cells and macrophages. Mucosal Immunol. 1, 460–469. doi: 10.1038/mi.2008.61

Khan, K. N., Fujishita, A., Masumoto, H., Muto, H., Kitajima, M., Masuzaki, H., et al. (2016). Molecular detection of intrauterine microbial colonization in women with endometriosis. Eur. J. Obstet. Gynecol. Reprod. Biol. 199, 69–75. doi: 10.1016/j.ejogrb.2016.01.040

Kim, S., Covington, A., and Pamer, E. G. (2017). The intestinal microbiota: antibiotics, colonization resistance, and enteric pathogens. Immunol. Rev. 279, 90–105. doi: 10.1111/imr.12563

Kleinewietfeld, M., and Hafler, D. A. (2013). The plasticity of human Treg and Th17 cells and its role in autoimmunity. Semin. Immunol. 25, 305–312. doi: 10.1016/j.smim.2013.10.009

Koenig, J. E., Spor, A., Scalfone, N., Fricker, A. D., Stombaugh, J., Knight, R., et al. (2011). Succession of microbial consortia in the developing infant gut microbiome. Proc. Natl. Acad. Sci. U.S.A. 108(Suppl. 1), 4578–4585. doi: 10.1073/pnas.1000081107

Kohling, H. L., Plummer, S. F., Marchesi, J. R., Davidge, K. S., and Ludgate, M. (2017). The microbiota and autoimmunity: their role in thyroid autoimmune diseases. Clin. Immunol. 183, 63–74. doi: 10.1016/j.clim.2017.07.001

Kopec, A. M., Fiorentino, M. R., and Bilbo, S. D. (2018). Gutimmune-brain dysfunction in Autism: importance of sex. Brain Res. doi: 10.1016/j.brainres.2018.01.009

Kosiewicz, M. M., Dryden, G. W., Chhabra, A., and Alard, P. (2014). Relationship between gut microbiota and development of T cell associated disease. FEBS Lett. 588, 4195–4206. doi: 10.1016/j.febslet.2014. 03.019

Kostic, A. D., Gevers, D., Siljander, H., Vatanen, T., Hyotylainen, T., Hamalainen, A. M., et al. (2015). The dynamics of the human infant gut microbiome in development and in progression toward type 1 diabetes. Cell Host Microbe 17, 260–273. doi: 10.1016/j.chom.2015.01.001

Kristensen, B. (2016). Regulatory B and T cell responses in patients with autoimmune thyroid disease and healthy controls. Dan. Med. J. 63:B5177.

La Fata, G., Weber, P., and Mohajeri, M. H. (2017). Probiotics and the gut immune system: indirect regulation. Probiotics Antimicrob. Proteins 10, 11–21. doi: 10.1007/s12602-017-9322-6


psychosis identifies preliminary associations with symptom severity and treatment response. Schizophr. Res. 192, 398–403. doi: 10.1016/j.schres.2017. 04.017


in experimental intra-abdominal sepsis. J. Clin. Invest. 96, 2727–2731. doi: 10.1172/JCI118340


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Opazo, Ortega-Rocha, Coronado-Arrázola, Bonifaz, Boudin, Neunlist, Bueno, Kalergis and Riedel. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Upregulation of intestinal Barrier Function in Mice with Dss-induced colitis by a Defined Bacterial consortium is associated with expansion of il-17a Producing gamma Delta T cells

#### *Edited by:*

*Wesley H. Brooks, University of South Florida, United States*

#### *Reviewed by:*

*Claudio Silva, Federal University of Uberlandia, Brazil Amariliz Rivera, New Jersey Medical School, United States Claudio Acuña-Castillo, University of Santiago, Chile, Chile*

#### *\*Correspondence:*

*Jieli Yuan zgwst@126.com; Xia Li lixia416@163.com*

*† These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 29 March 2017 Accepted: 29 June 2017 Published: 12 July 2017*

#### *Citation:*

*Li M, Wang B, Sun X, Tang Y, Wei X, Ge B, Tang Y, Deng Y, He C, Yuan J and Li X (2017) Upregulation of Intestinal Barrier Function in Mice with DSS-Induced Colitis by a Defined Bacterial Consortium Is Associated with Expansion of IL-17A Producing Gamma Delta T Cells. Front. Immunol. 8:824. doi: 10.3389/fimmu.2017.00824*

*Ming Li 1†, Bing Wang2†, Xiaotong Sun2 , Yan Tang1 , Xiaoqing Wei <sup>3</sup> , Biying Ge4 , Yawei Tang2 , Ying Deng1 , Chunyang He1 , Jieli Yuan1 \* and Xia Li <sup>2</sup> \**

*1Department of Microecology, College of Basic Medical Science, Dalian Medical University, Dalian, China, 2Department of Immunology, College of Basic Medical Science, Dalian Medical University, Dalian, China, 3 The Core Laboratory of Medical Molecular Biology of Liaoning Province, Dalian Medical University, Dalian, China, 4 Functional Laboratory, College of Basic Medical Science, Dalian Medical University, Dalian, China*

Bacterial consortium transplantation (BCT) is a promising alternative to fecal microbiota transplantation in treating inflammatory bowel disease (IBD). Here, we showed that a defined bacterial consortium derived from healthy mice was able to enhance the intestinal barrier function of mice with dextran sulfate sodium (DSS)-induced colitis. Interestingly, we found that the bacterial consortium significantly promoted the expansion of IL-17Aproducing γδT (γδT17) cells in colonic lamina propria, which was closely associated with changing of intestinal microbial composition. The increased IL-17A secretion upon treatment with microbial products derived from the bacterial consortium was accompanied with upregulation of TLR2 expression by γδT cells, and it might be responsible for the upregulation of mucosal barrier function through IL-17R-ACT1-mediated recovery of the disrupted occludin subcellular location. Changing of some specific microbial groups such as *Bifidobacterium* and *Bacillus* spp. was closely correlated with the promotion of TLR2<sup>+</sup> γδT cells. Our results support that BCT can restore the alliance between commensal microbiota and intestinal γδT cells, which contributes to the improvement of intestinal barrier function. This study provides new insight into the development of bacteria transplantation therapy for the treatment of IBD.

Keywords: inflammatory bowel disease, dysbiosis, bacterial consortium transplantation, **γδ**T cells, IL-17A, occludin, intestinal barrier function

## INTRODUCTION

The gastrointestinal (GI) illness inflammatory bowel disease (IBD) is characterized by chronic intestinal inflammation caused by immune responses against the patient's own organs. Several factors are found to play important roles in the development and progression of IBD, including host genotype, the composition of intestinal microbiota, immune disequilibrium, as well as environmental factors (1, 2). In recent years, intestinal dysbiosis has been increasingly linked to IBD (3, 4). Current research suggests that disruption of the alliance between the host immune system and commensal microbiota may contribute to the pathogenesis and status of IBD (5).

Many studies have supported the concept that disrupted intestinal microbiota may be a contributing factor in patients with IBD, which can be initiated because of a dysregulated epithelialimmune cell communication (6). The appreciation of commensal microbes as being beneficial for the host, especially in shaping the intestinal barrier, has been the justification for using fecal microbiota transplantation (FMT) therapy for the treatment of IBD (7). FMT has resulted in disease remission and improved quality of life (8). However, some unfavorable outcomes of FMT were also reported (9). Colonization of transplanted bacteria in the GI tract was found to differ significantly among patients (10). In addition, the safety issues and the non-standardization of procedures have limited the clinical application of FMT (6, 11). In contrast, compared with undefined fecal samples, a consortium of harmless, health-associated bacteria may be a viable therapeutic alternative for the treatment of IBD. Bacterial consortium transplantation (BCT) is regarded as being more stable and controlled than FMT (12–14). Using a consortium of 10 bacterial strains that were isolated from fecal samples of healthy mice, we have previously shown that BCT and FMT were comparable in re-establishing mucosal barrier function in mice with intestinal dysbiosis (15). Further studies in rats with trinitrobenzene sulfonic acid (TNBS)-induced colitis showed that this bacterial consortium exhibited anti-inflammatory activity in the intestine of rats and contributed greatly to the reduction of gut permeability, as well as the rapid re-establishment of intestinal microbial equilibrium (16). However, the underlying mechanisms remain unclear.

The intestinal epithelium is built of monolayered columnar epithelial cells that are tightly connected by tight junction proteins such as occludin and claudins. Impaired tight junction protein, which leads to the reduced epithelial barrier integrity, was found in both human IBD and a mouse model with dextran sulfate sodium (DSS)-induced colitis—a well-established model of mucosal inflammation used in IBD studies (17, 18). Although tight junction proteins are normally considered as a part of the physical barrier, they are largely affected by mucosal immune homeostasis within the lamina propria (LP) and the composition of gut microbes and thus can be regarded as a translator between microbiota and immune system (6, 19, 20). Among the immune cells dwelling in intestinal mucosa, γδT cells are rare T-cell subsets that were found to be involved in both pathogenic and protective networks in IBD (21). They have been the targets of recent investigation because of their spontaneous IL-17A expression (22) and interaction with intestinal microbiota (23, 24). γδT cells in intestinal LP were found to be the major source of gut-protective IL-17, and this γδT cell-derived IL-17 promoted the repair of damaged intestinal epithelium through adaptor molecule Act1-mediated regulation of occludin subcellular localization (25). Given the important regulatory role of gut microbes and γδT cells in regulating occludin, we therefore hypothesized that the IL-17A-producing γδT (γδT17) cells might be involved in the protection mechanism of the bacterial consortium on colon mucosa.

In this study, we showed that transplantation of the bacterial consortium resulted in the reduction of gut permeability and improvement of intestinal dysbiosis in mice with DSS-induced colitis; and it also resulted in the elevation of γδT17 cells in colonic lamina propria (cLP) of mice. The IL-17R-Act1-occludin signaling crosstalk was found to be upregulated in mice that received BCT. In addition, specific bacterial groups were found correlated with the expansion or reduction of γδT17 cells. Our results support the possibility that the alliance between gut microbiota and γδT17 cells plays an important regulatory role in IBD, and this study provides new insights for the development of microbiota transplantation therapy.

### MATERIALS AND METHODS

#### Animal Experiments

Male C57BL/6J mice were obtained from the Experimental Animal House of Dalian Medical University, China, where they were maintained under stress-free and specific pathogen-free conditions, under 12 h cycles of light and darkness. Food and water were provided *ad libitum* before experiments. The animal experimental procedures were approved by the Medical Ethics Committee of Dalian Medical University, China (SYXK2015- 0002) (16).

To induce acute colitis, 6- to 8-week-old mice were given drinking water containing 3.0% (w/v) DSS (MP Biomedicals) *ad libitum* for 7 days and distilled water for 1 additional day before sacrifice. All the mice were anesthetized before sacrifice. To detect gut permeability, FITC-Dextran (4 kDa, Sigma-Aldrich) were gavaged to mice as previously described (25) 3 h prior to fluorometric analysis of FITC fluorescence in serum.

For BCT, bacterial strains isolated previously from healthy mice were cultured, mixed, and re-suspended in 0.2 ml PBS, according to the method described previously (15). The population of each strain in the mixture is shown in Table S1 in Supplementary Material. The mixture was then gavaged to mouse once a day. Mice of the control and DSS group were gavaged with 0.2 ml PBS as vehicle. The strains used for bacterial transplantation are deposited in the Bacteria Collection of Dalian Medical University, China, and The University's Institutional Biosafety Committee approved applications to conduct only scientific research with these microorganisms.

#### Histopathological Analysis and Measurement of Colon Myeloperoxidase (MPO), Cytokines, and Serum FITC-Dextran

The Disease Activity Index of mice was evaluated according to a previous study (16). After the mice were sacrificed, the colon was extracted, cleaned, weighed, and measured. The colonic tissue samples were preserved in buffered formalin, embedded in paraffin, cut into 5 µm sections, and stained with hematoxylin and eosin (H&E) for histopathological analysis. The rest of the tissue was homogenized in a Greenburger buffer supplemented with protease inhibitors (complete Mini EDTA free; Roche), after sonication for 10 s, the suspension was centrifuged at 8,000 × *g* for 20 min at 4°C. The supernatant was used to quantify the MPO activity (which correlated with the degree of neutrophil infiltration) and the levels of cytokines using ELISA kits (USCN, USA). Peripheral blood of the mice was centrifuged at 1,500 × *g* for 15 min at 4°C to obtain serum. Serum FITC-Dextran was assayed by ELISA (USCN, USA) according to the manufacturer's instructions (16).

#### Immunofluorescence Microscopy

The distal colons of mice were first flushed with PBS, then embedded in Tissue-Tek O.C.T. compound (SAKURA Finetechnical Company) in cryomolds, and snap frozen in liquid nitrogen for cryosectioning. The cryosections were prepared at −20°C with 8 µm thickness on a Leica Cryostat (Leica Microsystems). Sections were mounted on glass slides and fixed in 100% ethanol at 4°C for 30 min followed by 3 min of 20°C acetone fixation at room temperature. After washing in PBS, the slides were blocked in FBS and goat serum for 1 h in room temperature. The tissue sections were stained with a monoclonal occludin antibody sc-5562 (SANTA CRUZ Biotechnology) at 4°C overnight. After washing in PBS, the sections were stained with a goat anti-rabbit IgG Alexa Fluor 488 conjugated secondary antibody (Fcmacs) for 60 min at room temperature. After washing in PBS, the tissue sections were treated with DAPI (Millipore) for 5 min and covered with a coverslip. Fluorescence was observed with a microscope (DM4000B; Leica, Germany) equipped with a digital camera at 40× magnification (25).

## RNA Isolation and Quantitative Real-time PCR (qPCR)

To analyze the mRNA expression of genes, total RNA in mouse colon tissues was extracted by using the RNeasy mini Kit (Qiagen, Hilden, Germany). The complementary DNA (cDNA) was synthesized using the AffinityScript Multiple Temperature cDNA synthesis Kit (Stratagene, La Jolla, CA, USA). Reverse transcription was performed to obtain cDNAs, which were used to detect mRNA expression of *Tjp1*, *Ocln*, *il17a*, and *ROR*γ*t* by the specific primers listed in Table S2 in Supplementary Material. The reactions were run on an ABI StepOne Plus Sequence Detection System (Applied Biosystems). The reaction mixture (10 µl) includes 4.5 µl SYBR Premix Ex Taq (Perfect Real Time, TAKARA, Japan), 0.5 µl of forward and reverse primers (30 mM), 2.5 µl of sterile distilled water, and 2.5 µl of cDNA (100 ng/ml). Each sample was run in triplicate, and the mean Ct was determined. The relative mRNA expression was expressed as ΔCt = Ct (target genes) − Ct (calibrator). The expression of the GAPDH gene was used as a calibrator after verification of its stability under current experimental conditions. The relative mRNA expression was calculated as ΔΔCt = ΔCt (test group) − ΔCt (control group) and expressed as fold change (=2−ΔΔCt).

## Western Blot (WB)

The total protein was extracted from colon tissues of mice. Equivalent amounts of protein were separated by sodium dodecylsulfate polyacrylamide gel electrophoresis (SDS-PAGE) and were transferred onto nitrocellulose membranes. Antibodies against Occludin, Act1, IL-17RC GADPH (USCN, USA) were used for blotting, and the secondary antibodies conjugated with horseradish peroxidase (HRP, USCN, USA) were used to show the bands. The immune complexes were detected with a WesternBright™ ECL Western Blotting HRP Substrate kit and analyzed with image lab software (Bio-Rad, USA).

## Cell Isolation from cLP and Spleen

The colons of mice were incubated in a 37°C water bath in cell dissociation solution made with HBSS and HEPES (Solarbio) to strip the epithelial cells. Supernatant was discarded, and colonic tissue was then incubated in a digestion cocktail containing HBSS, FCS (10%, Gibco), type IV1 collagenase (1 mg/ml), DNaseI (0.5 mg/ml), and dispase (0.5 mg/ml, all from Sigma-Aldrich) in a 37°C water bath. After that, the digested tissue was processed through a 70 mm filter (Falcon) and washed before lymphocytes were separated by the methods used previously (20, 24). γδT cells were purified from the spleens of mice using the Mouse TCRγδ+ T-cell Isolation Kit and the miniMACSTM (Miltenyi Biotec, Germany).

### Cell Culture

Cells were put into single-cell suspensions at 1–2 × 106 cells/ml and were plated for a final volume of 200 µl and re-stimulated with media alone or with media and bacterial components. A total of 1 ml of the bacterial consortium, or culture broth of *Bifidobacterium longum* ATCC 15707 and *Bacillus cereus* ATCC 14579, was sonicated to release the cellular components, after that, the mixture was centrifuged at 8,000 rpm for 2 min, and different volumes of the supernatant were added to the single-cell suspension after filtration by a 0.2 μm-filter. After co-culture for 20 h, the IL-17A level in the cell culture supernatant was detected by ELISA (USCN, USA).

## Flow Cytometry

For IL-17A staining, cells were first stimulated with 2 µl/ml cell stimulation cocktail (500, plus protein transport inhibitors; eBiosceince) in the presence of 100 µl Intracellular Fixation and Permeabilization Buffer Set (eBioscience) Golgi-plug (eBioscience) for 12 h in complete medium. Surface staining was then performed in the presence of Fc-blocking antibodies (albumin, bovine, fraction V-Coolaber) and using αCD4 (anti-mouse CD4 FITC, 11-0041; eBioscience), αCD3 (anti-mouse CD3 FITC, 11-0032-80; eBioscience), α-γδTCR (anti-mouse gamma delta TCR PE, 12-5711; eBioscience), and α-TLR2 (anti-mouse CD282 eFluor®660; 50-9021, eBioscience). Cells were then fixed and permeabilized with the Cytofix/Cytoperm kit (BD) followed by intracellular staining using antibodies against IL-17A (antimouse/rat IL-17A APC, 17-7177; eBioscience). All samples were collected (Accuri C6; BD Bioscience, USA), and data were analyzed with Flow Plus 1.0.264.15 Software.

## DNA Isolation from Colonic Contents

The metagenomic DNA in the colon content of mice was isolated by the QIAamp DNA stool mini kit (Qiagen, Germany), and a NanoDrop 2000 spectrophotometer was used to measure the purity and concentration of the DNA (Thermo, USA) (16).

## Library Construction, Sequencing, and Data Analysis

The V3–V4 region of 16S rDNA was amplified with universal primers. The PCR products were then quantified by electrophoresis on a 1.5% agarose gel and purified with the QIAquick Gel Extraction kit (Qiagen). Sequencing and data analysis were subsequently performed on an Illumina HiSeq platform by Novogene (Beijing, China) using a method described previously (26). Briefly, we used Ribosomal Database Project Classifier 2.8 to perform the assignment of all sequences at 50% confidence after the raw sequences were identified by their unique barcodes. OTUs present in 50% or more of the colon content samples were identified as core OTUs. PLS-DA of core OTUs was performed using Simca-P version 12 (Umetrics), and a heatmap was generated using Multi-Experiment Viewer software to visualize and cluster the bacterial community into different groups. Community diversity was measured by the Shannon–Weiner biodiversity index (Shannon index).

#### Statistics

All data were evaluated as mean ± SEM (*n* = 5). Statistical analysis of the quantitative multiple group comparisons was performed using one-way analysis of variance (and non-parametric), followed by Tukey's test (compare all pairs of columns); when two groups were compared, the non-parametric *t*-test was performed with the assistance of GraphPad Prism 5 (Graph Pad Software, La Jolla, CA, USA). Results were considered to be statistically significant with *p* < 0.05.

## RESULTS

## The Bacterial Consortium Protected against DSS-Induced Colitis in Mice

We first investigated the potential therapeutic action of the bacterial consortium in mice with DSS-induced colitis, although it had already shown protective effects in TNBS rat models (16). Mice subjected to 3.0% DSS developed severe illness characterized by profound and sustained weight loss (**Figure 1A**), bloody diarrhea, and wasting syndrome resulting in the increase of the disease index and mortality (**Figures 1B,C**). Transplantation of the bacterial consortium ameliorated the weight loss and disease severity, with a survival rate of 85.7%, which was significantly higher than the survival rate of DSS-treated mice (70.0%) (**Figures 1A–C**). The average colon length of the consortium-treated mice was longer than that of the DSS group (**Figures 1D,E**) with edema-induced shortening. Histological evaluation of the distal colon of the consortium-treated group showed no signs of macroscopic and transmural inflammation (**Figures 1F,G**); in contrast, the colon of DSS-treated mice showed significant transmural inflammation involving all layers of the colonic wall and marked increase in the thickness of the muscular layer. The neutrophil infiltration, which was correlated with a significant increased colonic MPO level, was detected in DSS-treated mice but was much mitigated in the bacterial consortium-treated mice (**Figure 1H**).

## BCT Decreased the Gut Permeability of DSS-Treated Mice by Upregulating the Expression of Tight Junction Protein Occludin

To further characterize the protective effects of the bacterial consortium on the epithelial layer in DSS-treated mice, we quantified the permeability by orally administering FITC-dextran to mice and measuring the serum levels. Results showed that the diffusion of FITC-dextran through the epithelium after DSS treatment was significantly increased in the intestines of mice (**Figure 2A**) on day 3 and day 7, and it was compromised in the BCT group of mice. The tight junction complex claudins and occludins play crucial roles in regulating gut permeability and the epithelial paracellular pathway (27). We therefore tested the transcription levels of *Cldn1* and *Ocln* in mouse colon tissues by quantitative real-time PCR (qPCR). DSS treatment resulted in a decrease in mRNA expression level of tight junction complex (**Figures 2B,C**), and it was enhanced in the bacterial consortium-administrated mice. Although the transcripts of *Ocln* mRNA in BCT mice were still lower than in control mice when tested by immunofluorescence, we found that the cellular localization of occludin was less impacted by DSS treatment, which was localized on the apical surface of the cell (**Figure 2D**), in contrast to the intracellular occludin staining observed in the DSS group of mice. Previous study showed that cellular localization of occludin could be regulated by the IL-17R-Act1 signaling pathway. We therefore tested the expression of IL-17RC and Act1 in colon tissue by WB and found that BCT improved the expression of these proteins, which was significantly decreased in DSS-treated mice (**Figure 2E**).

## The Intestinal **γδ**T17 Cells Were Upregulated by BCT

Previous study had shown that after acute intestinal injury, the intestinal-protective IL-17A could regulate the tight junction protein occludin to limit excessive permeability and maintain barrier integrity (25). We therefore tested the IL-17A levels in the colon mucosa by ELISA. Our results showed that DSS-caused injury induced an elevation of IL-17A in colon tissue, and this response could be detected as early as day 3 (**Figure 3A**). We detected an even higher concentration of IL-17A in mice that received BCT at both day 3 and day 7. In addition, the mRNA expressions of *il17a* and RORγt were significantly upregulated in BCT mice compared with that of the DSS group (**Figure 3B**), which confirmed the elevation of IL-17A secretion in colon mucosa. Studies have shown that γδT cells in colon mucosa are the primary producers of early protective IL-17A and, thus, play important roles in maintaining epithelial barriers. To investigate whether the protection effects of the bacterial consortium were associated with the upregulation of γδT17 cells, we analyzed the major intestinal IL-17-producing T-cell population in different mouse groups at day 3. Increase in γδT (CD4<sup>−</sup>TCR-γδ+) cells, as well as γδT17 (CD4<sup>−</sup>TCR-γδ+ IL-17A<sup>+</sup>) cells, was detected in BCT mice (**Figures 3C,D**). Further investigation revealed that the promoting effect of the bacterial consortium was only on γδT17 cells but not on the CD4<sup>+</sup> cells or Th17 (CD4<sup>+</sup> IL-17A<sup>+</sup>)

variance followed by Tukey's test, \**p* < 0.05, \*\*\**p* < 0.001.

cells (**Figures 3C,E**). We further isolated the total cLP cells and tested the effects of the bacterial consortium on the expansion of γδT cells and found that the population of γδT cells increased dramatically (*p* < 0.0001) upon treatment with bacterial products derived from the consortium (**Figure 3F**). An elevated IL-17A concentration was detected in the cell culture supernatant that was treated with the bacterial consortium when measured by ELISA (**Figure 3G**).

## The Bacterial Consortium Promoted Expansion of TLR2**<sup>+</sup> γδ**T Cells

Previous study (28) had shown that γδT17 cells express toll-like receptors TLR2, which selectively expands in response to bacterial products. We therefore examined the expression of TLR2 by cLP single-cell suspension with or without treatment with the bacterial consortium. Results showed that, upon treatment of the bacterial consortium (**Figure 4A**), the TLR2<sup>+</sup> γδT cells were un-regulated (**Figure 4A**). We further purified the splenic γδT cells of mice from different groups by MACS and found that BCT significantly increased the expression of TLR2 by γδT cells (**Figure 4B**). When purified γδT cells were challenged with the bacterial consortium *in vitro*, we detected an expansion of TLR2<sup>+</sup> γδT cells (**Figure 4C**), as well as an elevated IL-17A concentration in the cell culture supernatant (**Figure 4D**), which suggests that the TLR2 signaling pathway may be involved in the secretion of IL-17A from γδT cells.

## BCT Benefited the Re-Establishment of Intestinal Bacterial Equilibrium

We have previously shown that administration of the bacterial consortium promoted the recovery of intestinal microbial equilibrium in mice with ceftriaxione-induced dysbiosis, as well as in rats with

TNBS-induced colitis (15, 16). However, these results were based on PCR-DGGE technique, which gave a low resolution of intestinal microbiota. Here, we further adopted metagenomic analysis of the V3–V4 region of 16S rRNA gene sequences to systematically characterize the composition of the intestinal microbiota in mouse intestine post-BCT. Results showed that the overall OTUs of intestinal bacteria differ between groups (**Figure 5A**). Compared with the control group, the DSS-treated mice and BCT mice had a decreased number of OTUs. The three groups share the same 449 OTUs, but there are significant differences between every two groups (**Figure 5B**), with the most significant difference observed between the control group and the DSS group. Transplantation of the bacterial consortium resulted in obvious modification of the bacterial structure of mouse intestine as confirmed by the Principal Coordinate Analysis (**Figure 5C**). Changes were observed not only on the phylum level but also on the levels of order, class, family, and genus (**Figure 5D**; Figures S1 and S2 in Supplementary Material). A significant decrease in the abundance of *Bacteroidetes* (phylum) and *Bacteroidales* (order) and an elevated abundance of *Gammaproteobacteria* (class), *Enterobacteriales* (order), *Escherichia– Shigella* (genus), etc., were detected in DSS-treated mice (**Figure 5E**; Figure S3 in Supplementary Material). In contrast, transplantation

of the bacterial consortium resulted in the correction of these bacterial groups, especially the genus of *Escherichia–Shigella*, which may contribute to the re-establishment of intestinal equilibrium.

#### Biomarkers in Each Group

The metagenome analysis LEfSe approach was applied to identify the key phylotypes responsible for the differences between groups. *Bacteroidetes* and *Clostridia*, which were most abundant in the control mice, and *Proteobacteria*, *Diferribacteres*, and *Enterobactriales*, which were most abundant in the DSS mice, were the dominant phylotypes that contributed to the differences between the intestinal microbiota of control and DSS-treated mice (**Figure 6**). We did not detect any biomarkers in BCT mice, which suggest that the manipulation of microbiota by the bacterial consortium mainly may contribute to the recovery of disrupted intestinal homeostasis, rather than establishing a new unique microbiota.

## Changing of Specific Bacterial Groups Was Correlated with **γδ**T Cells

With the aim of clarifying how changing of intestinal microbial structure affects γδT cells, we analyzed the spearman correlation

γδT17 cells in γδT cells from cLP of different mice. (E) Percentage of CD4+ T cells and the percentage of Th17 cells in cLP of mice from different groups. (F) Representative flow cytometry plots and evaluation of the percentage of γδT cells among cLP cells treated or un-treated with 30 µl of the bacterial consortium *in vitro*. (G) IL-17A concentration in supernatant from cLP cells treated or un-treated with different volumes of the bacterial consortium *in vitro* detected by ELISA. All values are mean ± SEM (*n* = 5). Statistical analysis of the quantitative multiple group comparisons was performed using the one-way analysis of variance followed by Tukey's test, \**p* < 0.05, \*\*\**p* < 0.001, n.s., not significant.

between intestinal bacterial groups and γδT17 cells. It was found that bacteria belonging to the families of *Rhodospirillaceae*, *Flavobacteriaceae*, *Prevotellaceae*, etc., were negatively correlated with the upregulation of γδT17 cells in mouse intestines (**Figure 7A**), especially the genus of *Lachnospiraceae* UCG.005, *Lachnospiraceae*

NK4A136, *Prevotella* 9, *Prevotella* UCG.003, and the species of *Helicobacter ganmani*. In contrast, bacterial groups, such as the families of *Bifidobacteriaceae* and *Bacillaceae*, were positively correlated with γδT17 cells, especially the species of *Bifidobacterium animalis* and *Bacillus anthracis*, the correlation between these two bacterial

Figure 4 | The bacterial consortium promoted the expansion of TLR2<sup>+</sup> γδT cells. (A) TLR2<sup>+</sup> γδT cells in colonic lamina propria (cLP) of colitis mice were upregulated post-transplantation of the bacterial consortium. Left panel, representative flow cytometry analysis of TLR2<sup>+</sup> γδT (TLR2+TCR-γδ+) cells in cLP of mice from different groups at day 7. Right panel, the proportion of TLR2<sup>+</sup> γδT cells in total cLP γδT cells. (B) TLR2<sup>+</sup> γδT cells in spleen of colitis mice were upregulated post-transplantation of the bacterial consortium. Left panel, representative flow cytometry analysis of MACS-purified splenic γδT cells from mice of different groups at day 7; right panel, the proportion of TLR2<sup>+</sup> γδT cells in total splenic γδT cells. (C) Effects of the bacterial consortium on TLR2 expression of purified splenic γδT cells. Representative flow cytometry plots and evaluation of the percentage of TLR2<sup>+</sup> γδT cells among γδT cells treated or un-treated with 30 µl of the bacterial consortium *in vitro*. The γδT cells were purified by MACS from the spleen of C57BL/6J mice at day 7, which belonged to the control group. (D) IL-17A concentration in the supernatant of purified γδT cells treated or un-treated with 30 µl of the bacterial consortium *in vitro* detected by ELISA. All values are mean ± SEM (*n* = 5). Statistical analysis of the quantitative multiple group comparisons was performed using one-way analysis of variance followed by Tukey's test; when two groups were compared, the non-parametric *t*-test was performed by the assistant of GraphPad Prism 5. \**p* < 0.05, \*\**p* < 0.01,\*\*\**p* < 0.001.

species and γδT17 cells are statistically significant (*p* < 0.05). We further tested the effects of culture broth of single bacterial strains, including *Bifidobacterium longum* and *Bacillus cereus*, on purified splenic γδT cells. Interestingly, promotion of TLR2 expression and secretion of IL-17A by γδT cells were observed when the *Bacillus cereus* strain was treated, however, the *Bifidobacterium longum* strain did not show similar effects on γδT cells (**Figures 7B,C**).

#### DISCUSSION

The sustained immune responses caused by intestinal dysbiosis contribute significantly to the pathogenesis and development of IBD. Manipulation of the disturbed microbiota has therefore become a promising therapeutic means for IBD prevention and treatment. In recent years, FMT had been applied clinically in

(*n* = 5). \*adjusted *p* value <0.05; \*\*adjusted *p* value <0.01.

the US, China, and other countries; however, the outcomes were inconsistent (29). In addition, the potential risks of non-conditional pathogens in feces of donors and the attitude of patients toward FMT obstructed its development. As a new generation of bacterial therapy, BCT also has the potential for targeted restoration of the intestinal ecosystem, and the safety and manageability of this therapy suggested a significant step toward precision medicine (30). To this end, research has been carried out to test whether defined microbiota is protective in animal models infected by *C. difficile* (31) or *Salmonella* (32), and the results are promising. Our previous work also showed that a simple, defined microbial consortium can promote the recovery of intestinal microbial equilibrium in mice with ceftriaxone-induced dysbiosis (15). Transplantation of this bacterial consortium also showed effective anti-inflammatory activity on TNBS-induced colitis and upregulation of intestinal barrier function in rats (16), but the specific mechanism remains unknown.

In this study, we further analyzed the BCT effects on intestinal barrier function of DSS-treated mice and found that the tight junction protein occludin was specifically affected. Changing of the expression and the intracellular localization of occludin may contribute mainly to the improvement of the intestinal barrier function. This result suggests an emerging mechanism that the changing of intestinal microbial structure by BCT may affect the occludin regulation system. Intestinal T lymphocytes play a critical role in the mucosal immune system regulation by providing immune surveillance of the epithelium. Among them, the tissue-resident γδT cells are very important members that are considered as both protective and pathogenic T cells in IBD. The protective γδT cells were found to accumulate during DSS or pathogen-induced intestinal inflammation and produce KGF and IL-22, which promotes tissue repair and epithelial cell healing (33). The pathogenic role of γδT cells is mainly due to their production of IL-17, which can induce Th17 differentiation through the inflammatory DC-mediated production of IL-6 and IL-23. And, the Th17 cell-derived IL-17, IL-21, and IFN-γ promote MMP and NO production to induce inflammation and tissue damage (34, 35). However, a recent study demonstrated that IL-17A derived by γδT cells in the cLP promoted epithelial barrier function during DSS-mediated injury and protected the mice from excessive gut permeability (25). Given the previously suggested role of IL-17A in maintaining barrier function of epithelial tissues and the fact that the clinical trials targeting IL-17A were ineffective in treating CD (36, 37), this new finding therefore proves that IL-17A produced by γδT cells is protective, and its effects of maintaining barrier integrity may be exerted mainly through an IL-17R-ACT1-occludin pathway in epithelial cells. Our investigation into this pathway confirmed that the regulation of occludin and the enhancement of barrier function of colon mucosa after BCT are correlated with the expression changes of IL-17A, IL-17RC, and ACT1. And, to test whether γδT cells contributes to the higher levels of IL-17A in colonic tissue, we further tested the T-lymphocyte subpopulations by flow cytometry. The results confirmed that γδT cells but not the Th17 cells are the major source of IL-17A. DSS induced a compensatory accumulation of γδT17 cells in cLP, and this proportion was upregulated after transplantation of the bacterial consortium. The promotion of IL-17A producing γδT cells, therefore, highlighted the regulation mechanism of BCT on intestinal barrier function.

However, it is still unknown how BCT promoted IL-17A production from γδT cells. When tested *in vitro*, we found that the BCT-promoted IL-17A production from γδT cells was closely associated with the upregulation of TLR2 expression. This is in line with prior reports (38, 39) that the TLR2 pathway is critical for recognizing microbial products and activating the innate immune system in response to altered microbiota. Therefore, changing of the intestinal microbial structure may contribute to the mucosal immune response of γδT cells. This idea can be further supported by recent studies that commensal microbiota affects ischemic stroke outcome by regulating intestinal γδT cells (23). Gut commensal microbes were also found essential in

maintaining the homeostasis of liver-resident γδT17 cells (24). We therefore tested the intestinal microbial structure of mice from different experimental groups by 16S rDNA pyrosequencing. A dramatically dispersed intestinal microbiota in DSS-treated mice was detected, and this skewed homeostasis was corrected through transplantation of the bacterial consortium.

The deleterious roles of certain bacteria toward the intestinal damage and development of IBD have been proposed. For example, the clade of *Enterobacteriaceae*, particularly *Escherichia*/*Shigella*, has been found to be significantly increased in IBD patients and closely associated with intestinal inflammation (40). The genera *Escherichia*/*Shigella* were also found to be highly enriched in ileal CD patients above the general abundance of CD patients (41). Genera of *Mucispirillum*, which are increased during inflammation, have been suggested to be mucus-dwelling commensals that can cause disease, so calledpathobionts, because under some conditions, the immune system mounts an IgG response against them (42). Significant association of *Prevotella* with CD has been demonstrated by Said et al. (43), and their further studies supported the possibility that the increase of *Prevotella* contributes, at least partially, to the genetic susceptibility to CD (44). In addition, *Helicobacter ganmani*, which are some of the most prevalent bacterial contaminants of laboratory mice, were found associated with alterations in inflammatory cytokines in IL10-deficient mice (45). In accordance with these studies, we demonstrated that the abundance of these bacterial groups was suppressed in the gut of model mice received BCT. Importantly, we observed a significantly negative correlation between *Prevotella* spp. and *Helicobacter ganmani* with γδT17 cells. Interestingly, we also detected a negative correlation between certain *Lachnospiraceae* groups (UCG.005 and NK4A136) with γδT17 cells. The extensive diversity of this family (46) give them divergent resilience to colitis events (42), because promotion of *Lachnospiraceae* strains has been shown to restrict intestinal inflammation (47). Our results thus suggest a previously unidentified physiological difference in this group that may be responsible for the development of colitis.

(*n* = 5). The non-parametric *t*-test was performed between two groups by the assistant of GraphPad Prism 5. \**p* < 0.05, n.s., not significant.

In contrast, BCT promoted some bacterial groups, which were significantly depleted in DSS-treated mice. And, these groups have been implicated to be beneficial microbes in the gut, which may have evolved mechanisms to ameliorate intestinal inflammation and experimental colitis (48, 49). For example, IBD patients were found to have a decreased population of *Lactobacillus* compared to healthy controls. Many studies support that the colonization of *Lactobacillus* strains accounts for the resistance to DSS-induced colitis of animal models (50). Analogously, in our study, we found a decreased proportion of *Lactobacillus* in DSS-treated mice, while upon BCT, the *Lactobacillus* population was obviously increased (**Figure 5D**). Species of *Bifidobacterium* are also well-known as beneficial microbes in combating intestinal inflammation. Particularly low numbers and diversity of *Bifidobacterial* populations were found in pediatric IBD patients (51). *Bifidobacterium animalis* supsp. *lactis* was proved to have protective capacity on

acute and chronic colitis in mice (52). In our study, a positive spearman correlation between *Bifidobacterium animalis* and γδT17 cells was detected, which enhanced the important regulatory role of *Bifidobacteria* on colon mucosa. In addition, we also detected a positive correlation between *Bacillus* species with γδT17 cells. The *Bacillus* species has also been demonstrated to ameliorate DSSinduced dysbiosis and gut inflammation by balancing beneficial and harmful bacteria and associated anti- and pro-inflammatory agents (53). However, when tested *in vitro*, we found that only the *Bacillus* strain, but not the *Bifidobacterium* strain, showed promoting effects on TLR2 expression and IL-17A secretion by γδT cells, which suggested that some bacteria in the consortium may contribute to the expansion of intestinal γδT17 cells, and bacteria such as *Bifidobacterium* spp. may benefit from the re-established balance between gut microbiota and intestinal immune system and, in turn, exert beneficial effects on colon mucosa.

In conclusion, our results suggest that BCT can restore the alliance between commensal microbiota and intestinal γδT17 cells through the manipulation of intestinal dysbiosis in DSStreated mice and this contributes to the improvement of mucosal barrier function, which is closely associated with the upregulation of the IL-17A-Act1-occludin regulatory pathway. Bacteria, such as *Bacillus* and *Bifidobacterium*, may be responsible for or affected by the upregulation of γδT17 cells. However, further study is needed to clarify the underlying mechanisms.

## ETHICS STATEMENT

The animal experimental procedures were approved by the Medical Ethics Committee of Dalian Medical University, China (SYXK2015-0002). The strains used for bacterial transplantation are deposited in the Bacteria Collection of Dalian medical University (DMBC), China, and The University's Institutional Biosafety Committee (IBC) approved applications to conduct only scientific research with these microorganisms.

## AUTHOR CONTRIBUTIONS

XL and JY designed the research; ML, BW, XS, YT, XW, BG, YT, YD, and CH performed the experiments; ML and BW analyzed the data; ML and XL wrote the manuscript. All authors read and approved the final manuscript.

## REFERENCES


## FUNDING

This work was supported by the Nature Science Foundation of China (NSFC 81671606), the China Postdoctoral Science Foundation (2016M601317), the Nature Science Foundation of Liaoning Province, China (2015020262), and the Research Foundation from the Department of Education, Liaoning Province, China (L2016003).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu.2017.00824/ full#supplementary-material.

Figure S1 | The composition of bacterial composition in different experimental groups at Class and Order levels.

Figure S2 | Heatmaps of the abundance of the dominant Phyla (A), Family (B), and Genera (C) in different groups. The relative abundance of each microbial group was normalized, and the *Z*-value was presented and depicted by the color intensity.

Figure S3 | The specific bacterial groups that are significantly manipulated by BCT. The MetaStat method was used to identify the significantly reduced or elevated microbial groups by DSS treatment and was reversed by BCT. k, kingdom; p, phylum; c, class; o, order; f, family; g, genus; All values are mean ± SEM (n = 5). \*adjusted p value <0.05; \*\*adjusted p value <0.01.


inflammatory bowel disease. *Inflamm Bowel Dis* (2005) 11:481–7. doi:10.1097/ 01.MIB.0000159663.62651.4f


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Li, Wang, Sun, Tang, Wei, Ge, Tang, Deng, He, Yuan and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Reduced Mass and Diversity of the Colonic Microbiome in Patients with Multiple Sclerosis and Their Improvement with Ketogenic Diet

Alexander Swidsinski 1, 2 \*, Yvonne Dörffel <sup>3</sup> , Vera Loening-Baucke<sup>1</sup> , Christoph Gille<sup>1</sup> , Önder Göktas <sup>1</sup> , Anne Reißhauer <sup>1</sup> , Jürgen Neuhaus <sup>4</sup> , Karsten-Henrich Weylandt <sup>5</sup> , Alexander Guschin<sup>6</sup> and Markus Bock <sup>7</sup>

<sup>1</sup> Laboratory for Molecular Genetics, Polymicrobial Infections and Biofilms, Section of Gastroenterology and Hepatology, Department of Medicine, Charité Universitätsmedizin Berlin, Berlin, Germany, <sup>2</sup> Institute of Molecular Medicine, Sechenov First Moscow State Medical University, Moscow, Russia, <sup>3</sup> Outpatient Clinic, Charité Universitätsmedizin Berlin, Berlin, Germany, <sup>4</sup> Faculty of Veterinary Medicine, Centre for Infectious Diseases, University of Leipzig, Leipzig, Germany, <sup>5</sup> Charité Universitätsmedizin Berlin, Campus Virchow, Gastroenterology Berlin, Berlin, Germany, <sup>6</sup> Laboratory of Molecular Diagnostics, Russian Research Center for Molecular Diagnostics and Therapy, Central Research Institute, Moscow, Russia, <sup>7</sup> Experimental and Clinical Research Center, A Joint Cooperation between the Charité Medical Faculty and the Max-Delbrueck Center for Molecular Medicine, Berlin, Germany

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Undurti Narasimha Das, UND Life Sciences, United States Paolo Puccetti, University of Perugia, Italy

\*Correspondence: Alexander Swidsinski alexander.swidsinski@charite.de

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 28 May 2017 Accepted: 06 June 2017 Published: 28 June 2017

#### Citation:

Swidsinski A, Dörffel Y, Loening-Baucke V, Gille C, Göktas Ö, Reißhauer A, Neuhaus J, Weylandt K-H, Guschin A and Bock M (2017) Reduced Mass and Diversity of the Colonic Microbiome in Patients with Multiple Sclerosis and Their Improvement with Ketogenic Diet. Front. Microbiol. 8:1141. doi: 10.3389/fmicb.2017.01141 Background: Colonic microbiome is thought to be involved in auto-immune multiple sclerosis (MS). Interactions between diet and the colonic microbiome in MS are unknown.

Methods: We compared the composition of the colonic microbiota quantitatively in 25 MS patients and 14 healthy controls.Fluorescence in situ hybridization (FISH) with 162 ribosomal RNA derived bacterial FISH probes was used. Ten of the MS patients received a ketogenic diet for 6 months. Changes in concentrations of 35 numerically substantial bacterial groups were monitored at baseline and at 2, 12, and 23/24 weeks.

Results: No MS typical microbiome pattern was apparent.The total concentrations and diversity of substantial bacterial groups were reduced in MS patients (P < 0.001). Bacterial groups detected with EREC (mainly Roseburia), Bac303 (Bacteroides), and Fprau (Faecalibacterium prausnitzii) probes were diminished the most. The individual changes were multidirectional and inconsistent. The effects of a ketogenic diet were biphasic. In the short term, bacterial concentrations and diversity were further reduced. They started to recover at week 12 and exceeded significantly the baseline values after 23–24 weeks on the ketogenic diet.

Conclusions: Colonic biofermentative function is markedly impaired in MS patients.The ketogenic diet normalized concentrations of the colonic microbiome after 6 months.

Keywords: FISH, colonic microbiota, multiple sclerosis, biofermentation, ketogenic diet

## INTRODUCTION

There is a growing awareness of the significance of the human microbiome in health and disease. Microbial colonization of the skin and epithelial surfaces protects from pathogens. The biofermentation in the colon delivers energy from digestive leftovers and synthesizes a broad spectrum of vitamins and hormone-like substances, which regulate metabolism and neuronal

**462**

activity (Galland, 2014). The enormous variety of the healthy microbiome conveys antigenic diversity to the host shaping its immunity and autoimmunity (Berer and Krishnamoorthy, 2014).

An ever growing number of studies demonstrates the involvement of the colonic microbiome in obesity, digestive, endocrine, inflammatory, and auto-immune disorders including multiple sclerosis (MS) (Berer and Krishnamoorthy, 2014; Galland, 2014; Glenn and Mowry, 2016).

Topics, methods and results of these studies are well presented and should not be repeated here. However, the previous studies of the colonic microbiome in MS were restricted to identification of microbial patterns associated with disease (Miyake et al., 2015; Jangi et al., 2016; Tremlett et al., 2016). We found no literature on the quantitative evaluation of the colonic microbiome in MS patients. Microbial concentrations, however, are an important feature, which directly measures their functional contribution to colonic fermentation. The aim of this study was to compare the concentrations of different microbial groups in MS patients and healthy controls and to follow up changes in the colonic microbiome taking place during ketogenic diet.

The option of ketogenic diet was important for the following reasons: as long as complex microbiomes cannot be reliably transferred, maintained and tested in vitro, all data raised in vivo must remain observational. Therefore, interventions simultaneously affecting the microbiome and disease are necessary to unravel possible causality.

Ketogenic diet influences brain function, inflammation, immunity and the colonic microbiome. It is increasingly applied in clinical studies (Piccio et al., 2008; Kim et al., 2012; Choi et al., 2016). Different to fasting or mono-diets, ketogenic diet can be maintained over months and is well tolerated. Diet prescriptions are often circumvented in real life. The compliance of the ketogenic diet can be reliably verified through measurement of ketone bodies in blood and urine and cannot be falsified.

## MATERIALS AND METHODS

#### Patients/Samples

Fourteen healthy volunteers from the Laboratories of Centre for Infectious Diseases, Faculty of Veterinary Medicine, University of Leipzig and 25 patients with relapsing-remitting multiple sclerosis cared for at the Charité hospital were investigated for the composition of their colonic microbiome using fluorescence in situ hybridization ribosomal RNA based FISH probes available in public resources (Loy et al., 2016).

The study was reviewed and approved by institutional review board: Ethikkomission Ethikausschuss 1 an Campus Charite Mitte EA1/130/07.

After the baseline investigation, MS patients received a ketogenic diet for 6 months. The ketogenic diet was designed (1) to achieve a modest ketosis (≥500µmol/L ß-hydroxybutyrate) in the blood, self-measured after dinner twice a week (FreeStyle Precision, Abbott Diabetes Care Ltd.), (2) to achieve a modest ketosis (≥500µmol/L acetoacetate) in the urine, self-measured after dinner once a week (Ketostix, Bayer Consumer Care AG), and (3) to maintain compliance. Patients received a booklet with meal suggestions over 28 balanced days and were encouraged to ingest fat. An average daily intake of <50 g carbohydrates, >160 g fat, and <100 g protein was recommended. Patients received detailed information about nutritional facts, glycemic load and learned how to handle carbohydrates by an experienced nutritional coach during group based workshops on 3 weekends.

Ten of the MS patients who were recruited for the ketogenic diet were randomly selected for accompanying microbiome investigations. Stools samples were collected at baseline, week 2, 12, and after 6 months (week 23–25). The evaluation of the clinical effects on MS was not performed because of the low number of patients.

None of the enclosed probands received antibiotics or probiotics in the last 6 month preceding the study.

## Fish

Colonic microbiota were investigated using FISH with ribosomal RNA derived probes. Hybridizations were performed on sections of Carnoy fixated, paraffin embedded and otherwise not manipulated stool cylinders (Swidsinski et al., 2010). Four micrometers thick sections were placed on SuperFrost plus slides.

A Nikon e600 fluorescence microscope was used. The images were photo-documented with a Nikon DXM 1200F color camera and software (Nikon, Tokyo, Japan).

Bacterial concentrations of homogeneous populations were enumerated visually in one of the 10 × 10 fields of the ocular raster corresponding to 10 × 10µm of the section surface at magnification of 1,000. This number was assigned to a concentration of × 10<sup>9</sup> bacteria/ml, which was most equivalent to the calculation formula, that we had used previously (Swidsinski et al., 2010).

In case of uneven distribution of bacteria over the microscopic field, the positive signals were enumerated in 10 fields of the ocular raster along the gradient of distribution and divided by 10.

Bacteria were quantified using group specific C3 probes. The FITC marked universal probe was used in each hybridization to evaluate the number of all bacteria, C5 marked probes with a different specificity to C3 probes were used to determine the spatial relation of different bacterial groups to each other.

Only signals that hybridized with a specific FISH probe and the universal FISH probe, but did not hybridize with specific FISH probes from unrelated bacterial groups, were enumerated (Swidsinski, 2006).

#### FISH Probes

One hundred sixty-two bacterial FISH probes available from public resources were applied for the comparative analysis of the colonic microbiome in healthy controls and MS patients (**Tables 1A,B**). The names of the FISH probes are listed according to abbreviations of the probeBase resource (http://probebase.csb. univie.ac.at/node/8) (Loy et al., 2016) and the details to FISH probe specificity and hybridization conditions are given. The Fprau probe is described in reference (Suau et al., 2001).

Probes in **Table 1** are alphabetically ordered to subgroups according to abundancy and specificity as described in the result section.

TABLE 1 | Applied FISH-probes.

#### Part A

Substantial groups Essential (N = 3) Erec482 (Eubacterium rectale, Clostridium coccoides group) Bac303 (most Bacteroidaceae) Fprau (Faecalibacterium prausnitzii)

#### Individual pioneer (N = 4)

Bif153 Genus Bifidobacterium Cdif198 Clostridium difficile Clit135 Clostridium lituseburense group including C. difficile Ebac1790 Enterobacteriaceae

Individual substantial (N = 28) ACI623 Acidaminococcaceae sp. (not the Selenomonas species) AKK406 Akkermansia Ato291 Atopobium cluster Bbif186 B. bifidum Blon1004 B. longum Bputre698 Bacteroides putredinis Burkho Burkholderia spp. Ceut705 C. eutactus, Coprococcus sp. Chis150 Clostridium histolyticum Cor653 Coriobacterium group Cvir1414 Clostridium viride group Ecyl387 Eubacterium cylindroides Ehal1469 Eubacterium hallii Eram997 Eubacterium ramulus Lab158 Lactobacillus sp., Enterococcus sp. Muc1437 Akkermansia muciniphila Myc657 Mycobacterium subdivision (mycolic acid-containing actinomycetes) Phasco741 Phascolarctobacterium faecium Pnig657 Prevotella nigrescens ProCo1264 Ruminococcus productus Rbro730 Clostridium sporosphaeroides, Ruminococcus bromii, Clostridium leptum Rfla729 Ruminococcus albus SFB1 Segmented filamentous bacteria SNA Sphaerotilus natans Strc493 most Streptococcus spp. SUBU1237 Burkholderia spp., Sutterella spp. Urobe63a Ruminococcus obeum-like Veil 223 Veillonella Ver620 Verrucomicrobium

#### Part B

Includes probes with extremely low occurrence and concentrations and probes with uncharacteristic signals:

#### Accidental groups (N = 88)

To low in occurrence and concentrations for statistical analysis MIB661 mouse intestinal bacteria AER66 Aeromonas spp. Alac1438 Anaerococcus lactolyticus ARC1430 Arcobacter Avag1280 Anaerococcus vaginalis Bbrel198 B. breve

#### TABLE 1 | Continued

Bden82 B. dentium BFV530 Bacteroides forsythus Bif1278 Bifidobacterium spp. Burcep Burkholderia cepacia CAP365 Capnoytophaga sp. Capno Capnocytophaga sputigena Cj490 Campylobacter jejuni CLOBU1022 Clostridium butyricum, Cperf 191 Clostridium perfringens Cra757 Clostridium ramosum assemblage Csac67 Clostridium saccharogumia CST440 Group 1 clones closely related to Clostridium stercorarium DSV1292 some Desulfovibrio and Bilophila wadsworthia DSV687 most Desulfovibrionales E.bar1237 Eubacterium barkeri E.bif462 Eubacterium biforme E.con1122 Eubacterium contortum E.cyl461 Eubacterium cylindroides E.cyl466 Eubacterium cylindroides E.dol183 Eubacterium dolichum E.had579 Eubacterium hadrum E.len194 Eubacterium lentum E.lim1433 Eubacterium limosum E.mon84 Eubacterium moniliforme E.mul Eubacterium multiforme E.sab Eubacterium saburreum E.ven66 Eubacterium ventriosum Enfl84 Enterococcus faecalis Enfm 93 Enterococcus faecium Fnec996 Fusobacterium necrophorum Fnuc133 Fusobacterium nucleatum GAN1237 Helicobacter ganmani Haeinf Haemophilus influenzae HEP642 Helicobacter hepaticus Hpy-1 Helicobacter pylori Hyo1210 Brachyspira hyodysenteriae Lbuc668 Leptotrichia buccalis Lis1255 Genera Listeria, Brochothrix Lis637 Genus Listeria, Lpara Lactobacillus casei Lzeae Lactobacillus zeae Pae997 Pseudomonas spp. Pamic1435 Parvimonas micra Pana134 Peptostreptococcus anaerobius Pden654 Prevotella denticola Pilosi1405 Brachyspira pilosicoli Pilosi209 Brachyspira pilosicoli Pint649 Prevotella intermedia Pint657 Prevotella intermedia Pnasa1254 Peptoniphilus asaccharolyticus Pnhar1466 Peptoniphilus harei Pnivo731 Peptoniphilus ivorii POGI Porphyromonas gingivalis

(Continued)

(Continued)


Ppu646 Pseudomonas spp. PRIN Prevotella intermedia Saga Streptococcus agalactiae SAL3 Genus Salmonella Sau Staphylococcus aureus Saur327 Staphylococcus Saur72 Staphylococcus aureus Ser1410 Genus Brachyspira SGD229 Genus Desulfotomaculum Sita649 Candidatus Sphaeronema italicum Spn Streptococcus pneumoniae Spy Streptococcus pyogenes Staaur Staphylococcus aureus Staph747 Staphylococcus spp. STEBA1426 Sterolibacterium lineage Stemal Stenotrophomonas maltophilia Strpyo (Streppyo) Streptococcus pyogenes Sval428 Some Desulfobulbaceae TM7305 subdivision 1 of candidate division TM7 Trep-D3-4 32 PT1 Treponema refringens Trep-D4-4 32 PT3 clone DDKL-20 Treponema refringens Trep-HW 170 PT6 clone DDKL-4 Treponema phagedenis Trep-T5-4 32 PT2 Treponema refringens Urobe63b Ruminococcus obeum-like VEPA Veillonella parvula VIB572a Genus Vibrio Y Yersinia FISH probes with signals which could not be definitively assigned to a

# specific group of bacteria (N = 31)

Alf 1b (Alpha) Alphaproteobacteria, some Deltaproteobacteria, Spirochaetes Arch 915 Archaea B for Bacteroides forsythus B(T)AFO Tannerella forsythensis Bact for Bacteroides forsythus Bang198 B. angulatum Bfra602 most Flavobacteria, some Bacteroidetes Bmy843 Bacillus BORR4 Genus Borrelia Bvulg1017 Bacteroides vulgatus CF319a most Flavobacteria, some Bacteroidetes CFB560 subgroup of Bacteroidetes, CFB division Clept1240 Clostridium leptum Efaec Enterococcus faecalis, Enterococcus sulfuricus Enc131 Enterococcus spp and other Ent Enterobacteriaceae except Proteus spp. FUS664 most Fusobacterium sp FUSO Fusobacterium sp. MIB724 mouse intestinal bacteria PBR2 Bifidobacterium breve PseaerA Pseudomonas aeruginosa PseaerB Pseudomonas aeruginosa

(Continued)



We performed hybridizations with all probes but excluded from analysis 31 of these probes, because they showed multiple uncharacteristic signals, in form and distribution not resembling bacteria or cross-reacting with non-related bacterial groups and eight FISH probes that were identical to related probes for the same species.

To reduce the number of unnecessary investigations, while following the impact of the ketogenic diet on the colonic microbiome, only 35 bacterial groups which were found to have substantial occurrence (in at least 20% of individuals) and concentrations (>10<sup>9</sup> in at least one of the stool samples of one individual) were applied.

#### Statistical Analysis

Differences between groups were evaluated using the two sided t-Student U-test. Data are presented as means ± SD, P <0.05 was considered statistically significant.

#### RESULTS

#### Eligibility of the FISH Probes for Analysis of the Stool Microbiome

Three bacteria detected with EREC (mainly Roseburia), Bac303 (Bacteroides), and Fprau (Faecalibacterium prausnitzii) probes were always present in healthy human controls and MS patients and contributed to about half of the colonic microbiome in each subject. We called these groups **essential bacteria**.

All other investigated bacterial groups were individual, detectable only in a subset of patients. We called them **individual bacterial** groups.

Twenty-eight of the individual bacterial groups were found in at least 30% of the probands (mostly 50–70%) in concentrations of higher than 10<sup>9</sup> bacteria/ml. They contributed substantially to the colonic microbial mass. We called them **individual substantial groups**.

Four of the **individual** bacterial groups including Bif (Bifidobacteriacae), Ebac (Enterobacteriaceae), Clit (Clostridium lituseburense), and Cdif (Clostridium difficile) are often found prevalent in newborns, after antibiotic treatment and convalescence patients and thus represent bacterial groups with pioneer function. We evaluated these groups separately.

Individual bacterial groups detected with 88 FISH probes (**Table 1B**) were observed in one, maximal two individuals in marginal concentrations of ≤0.1 ×10<sup>9</sup> ml. We called these **individual marginal** bacterial groups. Because of the uneven distribution of such bacteria over the stool cylinder, their quantification was highly unreliable.

Thirty-one of investigated FISH probes showed multiple uncharacteristic signals, in form and distribution, cross-reacting with unrelated bacterial groups. The appearance of these signals was not visually different in MS and healthy controls. Because of the uncertainty of what exactly is measured, we did not perform the quantitative analysis of the results detected with these probes.

Eight FISH probes, showed identical results with related probes for the same species/bacterial group, were also not quantitatively evaluated.

#### Microbiome in Healthy Controls and MS Patients Prior to Diet Intervention

The morphologic appearance of single bacteria detected with corresponding FISH probes was the same in MS patients and healthy controls. The distribution of bacteria over the stool cylinder surface was not noticeably different. None of the investigated bacterial groups, including groups with unspecific signals, demonstrated prevalence or absence in MS patients, which could be interpreted in terms of Koch's postulates.

As long as bacterial groups were compared pairwise, the differences between MS and healthy patients were discordant, gradual and moderate, reaching only in 9 groups statistical significance (**Table 2**). The concentrations of most investigated groups in MS were decreased. Six groups were slightly increased and included Cor653 (Coriobacterium group), Cvir1414 (Clostridium viride group) Ehal (Eubacterium hallii), Ecyl387 (Eubacterium cylindroides), Lab158 (Lactobacillus sp., Enterococcus sp.), Rfla729 (Ruminococcus albus) bacterial groups. Seven substantial individual bacterial groups had similar high concentrations in MS and healthy controls.

Due to uncertain occurrence and low concentrations, which was difficult to quantify, we did not compare the single marginal bacterial groups in MS patients and controls quantitatively. No rise of single marginal groups in MS patients was observed.

The difference in the microbiome of MS patients and healthy controls became striking, when concentrations of numerically substantial groups were summarized and considered as a whole (**Table 2**, marked in bold). The diversity of all substantial groups in MS patients was reduced by 36% (P < 0.001), the mean sum concentrations of all substantial bacterial groups were reduced by 24% (65 vs. 85.4 ×10<sup>9</sup> /bacteria/ml., P < 0.001) compared to healthy controls. The decrease in concentrations was most profound in the essential bacteria group (32%) and less impressive in individual substantial (19%) and pioneer groups (14%).

#### Changes of the Colonic Microbiome with the Ketogenic Diet in 10 MS Patients

The changes in the microbiome during the ketogenic diet were univocal. Except for Akkermansia, all groups of bacteria demonstrated consistently more or less marked decreases, leading to a reduction of the total bacterial concentrations of the substantial bacteria from 65 to 25 × 10<sup>9</sup> bacteria per ml at week 2. Some of the bacterial groups fell below detection level, resulting in further decline of the bacterial diversity from 48 to 36%.

However these tendencies were temporary.

The total bacterial concentrations in MS patients started to increase at week 12, reaching values typical for healthy controls at week 23/24, being then significantly higher than bacterial concentrations in MS patients prior to diet and statistically not different to mean bacterial concentrations in healthy controls (P = 0.7). This increase was consistent for all but the pioneer bacterial groups and Akkermansia. The concentrations of pioneer bacterial groups fell and remained low, the concentrations of Akkermansia increased initially but then declined during the ketogenic diet.

#### DISCUSSION

Previous investigation, using high throughput DNA sequencing technologies in large scale 16S rRNA or shotgun metagenomic sequencing, demonstrated miscellaneous changes in the composition of the colonic microbiome, which correlated with MS, MS-onset, -therapy or -relapse (Kim et al., 2012; Berer and Krishnamoorthy, 2014; Galland, 2014; Miyake et al., 2015; Glenn and Mowry, 2016; Jangi et al., 2016; Tremlett et al., 2016). Both single bacterial groups were found differently represented and also the whole microbiome was aberrantly composed. While the alpha diversity of MS and the healthy microbiome was similar, the beta diversity differed significantly. In ecology, alpha diversity expresses the mean species diversity in sites or habitat at a local scale, while beta diversity is the ratio between regional and local diversity. The differences indicated shifts in composition of the MS microbiome. The meaning of these observations is unclear. The pure occurrence of bacteria does not automatically mean that they are relevant or biochemically active. The vacant niches may be occupied by chance.

Although we applied publically available FISH probes as broadly as possible and included all groups covering numerically substantial components of the colonic microbiome, no conclusions to the entire biodiversity are possible, since FISH reliably detects only bacteria in concentrations of higher than 10<sup>5</sup> per ml.

However, while sequence analysis is perfect for identification of specific occurrence patterns, its information on physical abundance and contribution of bacteria to biofermentation is poor. Abundance of bacteria within the fecal mass however directly expresses their biofermenting power.

TABLE 2 | Colonic microbiome in healthy and MS patients prior to and during the ketogenic diet.


(Continued)


Our data quantifying microbial participants clearly demonstrate the impaired colonic function in patients with MS. Both concentrations and biodiversity of numerically substantial bacterial groups were markedly reduced in MS. Essential bacteria were most, individual substantial bacteria less depleted, while pioneer bacteria were nearly unchanged.

This grading of suppression matches well with the proposed role of these groups for colonic function. Essential bacteria are present in every healthy person in large concentrations, contributing roughly to approximately half of the mass of the colonic microbiome. They are obviously important for colonic fermentation and represent main fermentative groups in human. The individual substantial bacterial groups are present only in subsets of healthy persons in varying concentrations, which are each distinctly lower than those of the essential bacterial groups. Their presence is dispensable for colonic fermentation. However, their diversity is high, composition specific for each subject, indicating that they fulfill special tasks not covered by the essential bacterial groups. Despite markedly lower concentrations, when compared to essential bacterial groups, the individual substantial groups constitute together another half of the colonic biomass.

Pioneer bacteria are usually found in low concentration in healthy adults. Their concentrations are high in newborns after antibiotic treatment and in convalescence, while the colonic microbiome is reshaped (Swidsinski et al., 2016).

The fall in concentration in MS patients was gradual, decreasing from essential to individual substantial and further to pioneer bacterial groups. This suggests that the suppression is not due to the loss of responsible microbial groups or reshaping of the microbiome. It presumably results from the general downregulation of the colonic biofermentative function and affects mainly biofermentative active groups, leaving bacterial groups with other specific tasks untouched.

We observed no changes in the microbiome that could be specific for MS.

While mean concentrations of all essential bacteria detected with EREC (mainly Roseburia), Bac303 (Bacteroides), Fprau (F. prausnitzii) probes were consistently reduced in MS, the shifts in individual substantial bacterial groups were multidirectional with concentrations of some bacterial groups unchanged, increased or reduced when compared to healthy controls. Although some of the differences reached the level of statistical significance, the fluctuations in concentrations of individual substantial bacterial groups were moderate and in the range of those, when unmatched groups of subjects are compared with each other. Similar fluctuations are documented while describing microbiome composition in other pathologies such as diarrhea, irritable bowel syndrome, inflammatory bowel disease, and others (Mai et al., 2016; Swidsinski et al., 2016).

Since the microbiome is influenced by a multiplicity of racial, occupational, social, regional, and geographic factors, matching for all of them would be an impossible task. Mean values raised in small cohorts should therefore be critically evaluated, even when they occasionally prove to be statistically highly significant (Mai et al., 2016). Our data allow definitively no conclusions to whether the observed alterations in concentrations of colonic bacteria precede, result, specifically accompany MS or rely on independent coincident processes such as aberrant immunity, impaired digestion or simply changed behavior. However, they clearly demonstrate that the perturbation of the microbiome in MS is not inherent and inevitable but can be definitively corrected with diet, supplementation and other means as assumed previously (Tanca et al., 2015).

After 6 months on the ketogenic diet, the sum concentrations of the substantial microbial groups in MS patients increased significantly when compared to the period prior to intervention (83 vs. 65 × 10<sup>9</sup> bacteria/ml; P = 0.02) and became indistinguishable from the healthy group (83 vs. 85.4 × 10<sup>9</sup> bacteria/ml. P = 0.7).

Our data also demonstrate that the tracking of the changes in the microbiome should not be restricted to a single time point, but needs surveillance over longer periods of time.

The reduction of microbial concentrations observed 2 weeks after the start of the diet was dramatic, the ranges of suppression were comparable to antibiotics effects (Swidsinski et al., 2016). Concentrations of some individual substantial groups fell under detection limit, leading to further drop in the microbial diversity from mean 48 to 35 percent. However, the processes behind dietetic and antibiotic effects are principally different.

The improvement with antibiotic treatment occurred only after the end of treatment, while with the ketogenic diet, the improvement occurred in succession of the diet and the longterm effects of diet were opposite to the immediate response to the intervention. Although the diversity of the microbiome did not reach the values typical for the healthy population, at the end of the observation period, after 6 months on the ketogenic diet, it completely recovered as compared to basic values prior to treatment.

Obviously the increase in microbial concentrations was mainly due to improvement of colonic function and achieved by preexisting microbial groups. Typical for convalescence after depletion of the colonic microbiome following antibiotic use, stroke or inflammation is a temporary excess of pioneer bacterial groups. Such excess was not observed with the ketogenic diet. In contrast, the pioneer bacterial groups remained reduced over the whole duration of the ketogenic diet (p < 0.05–0.008), indicating an absence of substantial microbiome reshaping.

Although the concentrations and the biodiversity of colonic microbiota are strong markers of the intensity of the microbial metabolism, the shifts in bacterial groups per se do not reveal the exact metabolic changes taking place. The role of single substances and metabolites in neurologic disorders is still to be unraveled and follow our preliminary observations.

Summarizing, we state that colonic microbiome and neuropathology are closely interrelated. Concentrations of numerically substantial biofermentative bacteria are significantly reduced in MS patients. The microbial shifts can be reliably quantified and monitored by FISH under ambulatory conditions. The ketogenic diet for 6 months completely restored the microbial biofermentation mass and is an interesting interventional tool for prospective clinical studies.

## AUTHOR CONTRIBUTIONS

AS, MB,YD, and AG designed the study, YD, JN, AR, KW, and ÖG conducted the study, ÖG and VL critically revised the manuscript, and AS, AG, CG, and AR. performed FISH, CG, JN statistically analyzed the data. All authors contributed to conception of the work, revising of the data, shaping of the manuscript and approved the final draft submitted.

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Swidsinski, Dörffel, Loening-Baucke, Gille, Göktas, Reißhauer, Neuhaus, Weylandt, Guschin and Bock. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Associations between Viral Infection History Symptoms, Granulocyte Reactive Oxygen Species Activity, and Active Rheumatoid Arthritis Disease in Untreated Women at Onset: Results from a Longitudinal Cohort Study of Tatarstan Women

#### *Edited by:*

*Juarez Antonio Simões Quaresma, Universidade Federal do Pará, Brazil*

#### *Reviewed by:*

*Paul Proost, KU Leuven, Belgium Ali Mobasheri, University of Surrey, United Kingdom Antonio C. R. Vallinoto, Institute of Biological Sciences (ICB) of Federal University of Pará, Brazil*

> *\*Correspondence: Marina I. Arleevskaya marleev@mail.ru*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 17 July 2017 Accepted: 22 November 2017 Published: 05 December 2017*

#### *Citation:*

*Arleevskaya MI, Shafigullina AZ, Filina YV, Lemerle J and Renaudineau Y (2017) Associations between Viral Infection History Symptoms, Granulocyte Reactive Oxygen Species Activity, and Active Rheumatoid Arthritis Disease in Untreated Women at Onset: Results from a Longitudinal Cohort Study of Tatarstan Women. Front. Immunol. 8:1725. doi: 10.3389/fimmu.2017.01725*

*Marina I. Arleevskaya1 \*, Albina Z. Shafigullina1 , Yulia V. Filina1,2, Julie Lemerle3 and Yves Renaudineau1,3*

*1State Medical Academy, Kazan, Russia, 2Kazan Federal University, Kazan, Russia, 3 Laboratory of Immunology and Immunotherapy, INSERM U1227, Hôpital Morvan, Centre Hospitalier Régional Universitaire (CHU) de Brest, Brest, France*

To evaluate the effects of infectious episodes at early stages of rheumatoid arthritis (eRA) development, 59 untreated eRA patients, 77 first-degree relatives, from a longitudinal Tatarstan women cohort, were included, and compared to 67 healthy women without rheumatoid arthritis (RA) in their family history. At inclusion, informations were collected regarding both the type and incidence of infectious symptom episodes in the preceding year, and granulocyte reactive oxygen species (ROS) were studied at the basal level and after stimulation with serum-treated zymosan (STZ). In the eRA group, clinical [disease activity score (DAS28), health assessment questionnaire] and biological parameters associated with inflammation (erythrocyte sedimentation rate, C-reactive protein) or with RA [rheumatoid factor, anticyclic citrullinated peptide (anti-CCP2) antibodies] were evaluated. An elevated incidence of infection events in the previous year characterized the eRA and relative groups. In addition, a history of herpes simplex virus (HSV) episodes was associated with disease activity, while an elevated incidence of anti-CCP2 autoantibody characterized eRA patients with a history of viral upper respiratory tract infection symptoms (V-URI). Granulocyte ROS activity in eRA patients was quantitatively [STZ peak and its area under the curve (AUC)] and qualitatively (STZ time of peak) altered, positively correlated with disease activity, and parameters were associated with viral symptoms including HSV exacerbation/recurrence, and V-URI. In conclusion, our study provides arguments to consider a history of increased viral infection symptoms in RA at the early stage and such involvement needs to be studied further.

Keywords: rheumatoid arthritis, infection symptoms, viruses, women, reactive oxygen species

## INTRODUCTION

Rheumatoid arthritis (RA) development results from an inappropriate immune response to environmental challenges in genetically predisposed patients. Accordingly, the list of microorganisms associated with RA is still growing, and several hypotheses support their causative role, as recently reviewed (1). On one hand, the immunological hypothesis proposes that RA development results from an inappropriate immune response to infections, which can lead to loss of tolerance to self-antigens. In support of such a hypothesis, up to 40 risk factors have been associated with RA, and a significant part of the genetic contribution is associated with the major histocompatibility complex locus (2). While on the other hand, the environmental hypothesis proposes that RA development may result from a cumulative effect of microbial/ viral/environmental factors and thus explains the absence of a single defined pathogen. In this case, environmental factors may be accelerators or protective (3).

In order to decipher the respective contributions of the immune system and infections to RA, the longitudinal Tatarstan women cohort study was established in 2002 in order to include the follow-up of RA families (4). Our preliminary results from this cohort have established, first, an elevated rate of infection symptoms in the year preceding RA onset. Second, there is an elevated rate of bacterial colonization in feces, urine, and skin that remains after several years in treated patients with RA. Third, there are impaired innate (phagocyte) and acquired (antibacterial IgG response) immune responses in RA patients. In this study, our main objective was to study the infectious spectrum preceding RA onset and to test its impact on disease activity and innate immunity in comparison to controls and healthy first-degree relatives.

## MATERIALS AND METHODS

#### Subjects

In this prospective study conducted between 2002 and 2016 at the Kazan Rheumatology department, RA women were included in the Tatarstan cohort at time zero when they met RA diagnostic criteria according to the 2010 ACR/EULAR classification criteria (5, 6). In the cases of 14 of the patients who were diagnosed before 2010, the RA onset was diagnosed by consensus of three experienced rheumatologists. For this study, a total of 208 women were studied semiannually including 59 therapy naïve early stage patients [early stages of rheumatoid arthritis (eRA), defined as <6 months of RA duration, time since diagnosis 0.3 ± 0.1 years; age range: 40.8 ± 16.2 years old], 77 first-degree relatives, and a healthy control group (control) comprised of 67 women with no chronic disease and no RA among close relatives. All the individuals included in this study were described previously (4). Exclusion criteria were based on risk factors for infection as previously discussed (7). The study was approved by the Ethical Committee of the Kazan State Medical Academy, Kazan, Russia (Permit nr 1/2002). Consent was received from all patients involved in the study, including consent to participate in the study and consent to allow publication of the results.

When the RA diagnosis was verified and before starting therapy on the day of the chemiluminescent study, we also performed a standard clinical and laboratory examination including the erythrocyte sedimentation rate (ESR)-based 28 joints disease activity score (DAS28), the Health Assessment Questionnaire (HAQ), C-reactive protein (CRP), rheumatoid factor (RF), and antibodies to citrullinated peptides (CCP2), as previously described (8–10). All patients and those relatives with joint symptoms (pain and morning stiffness) in the small joints of the feet and hands underwent magnetic resonance imaging. Patients with a DAS28 ≥5.1 at the baseline were defined as highly active (11). The control group showed no symptoms of arthritis, no ESR, or CRP elevation (except during infection), and no reactivity of RF and anticyclic citrullinated peptide (anti-CCP2).

#### Infection

Information on infections within the last year and according to the criteria defined in **Table 1** was collected from the RA patients during 2-day hospital visits by a physician qualified in rheumatology (AMI). The subjects were biannually or annually questioned about any symptoms suggestive of infections experienced during the 6–12 months, and infection histories for the two semesters were compiled. For the control group, information was from the 1 year preceding enrollment. Information from documents at other clinics was sought whenever a general practitioner had been visited. All subjects were tested and were negative for serological markers of chronic viral infections including human viral hepatitis (HAV, HBV, and HCV), and human immunodeficiency virus (HIV). A history of allergic disease and/or detection of allergen-specific IgE was considered as an exclusion criterion.

The following parameters of the infectious syndrome were analyzed: no infection, viral upper respiratory tract infection symptoms (V-URI), bacterial upper respiratory tract infection symptoms requiring antibiotic therapy (B-URI), herpes simplex virus (HSV) exacerbation/reactivation, chronic tonsillitis exacerbation, and acute bronchitis. The incidence of infectious episodes and their overall duration per year were also collected. Only those episodes judged by the rheumatologist to truly indicate an infection were scored. In the case of exacerbation of a chronic infection and/or doubt regarding an infection, the diagnosis was established by a specialist in the corresponding medical area.

### Granulocyte Reactive Oxygen Species (ROS)

Peripheral blood granulocytes were collected from 59 untreated and eRA patients, 77 relatives, and 58 controls. Since an active infection may introduce a bias in the analysis, samples from all groups were collected in a period when there were no clinical symptoms of an infection and without any routine laboratory signs of inflammation. Granulocytes were isolated on a Ficoll– Urografin density gradient. The proportion of granulocytes in the cell suspension was 90–95% and the percentage of viable cells was 95–98%. The production of ROS was assessed at the basal level and after stimulation with serum-treated zymosan (STZ,

#### Table 1 | Criteria used in the study to establish infectious episodes within the last year.


Sigma) by the luminol-dependent chemiluminescence technique (chemiluminometer designed by Santalov BF, Pushchino, Russia). Real-time registration was performed every 4 s in thermostat plastic chambers with continuous mixing of granulocyte suspensions (sample volume 0.2 ml, cell density 106 cells/ml, and concentration of STZ 0.25 mg/ml). The following parameters were measured: spontaneous level of ROS production [arbitrary units (au)], peak of ROS production after STZ (in au), area under the curve (AUC, in au × min), and time of occurrence of peak ROS production (min).

#### Statistical Analysis

Continuous data are described as mean ± SEM. Differences among groups were analyzed by one-way ANOVA in a nonparametric test and the Dunn's test was used for *post hoc* comparisons, or the Fisher's exact test for categorical data. Following normality and equality of variance tests, nominal values were compared to controls using the Student's *t*-test or alternatively by using a nonparametric test (Mann–Whitney rank sum test) with Bonferroni corrections. For correlation analysis, the Pearson's coefficient *r* was calculated and Bonferroni corrections applied. *P*-values under 0.05 were considered significant. Statistical analyses were performed using GraphPad Prism 7.0 (La Jolla, CA, USA).

#### RESULTS

#### Infections

In order to evaluate the incidence of infectious episodes in the year preceding RA onset, 59 untreated eRA women and their first-degree relatives (*n* = 77) were selected and compared to 67 healthy controls. In these 203 women, the following parameters were assessed: viral upper respiratory tract infection symptoms (V-URI), bacterial upper respiratory tract infection symptoms (B-URI), acute bronchitis, HSV exacerbation or reactivation, and chronic tonsillitis exacerbation. When compared to controls and first-degree relatives (**Figure 1A**), eRA patients included in this study had more frequent infection symptoms in the previous year than healthy controls (4.2 ± 0.4 events/year in eRA versus 2.7 ± 0.2 in controls, *P* = 0.03), but less than what was observed in relatives (6.0 ± 0.4 events/year, *P* = 0.004 versus eRA, and *P* < 10<sup>−</sup><sup>4</sup> versus controls).

Next, and as presented in **Figure 1B**, the incidences of each specific type of infection in the subjects' histories were compared between the three groups, revealing differences for V-URI (Kruskal–Wallis test, *P* = 0.04), HSV exacerbation/reactivation (*P* = 0.01), and chronic tonsillitis reactivation (*P* = 0.001). An increase for HSV in the infection exacerbation/reactivation history was observed in eRA patients (**Figure 1C**, 1.5 ± 0.2 events/ year in eRA versus 0.7 ± 0.2 in controls and versus 0.8 ± 0.2 in relatives, *P* = 0.005 and *P* = 0.02, respectively). Chronic tonsillitis

Figure 1 | The number (nb) of infectious symptom events during the 1-year period preceding the arthritis onset (early [e]RA) or the first examination [healthy control (Cont) and first degree relatives (Rel)]. The following parameters were assessed: viral upper respiratory tract infection symptoms (V-URI), bacterial upper respiratory tract infection symptoms requiring antibiotic therapy (B-URI), acute bronchitis, herpes simplex virus (HSV) exacerbation/reactivation, and chronic tonsillitis exacerbation. (A) Total infectious symptom event incidence in the preceding year; (B) specific infectious symptom incidence history; (C) HSV exacerbation/reactivation; and (D) chronic tonsillitis exacerbation. Statistics are indicated when *P* ≤ 0.05 and (\*) denotes the calculated *P-*values \**P* ≤ 0.05, \*\**P* ≤ 0.01, \*\*\**P* ≤ 0.001.

reactivation in the history was almost absent from controls, present in eRA and the highest prevalence was observed in relatives (**Figure 1D**, *P* = 0.03 and *P* = 0.002, respectively). When considering the V-URI history, there was a trend for V-URI incidence reduction in eRA patients (data not shown). As the incidence of B-URI was similar between the three groups, and as the prevalence and incidence of acute bronchitis was modest in relatives and in eRA patients, both parameters were not considered further.

#### Clinical Data

Next, and as presented in **Table 2**, the impacts of V-URI, HSV exacerbation/reactivation, and chronic tonsillitis reactivation in the history of episodes were analyzed with regards to the clinical and biological parameters of RA. First, eRA patients who presented V-URI symptoms in the previous year were characterized by an elevated prevalence of anti-CCP2 autoantibodies (*P* = 0.006). Second, for those early stage patients who suffered from HSV infection exacerbations during the year preceding the examination, they had more active RA disease (DAS28 > 5.1, *P* = 0.0005), and no difference when considering HAQ, ESR, CRP, RF, and anti-CCP2 autoantibodies. Third, no associations were related to a chronic tonsillitis exacerbation episode history. These results support the notion that V-URI and HSV symptoms in the history influence RA activity, but probably through distinct pathways.

#### Granulocyte ROS

Granulocyte ROS production plays a crucial role in controlling viral and bacterial infections and in the pathophysiology of RA by promoting inflammation and, in turn, by inducing cartilage and joint tissue damage. Accordingly, we measured the production of ROS at basal level (spontaneous), and after stimulation by using STZ in 59 eRA patients, 77 relatives, and 58 controls. In relatives and in eRA patients (**Figures 2A,B**), differences in relation to controls did not reach significance with regards to the basal level and the maximum peak of ROS production. In contrast (**Figures 2C,D**), for relatives and even more for eRA, a reduction in ROS AUC (5,924 ± 766 au × min in eRA, 10,927 ± 2,212 au × min in relatives versus 16,184 ± 2,489 au × min in controls, *P* = 0.003 and *P* = 0.0003, respectively) and a remarkably delayed time until the peak (17.6 ± 1.0 min in eRA versus 13.2 ± 0.7 min in relative and versus 7.9 ± 0.4 min in controls, both *P* < 10<sup>−</sup><sup>4</sup> ) were observed.

Next, to test the impact of altered granulocyte ROS production on RA's pathophysiology, spontaneous ROS production and the ROS indexes after STZ stimulation were tested for any correlations with DAS28, HAQ, CRP, and RF in new RA patients (**Figures 3A,B**). Both STZ peaks and STZ AUC correlated with DAS28 (*P* = 4 × 10<sup>−</sup><sup>5</sup> and *P* = 2 × 10<sup>−</sup><sup>7</sup> , respectively), and CRP (*P* = 0.003 and *P* = 0.002, respectively). Regarding the time to reach the peak of ROS production, correlations were observed with CRP (*P* = 0.01). We conclude from these experiments that granulocyte ROS production upon STZ stimulation is altered in untreated eRA patients, and that a positive correlation exists between disease activity and STZ-dependent ROS production.


Table 2 | Characteristics of early (e)RA patients according their infection symptoms reported in the previous year: all studied infection symptoms (All), viral upper respiratory tract infection symptoms (V-URI), Herpes simplex virus (HSV) exacerbation or reactivation, and chronic tonsillitis (Ton) exacerbation.

*Mean* ± *SEM.*

*Statistics \*0.01* < *P* < *0.05; \*\*0.001* < *P* < *0.01; \*\*\*0.0001* < *P* < *0.001.*

## Linkage of Certain Infections and ROS Parameters

Taking into account that ROS are major pathogenic molecules produced during viral infections, including V-URI and HSV, and at the same time, important players in the immune response against bacterial infections presented by tonsillitis in our groups, we next tested whether viral and/or bacterial infection symptom episodes might influence RA through the control of the NADPH oxidase (NOX) activity. Accordingly, we further evaluated the association between V-URI, HSV, and tonsillitis exacerbation episodes reported in the previous year with the basal level and STZ-dependent ROS production in new RA patients, relatives, and controls.

When eRA patients were dichotomized according to the appearance of HSV during the year before examination (**Figure 4**), the HSV exacerbation/reactivation episode history was associated with both increased spontaneous ROS production (*P* < 10<sup>−</sup><sup>4</sup> ) and, upon STZ stimulation, an increase in the peak of ROS production as observed in controls (*P* = 0.05 and *P* = 0.03, respectively). In first-degree relatives, an opposite association between an HSV episode history and spontaneous ROS production (*P* = 0.01) was highlighted, while no differences were reported following STZ stimulation.

V-URI episodes during the year preceding the study were associated in new RA patients with increased spontaneous ROS production (*P*< 10<sup>−</sup><sup>4</sup> ) and, upon STZ stimulation, to a normalization of the time to reach the peak (*P* = 0.0006) (**Figure 5**). No difference was observed for controls and relatives when considering the V-URI history.

For chronic tonsillitis exacerbation, controls were not considered since only 2/58 controls presented episodes during the previous year. In contrast to HSV and V-URI, chronic tonsillitis exacerbation was associated in eRA patients with a negative impact on the spontaneous ROS production (*P* = 0.01) and an increase in the time to reach the peak (*P* = 0.001) (**Figure 6**). No differences were reported for the first-degree relative group.

## DISCUSSION

Results from this study suggest an important contribution of the viral infectious episodes in the year preceding RA in women on both onset and activity of the disease after excluding patients with allergy and chronic viral diseases (viral hepatitis, HIV). First, and based on the incidence of the minor infectious episode history in patients with eRA, a critical role for HSV events may be suspected. Second, a history of HSV and V-URI episodes was associated with increased granulocyte ROS production and disease activity (HSV) or specific antibody production (V-URI) in untreated eRA patients, while a decrease was observed in the case of chronic tonsillitis exacerbation. Third, the link between ROS response and disease activity in eRA was further confirmed. Fourth, firstdegree relatives of patients with eRA such as siblings, parents, and children presented an elevated incidence of viral infectious episodes with a normal immune response, characteristics that can help to better understand the mechanisms leading to RA.

#### Infections and RA

Dangerous connections exist between infections and RA since, on the one hand, defective anti-infectious activity characterizes patients with RA while, on the other hand, infections are

(AUC), time to reach peak ROS production, and rheumatoid arthritis (RA) parameters: erythrocyte sedimentation rate-based 28 joints disease activity score (DAS28), the Health Assessment Questionnaire (HAQ), C-reactive protein (CRP), and rheumatoid factor (RF) levels in RA patients. (A) Correlation table of RA parameters. Colors represent the correlation coefficients (red being the highest and blue the lowest), whereas statistical significances (\*) denotes the calculated *P* values \**P* ≤ 0.05, \*\**P* ≤ 0.01, \*\*\**P* ≤ 0.001, and \*\*\*\**P* ≤ 10−<sup>4</sup> . (B) Significance of correlations between ROS and RA parameters, the Spearman's *P*-value is indicated for each panel.

suspected of promoting autoimmunity (1). Among viral infections associated with autoimmunity, HSVs are strongly suspected of contributing to RA development, and we have observed an elevated rate of HSV events in the history of eRA patients. We have further established an association between HSV exacerbation/reactivation in the year preceding RA onset and disease activity (DAS28). These data are in agreement with previous reports. First, with regards to the patient's personal and family history of RA, a higher personal history of HSV (OR = 2.4) is observed (12). Second, the incidence rate of HSV has already been demonstrated to be two times greater in RA than the rate observed in age matched controls (13). Third, a higher HSV viral load is also reported in RA patients (14), and this is interpreted as the result of impaired cellular immunity that characterizes RA patients. Fourth, the defective capacity of the immune system to control an HSV infection is considered to favor reactivation in patients with RA receiving biological or conventional diseasemodifying antirheumatic disease drugs (15).

Similarly, among rhinovirus, enterovirus, respiratory syncytial virus, and influenza virus known to be associated with V-URI, the incidence of the latter is reported to be higher in RA than in controls and there is a 2.75-fold increase in pneumonia complications in RA patients infected with influenza (16). RA patients with interstitial pneumonia have elevated levels of anti-CCP and RF as recently reported (17). This is in agreement with our observation that a history of V-URI is associated with higher antibody prevalence (CCP2) in eRA patients supporting the concept that V-URI may be an actor in RA activity. However, we failed to confirm the higher prevalence of V-URI in eRA and such a discrepancy may be related to the seasonality of the infections and to the criteria for selection of the control cohorts between the studies.

The most studied bacterial strain associated with RA is related to *Porphyromonas gingivalis*, which is suspected of promoting autoimmunity by inducing protein citrullination, promoting HLA-DR overexpression and citrullinated peptide presentation to CD4 T cells, and by interfering with the production of cytokines and chemokines such as Jak/STAT and NF-kappaB (18, 19). In line with this model, case reports have been published indicating that RA disease can be improved following tonsillectomy (20), and that antibodies against a carbohydrate antigen, Strep A, of *Streptococcus pyogenes* are produced during chronic tonsillitis in RA patients (21). However (22), the analysis of 1,524 RA patients

with associated antecedent tonsillectomy or appendectomy have failed to show any association suggesting that chronic tonsillitis is at the best an activator but not an inducer of RA, a conclusion that is in agreement with our observations showing that chronic tonsillitis exacerbations were increased in both new RA patients and their relatives.

## Infections and ROS

Host defense against microbial infection involves ROS production, and toll-like receptors (TLR) have been identified as a major class of pattern-recognition receptors necessary for triggering ROS. TLR family members expressed by granulocytes facilitate the recognition of pathogen-associated molecular patterns, as was demonstrated with bacterial pathogens that colonize tonsils (23). In the case of chronic inflammation and tissue injury, the host can in addition produce endogenous TLR ligands, known as damage-associated molecular patterns (DAMPs), and DAMPs have the potential to activate inflammation, initiating a vicious cycle in germ-free conditions (24). Abnormal expression of DAMPs is reported in human RA tissues, and it has been speculated that DAMPs are critical for RA pathogenesis by controlling granulocyte functions including the oxidative burst (25, 26). The DAMPs hypothesis is attractive since it provides an explanation to the major defective ROS activity observed in the basal level and after STZ activation (but not when using PMA for activation, data not shown) in RA patients. Another hypothesis is related to the fact that zymosan is a TLR2 ligand (27), and that earlier investigations have highlighted the importance of TLR2 function in RA pathogenesis (28, 29). Accordingly, the DAMP and/or TLR2 hypothesis can both explain the ROS dysregulation observed in eRA although cell samples were collected in periods without any clinical symptoms of an infection and without any routine laboratory signs of inflammation. Future experiments are necessary to test whether or not the abnormal STZ capacity to induce ROS in eRA patients is related to DAMPs and/or to an abnormal TLR2/ NF-kB pathway.

Regarding viruses and, in particular, HSV, some of them alter granulocyte functions. Indeed, HSV can directly attach to granulocytes *via* the herpes virus entry mediator (HVEM) and formyl peptide receptors, penetrate into the cells, and subsequently increase ROS production (30). RA is characterized by HVEM overexpression on various cells due to the increased levels of phagocytosis, ROS production, as well as production of interleukin-8 (neutrophil chemoattractant) and TNF-alpha (31). ROS production might inhibit the antiviral immune response since, in particular, ROS downregulates NK cell function, NK cells being the principle players in the antiviral immune response and maintenance of the latent state of HSV (32, 33). The immune response during influenza infection, being an acute upper respiratory tract viral infection, includes

such a strong granulocyte stimulation, and, in particular, the ROS production by these cells, that these cytotoxic factors become the most important drivers of the inflammatory process and its complications (34–37). In our study, we have observed an increased ROS activity when considering both spontaneous and STZ stimulated peak ROS activity of granulocytes from RA patients presenting a V-URI or HSV history in the preceding year. Since such an effect is also associated with higher RA activity (HSV) or a higher immune response (V-URI), three non-exclusive hypotheses can be proposed. First, there is a direct effect of the latent virus on granulocyte ROS production. Second, the effect is indirect and may be a consequence of the intensive inflammatory process, in this way, the HSV capacity to produce DAMPs has been reported (38). Third, since the pathogenic role of anti-Fc gamma receptor IIIb autoantibodies on granulocytes is well documented in RA, our data allow the assumption that there is some link between V-URI and RA associated autoantibodies (39–41).

While studying one of the essential mechanisms of antiinfectious activity, granulocyte ROS production in RA, a remarkable feature of NOX activity was revealed at eRA with a twofold decrease in the response to STZ stimulation. Granulocytes are known to be first responders to infections, and immediate ROS release in response to a pathogen is an important condition for curbing an infection. So, the NOX response rate against a pathogen might be essential for its successful immobilization. This is especially important during an aggressive bacterial infection. As a consequence, the slowing of the granulocyte response to STZ is suspected to play a role in the relatives' and eRA patients' susceptibility to the exacerbations of chronic tonsillitis, mainly caused by bacteria. At the same time, there was no difference in this index in the patients who underwent or did not undergo HSV event exacerbation since the rapid granulocyte ROS production does not play a significant role in the neutralization of viruses. It should be noted that previously we demonstrated the delayed monocyte NADPH oxidative response upon STZ stimulation as well (42).

#### ROS and RA

In addition, we have observed an association between spontaneous ROS production, STZ stimulated ROS parameters and disease activity (DAS28, CRP) in patients with eRA at onset. However, the role of ROS in RA is not sufficiently clear. In the initial view, increased ROS activity has been documented in synovial joints to directly contribute to the inflammation, to induce the hyperplasia of synovial tissues, and to damage cartilage, bone, and ligaments (43). In addition, it was demonstrated that abnormal ROS activity controls antigen presentation and reduces T cell responsiveness in RA through effects on cellsurface proteins such as CD4 and signal transduction proteins

such as LAT or Zap70 that are present close to the plasma membrane (44, 45). In addition, ROS, by damaging endothelial cells, increase the permeability of the vasculature and promote granulocyte migration to inflammation sites (90%). Last but not least, cytokine overproduction, including TNF-alpha and IL-1, is thought to be the main contributor to ROS in RA. However, the paradigm has changed with the discovery that allelic polymorphism in the respiratory burst oxidase component neutrophil cytosolic factor (ncf)1 in rat and ncf4 in human was associated with a more severe and a higher incidence of RA (46). The exact mechanisms are incompletely understood but may rely on an enhanced activation of autoreactive T and B cells as observed in ncf1 transgenic mice (47). In our study, a strong and positive correlation between disease activity (DAS28) and the level of ROS was observed, which is in agreement with a previous report that proposed measurement of ROS for monitoring disease severity in RA (48).

In conclusion and although patients with allergy and viral chronic infections were excluded from the study, we are aware of the limits of our study in terms of sample size, in terms of sex selection, and the fact that infections were established mostly on symptoms rather than biological parameters. Thus more studies including serological and an extended microbiota analysis are needed to support our observations and for that the longitudinal Tatarstan cohort study, which provides a longitudinal analysis of RA patients from the cohort of Tatarstan women and their first-degree relatives, is appropriate for the study of interactions between the multiple factors associated with RA.

## ETHICS STATEMENT

The study was approved by the Ethical Committee of the Kazan State Medical Academy, Kazan, Russia (Permit nr 1/2002).

## AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.

## ACKNOWLEDGMENTS

We are thankful to Aida Gabdulkhakova (Central Research Laboratory, Kazan State Medical Academy, Russia) for assistance in selecting the conditions of the chemiluminescent study. The authors also express thanks to Dr. Wesley H. Brooks (University of South FL, USA) for editorial assistance, and to Simone Forest and Genevieve Michel for secretarial assistance.

## FUNDING

This study was supported by research funding from the "Russian Science Foundation" (No. 17-15-01099).

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer AV and handling Editor declared their shared affiliation.

*Copyright © 2017 Arleevskaya, Shafigullina, Filina, Lemerle and Renaudineau. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

#### *Zoe Rutter-Locher1 , Toby O. Smith2 , Ian Giles <sup>3</sup> and Nidhi Sofat <sup>1</sup> \**

*1Musculoskeletal Research Group, Institute of Infection and Immunity, St George's University of London, London, United Kingdom, 2 Faculty of Medicine and Health Sciences, University of East Anglia, Norwich, United Kingdom, 3Center for Rheumatology Research, Rayne Institute, University College London, London, United Kingdom*

Background: Systemic lupus erythematosus (SLE) is a chronic systemic inflammatory autoimmune disease, the etiology of which remains only partially characterized. Strong evidence implicates chronic infections in the development and chronicity of autoimmune conditions. Recently, an association has been demonstrated between periodontitis and rheumatoid arthritis. Such observations have led to the investigation of the possible role of periodontitis and oral dysbiosis in other systemic inflammatory conditions, including SLE. The aim of this study was to examine whether there is an association between SLE and periodontitis.

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Maximilian F. Konig, Massachusetts General Hospital, United States Giovanni Cizza, Novo Nordisk, United States*

> *\*Correspondence: Nidhi Sofat n.sofat@sgul.ac.uk*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 02 July 2017 Accepted: 27 September 2017 Published: 17 October 2017*

#### *Citation:*

*Rutter-Locher Z, Smith TO, Giles I and Sofat N (2017) Association between Systemic Lupus Erythematosus and Periodontitis: A Systematic Review and Meta-analysis. Front. Immunol. 8:1295. doi: 10.3389/fimmu.2017.01295*

Methods: MEDLINE *via* OVID, EMBASE *via* OVID, and PsycINFO *via* OVID databases were searched to identify eligible studies, screened by two independent authors and verified by a third. Studies comparing presence of periodontitis in SLE cases to controls without SLE were included. Data were extracted using a predefined table and papers were appraised using Down's and Black tool. Mantel–Haenszel meta-analysis was performed using RevMan.

results: Eight case–control studies were included, with 487 SLE cases and a total of 1,383 participants. On meta-analysis of four studies, risk of periodontitis in SLE cases compared to controls was significantly greater with a risk ratio of 1.76 (95% CI 1.29–2.41, *p* = 0.0004). No statistical difference was found in individual measures of periodontitis, such as probing depth or clinical attachment loss, between SLE cases and controls.

conclusion: Our study found a statistically significant increased risk of periodontitis in patients with SLE compared to controls. This finding suggests a possible association between these two conditions. Larger longitudinal studies are needed to confirm this possible association.

Keywords: systemic lupus erythematosus, autoimmune and inflammatory diseases, microorganisms, periodontitis, periodontal disease, meta-analysis

## INTRODUCTION

Systemic lupus erythematosus (SLE) is a systemic, chronic inflammatory condition with diverse clinical manifestations, primarily affecting the joints, internal organs, and the skin (1).

The etiology of SLE is incompletely understood, but it is thought to occur in genetically primed individuals in whom the inflammatory response is triggered by an environmental stimulus. Immunosuppressant medications are the mainstay of treatment, but these are limited both in efficacy and by multiple side effects which lead to significant morbidity and mortality. Oral manifestations of SLE are common and typically take the form of painless oral ulcers that are frequently present during disease flares and are included in current SLE classification criteria (2).

Periodontitis is an infectious-inflammatory condition, affecting the periodontal ligament and alveolar bone (3). Most cases are due to the chronic accumulation of oral plaque which initiates inflammation, further bacterial colonization and tissue destruction. Gingivitis usually occurs first and can be reversed with oral hygiene methods. However, once the inflammation extends past the gums to the deeper tissues, the loss of periodontal attachment and bone causes progressive loosening of teeth, eventually leading to their loss (4). The "red complex" organisms, *Porphyromonas gingivalis*, *Tannerella forsythia*, and *Treponema denticola*, have a key role in the development of periodontitis (5, 6).

There is strong evidence that periodontitis, and specifically oral dysbiosis is associated with autoimmune inflammatory disease, principally rheumatoid arthritis (RA). A recent meta-analysis including 153,492 participants showed a significant association between periodontitis and rheumatoid arthritis (7). Two specific bacteria have been implicated in triggering the underlying inflammatory process, *P. gingivalis* (8, 9) and *Aggregatibacter actinomycetemcomitans* (10). Clinical trials are underway to investigate the effect of non-surgical treatment of periodontitis, such as oral hygiene and mechanical removal of plaque in RA.

There is increasing interest regarding the possible role of oral dysbiosis in the etiology of other autoimmune inflammatory conditions, including SLE. Further understanding of any potential association between periodontitis and SLE would expand current knowledge of the etiology of SLE and may lead to novel management strategies.

In this review, we hypothesized that there may be an association between SLE and periodontitis. To evaluate this hypothesis, we conducted a systematic review and meta-analysis of relevant publications.

#### MATERIALS AND METHODS

The protocol for the review was registered with PROSPERO (Registration number: CRD42016053490) an international register of systematic reviews (http://www.crd.york.ac.uk/PROSPERO/). We used the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2009 checklist to report the review (11).

#### Eligibility Criteria for Population

Participants in the studies needed to have a diagnosis of SLE based on internationally recognized criteria, ACR 1982/1997 revised classification criteria (12) or clinical diagnosis by a rheumatologist. Studies which included participants of all ages, gender, and disease severity were eligible.

#### Eligibility Criteria for Study

To be included, studies needed to be observational studies of cross-sectional, case–control, or cohort design. Journal articles and conference proceedings were included. Review articles, case reports, animal model studies, and those with unavailable abstracts were excluded. Non-English language papers were excluded. There were no restrictions on date of publication or publication status.

#### Eligibility Criteria for Outcome Measure

To be included, prevalence of periodontitis using standardized measures needed to be reported in both SLE population and non-SLE population.

#### Search Strategy

MEDLINE *via* OVID, EMBASE *via* OVID, and PsycINFO *via* OVID databases were searched using the following terms: systemic lupus erythematosus, SLE, lupus erythematosus, systemic or lupus nephritis, lupus vasculitis and Periodont\*, gum disease, gingivitis, tooth decay, oral health, dental health, oral plaque index (PI), probing pocket depth, bleeding on probing (BOP), and clinical attachment loss (CAL). In addition, Google Scholar was searched using the following term "Periodontitis and systemic lupus erythematosus." The searches were re-run just before the final analyses on the 03/08/2017.

#### Study Selection

The Titles and/or abstracts were screened by two independent authors (Zoe Rutter-locher and Toby O. Smith) to identify potential eligible studies. Final selection of studies was performed by two independent authors (Zoe Rutter-locher and Toby O. Smith) and verified by a third author (Nidhi Sofat) by reviewing the full text based on inclusion criteria above. Any disagreements were resolved by discussion.

#### Data Extraction

A standardized, pre-piloted form was used to extract data from the included studies. Extracted information included: (1) study design including, publication journal and date, inclusion and exclusion criteria, diagnostic criteria of SLE, definition of periodontitis, (2) study participant demographics including % females, mean age, years of SLE disease, therapies, measure of severity of SLE, and (3) periodontal measures including prevalence of periodontitis, oral plaque index (PI), probing depth (PD), clinical attachment loss (CAL), and bleeding on probing (BOP). Two authors (Zoe Rutter-locher and Toby O. Smith) extracted the data independently, discrepancies were identified and resolved through discussion (with a third author, Nidhi Sofat, where necessary). Authors were contacted by email to obtain missing information and included when received.

#### Quality Assessment

Risk of bias and quality assessment was reviewed using Downs and Black tool for non-randomized control trials (13). This 27-point tool assesses studies on five key sections: (1) study quality (10 points), (2) external validity (3 points), (3) study bias (7 points), (4) confounding and selection bias (6 points), and (5) Power of the study (1 point). Disagreements between the review authors over the risk of bias in particular studies were resolved by discussion, with involvement of a third review author where necessary.

#### Data Analysis

As there was homogeneity in participants, study design and outcome measure on visual assessment of the data extraction table, a meta-analysis was performed. Primary outcome was to calculate relative risk of periodontitis in participants with SLE compared to participants without SLE. Secondary outcomes were to calculate relative risk or mean difference for measures of periodontal disease. These measures of periodontal disease included PD, oral PI, BOP, bleeding gingival index, and CAL.

Median and interquartile range were converted to mean (SD) to allow comparison of studies, under the assumption of normal distribution. A fixed effect meta-analysis was performed when the inconsistency value (*I*-squared) was ≤50% and Chi-squared equates *p*= 0.10 and a random-effect meta-analysis when *I*-squared was >50% and Chi-squared equates to *p* < 0.10. Risk ratio with 95% confidence intervals was calculated for the prevalence rates of periodontitis and the mean difference was calculated for continuous variables. Risk of bias was identified in Down's and Black tool. Publication bias was not performed as it is convention to only present a funnel plot for 10 or more data points in a meta-analysis (14).

All analysis and forest plots were performed on RevMan Version 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

#### RESULTS

#### Search Results

As shown in **Figure 1**, a total of 485 studies were identified using the search strategy. Of these, 454 were excluded as they were duplicates, non-English articles, review articles, or non-relevant. The 31 remaining full length articles were screened and eight were deemed appropriate to be included in the qualitative and quantitative analysis.

#### Study Design

The study characteristics are shown in **Table 1** and Table S1 in Supplementary Material. All studies included in the meta-analysis were case-control in design. Seven were published in peer review journals and one was taken from conference proceedings (15). Most studies (six of seven) were published since 2015. The remaining articles were published between 1993 and 2007. Three studies were based in Europe (two in the UK and one in Germany), two in Brazil, one in Saudi Arabia, one in Taiwan, and one in China.

The total number of study participants was 1,383 which included 487 cases of SLE. All studies defined cases as fulfilling the ACR 1982/1997 revised classification criteria for diagnosis of SLE (12). Cases were recruited mainly from Rheumatology clinics, with the exception of Meyer et al. who recruited from the "Department of Internal Medicine" (19). All studies defined controls as either "Healthy" or "Individuals without a history of rheumatic conditions or autoimmune disease." Controls were recruited from a variety of sources including dental clinics, staff at the medical school and epidemiological survey. Most studies included exclusion criteria in order to reduce the effect of confounders such as age, smoking, recent antibiotics usage, but this varied between studies.

#### Quality Assessment

Quality assessment was hindered by the limited information available for some studies, especially those that came from conference proceedings. As shown in **Table 2**, studies included in this meta-analysis had clear hypothesis, outcome measures and aims (*N* = 8, 100%). The participants were representative of the population (*N* = 7, 88%) and the outcome measures were reliable and clearly reported (*N* = 8, 100%) with estimates of random variability (*N* = 7, 88%). Particular limitations were that assessors were not blinded (*N* = 1, 13%) and sufficiently powered cohort size was present in only half of studies (*N* = 4, 50%).

#### Study Participant Demographics

**Table 3** shows the demographic of the cohorts. A total of 1,383 participants were included in the meta-analysis, with 487 participants with SLE and 896 participants without SLE. Mean age of participants with SLE was 41.3 years and controls was 42.8 years, excluding juvenile SLE. All studies had a predominance of females. Mean duration of disease was, as expected, lowest in study by Fernandes et al. looking at juvenile SLE (18). Excluding this study, mean disease duration ranged from 4.5 to 11 years. Use of immunosuppressant's varied from 36 to 100%. There was also variability in what medications were included under the term "immunosuppression." There was insufficient detailed data in order to control for immunosuppressant use in meta-analysis. Importantly, smoking status, a known risk factor for periodontitis, was the same in cases and controls in those studies with available data.

#### Meta-analysis: SLE and Periodontitis Prevalence

As shown in **Table 4**, all studies used similar measures of periodontitis. Four of the seven studies, which included 1,064 participants, defined periodontitis (15, 17, 21–23). Three studies (15, 17, 22) differentiated between mild and severe periodontitis.

The prevalence rates of periodontitis were similar in these four studies. The exceptions were the lower prevalence rate of periodontitis in the controls of the study by Wang et al. and Zhang et al. (21, 22). On meta-analysis, risk of periodontitis in SLE patients compared to controls was significantly greater with a risk ratio of 1.76 (95% CI 1.29–2.41, *p* = 0.0004, **Figure 2**). Furthermore, this greater risk ratio remained significant when the study published as a conference proceeding was excluded from the meta-analysis (RR 2.05 95% CI 1.15–3.66, *p* = 0.02) and when the study by Wang et al. was excluded from meta-analysis (RR 1.50 95% CI 1.28–1.75, *p* < 0.00001). Interestingly two studies which differentiated between mild and severe periodontitis found no difference in the prevalence of severe periodontitis between cases and controls, whereas one did (22).

#### Meta-analysis: SLE and Measures of Periodontal Disease

All studies provided data on measures of periodontal disease. Six of the eight studies reported oral PI (16, 18–20, 22, 23). Six of the studies reported on presence of gingivitis (16–20, 22, 23), with five reporting BOP and two Loe and Silness gingival index (20, 22). Six studies provided data on PD (15–17, 20–23) and four

on CAL (16, 17, 21–23) which are the gold standard measures of periodontitis. Four studies reported on residual teeth (16, 19–21), and two reported the WHO DMFT index (decayed-missing-filled teeth index) (18, 19). Meyer et al. and Zhang et al. were the only studies to use radiographs to measure bone loss, an accurate measure of periodontitis (19, 22).

Number of residual teeth was very similar in all studies. Overall, the mean residual teeth count was 23.68 for SLE and 23.65 for controls. On meta-analysis there was no significant difference in mean of PI (0.03, 95% CI −0.09–0.16, *p* = 0.62, **Table 5**, Figure S1 in Supplementary Material) or BOP (1.60, 95% CI −0.72–3.92, *p* = 0.18, **Table 5**, Figure S2 in Supplementary Material).

We also performed meta-analysis on the gold standard measures of periodontitis. Risk of PD ≥ 5 mm was greater in SLE patients compared to controls but this was not statistically significant with risk ratio of 1.34 (95% CI 0.99–1.82, *p* = 0.06, **Table 5**, Figure S3 in Supplementary Material). Risk of CAL ≥ 2 and difference in means for PD and CAL were not statistically different between cases and controls; Risk of CAL ≥ 2 risk ratio 1.09 (95% CI 0.99–1.21, *p*= 0.08, **Table 5**, Figure S4 in Supplementary Material), PD difference in means 0.08 (95% CI −0.27–0.43, *p* = 0.66, **Table 5**, Figure S5 in Supplementary Material), CAL difference in means 0.41 (95% CI −0.12–0.95, *p*= 0.13, **Table 5**, Figure S6 in Supplementary Material). Excluding the study in Juvenile SLE by Fernandes et al. did not make a significant difference to any outcome.


#### Table 2 | Down and Black's appraisal.


*S, not stated; N/A, not applicable.*

*1, Al-Mutari et al. (16); 2, Calderaro et al. (17); 3, de Pablo et al. (15); 4, Fernandes et al. (18); 5, Meyer et al. (19); 6, Mutlu et al. (20); 7, Wang et al. (21); 8, Zhang et al (22).*

#### DISCUSSION

Our report is the first systematic review to examine the association between periodontitis and SLE. On meta-analysis we found a statistically significant overall increased risk of periodontitis in patients with SLE compared to controls, suggesting an association between these two conditions. However, there was no statistical difference in individual measures of periodontitis, such as PD or CAL, between SLE cases and controls.

We found no significant difference in oral PI or BOP between SLE cases and controls. Plaque induced periodontitis is the most common form of periodontitis, and so a high oral PI would suggest

#### Table 3 | Demographic of cohorts.


*DM, data missing.*

*a Use prednisolone or immunosuppression.*

Table 4 | Definition of periodontitis and measures of periodontitis.


*PI, oral plaque index; BOP, bleeding on probing; GI, gingival index; MT, residual teeth; PD, probing depth; CAP, clinical attachment loss; DMFT, decayed-missing-filled index; CAL, clinical attachment loss.*

Figure 2 | Forest-plot representing risk ratio of periodontitis between cases with systemic lupus erythematosus (SLE) and healthy controls.

a greater risk of developing periodontitis. BOP is a measure of gingivitis, the reversible gingival inflammation which precedes periodontitis. Therefore, although these measures do highlight potential risk of periodontitis, the absence of a significant difference between cases and controls does not preclude an association between SLE and periodontitis.

There was also no difference in means of PD. PD calculates the depth of periodontal pockets and is a measure of current periodontal disease. The British Society of Periodontology delineates a healthy sulcus as <3.5 mm and a periodontal pocket as ≥3.5 mm (24). We propose that calculating differences in mean PD < 3.5 mm is, therefore, not clinically significant and risk ratio of PD above a certain level is more appropriate. In this case, the finding that PD ≥ 5 mm was higher in SLE potentially supports the evidence that periodontitis is associated with SLE. However, as this was not significant further studies with larger sample size will be needed to elucidate this further.

Clinical attachment loss is the other gold standard measurement for periodontitis (25) and is representative of cumulative destruction. Although CAL ≥ 2 mm and mean difference in CAL was higher in SLE compared to controls, this not significantly different. Again, further studies are needed to investigate this relationship.

The finding that periodontitis is associated with SLE is in agreement with other studies examining this relationship. Case studies since the 1980s have suggested a link between SLE and gum disease. Rhodus and Johnson found 93.8% of SLE patients had periodontitis (26), while a Japanese study reported a

#### Table 5 | Results from meta-analysis.


*a Mean difference analysis.*

*CI, confidence intervals.*

prevalence of periodontitis of 70% in SLE compared to 30% in the general population (27). Higher disease activity, measured by SLE Disease Activity Index (SLEDAI) (28), predicts worse periodontal disease (24, 29) and non-surgical treatment of periodontitis improves SLEDAI scores at 3 months (30). These findings suggest that oral dysbiosis may contribute to maintenance of the inflammatory process in SLE. Recently, differences in the composition of the oral microbiota, independent of periodontal status, have been elucidated using 16s ribosome sequencing in 52 cases of SLE (23). Interestingly, the periodontal pathogen *Aggregatibacter actinomycetemcomitans*, which has been identified as a potential trigger in RA (10), has also been implicated in SLE (31).

Systemic lupus erythematosus is thought to occur when an environmental stimulus triggers inflammation in a genetically primed individual. Initial studies have highlighted possible mechanisms to explain the potential association between periodontitis and SLE. Genetic variants in the Fcy receptor have been implicated in susceptibility to both SLE and periodontitis in a small Japanese study (32). Periodontitis and SLE are both inflammatory conditions, and share similar inflammatory profiles (33, 34). Specifically, a possible role for TLR-4 has been implicated. These molecules are activated by specific pathogenassociated molecular patterns produced by bacteria and stimulation of TLR-4 leads to autoimmune lupus in mice (35). However, it must be emphasized that these studies are small and much more work is needed to elucidate if there is any biological plausibility.

#### Study Limitations

A significant limitation to our meta-analysis is the lack of clear and recognized criteria to define periodontitis. Although all definitions of periodontitis used PD and CAL, they did vary in their thresholds, limiting the ability to directly compare outcomes. However, a minimum diagnostic threshold of CAL ≥ 2 mm and PD ≥ 3 mm has been suggested (25) and all studies used thresholds which surpassed these cutoffs.

Another significant limitation is the paucity of data available. Only eight studies could be included in the meta-analysis. The overall periodontitis risk ratio included only four studies, and one of these was from conference proceedings. Meta-analysis involving greater number of studies analyzing individual measures of periodontitis were not significant. This may suggest that the significant risk ratio in overall periodontitis cannot be extrapolated into larger study populations. However, previous similar meta-analysis in other conditions, which have included larger numbers of studies, have also found that overall periodontitis risk is significant, while individual markers are not (7).

There are only a small number of studies to date investigating this association and so we were unable to test for publication bias. We included data from conference proceedings try to mitigate this but it will be important in the future to ascertain the risk of small sample size publication bias.

There are a number of factors such as smoking, educational level, and immunosuppressant medications which increased risk of both SLE and periodontitis (1). Variation in the prevalence of these factors between cases and controls could, in part, be responsible for the differences seen. These factors were partly controlled for by the use of exclusion criteria in some studies. However, there was limited detailed information regarding smoking status and immunosuppressant use in most studies, and information that was available showed variations in immunosuppressant use from 36 to 100%.

Finally, all studies included were cross-sectional in nature and investigated an association at a given time point. We are, therefore, unable to make any conclusions regarding causality from this study. Further longitudinal studies are needed to delineate a temporal association and causality of periodontitis in the development of SLE.

## CONCLUSION

The results of this meta-analysis show a significant association between SLE and periodontitis. However, the meta-analysis was hampered by paucity of data and significant limitations. Therefore, these findings can only suggest a possible association and larger longitudinal studies are needed to confirm this association and investigate causality of periodontitis in SLE.

## AUTHOR CONTRIBUTIONS

ZR-l, TS, and NS conceived, analyzed, and drafted the manuscript. IG conducted literature searches and reviewed the manuscript. All authors approved the manuscript for publication.

#### FUNDING

This work was supported by an NIHR Academic Clinical Fellowship Award to ZR-L.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://www.frontiersin.org/article/10.3389/fimmu.2017.01295/ full#supplementary-material.

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Rutter-Locher, Smith, Giles and Sofat. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Microbes and Viruses Are Bugging the Gut in Celiac Disease. Are They Friends or Foes?

Aaron Lerner 1, 2 \*, Marina Arleevskaya<sup>3</sup> , Andreas Schmiedl <sup>2</sup> and Torsten Matthias <sup>2</sup>

<sup>1</sup> The Ruth and Bruce Rappaport Faculty of Medicine, Technion Israel Institute of Technology, Haifa, Israel, <sup>2</sup> Department of Research, AESKU.KIPP Institute, Wendelsheim, Germany, <sup>3</sup> Central Research Laboratory, Kazan State Medical Academy Kazan, Kazan, Russia

The links between microorganisms/viruses and autoimmunity are complex and multidirectional. A huge number of studies demonstrated the triggering impact of microbes and viruses as the major environmental factors on the autoimmune and inflammatory diseases. However, growing evidences suggest that infectious agents can also play a protective role or even abrogate these processes. This protective crosstalk between microbes/viruses and us might represent a mutual beneficial equilibrium relationship between two cohabiting ecosystems. The protective pathways might involve post-translational modification of proteins, decreased intestinal permeability, Th1 to Th2 immune shift, induction of apoptosis, auto-aggressive cells relocation from the target organ, immunosuppressive extracellular vesicles and down regulation of auto-reactive cells by the microbial derived proteins. Our analysis demonstrates that the interaction of the microorganisms/viruses and celiac disease (CD) is always a set of multidirectional processes. A deeper inquiry into the CD interplay with Herpes viruses and Helicobacter pylori demonstrates that the role of these infections, suggested to be potential CD protectors, is not as controversial as for the other infectious agents. The outcome of these interactions might be due to a balance between these multidirectional processes.

#### Edited by:

Kuldeep Dhama, Indian Veterinary Research Institute (IVRI), India

#### Reviewed by:

Mario M. D'Elios, University of Florence, Italy Maryam Dadar, Razi Vaccine and Serum Research Institute, Iran

\*Correspondence:

Aaron Lerner aaronlerner1948@gmail.com

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 15 May 2017 Accepted: 10 July 2017 Published: 02 August 2017

#### Citation:

Lerner A, Arleevskaya M, Schmiedl A and Matthias T (2017) Microbes and Viruses Are Bugging the Gut in Celiac Disease. Are They Friends or Foes? Front. Microbiol. 8:1392. doi: 10.3389/fmicb.2017.01392 Keywords: celiac disease, bacteria, viruses, gut, microbiome, environmental inducer, environmental protectors

## INTRODUCTION

#### Infection and Autoimmunity

The relationship between infections and autoimmunity is complex. Microbial and viral infections might act as environmental triggers inducing or propagating autoimmune and inflammatory processes, resulting in symptomatic presentation of a disease in genetically high risk individuals (Lerner, 2015; Arleevskaya et al., 2017). An autoimmune disease onset following an infectious agent exposure has been well-documented (Pordeus et al., 2008; Bogdanos et al., 2015; Sakkas and Bogdanos, 2016). At least, for CD, the following infections were suggested to be associated with the disease: viruses: enterovirus, Epstein-Barr virus (EBV), Cytomegalovirus (CMV), hepatitis C virus (HCV), hepatitis B virus (HBV), and rotavirus, microbes: Bacteroides species, Campylobacter jejuni, Pneumococcus, Mycobacterium tuberculosis, and Helicobacter pylori (Lerner, 2015). However, recent serological evidence suggests the opposite outcome, which is the protection against autoimmune conditions following bacterial/viral exposure (Christen and von Herrath, 2005). At least the suggestive protector agents for CD were CMV, EBV, Rubella, and Herpes simplex type 1 virus (HSV1) when compared to healthy people (Plot and Amital, 2009; Jansen et al., 2016). According to the "hygiene hypothesis," the excessively sterile environment leads to the enhanced incidence of autoimmune disorders, asthma, and allergies, thus, associating surge of CD incidence with decreased infectious environment (Lerner et al., 2015b,e; Bloomfield et al., 2016).

#### Celiac Disease

Celiac disease is a life-long autoimmune disease (Lerner et al., 1996) mainly of the proximal intestine, affecting genetically predisposed individuals. Gluten, the storage protein of wheat, is the environmental inducer of the disease in addition to other structurally related molecules found in barley, rye, and oat (Lerner, 2014). Many environmental factors were suggested to induce or enhance the disease: multiple infections (Lerner and Reif, 2015), early infections (Myléus et al., 2012), early gastrointestinal infections (Beyerlein et al., 2017), lack of breast feeding (Lerner and Matthias, 2016c), time and amount of gluten consumption (Chmielewska et al., 2015), microbiome/dysbiome repertoire (Lerner et al., 2015a; Lerner and Matthias, 2017a,b), mode of delivery (Decker et al., 2011), early vaccination (Kemppainen et al., 2017) or early consumption of antibiotics (Canova et al., 2014) and geo-epidemiological influences (Lerner, 1994, 2015; Reif and Lerner, 2004b; Lerner and Matthias, 2015a). The abnormal immune response is directed, in particular, against tissue transglutaminase (tTG), representing the autoantigen, (Reif and Lerner, 2004a; Lerner et al., 2015c) and the two main autoantibodies, anti-endomysium and anti-tTG antibodies, are the most prevalent serological markers used to screen for the condition (Shamir et al., 2002; Lerner and Matthias, 2015d). Recently, the list of CD serological markers was expanded by two additional autoantibodies: anti-deamidated gliadin peptide and anti- tTG neo-epitope antibodies, found to be reliable for CD diagnosis (Rozenberg et al., 2012; Lerner and Blank, 2014; Lerner et al., 2015e). As yet, HLA-DQ2 and HLA-DQ8 are known predisposing genetic factors. The sequential events in disease progression were unraveled in the last years and gave rise to multiple future therapeutic strategies (Lerner, 2010). Notably, its epidemiological, incidental, and clinical presentation are changing continuously, and new clinical pictures are reported and expand the abundance of clinical variance of the disease (Lerner et al., 2015d). In fact, age of disease onset increases and the traditional enteric presentation is more and more replaced by extraintestinal manifestations. Skin (Lerner et al., 2015d), endocrine (Lerner and Matthias, 2016d; Lerner et al., 2017), hepatic (Anania et al., 2015), metabolic (Eliyah Livshits et al., 2017), skeletal (Lerner and Matthias, 2016a), rheumatic (Lerner and Matthias, 2015b), geriatric (Lerner and Matthias, 2015c), hematological (Branski et al., 1992), neurological (Zelnik et al., 2004; Lerner et al., 2012), gynecological and infertility (Mårild et al., 2012; Casella et al., 2016), oral and dental (Cantekin et al., 2015), hypercoagulability (Lerner and Blank, 2014), cardiac (Lerner et al., 2015d), and behavioral abnormalities (Zelnik et al., 2004) are often described. Those epidemiological and clinical changes can explain why the disease is diagnosed during the whole human life-span including in the elderly (Lerner and Matthias, 2015c). There is no doubt that in the last decades its incidence is constantly increasing, ranging between 1 and 3% nowadays (Lerner, 2014; Lerner et al., 2015b). The present review will concentrate, expand and update on the multiple faces of the inductive/protective roles that infectious agents might play in CD pathogenesis. This aspect is further interesting since pathogens are the major drivers of human selective genetic adaptation during evolution (Vatsiou et al., 2016), and the question of microbes that are bugging the celiac patient "are they friends or foes?" is the subject of the current review.

## Infections and CD

It should be clarified that, although the trigger role of microorganisms and viruses in the CD development was undoubtedly traced in numerous investigations, it substantially differs from other immune pathogenesis like rheumatoid arthritis (as a classic model of an autoimmune disease; Arleevskaya et al., 2016; Kemppainen et al., 2017).

The induction of rheumatoid arthritis most likely occurs under the influence of the burden of many trivial infections, influencing the patient's immune system due to the frequent and prolonged infectious episodes (Arleevskaya et al., 2014). Individuals at CD risk apparently do not have such features in their mucosal immunity, nor significant defects in systemic antiinfective protection, impacting infection susceptibility. Since all the CD studies are focused only on the disease link with various gastrointestinal infections, such association is different from what was shown in rheumatoid arthritis (Riddle et al., 2012).

The number of infectious agents related to CD is continuously increasing (**Figure 1**). Examples for viruses enterovirus, EBV, Cytomegalovirus (CMV), HCV, HBV, and rotavirus And for

microbes Bacteroides species, C. jejuni, Pneumococcus, M. tuberculosis, and H. pylori (Lerner, 2015). However, links between CD and infections were more associative and less causative, thus, far from being elucidated. Moreover, the mutually exclusive hypotheses about the provocative and protective role of a particular microorganism/virus in CD pathogenesis were suggested and discussed in various publications.

It appears that the essential condition for the CD induction is a gastrointestinal infection (Kemppainen et al., 2017). Apparently, the other major condition is an early childhood—namely immature gastrointestinal tract, immature immune system, and gastrointestinal microbiome at the early phase of formation. The beginning of gut colonization by microorganisms set the stage for the cross talks between the epithelium, enteric lymphoid tissue, and microflora, all together establishing the intestinal barrier and as a consequence, a strictly dosed delivery of macromolecules into the internal environment and shaping of mucosal tolerance to food antigens and normal flora (Makarova et al., 2014).

So, in the vulnerable infant period any gastrointestinal infection, even a transitory one, is potentially able to disturb these processes of gut microbiome maturation and the establishment of local immunity and immune tolerance, including that against food and microbial antigens. Apparently, deleterious coincidence of these circumstances leads to an error in the negative selection of gluten-reactive lymphocyte clones.

Thus, the community of microorganisms, being extremely vulnerable during the ripening period, appears to be an inert system in adults. For example, in adults it needs no more than 30– 60 days to restore gut microflora after the exposure to antibiotics (Spanhaak et al., 1998; Tannock et al., 2000). At the same time, a weekly clindamycin treatment of a newborn reduces Bacteroides diversity for the next 2 years (Jernberg et al., 2007). The same infections, capable of altering the fate of a sick infant, are merely a light ripples on the ocean surface for an adult microbiome.

In conclusion, there is an undoubted link between CD development and microorganisms, and this link looks to be rather specific. Gastrointestinal infections in predisposed infants with an immature gastrointestinal tract and immune system might shape gut microbiome in the immature and therefore labile circumstances. Such an unfortunate combination could trigger early CD development or becomes a ticking time bomb, represented by the structural features of the gut microbiome and persistence of gluten–reactive lymphocyte clones with latent basal cell proliferation without overt disruptive inflammatory activation.

#### GUT MICROBIOME SIGNATURE IN CELIAC DISEASE

Gut microbiome analysis in the healthy adult human populations revealed about 1150 bacterial species, the majority (50–75%) being represented by Firmicutes, and then Bacteroidetes (10– 50%), Actinobacteria (1–10%), with fewer than 1% being Proteobacteria (Manichanh et al., 2012). Apparently, the HLA system, to a certain extent, shapes microbiome structure. Besides, polymorphisms of some other non-HLA genes were found to correlate with a certain microbiome structure (Spor et al., 2011). Interestingly, in addition to the HLA system, microbiome composition may be due to CD-associated polymorphisms of defensin, some molecules of Toll-like receptor signaling pathways and vitamin D receptor genes (San-Pedro et al., 2005; Fernandez-Jimenez et al., 2010; Wang et al., 2016). Whole genome study of 93 individuals and 16S rRNA gene pyrosequencing of their body microflora revealed 83 alliances between genetic variance in host sequence and plethora of specific microbial taxa (Blekhman et al., 2015). In particular, the links with CD-associated host genes were revealed. In addition to the host genes related to immunity, a link was found between the microbiome composition and SNP of the genes not related to immunity. For example, the authors revealed an interesting correlation between the abundance of Bifidobacterium in the gastrointestinal tract and host genetic variation in LCT gene, encoding the lactase enzyme hydrolyzing dietary lactose. This gene SNPs are known to be associated with lactose intolerance, which is frequently associated with celiac disease (Ojetti et al., 2005). Bifidobacterium is able to metabolize lactose, and there are some strains preferring lactose instead of glucose. The authors suggested that the problems with individual's consumption of milk products might impact the richness of Bifidobacterium in the gastrointestinal tract.

The results of gut microbiome structure investigations in the infants at CD risk as well as in the therapy-naïve patients at the disease onset are somewhat contradictory. Besides the bulk of the results was obtained by study of feces, while the principal for CD microbe community in the small intestine boundary layers has its own peculiarities, although in a certain extent it is associated with fecal microbes. However, a certain tendency can be traced. A single and rather limited study of mucosa-associated microbiota in the proximal gut—using enteric samples from 45 children with CD and 18 clinical controls born during the "Swedish CD epidemic"—demonstrated only marginally differences between the groups. Enrichment with Clostridium, Prevotella, and Actinomyces was revealed in the most of the CD samples (Ou et al., 2009). Feces studies demonstrated, that infants genetically predisposed to CD, had significantly higher abundance of Firmicutes (Clostridium) and Proteobacteria (Escherichia/Shigella) and desreased proportions of Actinobacteria (Bifidobacterium) and Bacteroidetes compared to low-risk infants (Sellitto et al., 2012; Olivares et al., 2015). In a proof of concept study, Sellitto and co-workers have traced the longitudinal changes in the microbial populations colonizing from birth to 24 months in 30 genetically predisposed infants. They demonstrated that even at 2 years of age their microbiota do not resemble that of adults, while in not at-risk infants the maturation was complete at 1 year of age. The group was divided into an early and a late gluten exposure groups (17 and 13 infants, respectively). The authors showed that genetically susceptible infants may benefit from delayed gluten exposure not before 12 months of age. A hypothesis was forwarded that lack of maturity of the enteric microbiota faced with early gluten consumption can induce or accelerate the autoimmunogenetic process. Not less interesting was the infant's metabolome. When solid food was introduced at 6 months of age, succinate, acetate, propionate, and butyrate accumulated in their stools. However, by 2 years of age, butyrate, and acetate were the dominant short-chain fatty acids (SCFA). Since bacteroidetes are associated most strongly with propionate, while Firmicuters are negatively correlated to SCFAs production (Koenig et al., 2011), Sellitto's group envisioned that the high Firmicutes and low Bactriodetes abundance in CD infants results in down production of those protective SCFAs, thereby abrogating enteric health and predisposing them to autoimmune diseases. The decrease in Lactobacillus spp. associated with lower lactate production, observed in between 6 and 12 months of age, accompanied by decreased SCFAs' feces repertoire, during a vulnerable time of mucosal immune maturation and microbiome compositional changes might lead to loss of tolerance to non-self-antigens like gluten in those CD genetically predisposed infants (Sellitto et al., 2012). Once again it points to the important relations connecting nutrition to microbiome composition and diversity, its metabolome and the local maturation and functioning of the immune system.

From this perspective, it is interesting to compare the composition of the gut flora in infants at CD risk and premature babies with a priori immature gut. Arboleya and coauthors compared the gradual establishment of the intestinal microbiome in very-low birth-weight preterm infants with that of healthy full-term, vaginally born, breast-fed neonates using 16S rRNA gene profiling and quantitative PCR for the various microbial taxa. It was demonstrated that preterm neonates sheltered a higher relative proportion of Firmicutes at 2 days of age, and of Proteobacteria in the later sampling times, compared to control babies. Prematurity reflected reduced levels of Bacteroidetes at day 2 and as well as in later sampling times together with Actinobacteria (Arboleya et al., 2016). In addition, verylow birth-weight preterm infants frequently displayed a lag in establishing an adult microbiota compared to full-term children (Weng and Walker, 2013). So, the parallels to the peculiarities of the gut flora in CD prone individuals seem to be obvious. Combining the results of Arboleya et al. and Sellitto et al. about the enteric mucosal immaturity and the unbalanced microbiome found early in life and in CD high-risk infants, it can be suggested that by ingesting gluten peptides, the disease will progress in these individuals, unlike in non-CD high-risk premature infants, that will never develop CD. The features of CD gut microbiome, incorporated in early childhood seem to persist in the adulthood even despite gluten withdrawal (Nistal et al., 2012; Wacklin et al., 2014).

Caminero and co-workers demonstrated that gluten amounts in feces of healthy volunteers, CD patients and individuals under risk receiving gluten-free or normal diet depended on gluten intake. The greatest amount of gluten was found in fecal samples from healthy volunteers being on normal diet, with a significant decrease in the untreated CD patients and the individuals under risk. It is noteworthy that in all the groups fecal peptidase activity against the gluten-derived peptide 33 mer inversely correlated with gluten amount in the samples (Caminero et al., 2015). It looks like the increased functional proteolytic activity of gut microflora in CD patients can affect gluten excretion. The same research group isolated 144 strains belonging to 35 microbial species that might be involved in gluten digestion in the human intestine. Most of the strains were part of the phyla Firmicutes and Actinobacteria, mainly from the genera Lactobacillus, Streptococcus, Staphylococcus, Clostridium and Bifidobacterium (Caminero et al., 2014). Ninety-four of these strains were capable to metabolize gluten, 61 of them showed an extracellular proteolytic activity of gluten proteins, and several strains showed a peptidase activity toward the "supra-molecule" 33-mer peptide, the luminal immunogenic molecule in CD patients (Caminero et al., 2014). At the end of the day, there is a certain difference in gluten proteolysis by various bacteria, and the immunogenicity of the generated peptide fragments might be different. It should be noted that these studies were carried out in cultures, in which the glutenase activity of aggregate microbial community as a whole (biofilm) in the gut boundary layer might significantly differ from the isolated activity of the individual members of this community.

It is known that the microbiome impact gastrointestinal and systemic functions by its metabolome, the most studied one being the short chain fatty acids (SCFA). Most recently, the topic of nutrition, microbiome, and SCFA associations in CD was updated (Lerner et al., 2016). Multiple beneficial effects to the host were attributed to them. Changes in microbiota and their SCFA production is clearly related to the pathogenesis of CD. Interestingly, peculiar dysbiosis and significant changes in stool SCFA profile were described in several autoimmune diseases, one of those is in Behcet's disease where decreased butyrate production was suggested to play a role in its pathogenesis (Consolandi et al., 2015).

Taken together, the microbiota/dysbiota disbalance may present a risk factor for CD either directly by influencing the mucosal immune responses or by intensifying inflammatory responses to gluten. In contrary, several microbial species are capable to break down gliadin and perhaps therefore decrease the immunopathogenicity of consumed gliadin (Sjöberg et al., 2013; Carding et al., 2015; Lerner and Matthias, 2015a; Rostami-Nejad et al., 2015).

#### COEXISTENCE OF CERTAIN INFECTIONS AND CD: A CRITICAL LOOK AT THE ISSUE

It is of interest how CD affected individuals survived in the presence of harmful conditions (increased gluten content and toxicity in wheat, increased gluten consumption worldwide; Lerner et al., 2017) and despite them, thrived and expanded? (Lerner, 2011; Lerner et al., 2015b). Despite being underprivileged with nutritional deficiencies, failure to thrive and high morbidity and mortality, a substantial increase in the disease incidence is observed in the last decades. There are several theories explaining this paradox. One is positive evolutionary selection, in which the celiac patient accumulates protective genes (Lerner, 2011). The other is due to some pathogens which are the major drivers of human selective genetic adaptation (Vatsiou et al., 2016) that could have been beneficial environmental factors, protecting the CD populations. Indeed, in contrast to the observations that infections may induce CD, some infection agents were assumed to have a protective impact (Christen and von Herrath, 2005; Gaisford and Cooke, 2009; Kivity et al., 2009; Plot et al., 2009).

This assumption was based in particular on a lower incidence of the serum antibodies against Cytomegalovirus (CMV), EBV, and Rubella in CD patients when compared to healthy people (Plot and Amital, 2009). Another argument for the protective effect is the inverse correlation between serum anti-CMV, -EBV, and/or -Herpes simplex type 1 virus (HSV1) IgG levels and anti-tTG antibodies (Jansen et al., 2016).

#### Viruses

It should be specified that Jansen and coauthors determined the antiviral antibody levels in the sera samples of 6-year old children (Jansen et al., 2016). CMV single infection and combined CMV, EBV, and/or herpes simplex virus type 1 infection antibodies were inversely associated with strongly tTg-IgA positivity. The authors suggested that the serological profile may indicate a protective effect of herpesvirus infections in the pathogenesis of celiac disease autoimmunity. It is accepted that by this age IgG and IgM production is close to that of the adults. However, the immune system is still immature and its reaction is not fully functional at this age even in healthy children, whereas the immune system of CD prone children, at least, the local immune defense, is somewhat delayed in the maturation. Thus, Jansen et al. suggested explanation should be taken with a grain of salt.

In Plot's publication, the age profile of the studied cohorts was not presented (Plot and Amital, 2009). There are some curious details in this publication which need to be interpreted. First, while the prevalence of the anti-EBV capsid antigen and anti-EBV nuclear antigen IgG in the CD patients was significantly lower than that in the controls, the prevalence of the anti-EBV early antigen IgG was comparable in the two groups and the prevalence of anti-EBV capsid antigen IgM, though unreliably, was more than twice higher in the CD group. In general, the anti-EBV capsid antigen IgMs are known to be produced during the acute phase (the first days—6 months from the onset of the disease) or during acute exacerbation of chronic EBV infection (http://www.tiensmed.ru/news/epstein-barr-bc1.html#nov1). In addition, if the anti-EBV early antigen IgG is not concurrently revealed, it may indicate incubation period or the very beginning of the infection (up to 1 week of symptoms). So, either in the CD patients this generally latent infection is apt to more frequent exacerbations, or, given the reduced frequency of the studied IgG antiviral antibodies, the existence of some features of the antiviral antibody production in CD can be assumed. Secondly, revealed by PCR, CMV DNA presence in the samples is not always accompanied by the presence of serum specific antibodies. This situation is typical in particular for the infants with immature immune system (Dong et al., 2004, 2005).

A number of publications on the HBV vaccine nonresponsiveness in CD might attest for possible unique features of the antiviral antibody response formation (Noh et al., 2003; Park et al., 2007; Urganci and Kalyoncu, 2013), further criticizing the assumption that decrease antibodies activities in CD patients represent a protective effect.

As for the inverse correlation of the antiviral and antitTG antibodies demonstrated by Jansen et al. (2016), similar patterns are not uncommon in autoimmune diseases, and this fact by no means demonstrates a lesser exposure to the infections. For example, high levels of autoantibodies against double-stranded DNA, reflecting the activity and severity of systemic lupus erythematosus, are quite often combined with lower levels of antibodies to particular bacterial DNA (Pisetsky and Drayton, 1997). The inverse correlation of those indexes is usually explained by a distorted immune humoral response.

Additionally, it should be specified that, in general, in Herpesvirus infections and viral hepatitis the specific antibodies are not the principal players in the antiviral defense but are accepted to be the reliable serological markers for an infection (Grinde, 2013). It is advisable to note that caution should be used for the interpretation of the protective role of these infections in CD, relying only on data on the incidence of the corresponding antibodies.

As for Rubella infection, the specific antibodies are the major antiviral response players. However, the incidence of CD in children vaccinated with inactivated rubella virus as part of polio vaccine was close to that in the unvaccinated children (Myléus et al., 2012). Thus, a reduced incidence of the anti-rubella IgG antibodies demonstrated by Plot and coworkers can mean an equally probable lower exposure of CD patients to Rubella and the above-discussed features of the specific IgG antibody production. Moreover, since the antiviral immune response is always multi-componental, the disturbed antibody formation, although being a weak link of an antiviral defense, does not necessarily entail an increased susceptibility to Rubella.

The data which might testify for the direct and reverse links of CD to Herpes virus infection are summarized in **Table 1**. The analysis of the data demonstrates that the links of CD and herpes infections are multi-directional. On one hand there are some peculiarities, which can promote the viral infecting: (1) CD-associated DQA1<sup>∗</sup> 0501/DQB1<sup>∗</sup> 0201 genotype, which is also due to the imperfect response against Herpes viral infection; (2) the typical CD immature gastro—intestinal tract and the delayed process of microbiome maturation, which might be risk factors for the virus infecting; (3) the typical CD mucosal overexpression of epidermal growth factor receptors, by which Herpes viruses enter the cells; (4) increased expression of IL-33, suppressing local antiviral immunity in CD patients. On the other hand, the increased levels of several humoral factors with the antiviral activity; increased expression of some cytokines, which promote mucosa maturation and thus increase its' resistance to the viruses; as well as some potential features of CD microbiome that might indicate a backward link between CD and Herpes viral infection. The protective effect of the infection on atopic manifestations was demonstrated in the case of the early (infancy or early childhood) EBV exposure, while the later infection predisposes to the atopic disease (Nilsson et al., 2005, 2009). So, if to extrapolate the data on the links of herpes and atopic diseases to CD, it is likely that the early (infancy or early childhood) EBV exposure might play a protective role, while the later infection might trigger CD or have no impact on it at all.

#### TABLE 1 | Direct (⇑⇑) and reverse (⇑⇓) links of CD and Herpes virus infections.

## ⇑⇑ ⇑⇓

Markedly impaired binding and presentation of some herpes antigens to the TCRs in CD-associated DQA1\*0501/DQB1\*0201 carriers (Koelle et al., 1997; Reichstetter et al., 1999) might be due to the imperfect antiviral immune response as well as to the peculiarities of antibody production.

Increased HSV2, CMV and EBV DNA levels in the stool samples were observed among premature neonates with intrauterine growth restriction compared with those infants born appropriate for gestational age (Naing et al., 2013). The immature gastro—intestinal tract and the delayed process of microbiome maturation might be the typical signs of CD (Sellitto et al., 2012), that might be risk factors for the virus infecting in infancy as well as for the higher rates of the clinical manifestations of the infection.

Herpes viruses were revealed in the inflamed gastrointestinal tract mucosa, but never in the endoscopically healthy tissue (Ramanathan et al., 2000; Roblin et al., 2011). It is unclear whether the inflamed mucosa is a consequence of the viral infection or the inflamed tissues "draw" viruses, due to the expressing of the corresponding receptors. Epidermal growth factor receptors, by which Herpes viruses enter the cells are overexpressed in CD gut mucosa, that being due to gliadin stimulatory effect (Barone et al., 2007; Juuti-Uusitalo et al., 2009).

Microbiome features might impact antiviral immunity via stimulation of IL-33 (alarmin) released by mucosal epithelium, which suppresses local antiviral immunity by blocking the migration of effector T cells to mucosa, thereby inhibiting the production of IFN-γ, a critical cytokine for antiviral defense, at local infection sites (Oh et al., 2016). Serum levels and intestinal tissue expression of IL-33 and its receptor in CD patients were found to be increased (López-Casado et al., 2015).

–

Virus-shaped cytokine levels might to some extent promote the maturation of the local immune system and intestinal tissues, lagging behind in the CD-prone individuals: IL-6 promotes enterocyte differentiation and inhibits enterocyte apoptosis, TNF-alpha promotes intestinal growth (Rollwagen et al., 1998; Maheshwari, 2004), IFN-gamma increases macromolecular transport in the immature gut particularly across Peyer's patches. This Peyer's patch-targeted effect can be important for setting mucosal immune responses against dietary antigens early in life and aiding their immune exclusion (Sütas et al., 1997).

In CD the level of various humoral factors with a pronounced diverse direct and indirect antiviral activity in the inflamed intestinal tissues are increased (defensins, IFN-gamma, IFN-alpha, TNF-alpha, IL-6, IL-15 the latter being necessary for the development and function of NK/NKT cells and maintenance of naive and memory CD8(+) T cells; (Forsberg et al., 2004; Hazrati et al., 2006; Di Sabatino et al., 2007; Brottveit et al., 2013; Meresse et al., 2015)), that might have a protective effect on the infection.

Pre-treatment of HeLa monolayer with inactivated Staphylococcus aureus cells before HSV infection increases expression of TNF-a, IL-6, and IL-8 genes, that being due to the protection from the occurrence of virus mediated cytopathic effect and to decrease of viral multiplication rate (Bleotu et al., 2015). Vaginal Lactobacillus strains neutralize lactic acid and thus, acidic pH values needed for the viral replication, as well as to macrophage activation (Conti et al., 2009; Khani et al., 2012). Microbiota might impact the antiviral defense via regulation of the natural killer T cells at the frontiers of the mucosal immune system (Zeissig and Blumberg, 2014). So, certain shifts in the structure of the microbiome mighty inhibit viral infection.

#### Helicobacter pylori (Hp)

The permanent interest in CD and Hp infection coexistence is quite natural, due to the gut-stomach axis (Lerner and Matthias, 2016b). The infectious inflammatory process directly in the gastrointestinal tract—CD epicenter, which might shape the local immune system and microbiota, might obviously play a role in CD pathogenesis. In addition, both CD and Hp infection in a number of cases are associated with the diffuse lymphocytic gastroenteropathy (Lynch et al., 1995; Broide et al., 2007; Pai, 2014). However, diffuse lymphocytic gastroenteropathy is far from being obligatory attributed only to both entities (Wu and Hamilton, 1999; Nielsen et al., 2014). Besides, lymphocytic gastritis and a subsequently villous atrophy are accepted to be a non-specific manifestation of many pathological conditions in the gastrointestinal tract, due to a wide variety of infectious, immunologic or any inflammatory stimuli raising intraepithelial lymphocyte numbers. Lymphocytic duodenitis and increased intraepithelial lymphocytosis are known to be associated with diseases that are completely different in their pathogenesis, such as autoimmune disorders like CD (Broide et al., 2007; Rostami et al., 2010), tropical sprue, food protein intolerance, Hp-induced duodenitis, peptic duodenitis, parasitic, and viral infections, intestinal lymphoma (Chang et al., 2005; Brown et al., 2006; Pallav et al., 2012; Rosinach et al., 2012; Shmidt et al., 2014) drugs' induced duodenitis (non-steroidal anti-inflammatory drug; Shmidt et al., 2014) and small-intestine bacterial overgrowth (Lappinga et al., 2010). The differential diagnosis of lymphocytic gastritis is not less restricted. CD and HP are not the only ones. Various non-HP infections, inflammatory conditions and several non-celiac autoimmune diseases were described (Broide et al., 2007; Polydorides, 2014).

The interest in the CD—HP infection link is fueled by the well-known data, indicating that childhood infection with HP could protect against the development of Crohn's disease, severe gastric-reflux disease, Barrett's esophagus and esophagus adenocarcinoma (Chen and Blaser, 2008). Yet, as for the protective role of Hp in CD, the available data are limited and quite contradictory. The prevalence of CD among Hp-positive adults was 0.05% compared with 0.09% among Hp-negative individuals (statistically non-significant) while the prevalence of Crohn's disease among Hp–positive patients was 0.07% compared with 0.24% among Hp-negative patients (Bartels et al., 2016). Based on these data, at least in adults, the protective effect of Hp on CD is minimal, if at all existing. However, given the fact that in childhood gastrointestinal infection appears to be a more important condition for CD triggering than in adults, this conclusion is not necessarily true in the case of early exposure to Hp (Kemppainen et al., 2017). Unfortunately, we failed to find the similar data on the CD frequency in children infected with Hp early in life.

The incidences of Hp and CD worldwide vary enormously (9–100%, 0.3–3.9%, respectively; Rostami-Nejad et al., 2016). Most of the studies aimed to determine the ratio of Hp infected individuals in CD and non-CD control groups. In our opinion, this study design gives less information about the inductive/protective effect of Hp infection on the CD development as on the susceptibility of the CD patients for the infection. The results of various studies on Hp both in children and adults are contradictory (**Table 2**). The spread in the ratios of both Hp infected CD patients and non-CD controls in these publications might be due to the national and age-related characteristics of the studied cohorts (Eusebi et al., 2014). As for the diametrically opposed regularities of HP incidence in CD and non-CD groups, a meticulously analysis of these publications, did not lead us to any reason for the conflicting results, but to the possible differences in the poorly described clinical characteristics of the controls. In these studies, the authors examined the collections of endoscopically obtained biopsies and sera allocating the cases into CD and non-CD (control) groups. It is obvious that the persons accepted as controls underwent the relevant examination because they had any gastroenterological problems associated or not associated with HP infection.

The analysis of the literature data which might testify the direct and reverse links of CD and Hp infection (**Table 3**) shows that the interaction of the two diseases represents an interweaving of differently directed processes. Perhaps the end result might depend on the balance of these processes, being deeply individual in each specific case. Important is the assumption that the direct or reverse CD/HP link may depend on the age at which the encounter with the bacterium occurred. At least, it is important for allergies–Hp links. Our attempt to test this hypothesis failed, because the literary data accumulated to the present moment are largely insufficient.

### POTENTIAL MECHANISMS FOR BENEFICIAL BUGS' EFFECTS IN CELIAC GUT

Multiple potential pathophysiological avenues were suggested to understand the microbial-gut cross-talks in CD.

#### Post-translational Modification of Protein (PTMP) from Non-self- to Self-proteins

Endogenous and microbial enzymes are capable to generate intestinal enzymatic neo-antigens via PTMP. The modifications taking place in the intestinal lumen include peptides crosslinking, de/amination/deamidation by the transglutaminases, de/phosphorylation, a/deacetylation, de/tyrosination, and many other enzymatic modifications exist (Lerner et al., 2016). Related to the present topic, the human endogenous intestinal enzymes, tTG and its family member, the exogenous microbial transglutaminases, induce multiple neo-epitopes on the TGgliadin cross-linked complex resulting in the formation of antibodies against the complexes in CD. CD is a classical disease where luminal PTMP is driving the disease. It seems logical that a microbial agent might modify non-self-peptide to self-one reducing its immunogenicity. Additionally, in CD, some microbe strains might modify gluten in the lumen, thus preventing or aggravating the inflammatory cascade and the intestinal damage progression via PTMP (Caminero et al., 2016).

## Horizontal Gene Transfer in the Human Gut Lumen

Given the extensive influence of the microbiota on human health, the gut-microbiome integrity is of prime importance for host health and survival. In this regards, our bodies' "second genome" cohabit with the human one to form a stable equilibrium for the two kingdom's long term survival. As opposed to longterm evolutionary events, newer genetic manipulations with microorganisms, plants, animals, or nutrients, applying new food technologies and/or microbial engineered delivery systems or novel mode of therapies are rapidly evolving.

Due to the close relationship and intimate cross-talks between the human and the gut's biospheres, consumption of the modified genetic cargo into the human intestinal ecosystem might occur. It was hypothesized that modern probiotic ingestion, genetically manipulated food consumption and genetically manipulated microorganism usage are potential genetic driving forces for changing the evolutionary equilibrium established during the last millions of years (Cho and Blaser, 2012). Horizontal gene exchange is the ability to transfer genetic material between contacting biological domains, including eukaryote (plants, animals, and man), prokaryote (microbes), and viruses (Aminov, 2011; Ruggiero et al., 2015). Despite not being investigated in CD, various virulent genes, the most studied one is the antibiotic resistance gene, were described to be laterally transferred. It is hypothesized that the opposite might occur. This infectious genetic cargo might include anti-inflammatory/proapoptotic/Treg or other immune-modulatory genes, attenuating or abrogating autoimmunity.

### Infections as Tight Junction Closure Enhancers

The tight junction protein, Zonulin, is involved in the regulation of the intestinal permeability between gut epithelial cells. Several clinical trials with Zonulin antagonist (Larazotide acetate, AT-1001, Alba Therapeutics, USA) demonstrated the promising therapeutic effect in CD (Lerner, 2010; Khaleghi et al., 2016). Larazotide acetate—an octa-peptide derived from a cholera toxin ZO, antagonizing zonulin via receptor blockade—is aimed to decrease the paracellular transport caused by gluten and thus to suppress the activation of the pathological immune cascade. In addition to the above mentioned cholera toxin derivative, many other factors produced by microorganisms can improve tight junction performance, modulating intestinal permeability. Salmonella enterica serovar, Escherichia coli, and C. jejuni modulated enteric epithelial barrier functions in chickens (Awad et al., 2012, 2014, 2015), and the probiotics Lactobacillus casei DN-114001 and E. coli strain Nissle 1917 decreased intestinal epithelium permeability in human intestinal originated cell lines (Parassol et al., 2005; Zyrek et al., 2007; Trebichavsky et al., 2010).


TABLE 2 | The incidence of Helicobacter pylori infection (Hp-positive, %) in CD patients and non-CD controls.

Taken together, modulation of gut permeability by infectious agent, counteracting the breached tight junction integrity in CD, might represent a protective pathway.

#### Molecular Mimicry between Infectious Agents and Self-antigens

Generally, molecular mimicry between foreign (infectious/environmental) and self-antigens is a well-described pathway of autoimmune disease induction. Recently, it was suggested that antigen mimicry between foreign and self-antigens might be due to the long-term regulation of inflammation (Pontes-de-Carvalho et al., 2013). In a cohort of African patients, infected with Schistosoma it was found that the parasite inhibited production of anti-nuclear antibodies (Mutapi et al., 2011). More recently, apparent effectiveness of rotavirus vaccination was found to prevent the onset of CD autoimmunity (Silvester and Leffler, 2016). This finding gives indirect evidence for the persistence of regulatory cells in the lack of stimulation of the immune system by pathogen-derived processes. The strength of the immune adjustment, however, may increase with the uninterrupted presence of the pathogen or its antigens. Thus, molecular mimicry can represent a protective mechanism of autoimmunity.

#### Th2 to Th1 Shift

Inflammation, given rise by microbes, viruses and especially by parasites such as helminths, can shift the Th1 pathway to Th2 one, resulting in a more immunosuppressive state where regulatory T cells might be induced or be activated (Shor et al., 2013). Recent studies have supplied such an evidence for pathogen-specific regulatory cells in Leishmania major, Herpes simplex virus, and Friend retrovirus (murine leukemia virus) infections (Christen and von Herrath, 2005). During the acute phase of the infection this Th2 profile counter-regulates Th1 driven autoimmune pathologies. Along the chronic stage of infection, immune-regulatory networks arise, mainly led by regulatory T cells. These cells produce IL-10 and TGFβ, which has observative effect on Th1-related autoimmune diseases, such type 1 diabetes mellitus or CD. In fact, several helminths were tried in CD patients with encouraging results (Croese et al., 2015). Treatment with helminthes or helminthes ova ameliorated the clinical pictures of several autoimmune conditions in patients as well as in animal models (Smallwood et al., 2017). A major recent contribution to the field is the helminth phosphorylcholine proved to be an immunemodulatory molecule. Most recently, tuftsin-phosphorylcholine, a novel helminth-based compound was shown to reduce pro-inflammatory cytokine production and induced anti-inflammatory cytokine expression and Treg and Breg cell expansion in mouse models of rheumatoid arthritis, lupus nephritis, and colitis (Bashi et al., 2015b, 2016; Shor et al., 2015).

It is conceivable, that the ability of helminthic parasites to attenuate host immune responses into an antiinflammatory/regulatory phenotype is attributed to the endogenous component that the parasites secrete and/or excrete interacting with immune effector cells to regulate their function (Lund et al., 2014; Selmi, 2016).

An additional mechanism was suggested for the helminth's immunomodulation of autoimmunity, in addition to the Th1 to Th2 shift. Accelerated T and B regulatory phenotypes, decreased levels of the inflammatory cytokines like IFNg and Il-17 or vice versa, promoting IL-4, IL-10, and TGF-β release (Bashi et al., 2015a). Since CD is a Th1 profile disease, shifting the immune pathway to Th2 profile might reduce the intestinal damage (Lerner, 2010).

#### Immune Activation Induced Cell Death

Inflammation can cause a substantial hyperactivation of autoaggressive lymphocytes, leading to activation-induced cell death and attenuate the systemic load of aggressive T cells. It seems that repeated encounter with powerful antigenic stimuli leading to restriction of an immune response is well-established in viral infections, where the primary response undergoes a major restriction after antigen elimination. EBV, HBV, and CMV infections are some of the examples. Similarly, administration of mycobacterial products, such as bacilli Calmette-Guérin, prevented the onset and recurrence of type 1 diabetes mellitus in NOD mice by inducing apoptosis

#### TABLE 3 | Direct (⇑⇑), reverse (⇑⇓) or no (⊗) links of CD and Hp infection.


(Continued)


#CD209 is a dendritic and macrophage surface molecule involved in pathogen recognition and immune activation, Hp infection in particular (Bergman et al., 2006; Núñez et al., 2006). It was found to be overexpressed in the Hp infected gastric epithelial cells and to mediate Th1 differentiation, which may be involved in gastric mucosal injury (Wu et al., 2014).

of autoreactive T cells (Christen and von Herrath, 2005). In view of the fact that viruses are inducers of immune cells apoptosis while sparing the Treg cells (Che et al., 2015) and apoptosis is enhanced in CD (Shalimar et al., 2013), it is suggested that viruses, by abrogating immune activation, might attenuate intestinal autoimmune progression in CD.

### Infection at Another Location might Keep Auto-Aggressive Cells from Reaching the Site of Autoimmune Destruction

As suggested by Christen (Christen and von Herrath, 2005), an infectious inflammation elsewhere in the body might keep auto-aggressive cells from arriving into the sites of autoimmune destruction, that might be due to the abrogation of type 1 diabetes in NOD mice after LCMV infection. The authors suggested that this occurred because the "abrogative" virus grew predominantly in peripheral lymphoid organs and other sites rather than the pancreas or its islets themselves. Thus, the sites of severe inflammation might act as a filter for auto-aggressive T cells removing them from the circuit and depriving them from homing the pancreatic islets. Similar scenarios might operate where infection with B. coxsackievirus or Salmomella typhi murium protected against autoimmunity (Tracy et al., 2002; Raine et al., 2006).

## Immunosuppression by Extracellular Vesicles

Release of extracellular vesicles is a natural phenomenon of almost all cell types. They derive either from multivesicular bodies or from the cellular plasma membrane. Those vesicles contain a subset of cell derived proteins, lipids, including nucleic acids. Extracellular vesicles regulate immune responses against pathogens, as well as autoimmunity. It is suggested that these suppressive vesicles would prevent peripheral selfantigens and commonly encountered foreign antigens from causing chronic inflammation and autoimmunity. Following this lines, it is hypothesized that various infectious agents can induce those regulatory extracellular vesicles, counteracting autoimmune pathways, playing a protective anti-autoimmune role (Robbins and Morelli, 2014; Robbins et al., 2016).

### Infectious Agents' Secretion of Anti-Autoreactive T Cells Proteins

Infections with helminths can prevent or attenuate autoinflammatory/immune diseases. In addition to their Th1 to Th2 shift, most recently, Helminth secreted proteins were shown to prevent autoimmunity. The excretory/secretory products of Fasciola hepatica contain immune-modulatory molecules that arbitrate protection from autoimmune diabetes via the activation and provision of a regulatory immune environment (Lund et al., 2014). Such a mechanism was not studied in CD, but might explain the new potential therapeutic strategy to treat CD with Necator Americanus larvae (Croese et al., 2015; Giacomin et al., 2015).

## CONCLUSIONS

The cross-talks between infections and autoimmunity are complex (**Figure 1**). Most of the data indicate that microbes and viruses are major environmental factors in autoimmunity induction. However, growing evidences conversely suggest that infectious agent can abrogate or protect against autoimmunity. This protective evolutionary cross-talks between microbes/viruses and us might represent a mutual beneficial equilibrium relationship between two cohabiting ecosystems. The protective pathways might involve PTMP, decreased intestinal permeability, Th1 to Th2 immune shift, induction of inflammatory immune cell apoptosis, auto-aggressive cells relocation from the target organ, immunosuppressive extracellular vesicles and anti-autoreactive cell immune-regulatory proteins.

Yet, our analysis demonstrates that the interaction of the microorganisms /viruses and CD is always a set of multi-directional processes. With a detailed consideration

#### REFERENCES


of possible mechanisms of CD and CMV, EBV, Herpes simplex type 1, Rubella, H. pylori, it can be assumed that the role of these infections suggested to be potential CD protectors infections, is not so unambiguous positive and the outcome of this interactions might be due to a balance between these multi-directional processes. In summary, there are more publications on the inducer role of infections in CD, and the few ones advocating the protective role should be further explored. The present review expend on several avenues that can be studied to understand the protective cross-talks between infectious agents and CD. Apprehending them can potentially suggest new therapeutic strategies for CD.

#### AUTHOR CONTRIBUTIONS

AL: designed, wrote, edited, submitted, MA: designed, wrote, AS: literature search, revised, reviewed, TM: literature search, edited, and revised.

#### ACKNOWLEDGMENTS

The authors thank Mr. Alf Neu for the artistic design of the figure.


gluten-degrading microbiome and its potential implications in coeliac disease and gluten sensitivity. Clin. Microbiol. Infect. 19, E386–E394. doi: 10.1111/1469-0691.12249


DQ2-negative celiac disease in the Spanish population. World J. Gastroenterol. 12, 4397–4400. doi: 10.3748/wjg.v12.i27.4397


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Lerner, Arleevskaya, Schmiedl and Matthias. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Virome and Its Major Component, Anellovirus, a Convoluted System Molding Human Immune Defenses and Possibly Affecting the Development of Asthma and Respiratory Diseases in Childhood

Giulia Freer<sup>1</sup> , Fabrizio Maggi<sup>2</sup> , Massimo Pifferi<sup>3</sup> , Maria E. Di Cicco<sup>3</sup> , Diego G. Peroni<sup>3</sup> \* † and Mauro Pistello1,2†

<sup>1</sup> Retrovirus Center, Department of Translational Research, University of Pisa, Pisa, Italy, <sup>2</sup> Virology Unit, University Hospital of Pisa, Pisa, Italy, <sup>3</sup> Department of Clinical and Experimental Medicine, Section of Pediatrics, University of Pisa, Pisa, Italy

The microbiome, a thriving and complex microbial community colonizing the human body, has a broad impact on human health. Colonization is a continuous process that starts very early in life and occurs thanks to shrewd strategies microbes have evolved to tackle a convoluted array of anatomical, physiological, and functional barriers of the human body. Cumulative evidence shows that viruses are part of the microbiome. This part, called virome, has a dynamic composition that reflects what we eat, how and where we live, what we do, our genetic background, and other unpredictable variables. Thus, the virome plays a chief role in shaping innate and adaptive host immune defenses. Imbalance of normal microbial flora is thought to trigger or exacerbate many acute and chronic disorders. A compelling example can be found in the respiratory apparatus, where early-life viral infections are major determinants for the development of allergic diseases, like asthma, and other non-transmissible diseases. In this review, we focus on the virome and, particularly, on Anelloviridae, a recently discovered virus family. Anelloviruses are major components of the virome, present in most, if not all, human beings, where they are acquired early in life and replicate persistently without causing apparent disease. We will discuss how modulation of innate and adaptive immune systems by Anelloviruses can influence the development of respiratory diseases in childhood and provide evidence for the use of Anelloviruses as useful and practical molecular markers to monitor inflammatory processes and immune system competence.

Keywords: virome, microbiome, anelloviruses, torque teno virus, asthma, respiratory diseases, wheezing

#### Edited by:

Yves Renaudineau, Université de Bretagne Occidentale, France

#### Reviewed by:

Pei Xu, Zhongshan School of Medicine, China Irit Davidson, Kimron Veterinary Institute, Israel

\*Correspondence:

Diego G. Peroni diego.peroni@unipi.it †These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 01 August 2017 Accepted: 23 March 2018 Published: 10 April 2018

#### Citation:

Freer G, Maggi F, Pifferi M, Di Cicco ME, Peroni DG and Pistello M (2018) The Virome and Its Major Component, Anellovirus, a Convoluted System Molding Human Immune Defenses and Possibly Affecting the Development of Asthma and Respiratory Diseases in Childhood. Front. Microbiol. 9:686. doi: 10.3389/fmicb.2018.00686

## INTRODUCTION

fmicb-09-00686 April 9, 2018 Time: 16:39 # 2

At birth, both the digestive system and the airways are immediately exploited as portals of entry by a number of microbes, most of which are likely to persist and become part of the so-called "microbiome." This is a community of microorganisms that live on the human body without apparently affecting health (Whipps and Karen Lewis, 1988; Hooper et al., 2012; Tremaroli and Bäckhed, 2012). It has long been known that the microbiome is beneficial to hosts in a number of ways, and, in recent years, its interaction with the immune system has even been recognized as fundamental for immune system maturation, reactivity to specific antigens and development of tolerance (Scharschmidt et al., 2015; Ignacio et al., 2016; Kim et al., 2016). The microbiome, from this point of view, tunes immunity by acting as a constant source of stimuli (Belkaid and Hand, 2014; Belkaid and Segre, 2014).

Recently, with the advent of high throughput sequencing methods, the diversity of the microbiome inhabiting gut, lung, skin, and even blood in physiological conditions has been found to be much larger than first thought. In particular, a constantly fluctuating population of viruses have joined the list of infectious agents that are now considered part of the microbiome in several body sites (Blaser and Valentine, 2008; Shulman and Davidson, 2017). Very recent work has estimated that roughly 45% of mammalian viruses can be detected in healthy humans (Olival et al., 2017).

Most initial interactions between hosts and viruses are governed by the innate immune system, that prevents colonizing infectious agents from spreading systemically and maintains mucosal homeostasis (Medzhitov and Janeway, 2002; Lamkanfi and Dixit, 2011). Activation of the innate immune responses triggers a cascade reaction that results in secretion of cytokines and chemokines, and often engages different cells to control invasion (Freer et al., 2017). Following recognition of specific microbial, viral and damage stimuli, intracellular multiprotein complexes called inflammasomes assemble and induce downstream immune responses to specific pathogens. The effects of turning on immunity generally protects against pathogen invasion, but reactions to harmless antigens may lead to the establishment of disease in predisposed individuals. In this review, we discuss the multiple effects of the virome on host health, with special reference to Anelloviruses.

## THE HUMAN VIROME

Although viruses have long been considered "bad news in a protein coat" (Medawar and Medawar, 1983), many novel viruses are found to replicate in healthy individuals. So far, roughly 220 viruses are known to infect humans and only about half are pathogens (Parker, 2016). Truly apathogenic viruses can be grossly divided in viruses infecting bacteria, integrating into human chromosomes as endogenous retroelements, and persisting indefinitely. They are referred to as "commensal" viruses that are part of the virome without an apparent clinical outcome (Rascovan et al., 2016). Many viruses that infect humans may even have a beneficial role (Phan et al., 2016): in animal models, resident intestinal viruses were shown to reduce intestinal inflammation by inducing interferon (IFN)-β, secreted mainly by plasmacytoid dendritic cells (DCs) (Yang et al., 2016), or by providing resistance to infection by bacterial pathogens (Barton et al., 2007).

The number of apathogenic viruses includes many genera detected in various tissues of healthy people, especially infants (Allander et al., 2005; Wang et al., 2016; Moustafa et al., 2017). What role they play in human physiology is still unknown, although they are currently hypothesized to alter disease susceptibility. This is suggested by many epidemiological observations and findings in animal models (Roossinck, 2011; Virgin, 2014).

Resident viruses influence the immune system helping it to develop properly, similarly to bacterial microbiome. Indeed, Cadwell demonstrated that mouse norovirus, a commensal relative of a human pathogen, restored intestinal morphology and immunological functions in germ-free newborn mice, where it is normally perturbed (Cadwell, 2015). On the other hand, the immune system has been recently demonstrated to control virome expansion, similarly to bacteria: HIV-infected patients exhibited low peripheral CD4<sup>+</sup> T cell counts and dramatic expansion of enteric virome adenovirus titers, possibly contributing to AIDS-associated enteropathy and disease progression. These findings suggest that virome expansion is linked to the pathogenesis of AIDS and highlights the role of the immune system in controlling viral populations in the intestine (Monaco et al., 2016). In addition, enteric viral communities have been found to change during HIV infection and raises in Anelloviridae and other virus titers have been associated to increased pathology (Gootenberg et al., 2017).

#### Anelloviruses and TTV

A group of viruses discovered in 1997 (Nishizawa et al., 1997; Okamoto et al., 1998), now called Anelloviruses (AV), represents about 70% of total viruses detected in blood and in most tissues and organs (De Vlaminck et al., 2013). Their prototype, presently named torquetenovirus (TTV), is one of a vast spectrum of viral agents with similar genomes, like torquetenominivirus (Takahashi et al., 2000) and torquetenomidivirus (Ninomiya et al., 2007), both of which have smaller genomes than TTV. All these viruses are classified in the newly established family Anelloviridae (from anellus, Latin for ring, for their circular genome). AV are characterized by a small (2.2 to 3.7 kb), single stranded DNA (ssDNA) circular genome, which makes AV the genetically simplest of all known replication-competent animal viruses. In addition, they are extremely diverse genetically, more than any other viral family. They all lead to persistent, possibly life-long infections and they can be detected at very high levels in blood and in practically all tissues of almost 100% of people worldwide. Different genetic forms are found in a large proportion of individuals regardless of age, socio-economical standing and health conditions, being acquired very soon after birth or even prenatally (Maggi and Bendinelli, 2010).

No specific pathogenic effect has so far been pinpointed to any AV, although similarity of human Anelloviridae to avian

ones suggests that their pathogenicity might be underestimated (Davidson and Shulman, 2008). Increased viremia levels of AV have been found in immune suppressed individuals and in subjects with inflammatory diseases, suggesting that they are normally kept under immunological control, but may contribute to maintain the background level of inflammation chronically elevated in the body (Maggi et al., 2004; Li et al., 2013; Young et al., 2015; Abbas et al., 2016).

## How TTV Interacts and Modulates Host Defenses

TTV interacts with many pathogen-associated molecular pattern (PAMP) receptors (PRR) that fuel immune and inflammatory responses (Zheng et al., 2007; Rocchi et al., 2009; Kincaid et al., 2013). In vitro studies show that TTV ORF2 protein suppressed the activity of NFκB, crucial for the expression of many genes connected to inflammation. ORF2 protein of TTV is able to influence the activity of NFκB by inhibiting its translocation to the cell nucleus and, consequently, its ability to activate transcription of genes, such as IL-6, -8, and cyclo-oxygenase-2 (Zheng et al., 2007).

In addition, the genome of TTV and its replication intermediates may stimulate TLRs in infected cells and consequently synthesis of pro-inflammatory molecules. Unmethylated heterodimers of guanosine and cytosine (CpGs) in bacterial and viral DNA are absent in mammalian DNA and therefore seen as molecular signatures of foreign DNA. The importance of these molecules as PAMPs is demonstrated by the fact that one PPR, namely toll-like receptor (TLR)-9, is specialized to detect CpGs. Depending on the number or nucleotides flanking CpGs, it triggers production of inflammatory cytokines, such as IFN-α, Interleukin (IL)-6, and IL-12, or, alternatively, it may generate an inhibitory signal (Krieg, 2002). Both stimulatory and inhibitory CpGs are present in DNA of TTV and in most microbes, and their relative frequency may differ considerably, even within strains of the same species, thus probably influencing the way they interact and stimulate TLR-9. For instance, we have found that TTV genogroup 4, detected at higher levels in pediatric patients with bronchopneumonia compared to those with milder acute respiratory diseases (ARDs) (Maggi et al., 2003), was rich in stimulatory CpGs and activated TLR-9 in mouse spleen cells in vitro, causing abundant production of pro-inflammatory cytokines (Zheng et al., 2007; Rocchi et al., 2009).

MicroRNAs (miRNA) are ∼22 nt small, single-stranded, noncoding RNAs produced by hosts and pathogens. They are potent modulators of pathogen recognition and host defense in a vast array of cellular metabolic pathways. As regards microbe– host interaction, cellular miRNAs seem to modulate immune responses and inflammation and to play a direct antiviral role by blocking translation of viral genes, counteracting block of apoptosis and persistent replication. Very recent work shows that miRNA can polarize macrophages toward allergic reactions in animal models (Zhou et al., 2017). Their role in inflammation is probably very complex, since they may both up- and downregulate inflammation in several diseases, including asthma (Dissanayake and Inoue, 2016). Viruses, including small ones like TTV, encode their own miRNAs that cooperate with viral proteins to regulate the expression of viral genes, replication, pathogenesis and immune evasion, and the whole process of virus-related inflammation (Kincaid and Sullivan, 2012; Cullen, 2013; Sorel and Dewals, 2016). Of note, both cellular and viral miRNAs have been found to transmit information to distant cells by circulating within plasma exosomes.

Interestingly, TTV was also found to encode in vivo miRNAs possibly involved in viral immune evasion and that could be involved in the regulation of IFN signaling (Kincaid et al., 2013). Different TTV species have been shown to encode miRNAs and cause these molecules to be found as plasma exosomes in many infected individuals. Production and release was not correlated with virus replication, as monitored by measuring TTV viremia levels. Notably, TTV miRNA profiles differed depending on patients' health status; the type of miRNAs produced also differed within sick patients (Vignolini et al., 2016). Role and significance of TTV miRNAs are not yet understood and warrant further studies. An overview of the mechanisms exploited by TTV to stimulate or soothe host innate and adaptive defenses is shown in **Figure 1**.

#### Role of Microbial Exposure and Viral Infections in Wheezing and Asthma Development

Asthma is a multifactorial inflammatory disease of the lower airways, caused by environmental and genetic factors. The disease incidence seems to be increasing especially in industrialized countries (Ellwood et al., 2017). A long-standing theory, the hygiene hypothesis, suggests that insufficient interaction with microbes in early life leads to the development of allergic reactions (Liu, 2015). Significant differences in the prevalence of asthma were found between Amish and Hutterite schoolchildren, despite similar genetic ancestries and lifestyles. As compared with the Hutterites, the Amish practice traditional farming and are exposed to an environment rich in microbes. The significantly lower rates of asthma and the distinct immune profiles in the Amish suggest that environment and sustained microbial exposure have profound effects on innate immunity (Stein et al., 2016). Further data generated in an experimental model of asthma support this notion by showing that the protective effect of the Amish environment requires the activation of innate immune signaling.

On the other hand, there is a consensus on the notion that early respiratory viral and bacterial infections are potent triggers of wheezing-related disorders and development of asthma later in life (Lemanske, 2004; Gern et al., 2005). Growing evidence indicates that respiratory viral infections, especially when acquired in early life may provide the stimulus to inflammasomes assembly, and prime immature DCs toward a Th2 response that, eventually, may sensitize genetically predisposed individuals to local allergens (Holgate, 2011; Lee et al., 2014).

Virus and microorganisms in general may act through several PRRs including TLRs (TLRs and TLR-9 in particular) on DCs. Virus infections have been shown to modulate the responses

of TLRs and miRNAs on the balance of T cells toward Th1, Th2, Th17, or T-regulatory (Treg) subtypes in respiratory airways (Durrani et al., 2012; Holt and Sly, 2012). Microbial products may in turn bind TLRs on airway epithelial cells, leading to the release of the IL-7-like cytokines and thymic stromal lymphopoietin. They may also interact with TLRs on DCs and upregulate the expression of costimulatory molecules to enhance Th2 polarization, also activating mast cells for Th2 cytokine production (de Heer et al., 2004; Troy and Bosco, 2016). Recently, the role of respiratory syncytial virus (RSV) and its impact on bronchiolitis at the time of infection and respiratory morbidity later in life has been revisited (Rossi and Colin, 2017). It has been shown, although controversially, that patients with RSV infection receiving hospital-based care have a higher incidence of asthma and reduced pulmonary function in childhood and in adolescence (Sigurs et al., 2010). In addition, large-scale use of molecular diagnostic techniques pinpointed human rhinovirus (HRV) to infant wheezing and asthma development (Song, 2016). Indeed, HRV has been isolated in 90% of children with asthma exacerbations and found closely connected to hospitalization risk in cohort studies (Bizzintino et al., 2010; Foxman and Iwasaki, 2011). Further evidence on the role of HRV infections during infancy in asthma development has been found: the Childhood Origin of Asthma (COAST) cohort study showed that at least one HRV infection during infancy was the most significant risk factor for persistent wheezing at the age of 3 years (Lemanske et al., 2005). Also, having an HRV wheezing episode in the first 3 years of life was a strong risk factor for asthma not only in childhood (Hyvärinen

et al., 2005; Jackson et al., 2015), but also in adolescence (Rubner et al., 2017). In fact, in a high-risk birth cohort, HRV wheezing, associated with early life aeroallergen sensitization, had additive effects on asthma risk at adolescence (Rubner et al., 2017). Other detrimental effects have been associated to various respiratory viruses, such as metapneumovirus and bocavirus (Camargo et al., 2008; Söderlund-Venermo et al., 2009; Rudd et al., 2017). Indeed, atopy predisposition may be the individual driving factor involved in promoting asthma development via interaction with HRV, and possibly other respiratory viral infections, in infancy (Song, 2016). Allergic sensitizations and viral infections may in turn skew immune responses to produce Th2 cytokines that may at the same time amplify allergic inflammation and reduce the host antiviral responses, resulting in increased viral replication and cellular damage (Xatzipsalti et al., 2008).

## AV and Respiratory Allergy

Although TTV has been investigated to a reasonable extent, its role on asthma is far from clear. Previous studies of our group have shown that presence or viremia levels of TTV were significantly associated with ARDs in pediatric patients and that children with bronchopneumonia (BP) have considerably higher TTV loads than do children with milder respiratory disease. Further, high TTV loads correlate with a decrease in circulating CD3<sup>+</sup> and CD4<sup>+</sup> T cells, an increment in B cells, and increased activity of eosinophils, again emphasizing the immunomodulatory activity of TTV (Maggi et al., 2004; Pifferi et al., 2005). In another study, a positive association was found

between nasal TTV loads and levels of eosinophil cationic protein, a marker of bronchial inflammation. These markers were found particularly elevated in the children with asthma who had moderately to severely compromised spirometric indices. This study was the first performed in children with asthma and suggested that TTV might be a contributing factor in lung impairment (Pifferi et al., 2005).

Concerning a mechanistic role of TTV in respiratory dysfunction, it has been postulated that this virus, either alone or synergistically with other viruses, may act as an enhancer of inflammation systemically or at specific body sites such as upper and lower airways (Maggi and Bendinelli, 2009). One possible way can be envisioned via high amounts of immune complexes that form following TTV replication in blood. In infants with ARD, the airways were shown to be the sites of primary infection and continual replication by TTV, with higher viral loads in patients with more severe lower respiratory tract infections (Maggi et al., 2003). Furthermore, TTV may worsen the extent and the severity of the inflammatory response due to allergens, if sensitization is present in the subject. This hypothesis is supported, in children with ARD, by the positive correlation between TTV loads and serum concentration of eosinophil cationic protein (Maggi et al., 2004), and of exhaled nitric oxide, a sensitive marker of airway inflammation in asthmatic children (Li et al., 2013). In another study, we were able to demonstrate a high prevalence of TTV infections in children with bronchiectasis, a chronic respiratory disorder associated with several invalidating airway diseases: indeed, strong correlation between TTV loads and airflow limitation within the more peripheral airways was found, as well as between severity of the disorder and limitation of the lung function (Pifferi et al., 2006).

#### CONCLUSION

Most viral infections elicit robust immune responses but viral clearance is not always obtained. Consequently, there

#### REFERENCES


must be unidentified factors/conditions that determine a tolerogenic status toward non-pathogenic viruses, and strong immune opposition against pathogenic ones. Tolerance may depend on host genetic, life-style and environmental factors, or alternatively on the ability of "commensal" viruses not to stir up inflammasomes and innate immune effectors.

Increasing evidence shows that the virome is actually beneficial to the host, who seems to tolerate "commensal" viruses, although they replicate and circulate among individuals. AV infect and persist in nearly all mammals and represent a large body of the human virome. They continuously replicate with no overt damage to the host and, therefore, are a good example of commensal viruses in this respect. Several clinical studies suggest that TTV plays a role in the development and/or exacerbation of respiratory diseases in childhood. Although further studies are warranted to draw firm conclusions, the virome and AV are one the best examples of a virus–host relationship. Its understanding will help clarify the role of viruses in shaping human immune defenses and perhaps contribute to their derangement.

#### AUTHOR CONTRIBUTIONS

GF, FM, DP, and MauP contributed to the elaboration of this mini review. MasP and MDC performed TTV studies in infants and made some unpublished results available. All authors read and approved the final manuscript.

#### FUNDING

This work was supported by Progetti di Ricerca di Ateneo 2017– 2018 of University of Pisa, Project No. PRA\_2017\_38, DR n. 83/2017.


reduced by the high-affinity IgE receptor in allergic asthmatic children. J. Allergy Clin. Immunol. 130, 489–495. doi: 10.1016/j.jaci.2012.05.023


risk at adolescence. J. Allergy Clin. Immunol. 139, 501–507. doi: 10.1016/j.jaci. 2016.03.049


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Freer, Maggi, Pifferi, Di Cicco, Peroni and Pistello. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Prophylactic Supplementation of Bifidobacterium longum 5 1A Protects Mice from Ovariectomy-Induced Exacerbated Allergic Airway Inflammation and Airway Hyperresponsiveness

Edited by:

Yves Renaudineau, University of Western Brittany, France

#### Reviewed by:

Andreas L. Lopata, James Cook University Townsville, Australia Aude Remot, Institut National de la Recherche Agronomique (INRA), France

#### \*Correspondence:

Caroline M. Ferreira caroline.nu.ferreira@gmail.com

†These authors are co-first authors.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 26 April 2017 Accepted: 24 August 2017 Published: 11 September 2017

#### Citation:

Mendes E, Acetturi BG, Thomas AM, Martins FS, Crisma AR, Murata G, Braga TT, Camâra NOS, Franco ALS, Setubal JC, Ribeiro WR, Valduga CJ, Curi R, Dias-Neto E, Tavares-de-Lima W and Ferreira CM (2017) Prophylactic Supplementation of Bifidobacterium longum 51A Protects Mice from Ovariectomy-Induced Exacerbated Allergic Airway Inflammation and Airway Hyperresponsiveness. Front. Microbiol. 8:1732. doi: 10.3389/fmicb.2017.01732 Eduardo Mendes<sup>1</sup>† , Beatriz G. Acetturi<sup>2</sup>† , Andrew M. Thomas3,4,5 , Flaviano dos S. Martins<sup>6</sup> , Amanda R. Crisma<sup>7</sup> , Gilson Murata<sup>7</sup> , TárcioT. Braga<sup>8</sup> , Niels O. S. Camâra<sup>8</sup> , Adriana L. dos S. Franco<sup>9</sup> , João C. Setubal<sup>4</sup> , Willian R. Ribeiro<sup>1</sup> , Claudete J. Valduga10, Rui Curi<sup>7</sup> , Emmanuel Dias-Neto3,11, Wothan Tavares-de-Lima<sup>2</sup> and Caroline M. Ferreira<sup>1</sup> \*

<sup>1</sup> Department of Pharmaceutics Sciences, Institute of Environmental, Chemistry and Pharmaceutical Sciences, Universidade Federal de São Paulo, Diadema, Brazil, <sup>2</sup> Department of Pharmacology, Institute of Biomedical Sciences I, University of São Paulo, São Paulo, Brazil, <sup>3</sup> Medical Genomics Laboratory, CIPE/A.C.Camargo Cancer Center, São Paulo, Brazil, <sup>4</sup> Department of Biochemistry, Institute of Chemistry, University of São Paulo, São Paulo, Brazil, <sup>5</sup> Bioinformatics Graduate Program, University of São Paulo, São Paulo, Brazil, <sup>6</sup> Department of Microbiology, Institute of Biological Sciences, Federal University of Minas Gerais, Belo Horizonte, Brazil, <sup>7</sup> Department of Physiology and Biophysics, Institute of Biomedical Sciences, University of São Paulo, São Paulo, Brazil, <sup>8</sup> Department of Immunology, Institute of Biomedical Sciences IV, University of São Paulo, São Paulo, Brazil, <sup>9</sup> Post Graduate Program in Biophotonics Applied to Health Sciences, Universidade Nove de Julho, São Paulo, Brazil, <sup>10</sup> Department of Pharmacy and Biotechnology, Universidade de Anhanguera de São Paulo, São Paulo, Brazil, <sup>11</sup> Laboratory of Neurosciences (LIM-27), Institute of Psychiatry, Medical School University of São Paulo, São Paulo, Brazil

Asthma is a chronic inflammatory disease that affects more females than males after puberty, and its symptoms and severity in women change during menstruation and menopause. Recently, evidence has demonstrated that interactions among the microbiota, female sex hormones, and immunity are associated with the development of autoimmune diseases. However, no studies have investigated if therapeutic gut microbiota modulation strategies could affect asthma exacerbation during menstruation and menopause. Here we aimed to examine the preventive effects of a probiotic, Bifidobacterium longum 5 1A, on airway inflammation exacerbation in allergic ovariectomized mice. We first evaluated the gut microbiota composition and diversity in mice 10 days after ovariectomy. Next, we examined whether re-exposure of ovariectomized allergic mice to antigen (ovalbumin) would lead to exacerbation of lung inflammation. Finally, we evaluated the preventive and treatment effect of B. longum 5 1A on lung inflammation and airway hyperresponsiveness. Our results showed that whereas ovariectomy caused no alterations in the gut microbiota composition and diversity in this animal model, 10 days after ovariectomy, preventive use administration of B. longum 5 1A, rather than its use after surgery was capable of attenuate the exacerbated lung inflammation and hyperresponsiveness in ovariectomized allergic mice. This prophylactic effect of B. longum 5 1A involves acetate production, which led to increased fecal acetate levels and, consequently, increased Treg cells in ovariectomized allergic mice.

Keywords: probiotic, Bifidobacterium longum, airway inflammation, ovariectomy, microbiota

## INTRODUCTION

fmicb-08-01732 September 7, 2017 Time: 17:25 # 2

Asthma is a chronic inflammatory lung disease and a serious health concern (Cohn et al., 2004; Leynaert et al., 2012). The number of people with asthma worldwide increased from 235 to 334 million between 2011 and 2014 (Ellwood et al., 2017). At age 11, the prevalence of asthma is greater in males than in females; however, after puberty, this trend reverses. Furthermore, in adulthood, the mean duration of hospital stay is longer for women than for men, suggesting not only that the prevalence is greater in women after puberty but also that the disease is more severe (Hanley, 1981; Gibbs et al., 1984; Eliasson et al., 1986). Clinical evidence shows that a significant percentage of women with asthma exhibit worsening of symptoms during the perimenstrual phase (shortly before and during the first few days of the menstrual period). Terms such as (pre)menstrual, perimenstrual or circamenstrual asthma have been used to describe this phenomenon. It has been suggested that hormonal fluctuations during the menstrual cycle play a significant role in the periodic worsening of asthma severity in adult females. The hormonal fluctuations that occur in menopause can also trigger or exacerbate adult-onset asthma (Troisi et al., 1995; Bellia and Augugliaro, 2007; Zemp et al., 2012; Triebner et al., 2016). The roles of gender and sex hormones in asthma are very complex. Recently, experimental findings have demonstrated that interactions among the microbiota, female sex hormones, and immunity are associated with the development of autoimmune diseases (Markle et al., 2013; Yurkovetskiy et al., 2013). In this scenario, it has been clearly recognized that germ-free NOD female mice lack the commonly observed gender bias of diabetes, with enhanced type 1 diabetes development found in these females, showing a pivotal role of the gut microbiota in disease development. On the other hand, after castration, male mice exhibit a microbiota composition that is more similar to that found in female mice. Overall, these data demonstrate that female sex hormones, rather than X chromosome-associated factors, are involved in the modulation of microbiota composition (Markle et al., 2013; Yurkovetskiy et al., 2013). Thus, it is strongly suggested that hormones and the microbiota cooperate to modify the course of disease progression. Given the recognized role of the gut microbiota in regulating the immune response via hormone interactions, it is not surprising that the oscillating hormone levels observed during the menstrual cycle and in menopausal women could interfere with the gut microbiota profile, which in turn may affect female asthma symptoms.

The most popular probiotics are lactobacillus and bifidobacteria. Probiotic strains such as Lactobacillus reuteri (Miraglia Del Giudice et al., 2012) and L. casei (Yu et al., 2010) have been studied for their amelioration of allergic disease. Moreover, experimental studies have demonstrated the immunomodulatory role of probiotic bacteria, notably Bifidobacterium, in reducing the Th2 inflammation profile induced in mice (Abrahamsson et al., 2007; Iwabuchi et al., 2009; MacSharry et al., 2012). In general, the strategy has been to maintain or restore a healthy gut microbiota by regular supplementation with the abovementioned probiotics. Recently, studies have demonstrated that oral administration of probiotics protects female mice from bone density loss caused by ovary removal (Sjogren et al., 2012; Britton et al., 2014; Ohlsson et al., 2014; Parvaneh et al., 2015). Despite the role of probiotics in the Th2 inflammatory response, there are no experimental studies investigating the preventive action of probiotics on the exacerbation of asthma by low levels of sex hormones, which occurs in asthmatic women during the menstrual cycle and menopause. Considering that life expectancy has increased and that most women now undergo menopause, strategies to prevent the initiation or exacerbation of asthma in this life stage are highly relevant (Troisi et al., 1995; Gong et al., 2016). Thus, the purpose of this study was to determine whether Bifidobacterium longum 5 1A exerts a modulatory effect on the exacerbated lung inflammatory response induced in ovariectomized allergic mice. Here, we show for the first time that B. longum 5 1A administered to ovariectomized allergic mice prevented the exacerbation of lung inflammation and airway hyperresponsiveness.

#### MATERIALS AND METHODS

#### Animals

Female Balb/c mice (18–20 g) were obtained from the animal facility of the Institute of Biomedical Sciences, University of Sao Paulo. Animals were housed in groups of five per cage in a light- and temperature-controlled room (12 h light/dark cycles, 21 ± 2 ◦C) with free access to food (AIN93-M) (Reeves et al., 1993) and water. The local Animal Care Committee of the University of Sao Paulo Institute of Biomedical Sciences approved the experiments.

### Ovariectomy

Mice were anesthetized by an intraperitoneal (i.p.) injection of ketamine/xylazine (100 and 20 mg/kg, respectively). After an incision was made in the lower part of the abdomen, the ovaries were identified and removed from the adherent tissue. The effectiveness of the OVx procedure was assessed by analyzing the morphologic features of cells in vaginal smears and by quantifying the uterine weight. Mice subjected to similar manipulations except for the ovary removal were used as the sham-operated controls and labeled 'sham' animals. The same surgeon performed all surgical procedures.

### Microbial Community Profiling

fmicb-08-01732 September 7, 2017 Time: 17:25 # 3

Fecal samples were collected 1 day before and 10 days after surgery and fecal pellets were frozen in 2 mL tubes at −80◦C until DNA extraction. Pellets were placed in MoBio PowerSoil bead tubes and incubated at 70◦C for 10 min before proceeding with the protocol recommended by the manufacturer. Extracted DNA was quantified using a Qubit spectrophotometer (Thermo Scientific Technologies) and visualized in 2% agarose gels stained with ethidium bromide to evaluate DNA integrity. Microbiote analysis was performed essentially as described by Thomas et al. (2016). In summary, for PCR amplification and amplicon sequencing, the V4–V5 region of the 16S rRNA gene was amplified using the forward 5<sup>0</sup> -AYTGGGYDTAAAGNG-3<sup>0</sup> and reverse primer 5<sup>0</sup> -CCGTCAATTCNTTTRAGTTT-3<sup>0</sup> , corresponding to the positions 562 and 906, respectively, of the Escherichia coli 16S rRNA gene. Three 20 µl amplification reactions were performed per sample, each containing: 2.5 µM of each primer; 10 µl of Kapa Hotstart High Fidelity Master Mix (Kapa Technologies); and 25 ng of genomic DNA (gDNA). Thermocycling conditions were: 95◦C, 3 min; 98◦C, 15 s; and 40◦C, 30 s for 35 cycles. This was followed by a last extension step at 72◦C for 5 min. Amplicons of the three reactions of each sample were pooled and purified using a MinElute PCR Purification Kit (Qiagen). The purified products were run on 1.5% agarose gels and bands within the expected amplicon range were excised using sterile and disposable scalpels and purified using the Qiaquick gel extraction kit (Qiagen) to remove artifacts, primer-dimers, and non-specific bands. Amplicons were end-repaired and Ion Torrent adaptors with barcodes were ligated using the Ion Plus Fragment Library Kit (Thermo Scientific). Equimolar amounts of barcoded-amplicons from each sample were pooled, using the Ion Torrent qPCR quantitation kit (Thermo Scientific), and used for emulsion PCR using the Ion PGM Template OT2 400 Kit (Thermo Scientific). All samples were sequenced on the Ion torrent PGM platform using a v2 chip and the Ion PGM Sequencing 400 Kit (Thermo Scientific). Sequences processed by the latest version of the Ion Torrent server (v3.6.2) were used as input into the Qiime (Quantitative insights into microbial ecology) software package (Version 1.6.0) (Caporaso et al., 2010). We removed sequences with an average quality score <20 using a 50 nt sliding window. Then, we identified barcodes used for fecal sample, allowing a maximum of two mismatches, and discarded sequences with no barcodes, and <200 nt or >500 nt after barcode removal. PCR primers identified at the start or at the end of the reads, allowing a maximum of 4 nt mismatches, were trimmed and sequences with no identifiable primers were discarded. After primer trimming, we removed all sequences below 200 nt and the remaining sequences were used as input for downstream analysis. Filtered sequences were clustered with 97% identity using UPARSE (implemented in USEARCH v7 – Edgar, 2013) and the seed sequence of each cluster was picked as a representative. Chimeric sequences (and clusters) were identified using UCHIME (Edgar et al., 2011) and the Broad Institute's chimera slayer database (version microbiomeutil-r20110519) and excluded from further analysis. With the Qiime interface (default parameters), the RDP classifier (Wang et al., 2007) was used to assign a taxonomic rank to each sequence using a minimum confidence value of 80% and, subsequently, to each operational taxonomic unit (OTU). Unless otherwise stated, OTUs that occurred in fewer than 25% of all samples and with less than 3 reads were not considered. Using Qiime, we rarefied the OTU table to a depth of 13,851 sequences in order to calculate species richness and phylogenetic diversity, using the total number of observed OTUs and Faith's phylogenetic diversity (Faith et al., 2009), respectively. Rarefaction analysis was performed using Qiime and a cutoff of 13,851 for species richness. For betadiversity analysis, OTU-representative sequences were aligned using PyNAST (Caporaso et al., 2010) against the aligned green genes core set (DeSantis et al., 2006) using Qiime default parameters, and the alignments were lane-mask filtered to build a phylogenetic tree using FastTree (Price et al., 2009). Distance matrices were generated for four metrics, unweighted and weighted UniFrac (Lozupone and Knight, 2005) Bray–Curtis and Euclidean, to compare pairwise distances between the four groups using ADONIS (Oksanen et al., 2016). Due to unequal sequence sampling depth across samples, we normalized the raw number of reads for each OTU using metagenome Seq's (Paulson et al., 2013) cumulative sum scaling function. We then used normalized counts to investigate differences between the groups at the OTU, genera, and family level.

## Probiotic Supplementation

The species of bifidobacteria (B. longum 5 1A) used in this study was isolated and characterized at the Laboratory of Ecology and Physiology of Microorganisms, Institute of Biological Sciences, Federal University of Minas Gerais (Souza et al., 2013). The bifidobacteria were isolated from the feces of healthy children up to 5 years old in the city of Salvador (Bahia, Brazil) and identified by morfotintorial, respiratory, and biochemical tests, followed by multiplex PCR (Kwon et al., 2005). The bacteria were replicated in MRS broth medium (Difco) and grown under anaerobic conditions in an anaerobic jar at 37◦C without stirring for 48 h. For the administration of probiotic bacteria, the mice received a daily inoculum by gavage of 0.1 mL containing 10<sup>8</sup> bacterial cells, beginning 15 days before the first sensitization (when animals were 4 weeks old) and until ovariectomy for prophylactic administration and 1 day after ovariectomy for the treatment protocol.

#### Induction of Allergic Lung Inflammation

Mice were sensitized intraperitoneally with 10 µg of chicken egg OVALBUMIN (OVA) grade V (Sigma Chemical Co., Saint Louis, MO, United States) dissolved in 0.2 mL of Imject alum suspension (1 mg). At days 14, 15, and 16, the mice were exposed to aerosolized OVA (1% in phosphate-buffered saline PBS) for 15 min using an ultrasonic nebulizer device (Respira Max <sup>R</sup> , SP, Brazil) coupled to a plastic inhalation chamber (18.5 cm × 18.5 cm × 13.5 cm). Ten days after the last challenge, the animals were subjected to ovariectomy (OVx), and 10 days later, they were re-challenged over 3 days as described above. The experimental asthma protocols were started when mice were 6 weeks old. All measurements of inflammatory parameters were performed 24 h after the last aerosol re-challenge.

## Short-Chain Fatty Acid (SCFA) Measurements

fmicb-08-01732 September 7, 2017 Time: 17:25 # 4

Acetic acid (99.7%) and citric acid were purchased from Sigma– Aldrich (St. Louis, MO, United States), butanol from Carlo Erba (Cornaredo, Italy), acetonitrile from Merck (Darmstadt, Germany), and tetrahydrofuran from Acros Organics (Fair Lawn, NJ, United States). Chromatographic analyses were performed using an Agilent 6850 system with EzChrom software, equipped with a 7683B automatic liquid sampler, a flame ionization detector (FID) (Agilent Technologies, United States), and a fused-silica capillary DB-23 column (Agilent Technologies, United States) with dimensions of 60 m × 0.25 mm internal diameter (i.d.) coated with a 0.15 µm thick layer of 80.2% 1-methylnaphatalene. The initial oven temperature was 100◦C (hold 7 min), which was then increased to 200◦C at a rate of 25◦C/min (hold 5 min). The FID temperature was maintained at 260◦C, and the flow rates of H2, air, and the make-up gas N2 were 30, 350, and 25 mL/min, respectively. Sample volumes of 5 µL were injected at 250◦C using a split ratio of approximately 25:1. Nitrogen was used as the carrier gas at 1 mL/min (hold 4 min), reduced to 0.8 mL/min (hold 1 min) and then 0.6 mL/min (hold 1 min), and finally increased to 1 mL/min. The runtime for each analysis was 16.5 min. Acetic acid (HAc) was diluted to 1.0 mg/mL in butanol:tetrahydrofuran:acetonitrile (5:3:2), and further dilutions were made using the same solvent to generate a series of standard solutions. Blank human plasma was spiked with 1.0 mg/mL HAc. This solution was further diluted with blank plasma to prepare plasma standards of 0.015–1 mg/mL. To each tube, 40 mg of sodium chloride, 20 mg of citric acid, 40 µL of 0.1 M hydrochloric acid, and 200 µL of butanol:tetrahydrofuran:acetonitrile (5:3:2) were added. To quantify the acids, a calibration curve for the concentration range of 0.015–1 mg/mL was constructed. Probiotic culture medium (100 µL) was used to measure the acetate concentration. The concentration of plasma acetate was expressed in mM. Fecal samples were weighed into 1.5 mL tubes, crushed and homogenized in 100 mL of distilled water. Subsequently, 40 mg of sodium chloride, 20 mg of citric acid, 40 µL of 0.1 M hydrochloric acid, and 200 µL of butanol: tetrahydrofuran: acetonitrile (5:3:2) were added. The tubes were vortexed for 1 min and centrifuged at 10,000 rpm for 10 min. The supernatant was transferred to microtubes, and 5 µL was injected in triplicate into the gas chromatograph. The retention time was 7.2 min. The precision and accuracy of the assay were evaluated by analyzing samples (0.62, 0.25, and 0.75 mg/mL) in five replicates. The precision, expressed as relative standard deviation (RSD), was less than 14.34%, and the accuracy was between 87.51 and 113.56%. The standard acids, HAc, HPr, and HBu, were successfully separated in the gas-liquid chromatogram and eluted between 7 and 10 min in the temperature and gas flow program used here. The retention time was 7.2, 8.2, and 9.2 min for HAc, HPr, and HBu, respectively. The standard calibration curves showed linear relationships for all acids (r = 0.999). There was no interference present in the blanks at the retention time of the acids analyzed, and the analyte peaks were identifiable, discrete, and reproducible. The precision and accuracy of the assay were evaluated by analyzing samples (0.62, 0.25, and 0.75 mg/mL) in five replicates. The precision, expressed as RSD, was less than 14.34, 13.93, and 12.68%, for HAc, HPr, and HBu, respectively. The accuracy was between 87.51 and 113.56% for HAc, 91.44 and 111.60% for HPr, and 86.71 and 98.56% for HBu. The limit of quantification (LOQ) was 0.031 mg/mL for both acids analyzed, with a precision lower than 15% and accuracy within ±15%. The limit of detection (LOD) for both acids was 0.015 mg/mL, where the mean coefficient of variation (CV) value was under 20% for precision and under ±20% for accuracy. The feces extraction efficiency was 97.35 ± 5.23%. Therefore, the developed methods were shown to be reliable (Shah et al., 1991) for quantifying the SCFAs HAc, HPr, and HBu in mouse feces.

## Bronchoalveolar Lavage (BAL)

Female mice were anesthetized with ketamine/xylazine (100 and 20 mg/kg, respectively), the trachea was exposed, and a cannula was inserted. The lungs were washed three times with aliquots (each 0.8 mL) of saline injected through the cannula. From the BAL fluid (BALF), the total number of cells was counted microscopically using a Neubauer chamber. Differential cell counts were performed by cytospin analysis and prepared from aliquots of BALF (200 µL) centrifuged at 300 g for 1 min using a Citospin <sup>R</sup> (Fanem). The cells were stained with Instant Prov (Newprov@, São Paulo, Brazil), and a total of 300 cells were counted to determine the proportion of neutrophils, eosinophils, and mononuclear cells using standard morphological criteria.

## Lung Function Analysis

All animals were anesthetized with ketamine (100 mg/kg, i.p.) and xylazine (20 mg/kg) and paralyzed with pancuronium bromide, and a stable depth of anesthesia was maintained (Ferreira et al., 2012). After tracheostomy, the trachea was cannulated with a blunt 18-gauge metal tube, and the mouse was ventilated with a computer-controlled small-animal ventilator (flexiVent; Scireq, Montreal, QC, Canada) using a tidal volume of 10 mL/kg and a respiratory frequency of 150 breaths/min. A positive end-expiratory pressure (PEEP) of 2 cm H2O was applied throughout. An external jugular vein was isolated for an intravenous (i.v.) infusion of methacholine (MCh). At the outset, 6 µg of MCh was provided intravenously to ensure that the animal was indeed responsive to MCh and that airway resistance returned to the baseline value after its MCh-induced rise, which indicated that the mouse was in stable physiological condition. To obtain a dose-response curve, a bolus of MCh was then injected starting at a dose of 4 µg (200 µg/mL solution in PBS; i.v. boluses of 10–40 µL). Prior to each MCh dose, the expiratory path was obstructed for 15 s to produce a deep inflation, after which exhalation was immediately allowed. Ventilation was continued for approximately 2 min between consecutive MCh doses. Airway responsiveness was equal to the Newtonian resistance (Rn) (Ferreira et al., 2012).

## Analysis of Histological Changes to Lung Tissue

Lungs were removed from mice after BAL and fixed by immersion in 4% paraformaldehyde. The lobes were sagittally sectioned, embedded in paraffin, cut into 5-µm sections, and stained with periodic acid-Schiff (PAS); mucus production was measured as previously described (Tong et al., 2006).

#### Flow Cytometry

Mouse Treg cells were collected from the BALF and analyzed for CD4<sup>+</sup> CD25<sup>+</sup> Foxp3<sup>+</sup> expression using a mouse Treg cell staining kit containing APC-labeled anti-CD4, PE-labeled anti-CD25, and FITC-labeled anti-Foxp3 (eBioscience) according to the manufacturer's instructions. Briefly, prepared cells (1 × 10<sup>6</sup> ) were washed by centrifugation with cold PBS, resuspended in 1 mL of fixation/permeabilization solution, and incubated in the dark at 4◦C for 30–60 min. The cells were washed once with 2 mL of permeabilization buffer, collected by centrifugation, resuspended in 20 mL of blocking agent with 2% (2 mL) normal rat serum in permeabilization buffer, and incubated at 4◦C for 15 min. Next, 20 mL of a fluorochrome-conjugated antibody or isotype control in permeabilization buffer was added, followed by incubation in the dark at 4◦C for 30 min. Finally, the cells were washed with 2 mL of permeabilization buffer, resuspended in flow cytometry buffer (PBS with 2% FBS), and analyzed using a FACSCanto II cytometer (BD Bioscience, San Diego, CA, United States). The data were analyzed using FlowJo <sup>R</sup> software.

## Quantification of Cytokine Levels in BAL and of Female Sex Hormones in Serum

Commercial preparations of paired antibodies and protein standards for measurements of mouse IL-4, IL-5, IL-10, and IFN-γ (BD Bioscience) in the BAL were used to develop ELISAs according to the manufacturer's instructions. The determinations were performed in duplicate. Serum estradiol and progesterone concentrations were determined using an enzyme immunoassay kit according to the manufacturer's instructions (Cayman Chemical, Ann Arbor, MI, United States).

## Statistical Analysis

The data are presented as the mean ± standard error of mean, Student's t-test and ANOVA followed by a Bonferroni post-test were used for comparisons between 2 and 3 or more groups, respectively. Values of p ≤ 0.05 were considered significant.

## RESULTS

## Ovariectomy Efficacy

Ten days after ovariectomy, we measured the body and uterus weight of female mice and analyzed the vaginal cell morphology. The ovariectomized (OVx) animals had significantly higher body weights and lower uterus weights than animals subjected to false surgery (Sham). In addition, the vaginal cell morphological changes observed after ovariectomy were representative of diestrus phase and were corroborated by decreased plasma concentrations of estrogen and progesterone (**Table 1**). **Table 1** shows the parameters measured to analyze ovariectomy efficacy of animals studied in the **Figure 1**.

## Effect of Ovary Removal on the Gut Microbial Profile

To analyze the effect of ovariectomy on gut microbial diversity and composition, we performed 16S rRNA next-generation sequencing. We obtained a total of 1,246,075 QC-passed sequences with an average length of 329 ± 2.9 nt and an average of 56,639 sequences/sample. A total of 258 operational taxonomic units (OTUs) were obtained, which was reduced to 248 after filtering to include only those OTUs present in at least 25% of samples and with at least three reads. When analyzing species richness, we found no significant differences between groups for any of the comparisons (**Figure 1A**) but did find increased phylogenetic diversity when comparing OVx POS (post-ovariectomy) mice with Sham PRE (preovariectomy) mice and OVx PRE mice with Sham PRE mice. We found significant differences in pairwise distances using the weighted UniFrac metric when we compared OVx POS mice with Sham POS mice and all groups simultaneously (**Table 2**). Multidimensional scaling (MDS) revealed that the OVx POS samples clustered farther away from the other samples (**Figure 1B**). Bacterial community composition was also investigated, revealing that the most abundant bacterial phyla were (in decreasing order of abundance) Bacteroidetes, Firmicutes, and Proteobacteria. Despite seeing increases in Bacteroidetes abundance and decreases in Firmicutes abundance in OVx POS mice, these were non-significant. At the family level, the most abundant bacterial families among the groups were Porphyromonadaceae, unclassified Clostridiales, Bacteroidaceae, Lachnospiraceae, and Ruminococcaceae. Hierarchical clustering of rarefied OTU abundances clustered all samples from the OVx POS group together in one main tree branch, whereas samples from the other groups were scattered across various branches (**Figure 1C**). When we evaluated differences between the groups at different taxonomic levels, we found a significant increase of Bifidobacteriaceae, but not any other bacterial taxon, in OVx POS samples (**Figure 1D**).

#### Ovariectomy Exacerbates Lung Inflammation and Airway Hyperresponsiveness in Re-challenged Allergic Mice

We next investigated whether re-exposure of OVx allergic mice to antigen (OVA) exacerbates lung inflammation (**Figure 2A**). We found a significant increase in total cells recovered in the BALF from Sham and OVx allergic mice that was exacerbated by re-exposure of OVx allergic mice to OVA (**Figure 2C**). Moreover, the number of eosinophils in the BAL from re-challenged OVx allergic mice significantly increased compared with Sham allergic mice. The total cells in the BAL recovered from non-sensitized or OVx mice did not differ.

Next, considering that airway hyperresponsiveness is a hallmark of asthma and that its evolution is often associated


TABLE 1 | Animal characteristics, body and tissue weights, and hormone measures

Group Number animals Average body weight (g) Uterus weight (g) 17 Beta estradiol (pg/mL) Progesterone (ng/mL) Diestro phase Ovx 7 24.5 ± 1.0<sup>∗</sup> 0.015 ± 0.004<sup>∗</sup> 21.0 ± 9.9<sup>∗</sup> 0.78 ± 1.1<sup>∗</sup> Positive Sham 4 22.7 ± 1.0 0.028 ± 0.006 41.5 ± 3.3<sup>∗</sup> 1.7 ± 0.6<sup>∗</sup> Negative

Values are mean ± SD. <sup>∗</sup>p =< 0.05 Sham vs. OVX.

fmicb-08-01732 September 7, 2017 Time: 17:25 # 6

between mouse fecal samples. (C) Heatmap and hierarchical clustering of mouse fecal samples using rarefied OTU abundances. (D) Mean relative abundances of the most abundant bacterial families identified in the samples. (E) Mean relative abundances of the least abundant bacterial families identified in the samples. Error bars represent the standard error of the mean and <sup>∗</sup> indicates a p-value < 0.05.

with lung inflammation (Maddox and Schwartz, 2002; Tong et al., 2006), we investigated whether the ovariectomy of allergic mice would also lead to exacerbated hyperresponsiveness. We quantified the respiratory Newtonian resistance (Rn) changes in response to intravenous administration of a cholinergic agent, methacholine (MCh), after the last re-exposure to antigen challenge. Re-challenged OVx allergic mice exhibited significantly increased airway

hyperresponsiveness to MCh compared with Sham OVx allergic mice (**Figure 2B**).

We also determined the effect of ovariectomy on the Th2 cytokine profile. Our data revealed a significant increase in IL-5 levels in the BAL recovered from OVx allergic mice re-exposed to antigen challenge compared with that of Sham OVx allergic mice (**Figure 2D**). Although a low level of IL-4 was found in the BALF, no significant difference between

#### TABLE 2 | Alpha and beta diversity p-values for various group comparisons.


Bold values = P < 0.05.

FIGURE 2 | Ovariectomy exacerbates the lung inflammation and airway hyperresponsiveness of re-challenged allergic mice. (A) Ten days before being ovariectomized (OVx) or not (Sham), the animals were sensitized and challenged. Ten days after ovariectomy, the animals were re-challenged. All parameters were analyzed 24 h after the last re-challenge. (B) Measurement of AHR, as assessed by Newtonian airway resistance to increasing doses of methacholine. (C) Quantification of the total number of cells in the bronchoalveolar lavage (BAL). (D) Concentration of IL-10, IL-5, and INF-γ in the BAL. (E) Representative periodic acid–Schiff (PAS)-stained lung tissue from mice in the SHAM, OVx, SHAM OVA, and OVx OVA groups. Scale bars, 200 µm. All results are representative of data generated in two different experiments and are expressed as the mean ± SEM (n = 5 in all groups). Statistical significance was determined by a one-way analysis of variance (ANOVA), except in (B), in which a two-way ANOVA test was used. Hash symbol: non-allergic (sham and OVx) groups vs. respective allergic group (sham Ova and OVx OVA), ###P < 0.001; asterisk: Sham OVA vs. OVx OVA. <sup>∗</sup>P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

these two groups was observed (unpublished data). In contrast, IFN-γ and IL-10 were reduced in the BAL of re-challenged OVx allergic mice compared with Sham OVx allergic mice (**Figure 2D**).

Excessive mucus production is a characteristic of asthmatic airways; therefore, we evaluated the presence of PAS-stained cells in the lungs of re-challenged mice. As seen in **Figure 2E**, a significant increase in PAS-stained cells was observed in the lungs of re-challenged OVx allergic mice compared with re-challenged Sham OVx allergic mice. The lungs of non-sensitized mice did not show PAS-stained cells (**Figure 2E**).

### Preventive Oral Probiotic Supplementation Attenuates the Exacerbation of Lung Inflammation and Airway Hyperresponsiveness in Re-challenged OVx Allergic Mice

The role of probiotics in regulating the immune system and allergic diseases has been extensively debated (Abrahamsson et al., 2007). An oral probiotic (B. longum 5 1A) was administered daily by gavage for 15 days before the first OVA sensitization of mice and continued until 4 h before the mouse ovaries

t-test, except in (B), in which a two-way ANOVA test was used. <sup>∗</sup>P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001.

were removed (**Figure 3A**). The preventive effect of B. longum 5 1A supplementation on the exacerbation of lung inflammation and airway hyperresponsiveness in re-challenged OVx allergic mice was investigated. Oral supplementation with B. longum 5 1A attenuated the exacerbation of the MCh-induced airway hyperreactivity of re-challenged OVx allergic mice. The airway hyperresponsiveness of supplemented OVx-OVA mice was significantly reduced compared to the OVx-OVA group. Additionally, B. longum 5 1A treatment significantly reduced the number of eosinophils recovered from the BAL of re-challenged OVx allergic mice (**Figure 3B**). B. longum 5 1A administration also resulted in a significant reduction of IL-5 levels in the BAL of re-challenged OVx allergic mice relative to non-treated mice (**Figure 3C**). By contrast, B. longum 5 1A treatment significantly increased the levels of IFN-γ in re-challenged OVx allergic mice. The BAL levels of IL-4 and IL-10 were not changed (**Figure 3D**). The probiotic treatment also significantly reduced the mucus production in the lungs of re-challenged OVx allergic mice compared with non-treated re-challenged OVx allergic mice (**Figure 3E**).

#### Preventive Supplementation with B. longum 5 1A Increases the Production of SCFAs and the Number of Lung Treg Cells in Re-challenged OVx Allergic Mice

Because SCFAs, especially acetate, have anti-inflammatory properties, we hypothesized that SCFA production by B. longum 5 1A could modulate allergic lung inflammation. Thus, we measured SCFAs in the probiotic culture medium and in mouse feces. Our data indicated that acetate was the SCFA quantified in the bacterial medium (**Figure 4A**), whereas we did not find measurable levels of propionate or butyrate. In addition, B. longum 5 1A treatment significantly increased the levels of acetate in the feces of re-challenged OVA allergic mice compared to non-treated mice (**Figure 4B**). Butyrate was also measured but did not differ between groups (unpublished data). Since Treg cells play a role in the control of asthma, and SCFAs interfere with the functional activity of Tregs, we examined the levels of Tregs in the BALF of the re-challenged OVx allergic mice previously treated with B. longum. The re-challenged OVx allergic mice supplemented with the probiotic exhibited a significant increased percentage of Tregs in BALF compared with non-treated mice (**Figure 4C**).

## Oral Probiotic Supplementation after Ovariectomy Does Not Attenuate the Exacerbated Lung Inflammation and Airway Hyperresponsiveness of Re-challenged Allergic Mice

In this set of experiments, we investigated the effect of probiotic supplementation on exacerbated lung inflammation and airway hyperresponsiveness after ovariectomy, when lung inflammation is established. The mice were supplemented with B. longum 5 1A after ovary removal. As shown in **Figure 5**, when the probiotic was given after ovary removal, the profile of total inflammatory cells and eosinophils recovered in the BAL was not altered in the re-challenged allergic mice relative to the non-treated re-challenged OVx allergic mice (**Figure 5**).

## DISCUSSION

Clinical evidence shows that after puberty, asthma is more frequent in females than in males, which indicates that the oscillation of female sex hormones during the menstrual cycle is involved (Leynaert et al., 2012). In addition, asthma symptoms worsen in menopausal women (Keselman and Heller, 2015). Previous studies have investigated the beneficial effects of probiotics on allergic lung disease (Abrahamsson et al., 2007, 2009, 2011). However, the effect of probiotics on asthma symptoms or exacerbation during the perimenstrual and menopausal phases in women remains clinically debated. Many studies have shown that some Bifidobacterium levels are decreased in the presence of allergic disease (MacSharry et al., 2012). Menopause precedes senescence (Kirkwood and Shanley, 2010), and a negative correlation between senescence and Bifidobacterium use has been described in clinical studies (Arboleya et al., 2016). Additionally, B. longum has been associated with protection against asthma in clinical and experimental studies (Iwabuchi et al., 2009; Akay et al., 2014). The effects of B. longum 5 1A on allergic disease in particular have not been investigated yet. However, it has been demonstrated that oral B. longum 5 1A treatment resolved the inflammation due to lung infection caused by Klebsiella pneumoniae in mice (Vieira et al., 2016) and reduced the inflammatory response in an experimental murine model of gout (Vieira et al., 2015). Additionally, clinical studies have shown that B. longum 5 1A has an effect on pediatric functional constipation (Guerra et al., 2011).

In this study, we investigated the role of probiotic treatment of ovariectomized re-challenged allergic mice. The study was performed to evaluate the role of probiotics in asthma exacerbation during menopause. First, we confirmed that re-challenging ovariectomized allergic mice exacerbated Th2 mediated airway inflammation and airway reactivity, strongly suggesting that sex hormone deficiency after ovary removal modulates the severity of the asthma phenotype. Accordingly, after re-exposure to the antigen challenge, ovariectomized allergic mice showed exacerbated allergic inflammation, airway hyperresponsiveness, and excess secretion of mucus and the cytokine IL-5. Next, we treated the mice with B. longum 5 1A before or after the induction of allergic reaction and ovariectomy to evaluate the preventive and therapeutic effect of the probiotic in this model of allergic lung inflammation. Our data demonstrated that B. longum 5 1A prevented the exacerbation of Th2-mediated airway inflammation induced in re-challenged ovariectomized allergic mice. As the probiotic treatment began 15 days before the antigen sensitization and continued until the day of ovariectomy, we suggest that the probiotic B. longum 5 1A has a preventive effect on the exacerbation of allergic lung inflammation. In fact, our data revealed a significantly decreased eosinophil influx into the airways. In addition, reduced airway hyperresponsiveness, IL-5 levels, and mucus production were also observed. This is the first study investigating the effect of B. longum 5 1A on asthma exacerbation during menopause. Since allergic asthma is mediated by Th2 cytokines, and our data revealed an effective involvement of B. longum 5 1A in the reduction of airway hyperreactivity and eosinophil recruitment into the lungs, we hypothesized that the treatment of mice with the probiotic shifted the Th1/Th2 balance toward a Th1 profile. Indeed, in the present study, we found an association between B. longum 5 1A intake and an augmented release of the classical Th1 cytokine IFN-γ as well as decreased release

of the Th2 cytokine IL-5 by the lungs of re-challenged allergic mice. Considering that B. longum 5 1A treatment increased the number of regulatory T cells recovered in the BALF of re-challenged allergic mice, we note a possible role of B. longum in the induction of allergen-specific tolerance or immune suppression by regulatory T cells. Our data are in line with studies (Kwon et al., 2010) showing a suppressive effect of some probiotics on Th2 cytokines with a concomitant increased stimulation of Th1 cytokines. In this context, probiotic strains such as Lactobacilli and Bifidobacteria recognizably affect regulatory T cell development, modulating the Th1/Th2 balance. Experimental and clinical evidence has shown that fermentation of fiber by microbiota or probiotics mediates the production of SCFAs, such as acetate (De Filippo et al., 2010; Tan et al., 2014; Trompette et al., 2014). To verify the interaction between probiotic intake and the generation of SCFAs in our model of re-challenged allergic mice, we measured SCFAs in the feces of these mice. We found that B. longum 5 1A intake was positively correlated with the levels of acetate in feces from re-challenged ovariectomized allergic mice. Interestingly, acetate produced by the gut microbiota has been associated with an increased number of regulatory T cells in the lungs in a murine model (Thorburn et al., 2015). It already know that mice deficient in the short chain fatty acid receptor, G-protein coupled receptor 43 (GPR43) show exacerbated asthma response. In addition, acetate can regulate regulatory T cells by interfering with gene transcription, resulting in inhibition of histone deacetylases (Tao et al., 2007; Thorburn et al., 2015). Recently, it was also described that propionate, a SCFA, promotes the development of tolerogenic DCs to the draining mesenteric lymph nodes inducing the development of Treg cells (Trompette et al., 2014). This interaction between distal organ, as lung, and gut microbiota is currently named such as the lung-gut axis (Marsland et al., 2015). Besides this interaction between pulmonary and gut immunity happen due to the production of SCFAs by the gut microbiota, the oral microbiota serves as an inoculum for the intestine (Arimatsu et al., 2014), and possible lungs (Venkataraman et al., 2015; Wu and Segal, 2017). It has suggested that lung microbiota might be transiently recolonized through microaspiration and breathing (Venkataraman et al., 2015). Thus, gut microbiota modulation can be an important strategy to treat lung diseases.

Our data show that probiotic treatment of allergic mice started 1 day after ovariectomy and maintained until the last antigen challenge did not alter the profile of inflammatory cells recruited into the lungs. Thus, protection is not given when therapies are initiated after the disease has been established. Recently, it was demonstrated that preventive rather than therapeutic treatment with a high-fiber diet attenuates clinical and inflammatory markers of acute and chronic DSS-induced colitis in mice (Silveira et al., 2017). These results suggest that any therapy aimed at modifying the gut environment (e.g., prebiotic or probiotic strategies) should be given early in the course of disease. It is also possible that the effectiveness of the probiotic B. longum on the control of allergic lung inflammation may involve the activity of sex hormones. There are some bacteria affected by sex hormones; for example, Prevotella intermedia takes up estradiol and progesterone, which enhance its growth (Kornman and Loesche, 1982). These findings deserve further investigation.

We have not investigated the specific roles of estradiol in this study, but data from the literature show that the role of estradiol in allergic disease is very complex. Data from our group show that removal of the ovaries 7 days before sensitization to OVA significantly inhibited lung eosinophilia and IL-5 levels in lung lavage fluid (de Oliveira et al., 2007). However, if ovaries are removed 1 day before sensitization to OVA, lung eosinophilia and IL-5 levels are increased (Riffo-Vasquez et al., 2007). These data suggest that sex hormone levels during sensitization are relevant to asthma. Considering

that the worsening of pulmonary function observed in clinical studies during the perimenstrual and menopausal phases is a consequence of asthma, we investigated the repercussions of reexposure to the antigen after removal of the ovaries from animals with asthma already triggered. The experimental model proposed here is novel and important for studying therapeutic alternatives to treat asthma exacerbation during the perimenstrual and menopausal phases.

We examined microbiota composition and diversity 10 days after ovariectomy and, after this period we observed no differences in the microbiota diversity between the sham and ovariectomized groups or between the pre- and postsurgery groups. However, hierarchical clustering of rarefied OTU abundances clustered all samples from the ovariectomized group together in one main tree branch, while samples from the other groups were scattered across various branches. These results suggest that the microbiota in the ovariectomized animals may have a similar composition that would cluster them separately from the other groups. Importantly, the fecal microbiome was evaluated here is a single point, and this could maybe a too short period to observe significant differences in microbiota composition.

We also observed that mice in the same experimental group showed clear variations in gut microbiota composition, though they were housed in the same facilities and treated under the same conditions. Mice with the same maternal origin but from different litters do show some differences in gut microbiota composition (Ubeda et al., 2012). The mice analyzed in our experiments were not from the same litter or mothers and maybe this could explain some of the intragroup discrepancies observed here. It is also known that male and female- derived microbiota exhibit distinct circadian rhythmicity, with females showing a more obvious oscillation than males (Liang et al., 2015). In this study, we always collected feces at the same time, in the morning, to attenuate a possible circadian influence. Another important finding was that the ovariectomized group showed increased levels of bifidobacteria. Clinical studies have shown some differences in bifidobacteria species between atopic and healthy children (Stsepetova et al., 2007). B. adolescentis was isolated from allergic infants, and B. infants and B. bifidum were isolated from healthy infants (He et al., 2001; Stsepetova et al., 2007; Melli et al., 2016). Studies in vitro and in vivo

#### REFERENCES


have shown that the ability of Bifidobacterium to stimulate the immune system is species specific (Menard et al., 2008). B. adolescentis is an example of a bacteria that may not regulate the immune system in some situations, and it is not an important acetate producer (Menard et al., 2008). Due to the limitations of our microbiota analyses, we did not evaluate which specific strain of bifidobacteria was increased in ovariectomized mice.

In summary, our data show that the preventive supplementation of the probiotic B. longum 5 1A clearly attenuated the exacerbated airway inflammation and hyperresponsiveness in re-challenged ovariectomized allergic mice (**Figure 6**). However, the probiotic protection was lost when treatment was initiated after ovariectomy and disease was already established. These results suggest that therapeutic strategies to modulate the gut microbiota have the potential to prevent asthma exacerbation in the perimenstrual and menopausal phases. However, these strategies should be implemented before sex hormone levels decline.

## AVAILABILITY OF SUPPORTING DATA

Nucleotide sequences used for this study have been deposited in the SRA under accession, SRP105204.

### AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: CF, WT-L, ED-N, RC, and NC. Performed the experiments: EM, BA, AT, AF, AC, GM, and CF. Analyzed the data: EM, CF, TB, JS, and AT. Contributed reagents/materials/analysis tools: CV, WR, RC, and FM. Wrote the paper: CF, WT-L, ED-N, and RC.

### ACKNOWLEDGMENTS

This study was funded by São Paulo Research Foundation (FAPESP), project number 2012/50410-8 to CF, 2012, and CNPq (The National Council for Scientific and Technological Development), project number 486037/2012-6 to CF.


ovariectomized mouse model. J. Cell Physiol. 229, 1822–1830. doi: 10.1002/jcp. 24636


writing committee on the reformulation of the AIN-76A rodent diet. J. Nutr. 123, 1939–1951.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Mendes, Acetturi, Thomas, Martins, Crisma, Murata, Braga, Camâra, Franco, Setubal, Ribeiro, Valduga, Curi, Dias-Neto, Tavares-de-Lima and Ferreira. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Multivariate Analysis As a support for Diagnostic Flowcharts in Allergic Bronchopulmonary Aspergillosis: A Proof-of-concept study

*Joana Vitte1,2\*, Stéphane Ranque3,4, Ania Carsin2,5, Carine Gomez 4,6, Thomas Romain1 , Carole Cassagne3,4, Marion Gouitaa7 , Mélisande Baravalle-Einaudi <sup>5</sup> , Nathalie Stremler-Le Bel <sup>5</sup> , Martine Reynaud-Gaubert 4,6, Jean-Christophe Dubus 4,5, Jean-Louis Mège1,4 and Jean Gaudart <sup>8</sup>*

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Fernanda Ferreira Cruz, Federal University of Rio de Janeiro, Brazil Anita Hilda Straus, Federal University of São Paulo, Brazil Ashok K. Chaturvedi, University of Texas at San Antonio, United States*

> *\*Correspondence: Joana Vitte jvitte@ap-hm.fr*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 14 May 2017 Accepted: 08 August 2017 Published: 22 August 2017*

#### *Citation:*

*Vitte J, Ranque S, Carsin A, Gomez C, Romain T, Cassagne C, Gouitaa M, Baravalle-Einaudi M, Bel NS-L, Reynaud-Gaubert M, Dubus J-C, Mège J-L and Gaudart J (2017) Multivariate Analysis As a Support for Diagnostic Flowcharts in Allergic Bronchopulmonary Aspergillosis: A Proof-of-Concept Study. Front. Immunol. 8:1019. doi: 10.3389/fimmu.2017.01019*

*1Aix-Marseille Univ, APHM Assistance Publique Hôpitaux de Marseille, Hôpital de La Conception, Laboratoire d'Immunologie, Marseille, France, 2Aix-Marseille Univ, UMR INSERM 1067 CNRS 7333, Marseille, France, 3Aix-Marseille Univ, APHM Assistance Publique Hôpitaux de Marseille, Hôpital Timone, Laboratoire de Parasitologie, Marseille, France, 4Aix-Marseille Univ, URMITE, UMR 63, CNRS 7278, INSERM U1095, IRD 198, Marseille, France, 5Aix-Marseille Univ, APHM Assistance Publique Hôpitaux de Marseille, Hôpital Timone Enfants, Pneumo-pédiatrie, Centre de Ressources et de Compétences en Mucoviscidose, Marseille, France, 6Aix-Marseille Univ, APHM Assistance Publique Hôpitaux de Marseille, Hôpital Nord, Centre de Ressources et de Compétences en Mucoviscidose, Marseille, France, 7Aix-Marseille Univ, APHM Assistance Publique Hôpitaux de Marseille, Hôpital Nord, Service de Pneumologie, Marseille, France, 8Aix Marseille Univ, IRD, INSERM, SESSTIM UMR 912, Faculté de Médecine campus Timone, Marseille, France*

Molecular-based allergy diagnosis yields multiple biomarker datasets. The classical diagnostic score for allergic bronchopulmonary aspergillosis (ABPA), a severe disease usually occurring in asthmatic patients and people with cystic fibrosis, comprises succinct immunological criteria formulated in 1977: total IgE, anti-*Aspergillus fumigatus* (*Af*) IgE, anti-*Af* "precipitins," and anti-*Af* IgG. Progress achieved over the last four decades led to multiple IgE and IgG(4) *Af* biomarkers available with quantitative, standardized, molecular-level reports. These newly available biomarkers have not been included in the current diagnostic criteria, either individually or in algorithms, despite persistent underdiagnosis of ABPA. Large numbers of individual biomarkers may hinder their use in clinical practice. Conversely, multivariate analysis using new tools may bring about a better chance of less diagnostic mistakes. We report here a proof-of-concept work consisting of a threestep multivariate analysis of *Af* IgE, IgG, and IgG4 biomarkers through a combination of principal component analysis, hierarchical ascendant classification, and classification and regression tree multivariate analysis. The resulting diagnostic algorithms might show the way for novel criteria and improved diagnostic efficiency in *Af*-sensitized patients at risk for ABPA.

Keywords: allergic bronchopulmonary aspergillosis, *Aspergillus fumigatus*, immunoglobulin, molecular allergens, multivariate analysis

**Abbreviations:** ABPA, allergic bronchopulmonary aspergillosis; *Af*, *Aspergillus fumigatus*; CART, classification and regression tree; CF, cystic fibrosis; HAC, hierarchical ascendant classification; Ig, immunoglobulin; PCA, principal component analysis; s, specific.

## INTRODUCTION

*Aspergillus fumigatus* (*Af*) is a ubiquitous airborne fungus clinically relevant in asthmatic, cystic fibrosis (CF), and immunosuppressed patients. *Af*–human host interaction spans asymptomatic immunization with detectable immunoglobulin G to *Af*, sensitization with detectable IgE to *Af*, through severe pulmonary or systemic diseases such as allergic bronchopulmonary aspergillosis (ABPA), chronic pulmonary aspergillosis, invasive aspergillosis, and chronic lung allograft disease (1).

Diagnosis of *Af*-related disease relies on a body of clinical, radiological, immunological and mycological evidence. Among *Af*-related diseases in humans, ABPA or Hinson and Pepys' disease (2, 3) is the most frequent, preferentially targeting CF and asthmatic patients. The estimated prevalence of ABPA is 8–10% (range 1–25%) among CF patients and 1–2% among asthmatic ones (4–6). Lung transplantation does not completely prevent ABPA (7). ABPA diagnosis is particularly arduous in CF patients, as they may experience cough, bronchospasm, intercurrent infections, radiological abnormalities, *Af* colonization, and/or sensitization without true ABPA (4). In addition, mere colonization or sensitization by *Af* has been shown to contribute to the deterioration of respiratory function, in pediatric and adult CF patients (6, 8–11). A chronic disease made up of a succession of flare up and remission, ABPA has a major social burden impact, threatening people with asthma, 334 million people worldwide (12) and CF, around 70,000 people (13).

The diagnostic score for ABPA, established in 1977 and updated in 2013, includes succinct immunological criteria: total IgE, anti-*Af* IgE, anti-*Af* "precipitins," and anti-*Af* IgG (14). Progress achieved over the last four decades in the understanding and workup of *Af*—immune response cross talk led to the identification of multiple IgE and IgG biomarkers, with quantitative, standardized, molecular-level reports (15–17). Yet, the newly available biomarkers are not included in ABPA diagnostic criteria, either individually or in algorithms, despite favorable reports (1) and persisting underdiagnosis of ABPA (4). This prompted us to apply multivariate statistical analysis to *Af*-related immune biomarkers in search for a discriminant yet doctor-friendly diagnostic tool. We present here a proof-of-concept work on the potential benefit of multivariate algorithms for the diagnosis of ABPA.

## MATERIALS AND METHODS

The statistical analysis was performed retrospectively using flowcharts and laboratory data from 39 ABPA-free CF patients with detectable sIgE to *Af* extract (*Af*-CF group) and 10 ABPA patients (7 asthmatics, 3 CF; 2 children) from the Adult and Pediatric Regional Centers for Cystic Fibrosis (RCCF) and the Pulmonology Departments of Marseille, Southern France. Patient demography and detailed sIgE and sIgG4 data were previously described (17). Data from one ABPA patient (CF background, 12-year-old male) diagnosed with ABPA in July 2016 were included.

sIgE, sIgG, and sIgG4 for *Af* extract and recombinant allergens Asp f 1, Asp f 2, Asp f 3, Asp f 4, and Asp f 6 were measured with Thermo Fisher ImmunoCAP (Uppsala, Sweden). The detection thresholds were 0.01 mgA/L for sIgG and sIgG4 and 0.11 kUA/L for sIgE. In 44 of these 49 patients, determination of IgG to *Af* was also performed with an ELISA kit (Orgentec, Trappes, France), and *Af* precipitins were evaluated by immunoelectrophoresis (Sebia, Evry, France). ABPA patients were assayed at the initial diagnosis of ABPA or during a subsequent flare. The diagnosis of ABPA relied on (i) acute or subacute pulmonary function deterioration in patients with asthma or CF; (ii) total IgE levels of 500 kIU/L or higher; (iii) elevated levels of specific IgE and either sIgG or precipitins to *Af*; and (iv) chest or computed tomographic pulmonary infiltrates, with pathognomonic high attenuation mucous plugging. Therapeutic unresponsiveness to antibiotics, followed by resolution under corticosteroid treatment, was an additional criterion. Skin prick test reactivity for *Af* was not assessed because of discontinued availability of fungal *Af* extracts.

#### Ethics Statement

Recombinant *Af* allergen sIgE and sIgG (4) determination was part of the regular medical care since their commercial release. Patients received written laboratory workup reports. The study was based on a retrospective, non-interventional review of medical charts and laboratory results. According to the French law (18, 19), ethical approval and patient consent were not necessary for this type of study, while patients were informed and retained the right to oppose the use of their anonymized medical data for research purposes.

### STATISTICAL ANALYSIS

First, we performed a hierarchical classification on principal components, as previously described (20, 21). All the variables (biomarker measures) were included in this analysis and were equally weighted. A principal component analysis (PCA) was the first preprocessing step to explore the biomarker on the mixed data set, which takes into consideration relationships between biomarkers. Then, the coordinates of each variable in the first 20 principal components, which summarize 95% of the information, were used to perform a hierarchical ascendant classification (HAC). This method provides classes according to the immunological profile using an objective non-supervised classification technique that allows classifying independently from diagnosis. Furthermore, using the first 20 principal components is a way of denoising the data, thus yielding a more robust classification (21).

Next, we performed a classification and regression tree (CART) multivariate analysis. CART is a supervised, non-parametric and non-linear regressive approach (22), which classified the patients according to the outcome binary variable ABPA or *Af*-CF. Among all covariates, CART analyzed each possible threshold to split the sample in two opposite homogeneous groups. This process was recursively repeated until an optimal criterion was reached. The process enabled a tree to be built in which the terminal classes were groups with common biomarker findings. Statistical analysis was performed using R2.13.0 (R Foundation for Statistical Computing; http://cran.r-project.org/).

#### RESULTS

The *PCA-HAC* approach identified three clusters within the population, with a homogenous cluster 1 comprising 34/39 *Af*-CF patients, cluster 2 comprising 6/10 ABPA patients, and cluster 3 including both ABPA and *Af*-CF patients (**Figure 1**).

Classification and regression tree multivariate analysis method applied to current biomarkers of ABPA (total IgE, sIgE to *Af*,

Figure 1 | Hierarchical ascendant classification on principal component analysis (PCA) of the study sample. *X*- and *Y*-axes are the two first dimensions issued from the PCA. Patients are denoted 1–10 for the allergic bronchopulmonary aspergillosis (ABPA) group and 11–49 for the *Aspergillus fumigatus* (*Af*)-cystic fibrosis (CF) group. *Af*-sensitized patients without ABPA (*Af*-CF) are mostly found in the homogenous cluster 1 (34/39), while ABPA patients are mostly found in cluster 2 (6/10).

ELISA IgG to *Af*, and *Af* precipitins) resulted in a classification based on three parameters only: total IgE, ELISA IgG, and sIgE. Together, these three results allowed correct classification of 38/39 *Af*-CF patients and 3/10 ABPA patients. The remaining seven ABPA patients were correctly classified with a probability of 88% (Figure S1 in Supplementary Material).

Classification and regression tree was sequentially performed on datasets of either sIgE, IgG or IgG4 responses to *Af* molecules Asp f 1, Asp f 2, Asp f 3, Asp f 4, and Asp f 6. The molecular sIgE dataset allowed proper classification of 35/39 *Af*-CF patients and 8/10 ABPA, the latter with an 89% probability (Figure S2 in Supplementary Material).

Datasets of either IgG or IgG4 responses to molecules Asp f 1, Asp f 2, Asp f 3, Asp f 4, and Asp f 6 resulted in 70–80% of incorrect classification of ABPA patients (not shown), in line with previous IgG4 results (17).

Finally, the classification tree built using the three molecular datasets together (sIgE, sIgG, and sIgG4) performed better, and adequately classified 35/39 *Af*-CF and 7/10 ABPA patients. The number of equivocally classified *Af-*CF or ABPA patients was reduced to 4 and 3, respectively (**Figure 2**).

## DISCUSSION

The present study shows that multivariate analysis approaches can be applied to the analysis of complex *Af*-induced immune

response patterns and hold promise for improving diagnostic discrimination between ABPA and ABPA-free patients.

While the multivariate analysis of the current diagnostic criteria (total IgE, sIgE to *Af*, ELISA IgG to *Af*, and *Af* precipitins) correctly classified 38/39 of *Af*-CF patients, only 3/10 ABPA patients were correctly classified (Figure S1 in Supplementary Material). This result suggests that current criteria perform well at identifying ABPA-free patients and excluding ABPA diagnosis but are not optimally efficient for ABPA diagnosis. Shifting from sIgE against *Af* extracts to their molecular counterparts improved ABPA identification. Indeed, the best performance for identifying ABPA cases (8/10 correctly classified patients, Figure S2 in Supplementary Material) was obtained with multivariate analysis of molecular sIgE responses alone, a finding that supports the prominent pathophysiological and diagnostic significance of sIgE responses in ABPA.

The best compromise for identifying both *Af*-CF and ABPA patients was obtained with multivariate analysis of sIgE, IgG, and IgG4 against molecular Asp f 1, Asp f 2, Asp f 3, Asp f 4, and Asp f 6. The algorithm retained only four parameters (sIgE to Asp f 4, Asp f 1, and Asp f 6; IgG4 to Asp f 2) and correctly classified 35/39 *Af*-CF and 7/10 ABPA patients (**Figure 2**). This result is in line with previous reports on Asp f 4, Asp f 6, and Asp f 1 as ABPA biomarkers [reviewed in Ref. (1)].

In terms of overall diagnostic performance, multivariate analysis still needs improvement. One clue may come from the PCA results, which show that most *Af*-CF cases cluster together (34/39, cluster 1), but ABPA cases only partially do so (6/10, cluster 2, **Figure 1**). Biological heterogeneity resulting in a mixed *Af*-CF and ABPA cluster 3 needs further work. Differences might be underlain by sIgE or IgG (4) responses to further *Af* molecules currently not available. Conversely, it is likely that genetic, clinical, or radiological features not considered in our study may contribute to *Af*-CF and ABPA heterogeneity. Finally, increasing the size of the study population should improve the power of statistical analysis and yield more performant diagnostic flowcharts.

Taken together, our results add a diagnostic perspective to recent reports of multivariate analysis in allergic patients as a basis for personalized medicine (23). Further confirmation on large-scale populations and detailed analysis of variables are necessary. However, we believe that multivariate analysis of the complex *Af*-induced immune responses will pave the way for the

#### REFERENCES


discovery of clinically efficient diagnostic biomarkers and novel ABPA diagnostic algorithms.

## ETHICS STATEMENT

Recombinant *Af* allergen sIgE and sIgG (4) determination was part of the regular medical care since their commercial release. Patients received written laboratory workup reports. The study was based on a retrospective, non-interventional review of medical charts and laboratory results. According to the French law (18, 19), ethical approval and patient consent were not necessary for this type of study, while patients were informed and retained the right to oppose the use of their anonymized medical data for research purposes.

### AUTHOR CONTRIBUTIONS

JV, SR, and JG designed the research. AC, CG, MG, NB, MB-E, MR-G, and J-CD reviewed and collected clinical data. SR and CC interpreted and collected mycological data. TR and JV interpreted and collected immunological data. JG performed statistical analysis. JV, SR, J-CD, MR-G, J-LM, and JG wrote the manuscript. All the authors read and approved the final version of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu. 2017.01019/full#supplementary-material.

Figure S1 | Classification tree using the classical laboratory measures comprised in the current diagnostic score of allergic bronchopulmonary aspergillosis (ABPA) (redrawn). The diagnostic algorithm retains total IgE, ELISA IgG, and specific IgE, yielding 2 *Aspergillus fumigatus* (*Af*)-cystic fibrosis (CF)-only groups counting 38/39 patients, 1 ABPA-only group of 3 patients, and 1 mixed group of 8 patients with an 88% probability of having ABPA. Overall, 38/39 *Af*-CF patients, but only 3/10 ABPA patients are clearly identified.

Figure S2 | Classification tree analyzing sIgE responses to Asp f 1, Asp f 2, Asp f 3, Asp f 4, and Asp f 6 molecules. The resulting diagnostic algorithm retains IgE to Asp f 4, Asp f 1, and Asp f 6, yielding 1 *Aspergillus fumigatus* (*Af*)-cystic fibrosis (CF)-only group counting 35/39 patients, 1 mixed group of 9 patients with an 89% probability of having allergic bronchopulmonary aspergillosis (ABPA), and 2 mixed groups of undetermined clinical significance where other criteria are needed. Overall, 35/39 *Af*-CF patients, but no ABPA patients are clearly identified.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Vitte, Ranque, Carsin, Gomez, Romain, Cassagne, Gouitaa, Baravalle-Einaudi, Bel, Reynaud-Gaubert, Dubus, Mège and Gaudart. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Oral Bacterial and Fungal Microbiome Impacts Colorectal Carcinogenesis

#### Klara Klimesova\*, Zuzana Jiraskova Zakostelska and Helena Tlaskalova-Hogenova

Laboratory of Cellular and Molecular Immunology, Institute of Microbiology of the CAS, Prague, Czechia

Host's physiology is significantly influenced by microbiota colonizing the epithelial surfaces. Complex microbial communities contribute to proper mucosal barrier function, immune response, and prevention of pathogen invasion and have many other crucial functions. The oral cavity and large intestine are distant parts of the digestive tract, both heavily colonized by commensal microbiota. Nevertheless, they feature different proportions of major bacterial and fungal phyla, mostly due to distinct epithelial layers organization and different oxygen levels. A few obligate anaerobic strains inhabiting the oral cavity are involved in the pathogenesis of oral diseases. Interestingly, these microbiota components are also enriched in gut inflammatory and tumor tissue. An altered microbiota composition – dysbiosis – and formation of polymicrobial biofilms seem to play important roles in the development of oral diseases and colorectal cancer. In this review, we describe the differences in composition of commensal microbiota in the oral cavity and large intestine and the mechanisms by which microbiota affect the inflammatory and carcinogenic response of the host.

#### Edited by:

Yves Renaudineau, Université de Bretagne Occidentale, France

#### Reviewed by:

Matthias Hauptmann, Research Center Borstel, Germany Christian Furlan Freguia, Synthetic Biologics, Inc., United States

> \*Correspondence: Klara Klimesova

klimesov@biomed.cas.cz

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 28 September 2017 Accepted: 05 April 2018 Published: 20 April 2018

#### Citation:

Klimesova K, Jiraskova Zakostelska Z and Tlaskalova-Hogenova H (2018) Oral Bacterial and Fungal Microbiome Impacts Colorectal Carcinogenesis. Front. Microbiol. 9:774. doi: 10.3389/fmicb.2018.00774 Keywords: microbiome, mycobiome, pathobiont, dysbiosis, biofilm, Fusobacterium, oral diseases

### INTRODUCTION

Both the upper and lower parts of the human digestive tract harbor a complex ecosystem of bacteria, fungi, protozoa, and viruses, referred to as the microbiome. It begins to form even before birth, in the uterus, developing for another 2–3 years after birth to become a stable, fully functioning microbiome, until the physiological changes associated with senescence lead again to substantial shifts in its composition (Adlerberth and Wold, 2009; Aagaard et al., 2014; Maffei et al., 2017). The lower part of the digestive tract gets "inoculated" every day by about 10<sup>11</sup> bacteria from the oral cavity and microbial species detected in the oral and fecal microbiota overlap in about 45% of tested individuals (Socransky and Haffajee, 2002; Segata et al., 2012). Moreover, via the blood stream, these oral bacteria can disseminate all over the body. Fungal microbiota can colonize the gut perorally as well, with some strains detected in the gut likely to be contaminants from the environment or food, rather than commensals (Trojanowska et al., 2010). The composition and function of the microbiome change along the digestive tract, from the oral cavity to the rectum. These differences have been previously described in detail (Arumugam et al., 2011; Human Microbiome Project Consortium, 2012) and will be briefly outlined below.

Collectively, the genes encoded by the microbial genomes outnumber the genes in the human genome about 100-fold and this variation enables the commensal microbiota to use substrates indigestible by humans (Qin et al., 2010). Products of microbial metabolic activity include vitamins,

short-chain fatty acids (SCFAs), and other compounds important for host cell metabolism and survival. Moreover, the host's physical interaction with or sensing of the microbial components is important for proper mucosal barrier function and mucosal immune system development and homeostasis. On the other hand, recent studies have shown that some commensal microbes can under certain conditions become pathogenic – so called pathobionts. Mechanisms include expression of virulence factors, such as adhesion molecules or proteases, or formation of a biofilm, and such activity can lead to disease initiation or progression. One such example is Escherichia coli, a large and diverse group of various bacterial serotypes. E. coli strains differ in their activities and biological roles: some of them are gut commensals and commercially available probiotics, others can be pathogens causing gastrointestinal and urinary infections or pathobionts associated with inflammatory bowel disease (IBD) and colorectal cancer (CRC).

The jury is still out on what triggers this transformation of commensals into pathobionts; it might be that a change in the gut microenvironment simply allows the microbes to interact with the host in an aberrant way. The gut microbiota composition and function can be influenced by various factors (**Figure 1**). For instance, many substances produced by the host and secreted into gut lumen, such as antibacterial peptides, secretory IgA, mucins, cytokines, or neuromediators, can shape the microbial community. Their production depends on host gene polymorphisms and their expression in host cells is often driven by microbial stimulation, creating a positive feedback loop. Mechanisms responsible for cancer development include some of these factors but others are still under scientific investigation.

Colorectal cancer is the most prevalent type of cancer in developed and developing countries (Ferlay et al., 2015).

Only a small percentage of colorectal cancers is hereditary or associated with certain predisposing conditions, such as chronic intestinal inflammation in IBD patients. The majority of cases thus represents sporadic cancers (85–95%) and can be, to some extent, influenced by environmental factors. The composition and metabolic activity of gut microbiota may be therefore a crucial component of CRC pathogenesis (Arthur and Jobin, 2011).

Findings of significant colonization of cancer tissue by microbes usually found in the oral cavity have sparked a debate about a possible involvement of oral microbiota in CRC development process. Many experimental studies have provided evidence for a significant role of microbiota in carcinogenesis. However, due to the complexity of microbial cooperation and interaction with the host, all the underlying mechanisms are yet to be elucidated. Here, we review the properties of bacterial and fungal populations inhabiting the oral cavity and gut, with emphasis on their association with CRC pathogenesis.

## ORAL MICROBIOTA

The microorganisms found in the human oral cavity are referred to as the human oral microbiome and play a crucial role in maintenance of homeostasis in the mouth. Each individual's oral microbiome consists of a distinct set of microorganisms. The mouth supports one of the most diverse microbial communities compared with other body sites, such as the skin and vagina, due to its heterogeneity and the interrelationships between the different anatomic structures (Wade, 2013). The constantly humid environment of the oral cavity is maintained at a relatively stable temperature (34–36◦C), while the varying pH levels and different types of diet contribute to the substantial microbiome variability (Marcotte and Lavoie, 1998). Habitats of the oral cavity are represented by both hard (teeth) and soft tissues (cheek, tongue, lip, gingival sulcus, attached gingiva, hard and soft palate) and their interface (subgingival and supragingival margins, and gingival crevices around teeth). The contiguous extensions of the oral cavity, including the tonsils, pharynx, esophagus, Eustachian tube, middle ear, trachea, lungs, nasal passages, and sinuses, are also colonized by the oral microbiome but the majority of studies describing the oral microbiome composition include only samples from the oral cavity (Mager et al., 2003; Aas et al., 2005). Moreover, all these structures are constantly moistened by two physiological fluids, saliva and the gingival crevicular fluid, which help to maintain the oral microbiome homeostasis by providing water, nutrients, antibodies, and antimicrobial and adherence factors (Marcotte and Lavoie, 1998).

The National Institute of Health's Human Microbiome Project identified the most dominant phyla that account for over 95% of the entire oral microbiome: Firmicutes, with the Streptococcus as a dominant genus, Bacteroidetes, strongly represented by Prevotella and Proteobacteria, with highly abundant Haemophilus, Fusobacteria, and Actinobacteria (Human Microbiome Project Consortium, 2012; Huse et al., 2012; Zhou et al., 2013). The oral cavity displays greater alphadiversity (species richness) than either the skin or vagina,

characterized by uniform abundance of the major species (Huse et al., 2012). On the other hand, beta-diversity, i.e., differences in microbiome composition in oral sites among various subjects, is the lowest compared to the other body sites (Huse et al., 2012). Within the oral cavity, the highest microbiota richness has been found in the gingival plaque and in saliva, whereas the lowest richness has been described in keratinized gingiva (Huse et al., 2012; Zhou et al., 2013). It seems that saliva contributes unevenly to the microbial composition of different sites in the oral cavity. Due to its rapid turnover and low levels of nutrients, saliva itself does not contain a stable indigenous biota and owes its high alphadiversity primarily to bacteria shed from other oral tissues (Mager et al., 2003). In general, species like Streptococcus, Gemmella, Granulicatella, Veillonella, and Fusobacterium can be detected across almost all oral sites, while others are represented only at one or two oral sites, e.g., Prevotella, Bacteroides, Corynebacterium, Pasteurella, and Neisseria (Huse et al., 2012). Bacterial species in the oral cavity find the most stable environment at the supragingival or subgingival tooth surfaces. These non-shedding surfaces are covered by persisting biofilms composed mainly by Streptococcus and Actinomyces, representing the earliest colonizers of teeth, and Veillonella (Socransky and Haffajee, 2005; Teles et al., 2013). Taken together, although most microbial species found in the oral cavity differ in their abundance, their representation and activity is more important.

#### ORAL MICROBIOME AND ORAL DISEASES

A considerable number of oral conditions, including caries and periodontal diseases, endodontic infections, alveolar osteitis, and tonsillitis is connected to detrimental alteration in microbiota composition – a dysbiosis (Costalonga and Herzberg, 2014; Proctor and Relman, 2017). For instance, periodontal inflammation is directly induced by microbes colonizing the biofilm in the gingival sulcus and, at later stages, in the periodontal pockets. These oral biofilms contain 100s of different bacterial species in one periodontal lesion, with composition different from that in healthy periodontium (Socransky and Haffajee, 2005; Teles et al., 2013). Initially, the plaque forms only supragingivally but if it is not removed properly, after a few days it spreads below the gingival margin and into the sulcus. There, after the depletion of oxygen, a new environment is established, where anaerobic bacteria can flourish. Three most destructive anaerobic bacteria involved in severe periodontal disease, the so called "red complex," include Treponema denticola, Tannerella forsythia and Porphyromonas gingivalis (Socransky et al., 1998). Aggressive form of periodontitis is further associated with Aggregatibacter actinomycetemcomitans (Slots, 1976). The immunopathological mechanisms in periodontal disease development and course have been recently thoroughly reviewed by Hajishengallis and Korostoff (2017). Recurrent aphthous stomatitis is also connected to changes in microbiota composition; decreased abundance of Streptococcus salivarius and increased Acinetobacter johnsonii have been linked to the disease incidence (Bankvall et al., 2014; Kim et al., 2016).

As mentioned previously, the human microbiome diversity is not limited only to bacteria but also includes fungal species. However, this oral mycobiome have been only recently characterized and despite its potential great scientific importance, we found only few studies where its composition was analyzed using high throughput sequencing. Ghannoum et al. (2010) reported that healthy oral mycobiota contained 74 culturable and 11 non-culturable fungal genera. They have revealed great interindividual variation and proposed that the presence of certain fungal isolates (e.g., Candida, Aspergillus, Cryptococcus) probably predisposes the host to opportunistic infections. Malassezia species, previously described as commensals and pathogens of the skin and lungs, have been recently found as predominant commensals in saliva (Saunders et al., 2012; Dupuy et al., 2014). Many other species are likely still waiting to be discovered and were not detected earlier because of their special growth requirements (Nagano et al., 2010). The first evidence of interactions among members of the oral mycobiome community and their association with specific disease came from a study characterizing the oral mycobiome in HIV patients (Mukherjee et al., 2014). The authors found that a decrease in abundance of an indigenous fungus Pichia in uninfected individuals went hand in hand with an increase in abundance of Candida, suggesting an antagonistic relationship. Pichia inhibits Candida by different mechanisms, including competition for nutrients and secretion of factors that disrupt the latter's ability to adhere, germinate, and form biofilms. Moreover, they found a negative correlation between Candida and Campylobacter in HIV-infected subjects, whereas in healthy subjects, no correlation between Candida and bacterial species was detected (Mukherjee et al., 2014). Recently, Peters et al. (2017) published a pilot study describing oral mycobiome in healthy subjects and those with periodontal disease. In diseased subjects, they found a slightly increased abundance of Candida genera which is in agreement with previous culture-based studies (Urzúa et al., 2008; Canabarro et al., 2013). Despite those first studies aiming for a deeper understanding of the factors affecting the oral mycobiota composition, information about their direct and indirect effects on human health and the interactions between the fungi and bacteria is still lacking.

The involvement of microbiota in the pathogenesis of oral diseases has been documented using gnotobiotic rat and mouse models of human diseases. Germ-free animals can be obtained by delivering the young by sterile Cesarean section and raising them aseptically in isolators for germ-free rearing. Such animals made it possible to study the effects of commensal bacteria in the oral cavity, including the effects on periodontal tissues. In germ-free mice and rats, it was demonstrated that periodontal disease and caries, similarly to other human inflammatory diseases, cannot be experimentally induced in the absence of microbiota (Heijl et al., 1980; Tlaskalova-Hogenova et al., 2004). Experimental animal models (e.g., the gavage model of periodontal disease) and in vitro studies revealed that certain components of oral microbiota, mainly P. gingivalis, play a crucial role in the innate host defense of periodontium and that dysregulation of the immune response

in the presence of oral microbiota leads to inflammation and alveolar bone loss (Ivanyi et al., 1991; Darveau et al., 2012; Papadopoulos et al., 2013). The mechanisms by which microbiota triggers the pathological changes are not yet fully understood, however, new approaches promise to shed light on the role of oral microbiota (Kinane et al., 2017; Pitts et al., 2017).

### ORAL MICROBIOME AND EXTRA-ORAL DISEASES

Dysbiosis of the oral microbiome is not only connected with the incidence and maintenance of oral diseases, but has been also implicated in the pathogenesis of autoimmune, inflammatory, and neoplastic diseases (e.g., heart disease, respiratory illnesses, psoriasis, psoriatic arthritis, and carcinogenesis at various sites) (He et al., 2015; Roberts and Darveau, 2015; Egeberg et al., 2017; Ungprasert et al., 2017). Moreover, oral microbiota seems to affect the outcome of pathological pregnancy (preterm birth, abortions, etc.); reviewed by Cobb et al. (2017). Periodontal bacterial DNA has been found in atherosclerotic plaques of patients suffering from ischemic heart disease and atherosclerosis (Ford et al., 2005, 2006). Bacteria may initiate or exacerbate atherosclerotic processes through activation of innate immunity, direct involvement of mediators activated by dental plaque antigens in atheroma processes, or involvement of cytokines and heat shock proteins from dental plaque bacteria. There might be also genetic predisposing factors influencing both diseases (Bartova et al., 2014). Furthermore, patients with rheumatoid arthritis have a higher prevalence of periodontal disease and vice versa (Kasser et al., 1997; Greenwald and Kirkwood, 1999). Bacterial DNA has been detected in the synovial fluid of patients with rheumatoid arthritis or with failed prosthetic joints, suggesting the possibility of infection translocating from the periodontal tissue to the synovium (Temoin et al., 2012). The oral microbiome is not confined to spreading to contiguous epithelial surfaces, but can also be carried by the bloodstream to distant body sites, such as the heart, skin, and joints. Oral microbiota enters the bloodstream during routine daily activities like tooth brushing or through inflamed tissue in the course of oral diseases (Tomas et al., 2012). The mechanisms of dissemination of potentially pathogenic microbes from the oral cavity through bloodstream are still not clear. Potential connection with systemic low-grade inflammation has been also discussed (Potgieter et al., 2015).

In most conditions discussed so far, no particular microbe or microbes have been described as a causative agent. However, thanks to advances in molecular methods, presence of Fusobacteria and other oral bacteria has been recently demonstrated in various systemic pathological conditions, including digestive diseases, such as appendicitis, IBD and CRC. Moreover, a correlation of the presence of oral bacteria P. gingivalis and A. actinomycetemcomitans with an increased risk of developing pancreatic cancer has been observed in a large group of subjects with incident primary pancreatic adenocarcinoma (Fan et al., 2016). Several recent studies have repeatedly confirmed the presence of oral bacteria in the gut, especially in association with CRC mucosa. For instance, Nakatsu et al. (2015) have shown that a substantial part of gut microbiome associated with CRC is composed of oral bacteria and Fusobacterium in particular. Momen-Heravi et al. (2017) also proposed a role of oral bacteria in CRC development. They conducted a retrospective study in a huge cohort of women and discovered an association of periodontal disease and tooth loss with CRC morbidity and found that women with less than 17 teeth may be at a greater risk of incident CRC (Momen-Heravi et al., 2017).

#### ORAL MICROBIOTA AND COLORECTAL CARCINOGENESIS

It is generally accepted that the gut microbiome plays a role in CRC development. Experimental proof of gut microbiota involvement in CRC development came from gnotobiotic animal models (Reddy et al., 1975; Vannucci et al., 2008; Arthur and Jobin, 2011; Klimesova et al., 2013; Tlaskalova-Hogenova et al., 2014). Recent studies using next generation sequencing and polymerase chain reaction have shown that Fusobacterium nucleatum is frequently detected in stool and biopsy samples from CRC patients (Castellarin et al., 2012; Kostic et al., 2012; Flanagan et al., 2014). These adhesive, anaerobic Gram-negative bacteria attract considerable attention in search for possible mechanisms behind their inflammatory and tumorigenic activity. Some of the features of F. nucleatum and the host responses are already known: F. nucleatum modulates E-cadherin/β-catenin signaling via its FadA adhesin/invasin – a key virulence factor, and alters macrophage infiltration and methylation of the CDKN2A promoter in CRC lesions (Rubinstein et al., 2013; Park et al., 2017). It seems that Fap2 Gal-GalNAc lectin of F. nucleatum could be responsible for its tendency to bind to tumor cells displaying Gal-GalNAc moieties (Abed et al., 2016). Moreover, several other virulence proteins that might participate in inflammatory and neoplastic processes have been described in F. nucleatum, using proteomic approaches (Zanzoni et al., 2017). In the host, F. nucleatum activates numerous immune responses including human β-defensin production, lymphocyte apoptosis, and production of proinflammatory cytokines interleukin (IL)-6, IL-8, and TNF-α (Han, 2015).

The fact that Fusobacterium is highly abundant in patients with CRC led to various efforts to apply this finding for clinical purposes. For instance, a highly sensitive DNA test for F. nucleatum has been developed for screening and prognosis of CRC in Japan (Yamaoka et al., 2017). Fusobacterium has been recently shown to predict the aggressiveness and recurrence of CRC and its resistance to chemotherapy. It seems to be modifying the innate immune signaling and regulating specific microRNAs that activate the autophagy pathway (Yu et al., 2017). In connection with this finding, an argument has been made to use anti-F. nucleatum therapy together with chemotherapy.

The dietary patterns leading to CRC development have been studied for several decades. Recently, it was shown that individuals consuming a western-type diet have a higher incidence of Fusobacterium-associated CRC and that diets rich

in whole grains and dietary fiber are associated with a lower risk of F. nucleatum-positive CRC (Mehta et al., 2017).

Despite the clinical observation of increased abundance of Fusobacterium in patients with CRC (especially in chemoresistant forms of CRC), direct clinical evidence of a causal relationship is still lacking. A recent study revealed that fusobacterial abundance is not significantly increased in fecal samples of patients with adenomas, implying that the relationship between Fusobacterium and CRC might not be causal after all. Amitay et al. (2017) hypothesize that Fusobacterium is more likely just a passenger colonizing the favorable niche in a gut with CRC, rather than the driver of cancer development. Experimental evidence of its potential causal role is based on in vitro and animal models of CRC. For instance, it has been shown that F. nucleatum potentiates tumorigenesis in monoassociated Apcmin/<sup>+</sup> mice (Kostic et al., 2013).

While Fusobacterium is the most studied periodontal microbe in connection to CRC, other components of the oral microbiome may also be implicated in CRC pathogenesis and will be discussed later (**Table 1**).

#### GUT MICROBIOME, METABOLIC ACTIVITY, AND COLON CARCINOGENESIS

Thanks to advances in high-throughput sequencing and metabolomic approaches, we have a growing understanding of the composition and metabolic activity of the microbiome associated with colorectal carcinogenesis. Studies have shown that gut microbiome differs between healthy individuals and adenoma/carcinoma patients and that microbial diversity in cancer is reduced (Peters et al., 2016). The adenoma-carcinoma sequence in CRC development suggests an associated continuous alteration of the resident microbiome, which is supported by recent research. Analyses of fecal microbiome composition in patients with adenoma have shown increased normalized abundance of genera such as Actinomyces, Corynebacterium, Porphyromonas, Mogibacterium, and Haemophilus when compared with healthy individuals (Peters et al., 2016; Hale et al., 2017). Cancer patients have shown further differences in fecal microbiota composition, with a marked enrichment of Ruminococcus, Oscillibacter, and Roseburia, and Porphyromonas, Fusobacterium, and Peptostreptococcus, i.e., strains associated with periodontal disease (**Table 1**) (Shen et al., 2010; Flemer et al., 2017; Liang et al., 2017).

Moreover, fecal samples reflect the microbial colonization of tissues, as biopsies from adenomas and carcinomas have been similarly different from healthy mucosa (Flemer et al., 2017). The findings that there is very little difference in microbiota composition between diseased and adjacent unaffected tissue suggest that the microbiome undergoes a systemic change, affecting the whole community (Lu et al., 2016; Flemer et al., 2017). Such results support the driver/passenger model, the idea that certain strains disturb the microbial community and the mucosal microenvironment (drivers) and such changes lead to subsequent colonization by pathobionts and pathogens TABLE 1 | Oral microbiota and its possible mechanisms related to tumorigenesis.


(passengers) (Tjalsma et al., 2012). Both the drivers and the passengers modulate the local microenvironment through different means, such as virulence factors or metabolic activity. Drivers thus promote cancer initiation at the very beginning by their involvement in DNA damage, cell cycle regulation, apoptosis and epithelium proliferation, whereas passengers more likely promote tumorigenesis via chronic proinflammatory stimulation and direct tissue damage.

Drivers often include microbes that produce genotoxic substances, which damage DNA, or cyclomodulins, which can modulate the epithelial cell cycle – recently thoroughly reviewed by Gagniere et al. (2016) and El-Aouar et al. (2017). Bacteroides fragilis and Enterococcus faecalis have the potential to damage epithelial cells and initiate cancer formation by producing the enterotoxin fragilysin and reactive oxygen species, such as superoxide, respectively (Huycke et al., 2002; Toprak

et al., 2006). Several studies confirmed that E. coli strains encoding the genotoxic polyketide synthase (pks) island are associated with inflamed gut mucosa and CRC (Swidsinski et al., 1998; Nougayrede et al., 2006; Arthur et al., 2012; Raisch et al., 2014). Interestingly, members of the family Enterobacteriaceae, including E. coli, can produce several types of genotoxins or cyclomodulins. Cytotoxic necrotizing factors and cycle-inhibiting factor modulate the cell cycle and can lead to uncontrolled proliferation or cell cycle arrest, respectively (Taieb et al., 2006; Miraglia et al., 2007). Cytolethal distending toxins, similarly to pks, induce DNA double-strand breaks and apoptosis, but can also promote proinflammatory cytokine production in the host (Blazkova et al., 2010). In summary, the interplay of these factors with gut epithelium and immune cells can promote low-grade inflammation and cancer initiation. Moreover, we can assume that these molecules represent just the tip of the iceberg of as yet unknown microbial products of gut commensals with the potential to harm the gut epithelium.

Mycobiome, i.e., the fungal microbiome, forms an integral part of the gut microbial community, although it is much less investigated then the bacterial part. The most common genera residing in a healthy gut are Candida, Saccharomyces, and Cladosporium (Hoffmann et al., 2013). However, some non-commensal transient fungi, acquired with food or from the environment, can be also found in fecal samples and may comprise potentially pathogenic species. Trojanowska et al. (2010) have shown that the gut is colonized also by the oral mycobiome, as they found a genetically identical Candida albicans strain in the mouth and colon of patients with IBD. Unfortunately, data about fungal colonization of the digestive tract in relation to neoplastic diseases are still sparse. A disruption of the bacterial and fungal community – dysbiosis, has been observed in individuals with IBD (Sokol et al., 2017), who are known to be at increased risk of CRC development. Interestingly, reduced richness and diversity has been detected not only in bacterial, but also in fungal microbiome (Chehoud et al., 2015; Liguori et al., 2016; Sokol et al., 2017). For instance, the Cystofilobasidiaceae family, Dioszegia genus and Candida glabrata have been found to be enriched in Crohn's disease compared with healthy mucosa (Liguori et al., 2016). The only published study on fungal microbiota in CRC deals with comparison of adenomas and adjacent tissues. Luan et al. (2015) have observed an increased abundance of Phoma and Candida genera and Candida tropicalis in adenomas. As pathobionts, these genera may be involved in cancer initiation, but further studies are needed to investigate whether they work as drivers or passengers.

Microbiome in the colon makes use of various catabolic and anabolic pathways, which enable it to utilize a broad spectrum of substrates that are not absorbed in the small intestine. These pathways interact with the metabolism of xenobiotics and influence micronutrient bioavailability, lead to the production of essential vitamins and degradation of fibers, and regulate the secretion of various molecules (Arthur and Jobin, 2011). Different dietary components can shift the microbiome composition. For instance, a diet high in resistant starch increases the abundance of bacteria metabolizing non-digestible polysaccharides (Walker et al., 2011). Indeed, increased abundance of Prevotella and Bacteroides has been observed in individuals preferring high sugar and high protein diet, respectively (Wu et al., 2011). Our digestion pathways lack the enzymes for the degradation of resistant starch and dietary fiber but the distal gut microbiome encodes about 81 different families of glycoside hydrolases (bacterial polysaccharidases, glycosidases), which are not present in the human genome (Gill et al., 2006). The microbiome thus significantly contributes to the utilization of starch, primary fiber, host-derived secretions (mucus glycans), sucrose, and monosaccharides.

Subsequent fermentation of depolymerized molecules leads to the production of SCFAs, mainly acetate, propionate, and butyrate. Compared with other microbiomes in gene libraries, the human gut microbiome is enriched with genes involved in the pathways generating SCFAs (Gill et al., 2006). Main producers of butyrate within the human gut microbiome are Faecalibacterium prausnitzii and Eubacterium rectale/Roseburia group (Louis et al., 2010). SCFAs provide one of the most important sources of energy, not only for intestinal epithelial cells but also for muscles, kidneys, heart, and brain. Their physiologic production impacts the metabolism and transport through the epithelium, as well as epithelial cell renewal and differentiation. Moreover, SCFAs greatly influence the immune system, colonic functions, and carcinogenesis. Butyrate production, for example, improves gut barrier integrity and reduces local oxidative stress and inflammation (Macfarlane and Macfarlane, 2012). Recently, Kaiko et al. (2016) came with an interesting finding that butyrate levels are much lower at the intestinal crypt base than in the lumen. Differentiated enterocytes use butyrate as an energy source and thus reduce its concentration along the way to the lamina propria, where a low concentration of butyrate keeps the epithelial progenitors proliferating and stimulates tolerogenic immune response (Kaiko et al., 2016). The role of SCFAs in cancerogenesis is not fully understood but their concentration could be an important factor.

On the other hand, degradation and fermentation of dietary proteins, peptides, and amino acids by bacteria generates byproducts, such as phenols, indoles, ammonia, amines, and hydrogen sulfite, all of which are to some extent harmful to the host, being co-carcinogens, mutagens, and cellular toxins (Macfarlane and Macfarlane, 2012). Moreover, hydrogen released as the end-product of fermentation is processed by methanogenic species of Archaea (e.g., Methanobrevibacter) to methane, which changes local conditions (redox potential and pH) and thus regulates biochemical pathways (Gill et al., 2006).

Fungal metabolic activity includes the digestion of polysaccharides and fat residuals from the diet and host residuals, leading to the synthesis of a variety of secondary metabolites, which can substantially influence the surrounding prokaryotic and eukaryotic cells. An investigation of the relationship between fungal diversity and diet revealed a positive correlation of Candida with diet rich in saccharides and a negative correlation of Aspergillus with SCFAs (Hoffmann et al., 2013). Thus, close relationships between bacterial and fungal metabolic requirements can help structure the microbial community in the gut. For instance, antimicrobial treatment

can significantly disrupt the ecological balance of microbiota throughout the digestive tract. Antibiotics eradicate some sensitive bacteria and their niche can be subsequently invaded by other bacteria or fungi (Huffnagle and Noverr, 2013). Interestingly, a nested case-control study has shown that bacterial or fungal outgrowth after multiple penicillin treatments slightly increases the risk of CRC development (Boursi et al., 2015). And, last but not least, consumption of food-associated mycotoxins – secondary metabolites of fungi, has been linked to carcinogenesis throughout the digestive tract (De Ruyck et al., 2015). Several in vitro studies have shown that exposure to mycotoxins affects apoptosis, intestinal barrier integrity and mucus production and causes DNA damage, suggesting a possible role of mycotoxins in CRC development; reviewed by Maresca and Fantini (2010).

#### HOST–MICROBIOME INTERACTION AND DISEASE DEVELOPMENT

Host derived proteoglycans, forming the mucus layer, are an important part of the mucosal immune system. They protect the epithelium from an extensive contact with the microbiome and reduce the risk of microbial invasion. The oral cavity and esophagus harbor several layers of tight and largely inert squamous epithelium, whereas the remaining parts of the digestive tract are covered with a single layer of intensely active cells (Johansson et al., 2013). The structure of the mucus layers and types of mucin (MUC) vary widely along the digestive tract. The salivary glands in the oral cavity produce mainly MUC5B and MUC7, glands in the stomach and duodenum secrete gel-forming mucins MUC5AC and MUC6, and goblet cells in the gut specialize in MUC2 production (Khan et al., 1998; Wickstrom et al., 1998; Nordman et al., 2002). While in the small intestine, MUC2 forms a loose unattached mucus layer, in the colon it has two parts with different functions, an inner, attached layer and an outer, unattached one (Johansson et al., 2013). The inner layer, which is about 50–100 µm thick, is dense and impenetrable to most microbes, while the outer layer flows with the gut content. Mucus contains distinct products of epithelial cells, such as antimicrobial peptides and secretory IgA, which play an important role in the protection of gut mucosa against pathogen invasion or excessive inflammatory response to commensals (Johansson et al., 2011).

Interestingly, some members of the oral and gut microbiome can form a multilayer structure, composed of microbes and a polymeric matrix, termed a biofilm. Biofilm formation is one example of the mechanisms microorganisms use to evade antimicrobial defenses in the hostile environment of the host. Most biofilms are of polymicrobial nature and members of the biofilm community are distinct from the planktonic microbiota colonizing the mucosal surfaces throughout the body. Polymeric matrix formation and subsequent microbial colonization is the consequence of adhesion processes mediated by a wide spectrum of glycoproteins. Caries and periodontal disease are associated with biofilm formation by well-known periodontopathic bacteria. Biofilm in dental caries contains mainly streptococci, L. acidophilus, and Actinomyces and is secondarily colonized by anaerobic species, such as F. nucleatum and P. gingivalis (Chenicheri et al., 2017). In periodontal disease, early biofilm colonizers are mainly represented by streptococci and Actinomyces. Later on, more pathogenic bacteria such as F. nucleatum, P. gingivalis, T. forsythia, T. denticola, and A. actinomycetemcomitans appear (Socransky and Haffajee, 2005; Teles et al., 2013). The tendency of some microscopic fungi to form biofilms is also well-established in the literature. Recently, cooperation between Candida and oral commensal streptococci has been described as a significant factor in biofilm formation. Such cohabitation supports Candida growth and survival by providing it with an adhesive surface and the ability to invade tissue by promoting hyphae formation (Diaz et al., 2012).

Current research has confirmed the presence of polymicrobial biofilms on gut mucosa of CRC patients, suggesting their possible role in CRC pathogenesis. Even in healthy mucosa, biofilm formation is associated with oncogenic potential and might be used to predict susceptibility to cancer development (Dejea et al., 2014). These biofilms consist of periodontopathic bacteria – F. nucleatum and P. gingivalis, as well as oral commensals, such as Peptostreptococcus, Prevotella, and Parvimonas, and their metabolic products, which may contribute to CRC progression (Li et al., 2017). Microbial biofilms disrupt mucus layers, enabling potentially harmful microbes to attach to or even invade the mucosa and directly affect the epithelial cells by cytotoxic or genotoxic metabolites. Indeed, a recent study by Johnson et al. (2015) has provided evidence that the presence of biofilm increases polyamine metabolites in cancer tissues. Interestingly, fungal genera Phoma and Candida have been detected in higher quantities in adenoma biopsies (Luan et al., 2015). However, to date, the connection of these mixed-species biofilms with CRC has not been thoroughly studied.

When passing from the upper to the lower digestive tract, some previously mentioned bacteria change their oxygen requirements from facultative anaerobic to strict anaerobic, thereby switching to asaccharolytic and proteolytic metabolism (Eley and Cox, 2003). Microbial proteolytic enzymes break down the host's extracellular matrix and soluble factors to get nutrients and invade the tissue. Periodontopathic bacteria produce a wide spectrum of enzymes, including collagenases, elastases, peptidases, etc. For instance, gingipains are cysteine proteases secreted by P. gingivalis, classified as either arginine (Rgp) or lysine (Kgp) specific (Potempa et al., 2003). They play a key role in biofilm formation, consequent host tissue destruction and vascular permeability induction (Kadowaki et al., 2000; Eley and Cox, 2003). Some of the P. gingivalis proteases can degrade immunologically active molecules, such as immunoglobulins, cytokines and components of the complement, and thus modulate the antibacterial immune response. Similarly, oral streptococci produce proteases which have been shown to cleave IgA1 (Kilian et al., 1988). Interestingly, 88% of the streptococci that initiate plaque formation on dental enamel possess IgA1 protease activity. Moreover, oral streptococci attack human

immunoglobulin IgA1 not only by protease production but also by glycosidases (neuraminidase and beta-galactosidase). Oral streptococci thus cleave the alpha chains and also the carbohydrate moiety of IgA1. This finding suggests that the ability of streptococci to evade secretory immune mechanisms is one of the factors that enable them to colonize the oral cavity (Kilian et al., 1989).

Another important feature of the microbiota that protects the host against pathogens is resistance to outsider invasion. Microbiota presents a competitive barrier to pathogenic microbes by active struggle for existence, fighting for nutrients and niche occupation. Moreover, commensals express antimicrobial effector molecules (bacteriocins) that serve as an effective tool for community shaping by endogenous microbiota. The third mechanism is indirect through constitutive stimulation of the mucosal immune system by commensal microbes, which strengthens the mucosal barrier, thus reducing pathogen translocation (Robinson et al., 2010; Stecher and Hardt, 2011; Backhed et al., 2012).

Mucosal surface of the gut is in continuous contact with foreign compounds derived from diet as well as from commensal or pathogenic microorganisms. Therefore, maintaining balance between the inner and outer milieu is the hallmark of the whole mucosal immune system. Many different cell types and their products are involved in this complex dialog, including epithelial and immune cells, cells of supporting tissues, antimicrobial peptides, growth factors, cytokines, and other mediators. Various components of microbiota can differentially trigger cellular pathways that shape local as well as systemic immune response and physiological functions. Recognition of these microbe-associated molecular patterns (MAMPs) is one of the most important features of the mucosal immune system. Receptors facilitating this are known as patternrecognition receptors and can be divided into several families, such as retinoic acid inducible gene I-like receptors, nodlike receptors, toll-like receptors (TLR), and lectin receptors. Proinflammatory processes that are mediated by MAMPs, such as lipopolysaccharide, polysaccharides, peptidoglycan, flagella, and microbial DNA/RNA, activate pattern-recognition receptors on various host cells. These cells elicit local pro-inflammatory response and/or drive the differentiation of adaptive immune response (Sartor, 2008; Underhill and Iliev, 2014). Therefore, a persistent inflammatory reaction of the host, constantly challenging the mucosal immune system, can lead to disease initiation (Tlaskalova-Hogenova et al., 2004). Appropriate immune response requires that recognition of commensals on the apical side of the epithelium induces tolerance, while recognition of pathogens on the basolateral membrane or inside

the cell induces inflammation. Moreover, TLRs are important for stimulation of gut epithelium growth and barrier integrity as well as production of mucus, secretory IgA, antimicrobial peptides, and chemokines (Abreu, 2010). Generally, the expression of TLRs in the epithelium is low in the steady state but increases during inflammation. Indeed, recent studies suggested the association of TLR polymorphisms with the progression of IBD into CRC, as TLRs expression is changed during gut inflammation (Bank et al., 2015).

Chronic inflammation leads to massive accumulation of activated immune cells and their mediators (cytokines and chemokines), residues of damaged cells, and large amounts of oxygen and nitrogen reactive species. Isolated dysplastic cells are further modified by the local microenvironment, as proinflammatory cells and cytokines promote the progression of dysplasia into carcinoma. During chronic inflammation, innate immune cells produce excessive quantities of reactive oxygen and nitrogen species that cause DNA and cellular damage. Pattern-recognition receptors signalization in the milieu of chronic inflammation activates MyD88-dependent pathways that promote pro-inflammatory cytokine release and subsequent tumor progression (Tlaskalova-Hogenova et al., 2014).

#### CONCLUSION

There are two fundamental links between microbes and diseases. The first involves the host's recognition and immune response mechanisms and the second involves the microbiota itself, its presence and metabolic activity. Impaired barrier function, inadequate activation of the innate immune system and dysregulation of the appropriate mucosal immune response to gut microbiota (tolerance) are the primary elements of disease development. Low microbiome diversity seems to be a common feature in the pathogenesis of diseases of the

#### REFERENCES


digestive tract, suggesting that high microbial richness, both bacterial and fungal, is crucial for physiological homeostasis. Normally, a dynamic balance is maintained, where reduction of the bacterial community leads to outgrowth of the fungal one and vice versa. Disruption of this balance is associated with outgrowth of certain pathobionts at the expense of a complex community and may lead to disease initiation (**Figure 2**). Established dysbiosis is supported by local proinflammatory and tumor microenvironment which creates distinct nutritional conditions. A growing body of literature dealing with microbiome metabolic activities gradually unravels the complex host–microbe interactions. Nevertheless, collecting data about these interactions at the time of disease development is especially challenging in humans. In this respect, we still have to rely on retrospective clinical studies and animal models to generate helpful insights. Elucidation of the underlying mechanisms involving the interactions of the immune system with the microbiome, pathogens, or pathobionts can lead to the development of early screening and preventive interventions and, ultimately, reduce the risk of severe colon cancer.

#### AUTHOR CONTRIBUTIONS

KK, ZJZ, and HT-H wrote the manuscript and approved its final version.

#### FUNDING

The authors are supported by Czech Science Foundation (17- 06632Y and 16-06326S), Ministry of Health of the Czech Republic (15-27580A, 15-29336A, 15-28064A, and 15-30782A), and Institutional Research Concept (RVO: 61388971).




and CDKN2A methylation in microsatellite-unstable colorectal carcinoma. Virchows Arch. 471, 329–336. doi: 10.1007/s00428-017-2171-6



modulating autophagy. Cell 170, 548–563. doi: 10.1016/j.cell.2017.07.008 e16, doi: 10.1016/j.cell.2017.07.008


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Klimesova, Jiraskova Zakostelska and Tlaskalova-Hogenova. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Role of Type 2 Diabetes for the Development of Pathogen-Associated Cancers in the Face of the HIV/AIDS Epidemic

Melissa J. Blumenthal† , Sylvia Ujma† , Arieh A. Katz and Georgia Schäfer\*

Receptor Biology Research Unit, Division of Medical Biochemistry and Structural Biology, Institute of Infectious Disease and Molecular Medicine, Department of Integrative Biomedical Sciences, Faculty of Health Sciences, SA-MRC Gynecology Cancer Research Centre, University of Cape Town, Cape Town, South Africa

The contribution of HIV to the development of pathogen-associated cancers has long been recognized, as has the contribution of type 2 diabetes for the development of several types of cancer. While HIV/AIDS-associated immunosuppression reduces immunosurveillance and indirectly contributes favorably to cancerogenesis, diabetes directly increases cancer development due to chronic low-grade inflammation, dysregulated glucose metabolism, hyperactivation of insulin-responsive pathways, and anti-apoptotic signaling. Pathogen-associated cancers contribute significantly to the cancer burden particularly in low- and middle-income countries. In those countries, the incidence of type 2 diabetes has increased alarmingly over the last decades, in part due to rapid changes in diet, lifestyle, and urbanization. It is likely that the HIV/AIDS epidemic and the steadily increasing rate of type 2 diabetes display synergistic effects on oncogenesis. Although this possible link has not been extensively investigated, it might become more important in the years to come not least due to the stimulating effects of antiretroviral therapy on the development of type 2 diabetes. This review provides an overview of the current understanding of pathogen- and diabetes- associated cancers with focus on geographical regions additionally burdened by the HIV/AIDS epidemic. As both HIV and carcinogenic infections as well as the onset of type 2 diabetes involve environmental factors that can be avoided to a certain extent, this review will support the hypothesis that certain malignancies are potentially preventable. Deploying effective infection control strategies together with educational policies on diet and lifestyle may in the long term reduce the burden of preventable cancers which is of particular relevance in low-resource settings.

Keywords: type 2 diabetes, pathogen-associated cancers, HIV/AIDS, HPV, KSHV, low- and middle-income countries, sub-Saharan Africa

## INTRODUCTION

The development of cancer is a complex multistep process. It is well-understood that the lack of responsiveness to signals from the microenvironment together with the accumulation of aberrations in multiple cellular regulatory systems can eventually lead to the characteristic loss of growth control displayed by cancer cells (Hanahan and Weinberg, 2011). Oncogenesis is typically

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Brian James Morris, University of Sydney, Australia Rosario Le Moli, Università degli Studi di Catania, Italy

> \*Correspondence: Georgia Schäfer georgia.schafer@uct.ac.za †These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 23 July 2017 Accepted: 16 November 2017 Published: 29 November 2017

#### Citation:

Blumenthal MJ, Ujma S, Katz AA and Schäfer G (2017) The Role of Type 2 Diabetes for the Development of Pathogen-Associated Cancers in the Face of the HIV/AIDS Epidemic. Front. Microbiol. 8:2368. doi: 10.3389/fmicb.2017.02368

initiated by a non-lethal genetic or epigenetic alteration leading to abnormal proliferation of a single cell. Apart from the genetic constitution of an individual, this can be caused by multiple agents including radiation, chemicals and persistent infection with oncogenic microbes, the latter causing about 15–20% of all human cancers (Plummer et al., 2016).

Although necessary, initiation alone is not sufficient for tumor formation. In addition, tumor promoters are needed to induce the initiated cells to transform and become malignant. Tumor promoters are generally non-tumorigenic by themselves but rather reversibly stimulate cell proliferation and/or provide a tumor-friendly environment (e.g., hormones, chronic inflammation, and immunosuppression). The combined action of initiators and promoters can then eventually give rise to cells that become gradually and irreversibly malignant through a progressive series of alterations (Vincent and Gatenby, 2008).

Smoking and alcohol consumption, obesity and physical inactivity are the worldwide leading risk factors for cancer death, with oncogenic pathogens significantly contributing to cancer incidence in low-and middle-income countries which are often additionally burdened by the HIV/AIDS epidemic (Danaei et al., 2005). The leading types of cancer-causing infections worldwide are Helicobacter pylori, human papillomavirus (HPV), hepatitis B virus (HBV) and hepatitis C virus (HCV), while Kaposi's sarcoma-associated herpesvirus (KSHV) is particularly important in the sub-Saharan African context. Co-infection with human immunodeficiency virus (HIV) substantially enhances the incidence of pathogen-associated cancers (Working Group on the Evaluation of Carcinogenic Risks to Humans [IARC], 2012). Although HIV is not considered an oncogenic virus, it indirectly impacts on cancerogenesis through immune deficiency and impaired immune surveillance, thereby increasing the effects of oncogenic infections (Working Group on the Evaluation of Carcinogenic Risks to Humans [IARC], 2012). Cancerpromoting effects have also been ascribed to the chronic metabolic and hormonal disturbances as seen in type 2 diabetes (Giovannucci et al., 2010), a condition that is no longer considered a disease primarily affecting the industrialized world but has become increasingly common in low-resource settings (Hu, 2011).

All three conditions, namely type 2 diabetes, HIV-associated immunosuppression and oncogenic pathogen infections, not only contribute to cancer development, they are also largely preventable. Fundamental changes in public policies with regard to diet and lifestyle modifications, education and training on infection prevention as well as nation-wide prophylactic vaccination programs against HBV and HPV are predicted to have a significant impact on the cancer burden in low- and middle- income countries (Sylla and Wild, 2012).

### PATHOGEN-ASSOCIATED CANCER DEVELOPMENT IN THE CONTEXT OF HIV/AIDS

Cancers attributable to infectious agents are an important component of the global cancer burden. Assessments dating from 1990 to 2012 have attributed 1/6th of global cancer cases to infectious etiologies and in sub-Saharan Africa, which is heavily burdened with HIV/AIDS, this is greater than 30% (Pisani et al., 1997; Parkin, 2006; de Martel et al., 2012; Plummer et al., 2016). Oncogenic infectious agents include viruses: HCV, HBV, high-risk HPV, KSHV, Epstein–Barr virus (EBV), human T-cell lymphotrophic virus type-1 (HTLV-1) and Merkel cell polyomavirus (MCPyV) (Houben et al., 2010; Working Group on the Evaluation of Carcinogenic Risks to Humans [IARC], 2012; Schäfer et al., 2015); bacteria: Helicobacter pylori (Forman et al., 1991; Nomura et al., 1991; Parsonnet et al., 1991); and parasites: Opisthorchis viverrini, Clonorchis sinensis, and Schistosoma haematobium (International Agency for Research on Cancer Working Group, 1994; Honjo et al., 2005; Bouvard et al., 2009). Additionally, HIV is classified as carcinogenic, although its mechanism is indirect via cell-mediated immune deficiency and needs to be in conjunction with another infectious agent (Working Group on the Evaluation of Carcinogenic Risks to Humans [IARC], 2012). Not surprisingly, the exceptional elevation of pathogen-associated cancers in the developing world is, not least, exacerbated in the high HIV/AIDS context (Parkin et al., 2005; Grulich et al., 2007).

The introduction of the highly active antiretroviral therapy (HAART) strategy has substantially reduced the number of AIDS-related deaths and extended the lifespans of HIV infected individuals (Palmisano and Vella, 2011). This extended lifespan has led to the emergence of a range of HIV-associated malignancies, again particularly burdening sub-Saharan Africa, that were not often seen preceding the introduction of HAART when average lifespan following HIV infection was significantly less (Frisch et al., 2001; Grulich et al., 2007; Coghill et al., 2013; Antiretroviral Therapy Cohort Collaboration, 2017). In contrast, HAART also had a profound preventative effect on some, but not all, HIV-associated malignancies. Kaposi's sarcoma (KS) incidence in Western countries was reported to decrease by greater than 90% from 1994 to 2003, spanning the introduction of HAART in 1996 (Mocroft et al., 2004). Similarly, incidence of EBV-associated non-Hodgkin lymphoma decreased by greater than 40% after the introduction of HAART in Western countries (International Collaboration on HIV and Cancer, 2000). However, incidences of HPV-associated cervical cancer and non-AIDS defining cancers have not yet been seen to be reduced (International Collaboration on HIV and Cancer, 2000). Although promising, the reductions of incidences of particularly KS and non-Hodgkin lymphoma in Western countries, has not been mirrored in resource-limited, developing countries. In fact, GLOBOCAN age standardized incidence rates reported in 2002 and 2012, indicate an increase in incidence of non-Hodgkin lymphoma in males and females in Southern and Northern Africa (Parkin et al., 2005; Torre et al., 2015). This is corroborated in a Ugandan study over the period 1991–1995 to 2002–2006, which reports an annual increase in incidence of non-Hodgkin lymphoma by 6.7 and 11.0% in men and women, respectively (Parkin et al., 2010). Similarly, KS incidence in Ugandan women has increased (1.4% annually, 10% over the period), while in men incidence has slightly decreased (2.8% annually, 30% over the period); this in contrast to the large (>90%) reductions in

KS incidence in Western countries over a similar period, and concomitant with the large scale roll-out of HAART (Mocroft et al., 2004; Parkin et al., 2010; Casper, 2011). This disparity has been attributed to a delayed and stilted availability of HAART in low-resource settings; indeed, it has been noted that even in sub-Saharan countries with relatively well-established antiretroviral therapy (ART) programs, KS incidence has not decreased as expected (Casper, 2011). Further, Nguyen et al. (2008) found that up to half of AIDS-related KS patients treated with HAART and chemotherapy never achieved total remission, which was in concordance with other studies (Dupin et al., 1999; Dupont et al., 2000; Bihl et al., 2007; Nguyen et al., 2008). It is therefore predicted that cancers caused by infectious agents will become a burgeoning complication of long-term HIV infection (Grulich et al., 2007; Sasco et al., 2010; Casper, 2011).

Oncogenic organisms infect, but do not kill their target cells, leading to persistent infection. This is facilitated by various immune evasion strategies driven by the expression of pathogen-encoded proteins and subversion of cellular regulation of proliferation and apoptosis (McLaughlin-Drubin and Munger, 2008; Mesri et al., 2014). Perhaps the best example of this is oncogenic HPV infection of basal cells of the cervical epithelium, which precedes development of virtually all cases of cervical cancer by decades (Woodman et al., 2007; Castellsague, 2008). HPV oncogenes E6 and E7 mitigate host innate immune responses through inhibition of the interferon pathway (Ronco et al., 1998; Park et al., 2000) and similarly, E5 oncoprotein causes downregulation of MHC class I (Campo et al., 2010). Likewise, HBV and HCV infection, which together account for approximately 80% of hepatocellular carcinoma cases, promote cirrhosis through non-cytocidal chronic infection of hepatocytes (Parkin et al., 2005; El-Serag, 2012). HBVencoded X antigen (HBx) and HCV-encoded core, non-structural protein 5A (NS5A) and NS3 antigens inhibit innate antiviral signaling pathways as well as apoptosis, conferring a survival advantage to infected hepatocytes (Mesri et al., 2014). Persistent infection through immune evasion is further facilitated in an immunosuppressive environment, such as in HIV infection, where host-mediated anti-tumor responses are extinguished.

Persistent infection can be accompanied by chronic inflammation. Persistent HBV and HCV infection lead to chronic inflammation of the liver (hepatitis) which promotes cancer development (Mesri et al., 2014). Similarly, key to the oncogenic potential of Helicobacter pylori, to which approximately 90% of new stomach cancer cases are attributed, is its ability to persist in the gastric mucosa for decades, eliciting chronic inflammation through the expression of virulence factors CagA (cytotoxinassociated gene A) and VacA (vacuolating cytotoxin A) (Wang et al., 2014; Plummer et al., 2015, 2016). Chronic low-grade inflammation is also present in HIV infected individuals on ART treatment and is associated with cancer development (Marks et al., 2013; Maartens et al., 2014).

Persistent oncogenic infection is necessary but often not sufficient to trigger carcinogenesis, but rather requires precipitating co-factors for the development of malignancy (McLaughlin-Drubin and Munger, 2008; Mesri et al., 2010). Often, immunosuppression plays this role, made evident by the enhanced incidence of infection-related cancers and cancers suspected to have infectious etiology seen in both HIV-positive patients and immunosuppressed post-transplant cohorts (Grulich et al., 2007). Of those, KS is the most common AIDS-related malignancy and a significant public health burden in sub-Saharan Africa, where KSHV seroprevalence rates as high as 50% have been reported (Mesri et al., 2010). While KSHV infection accounts for 100% of KS cases, alone it is asymptomatic and not sufficient for tumorigenesis, requiring HIV-related or another form of immunosuppression [e.g., standardized incidence ratio (SIR) for KS in HIV/AIDS cohort versus transplant cohort is 3640.0 (3326–3976) versus 208.0 (114–349), respectively] amongst other potential co-factors, to trigger KS development (Grulich et al., 2007; Mesri et al., 2010; Plummer et al., 2016). Moreover, due to shared routes of transmission, HPV and HIV coinfection is common and HIV infection increases the probability of HPV persistent infection resulting in increased relative risk [5.8 (3.0–11.3)] of cervical cancer development compared to HPV infection in the absence of HIV (Strickler et al., 2005; Grulich et al., 2007; Working Group on the Evaluation of Carcinogenic Risks to Humans [IARC], 2012). Similarly, HIV co-infection with either HBV or HCV increases the rate of progression of HBV- or HCV-mediated liver damage and the SIR of HBV/HCV-mediated liver cancer [5.22 (3.32–8.20)] (Graham et al., 2001; Grulich et al., 2007; Chen et al., 2009; Nikolopoulos et al., 2009).

Furthermore, a number of infectious cancers show distinct geographical distributions or population-specific prevalence. Before the HIV epidemic, KS was considered rare except for hotspots in the Mediterranean and Eastern Europe (Classic KS) and Central and Eastern Africa (Endemic KS). Now, AIDSrelated KS is prevalent across sub-Saharan Africa, but its peculiar geographical epidemiology has led to speculation that host genetic factors may influence seroconversion after exposure to KSHV and/or subsequent KS development (Mesri et al., 2010; Cavallin et al., 2014). Interestingly, while HCV is an important risk factor for the development of hepatocellular carcinoma in the United States, Europe and Japan, HBV contributes more substantially to the development of liver cancer globally and particularly in countries with low human development index (Barth et al., 2010; Chung et al., 2010; Plummer et al., 2016).

Besides HIV immunosuppression as outlined above, environmental factors such as host diet, physical inactivity, behaviors such as smoking, reproductive factors, and co-factors leading to smoldering chronic inflammation are thought to further influence the natural history of oncogenic infections and as a consequence of globalization, these are thought to become more important in the developing world (Parkin et al., 2005; Ajuwon et al., 2009; Plummer et al., 2016).

## DIABETES-ASSOCIATED CANCER DEVELOPMENT IN THE CONTEXT OF HIV/AIDS

Type 2 diabetes is a non-communicable disease, characterized by insulin resistance and hyperglycemia (Zaccardi et al., 2016).

Globally, type 2 diabetes is a major problem, but prevalence of the disease is particularly rapidly increasing in low- and middle-income countries (Guariguata et al., 2014). This is largely due to the changes in lifestyle and diet brought about by rapid urbanization, such as decreased physical activity as well as increased consumption of refined carbohydrates and sugar sweetened beverages (Hu, 2011). On that note, it was shown that only 1–2 servings of sugar-sweetened beverages per day increase diabetes risk by 26% (Malik et al., 2010). In 2013, low- and middle-income countries had the highest prevalence of type 2 diabetes and it is projected that Africa will experience a 109% increase in cases over the next 22 years (Guariguata et al., 2014). The most undiagnosed cases of type 2 diabetes occur in low- and middle-income countries due to a lack of resources necessary for diagnosis, meaning that often the disease is only detected after complications have already developed (Beagley et al., 2014). The Lancet Diabetes & Endocrinology Commission on diabetes in sub-Saharan Africa provides a comprehensive and up-to-date analysis of the vast economic burden that diabetes places on the resource-constrained health systems found in those regions (Atun et al., 2017).

As outlined above, low- and middle-income countries are also highly burdened by the HIV/AIDS epidemic, and 2–4 fold higher prevalence of dysglycemia in HIV-infected individuals in South Africa has been reported (Dave et al., 2011; Levitt et al., 2016). Although the mechanisms are not well-understood, the presence of HIV infection itself may contribute to diabetogenesis, both directly through inflammation and immune activation, and indirectly through immunodeficiency (Kalra and Agrawal, 2013; Lake and Currier, 2013; Levitt et al., 2016). Moreover, treatment by HAART may be directly or indirectly linked to the development of type 2 diabetes in HIV-infected patients (Hadigan and Kattakuzhy, 2014). The increased life expectancy of HIVinfected patients on HAART treatment has dramatically changed the natural history of HIV infection (Samaras, 2009), resulting in the emergence of more chronic illnesses, including type 2 diabetes (Hasse et al., 2011). Moreover, certain nucleoside reverse transcriptase inhibitors (NRTIs) and protease inhibitors (PIs) have been associated with the development of insulin resistance and an increased incidence of type 2 diabetes in HIV-infected individuals, though the literature has been contentious about whether PIs do indeed play a role (Behrens et al., 1999; Brown et al., 2005; Tien et al., 2007; De Wit et al., 2008; Capeau et al., 2012). A large cross-sectional study recently published reported a significantly higher odds ratio (OR) of diabetes and metabolic syndrome (OR 3.85 and 1.45, respectively, 95% CI) among ART-exposed patients compared to their naïve counterparts, although the association between ART and diabetes was not interpreted as cause and effect (Nduka et al., 2017). HAART is also known to cause lipodystrophy and hyperlipidemia, conditions common to insulin resistance and type 2 diabetes, therefore indirectly increasing the risk of development of type 2 diabetes in HIV-infected individuals who are on HAART (Samaras, 2009). The high prevalence of HIV/AIDS in today's era of HAART may therefore further substantially contribute to the increase in type 2 diabetes predicted for low- and middle-income countries.

Of the various complications associated with type 2 diabetes, cancer development, incidence, prognosis, and mortality have been linked to the long-term effects of the disease (Giovannucci et al., 2010). For example, a meta-analysis of 23 studies reported increased mortality across all cancer types in diabetes patients (HR 1.41; 95% CI, 1.28–1.55) (Barone et al., 2008). Specifically, an increased risk of colon cancer has been associated with type 2 diabetes [summary relative risk (SRR) = 1.30] (Larsson et al., 2005), and a cohort study showed that diabetic patients with stages 2 and 3 colon cancer had higher mortality rates and cancer recurrence as compared to non-diabetic patients (Meyerhardt et al., 2003). Individuals with type 2 diabetes had a SRR of 1.94 for the development of pancreatic cancer (Ben et al., 2011) and a SRR of 2.01 for hepatocellular carcinoma development (Wang et al., 2012). Another study described synergistic interactions between diabetes mellitus and hepatocellular carcinoma (OR, 9.9; 95% CI, 2.5–39.3) which was further found to be strikingly associated with heavy alcohol consumption and chronic hepatitis virus infection (OR, 53.9; 95% CI, 7.0–415.7) (Hassan et al., 2002). Type 2 diabetes has also been associated with a 42% increase in kidney cancer risk (Larsson and Wolk, 2011) as well as a 20% increase in risk of developing breast cancer (Larsson et al., 2007). Overall, type 2 diabetes is associated with increased mortality in cancer patients and an overall poor prognosis of cancer (Barone et al., 2008).

Both type 2 diabetes and cancer are highly complex diseases affecting many cellular processes. However, they share several risk factors, such as alcohol consumption, smoking, obesity, diet and physical inactivity, but the possible biological links between the two diseases are not yet completely understood. Possible mechanisms for a direct link between cancer and type 2 diabetes include hyperglycemia, hyperinsulinemia, and inflammation. Firstly, hyperglycemia has been suggested to be one of the possible influences of type 2 diabetes on cancer, mainly due to the Warburg effect: a well-known observation that cancer cells tend to undergo aerobic glycolysis and therefore consume more glucose than normal cells in order to accumulate precursor molecules for biomass rather than energy production (Vander Heiden et al., 2009). Additionally, high levels of glucose are linked to increases in WNT signaling, which enhances proliferation (Garcia-Jimenez et al., 2014), as well as upregulation of the oxidative response genes, leading to increased reactive oxygen species and mutations (Turturro et al., 2007). Secondly, hyperinsulinemia may also be a potential link between type 2 diabetes and cancer, as insulin is not only a metabolic hormone but also a growth factor that has anti-apoptotic and mitogenic effects via activation of the insulin receptor (Vigneri et al., 2016). Of the different insulin receptor isoforms the A isoform, which has predominant mitogenic activity, is often overexpressed in cancer cells, providing a selective growth advantage to malignant cells when exposed to insulin (Vigneri et al., 2016). High levels of circulating insulin due to insulin resistance or insulin treatment also result in reduced insulin-like growth factor binding protein (IGFBP), leading to increased insulin-like growth factor 1 (IGF-1), which has more potent anti-apoptotic and mitogenic effects than insulin (Giovannucci et al., 2010). However, high levels of insulin can also spill over to IGF-1 receptors, and indeed

it was shown that both the insulin receptor and the IGF-1 receptor can act as identical portals to the regulation of gene expression with differences between insulin and IGF-1 effects due to a modulation of the amplitude of the signal created by the specific ligand-receptor interaction (Boucher et al., 2010). Hyperinsulinemia also indirectly leads to increased levels of estrogen and androgens by reducing the hepatic synthesis of sex hormone-binding globulin (Calle and Kaaks, 2004); aberrant steroid hormone metabolism is associated with a higher risk of certain cancers, including breast and endometrial cancer. Lastly, excessive caloric intake often associated with urbanization and an increasing Westernized lifestyle leads to obesity which is not only a well-known risk factor for the development of type 2 diabetes but is also linked to cancer development (Hjartaker et al., 2008). In South Africa, 68% of women and 31% of men are overweight or obese (Department of Health, 2016). Additionally, consumption of sugar sweetened beverages has approximately doubled in rural areas since 2005, with 56% of women and 63% of men consuming these beverages, which further contributes to the increasing prevalence of obesity here (Vorster et al., 2014). Chronic lowgrade adipose tissue inflammation is not only a symptom of obesity but is also known to be a recognizable feature of the metabolic syndrome and a major cause of the decreased insulin sensitivity seen in type 2 diabetes (van Kruijsdijk et al., 2009; Shu et al., 2012). Moreover, obesity can induce sustained systemic production of reactive oxygen species which can eventually lead to somatic mutations and neoplastic transformation (Manna and Jain, 2015). As an endocrine organ, excess adipose tissue secretes elevated levels of various cytokines, hormones and growth factors including interleukin-6 (IL-6), tumor necrosis factor alpha (TNF-α), plasminogen activator inhibitor-1 (PAI-1) and leptin that can alter the body's ability to appropriately respond to insulin (van Kruijsdijk et al., 2009). TNF-α has been shown to promote cell survival through the NF-κB pathway, and IL-6 mediates cell proliferation and survival through the JAK/STAT pathway (van Kruijsdijk et al., 2009). Type 2 diabetes is therefore associated with a general dysregulated innate immune response (Paich et al., 2013). Together with metabolic disruptions in insulin signaling, this manifestation of immune dysfunction common to obesity and type 2 diabetes present a favorable environment for cancer cell proliferation, invasion, and survival (Allavena et al., 2008; van Kruijsdijk et al., 2009).

#### THE LINK BETWEEN TYPE 2 DIABETES AND PATHOGEN-ASSOCIATED CANCERS IN THE CONTEXT OF HIV/AIDS

Low- and middle-income countries are projected to carry the majority of the world-wide cancer burden over the next decades, highlighting the emergence of cancer as a major public health problem (Sylla and Wild, 2012). While chronic infection with oncogenic pathogens has long been identified as a main contributor, the impact of HIV infection, particularly in the HAART era, as well as the ever-increasing incidence of type

FIGURE 1 | The relative contribution of oncogenic pathogen infection, HIV infection and type 2 diabetes for pathogen-associated cancers in low-income versus high-income countries and the potential prevention of these malignancies.

2 diabetes represent emerging risk factors for morbidity and mortality in those regions of the world. Indeed, patients living with HIV/AIDS have an increased burden of non-communicable diseases relative to HIV-uninfected individuals, with more than 25% predicted to develop three or more non-communicable diseases by 2030 (Smit et al., 2015).

Globally, two-thirds of infection-attributable cancers occur in less developed countries (Plummer et al., 2016) (**Figure 1**), although the true incidence rates within the African continent

are very uncertain due to underreporting and the data from GLOBOCAN often not being comprehensive. Nevertheless, lowresource settings also carry by far the largest burden of the HIV epidemic: in 2015, an estimated 36.7 million people were living with HIV globally of whom approximately 19.0 million (i.e., more than 50%) live in Southern Africa (Global AIDS, 2016). These numbers have been increasing over the years (from 31.0 million in 2002 to 36.7 million in 2016) due to the life-extending effects of HAART (Zaidi et al., 2013). Moreover, two-thirds of all diabetes cases as a consequence of rapid changes in lifestyle, urbanization and population aging are reported to occur in low- and middleincome countries (Wild et al., 2004; Hu, 2011), representing a growing public health issue in the HIV-infected population, especially as the clinical course of HIV infection in the HAART era has changed from a progressive illness with a fatal outcome to a chronic manageable disease (Deeks et al., 2013).

Both type 2 diabetes and the consequences of HIV infection share common characteristics with regard to facilitating cancerogenesis initiated by oncogenic pathogen infection, but is there a causal relationship between these two conditions? As outlined above, chronic inflammation and immunosuppression are two of the main cancer hallmarks associated with persistent oncogenic pathogen infection adverting the normal functioning of the cellular machinery. While immune dysfunction is characteristic for HIV infection, smoldering chronic inflammation is also associated with type 2 diabetes and HIV infection (particularly when treated by HAART), highlighting the importance of these two conditions as promoting environments for the development of pathogen-associated cancers. Although type 2 diabetes and HIV infection might constitute independent risk factors for pathogen-associated cancers in general, some striking synergism has been described for hepatitis virusassociated hepatocellular carcinoma. Indeed, overwhelming evidence in the literature points toward an association between co-infection of HIV and HBV/HCV together with insulin resistance and/or diabetes and a significantly enhanced risk of liver disease (Merchante et al., 2009; DallaPiazza et al., 2010; Howard et al., 2010; Salmon et al., 2012; Elkrief et al., 2014; Hadigan and Kattakuzhy, 2014; Lo Re et al., 2014; Oliver et al., 2016). This was found to be further exacerbated by certain antiretroviral drugs such as didanosine and/or stavudine (Blanco et al., 2011).

This well-documented link of HIV infection, type 2 diabetes and liver cancer is less evident for other pathogen-associated malignancies. For example, only case reports on HIV-related KS in the presence of type 2 diabetes are currently available, calling attention to the risk of delayed diagnosis of KS in patients on ART with a relatively high CD4 count (Chan and Pakianathan, 2011). Although no such case report on the link between HIV, type 2 diabetes and cervical cancer exists, an interesting recent study describes the combination of metformin, the worldwide most widely prescribed first-line therapeutic drug for type 2 diabetes and known for its antitumor properties (Kasznicki et al., 2014), together with nelfinavir, an HIV protease inhibitor. This drug combination showed promising effects on HPV-associated cervical cancer cell growth in vitro and in vivo (Xia et al., 2017). It is highly likely that other pathogen-associated cancers also benefit from the underlying cancer-friendly environment in the presence of both HIV infection and type 2 diabetes. Their cancerpromoting contributions are unlikely to be separate entities but rather present complex interactions facilitating cancerogenesis especially in a setting where these risk factors predominate.

In light of the low resources and the often deficient health care infrastructure in less developed countries, concerted efforts on cancer prevention rather than cancer treatment will likely have a sustainable effect on the reduction of cancer. The first step to prevention is the identification of risk factors and/or the etiological agent. Indeed, infections have been suggested to be one of the most important preventable causes of cancer in general, not least reflected by the significantly disparate incidence rates of pathogen-associated malignancies in industrialized versus lowand middle-income countries (**Figure 1** and Kuper et al., 2000). Highly effective prophylactic vaccines exist for some of the most common etiological agents of pathogen-associated cancers, namely HBV and HPV types 16 and 18, causing liver and cervical cancer, respectively, the two most common causes of cancer death in Africa (Sylla and Wild, 2012). Although the HBV vaccine is relatively inexpensive, national HBV infant immunization programs as seen in many African countries account only for 10% coverage in Africa (Plummer et al., 2016). The HPV vaccines are still rather expensive to produce, limiting efficient coverage of the population in the most affected areas (Hanson et al., 2015). Moreover, of the 13 HPV genotypes classified as carcinogenic (Working Group on the Evaluation of Carcinogenic Risks to Humans [IARC], 2012), the HPV vaccines currently available in low-resource settings only target the most common oncogenic genotypes 16 and 18 which account for approximately 70% of invasive cervical cancers globally. In sub-Saharan Africa however, other carcinogenic HPV types, such as HPV45 and 35, occur relatively more frequently than in other world regions as they seem to be more affected by changes in immunodeficiency levels (De Vuyst et al., 2013; Clifford et al., 2016). Therefore, even if there was saturation coverage of vaccinated populations in those regions of the world, the reduction in cervical cancer will fall well short of 100%. In this regard, voluntary medical male circumcision has a significant protective effect not only against HIV but also high-risk HPV, HBV and other sexually transmitted infections (Castellsague et al., 2002; Tobian et al., 2014; Wahome et al., 2017). Although highly cost-effective, implementation of the WHO recommended goal of 80% circumcision coverage among men aged 15–49 years in 13 countries in Eastern and Southern Africa with high HIV prevalence has been very slow and is far from being reached (Tobian et al., 2014). To reduce pathogen-associated cancers in low-resource settings both HPV/HBV vaccination and male circumcision should be advocated. Even so, efforts in the reduction of infection-related cancers are often offset by an increasing number of new cases that are associated with (among others) dietary and lifestyle factors (Bray et al., 2012). This is further exacerbated by the adverse effects of HAART on the development of type 2 diabetes and low-grade chronic inflammation seen in HIV/AIDS patients (Maartens et al., 2014). Therefore, programs to implement nation-wide prophylactic vaccination against HBV and HPV together with concerted efforts to reduce type 2 diabetes risk

and the risk of acquiring sexually transmitted infections such as HIV, HBV/HCV and HPV, would have a significant impact on cancer incidence and mortality (**Figure 1**). Indeed, several randomized clinical trials have demonstrated that diet and lifestyle modifications are highly effective in preventing type 2 diabetes in different ethnic and racial groups (Knowler et al., 2002; Lindstrom et al., 2006; Ramachandran et al., 2006; Li et al., 2008).

#### CONCLUSION

It has become evident over the last decades that the nature of comorbidities in HIV-infected individuals has changed substantially, particularly with regards to tumorigenesis and oncogenic disease progression. Although the descriptive data presented here may not lead to a definitive scientific interpretation, they clearly support the hypothesis that some of the highest cancer risk factors that predominate in resourcelimited settings (such as oncogenic pathogen infection, HIV infection and type 2 diabetes) are potentially avoidable. It is suspected that there might be a link between these risk factors, particularly between HIV infection, its treatment and the onset of non-communicable diseases such as type 2 diabetes. Effective interventions including population-based vaccination against

#### REFERENCES


HBV and HPV together with HIV prevention and cervical cancer screening programs as well as awareness, counseling and educational programs on changes in diet and physical activity would lead to significant reductions of the cancer burden in those areas of the world. Early intervention programs not only prevent disease onset and complications but are clearly much simpler and cheaper than treating later stage disease. Research activities to understand the synergistic effects between the risk factors discussed in this review are needed and should be a focus of future scientific efforts.

#### AUTHOR CONTRIBUTIONS

GS led the conception and design of this article, drafted, revised and approved its final version. MB and SU equally contributed to drafting individual sections, while AK critically revised the article.

#### ACKNOWLEDGMENT

This work was supported by funding from the Poliomyelitis Research Foundation (PRF), the Cancer Association of South Africa (CANSA), and the National Research Foundation (NRF) of South Africa.

reconstitution and antiviral effect of combined HAART/chemotherapy in HIV clade C-infected individuals with Kaposi's sarcoma. AIDS 21, 1245–1252. doi: 10.1097/QAD.0b013e328182df03





**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Blumenthal, Ujma, Katz and Schäfer. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Commentary: High Glucose Induces Reactivation of Latent Kaposi's Sarcoma-Associated Herpesvirus

Fabrizio Angius, Maria A. Madeddu and Raffaello Pompei\*

Biomedical Sciences, Università degli Studi di Cagliari, Cagliari, Italy

Keywords: Kaposi sarcoma, cell metabolism, human herpesvirus 8, diabetes mellitus, oncogenic viruses

#### **A commentary on**

#### **High Glucose Induces Reactivation of Latent Kaposi's Sarcoma-Associated Herpesvirus**

by Ye, F., Zeng, Y., Sha, J., Jones, T., Kuhne, K., Wood, C., et al. (2016). J. Virol. 90, 9654–9663. doi: 10.1128/JVI.01049-16

In their recent work Ye et al. claimed that high glucose concentration enhances Kaposi sarcoma Herpes virus (KSHV) lytic gene expression and replication in different types of cells and that induction of the KSHV lytic gene expression by high glucose is mediated by H2O2. They suggested that H2O<sup>2</sup> mediates down-regulation of the silent information regulator 1 (SIRT1), a class III histone deacetylase, in cells that are cultured in media containing a high concentration of glucose and that, as a consequence, high glucose also transactivates KSHV lytic gene expression via epigenetic modifications of the transcription activator promoter region RTA.

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Keiji Ueda, Osaka University, Japan Consolato Sergi, University of Alberta, Canada

> \*Correspondence: Raffaello Pompei rpompei@unica.it

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 19 May 2017 Accepted: 05 September 2017 Published: 15 September 2017

#### Citation:

Angius F, Madeddu MA and Pompei R (2017) Commentary: High Glucose Induces Reactivation of Latent Kaposi's Sarcoma-Associated Herpesvirus. Front. Microbiol. 8:1796. doi: 10.3389/fmicb.2017.01796

The authors found increased levels of KSHV lytic gene expression when KSHV-infected BCBL1 and TIVE-KSHV cells were cultured in media containing diabetic levels of glucose. They also demonstrated that cells cultured in high concentrations of glucose produce increased levels of intracellular H2O2. In their study, Ye et al. considered that the high prevalence of Kaposi sarcoma (KS) in diabetes type 2 (DM2) has been widely reported in various studies carried out over the last 40 years. In their early work, Laor and Schwartz (1979) described a series of 37 patients with KS, among whom a significant number of 12 (32%) were found to have concurrent diabetes mellitus. Caprio et al. (1985) also described 3 clinical cases of KS with concomitant DM2. More recently, Weissmann et al. (2000) study on KS included 125 patients diagnosed and followed at a Dermatological Clinic; a significant incidence of up to 22.4% of these patients had DM2 as compared to a general prevalence of 5–8% in the non-diabetic controls. On the basis of the reported cases, a possible role of DM2 in the onset of KS was hypothesized.

In conclusion, Ye et al.'s study provides evidence for a link between type 2 diabetes and higher levels of KSHV replication, which may lead to development of classic KS. These results highlight H2O<sup>2</sup> as the mediator for the high glucose induction of KSHV lytic replication through multiple mechanisms.

However, recent studies published over the last decade have proposed a possible cause-effect of KSHV infection and DM2 onset. A prevalence of KSHV DNA and anti-KSHV antibodies in DM2 subjects was detected by Piras et al. (2016) in Southern Sardinia with about 58% (43 out of 74) of KSHV positive patients against 27% (10 out of 36) of controls. About 50% of the DM2 subjects examined in this study were permanently infected by KSHV, whilst the other 50% were free from this virus. This observation implies that other factors (other known or unknown viruses? Or other environmental causes?) could be involved as additional risk factors for DM2 onset. Recently, Sobngwi et al. (2008) also examined a series of diabetes mellitus patients in sub-Saharan Africans and found a number of 71 out of 81 subjects (87%) who were significantly positive for KSHV. They also claimed that HHV8 was able to infect and replicate in the pancreas islets. It is also important to consider that KSHV causes a series of general metabolic modifications which can lead to altered insulin uptake and neutral lipid accumulation in primary endothelial cells as shown by Angius et al. (2015). In addition, in persons with normal glucose-concentration, KSHV induces a general impairment of the immune system, a decrease in adaptive immunity, and an up-regulation of both insulin and glucose consumption (Delgado et al., 2012; Gregory et al., 2012; Angius et al., 2015). Most viruses examined to date induce aerobic glycolysis, also known as the Warburg effect, in normal glucose-concentration medium. These modifications of carbon source utilization by infected cells can increase the available energy for virus replication and virion production (Delgado et al., 2012). In addition, Bottero et al. (2012) have shown that KSHV alone was able to induce ROS and H2O<sup>2</sup> production very early during the infection of HMVEC-d cells grown in normal-glucose medium to facilitate virus entry and replication.

The obvious questions that arise from these interesting studies are the following: (i) is Kaposi's sarcoma a frequent consequence of diabetes, due to the pathological alteration of systemic glucose concentrations? Or, conversely, (ii) is KSHV the actual cause of a possible diabetes onset, since it can permanently modify the general cell metabolism, impairing both insulin and glucose utilization? Or again, (iii) should both possibilities be taken into consideration as far as the relationship between KS and DM2 is concerned? This doubt is extremely intriguing and calls for the attention and efforts of the scientific community. Both hypotheses can be taken into consideration, since it is well established that all immune deficiencies (AIDS, transplantation, iatrogenic therapies) lead to increased KS incidence, and diabetes is also known to be a cause of severe impairment of immunity. But what causes diabetes? This is the problem. Can the strong decrease in immune system function and the general metabolic

#### REFERENCES


modification induced by KSHV infection be causes that lead to DM2 each time? Can KSHV induce insulin resistance? Can this virus impair glucose utilization by peripheral body cells? This task calls for interdisciplinary research requiring the competences of clinicians, infectious disease physicians, hygienists and pharmacologists to clarify this still unknown problem. To date there is robust epidemiological and clinical evidence of a possible association between KSHV infection and DM2 (Sobngwi et al., 2008; Angius et al., 2015; Piras et al., 2016). Insulin and glucose alterations by HHV8 are well documented in infected cells by several in vitro studies (Delgado et al., 2012; Angius et al., 2015). Interestingly, Piras et al. (2016) observed an increase in KSHV infection in the general Sardinian population and, at the same time, DM2 was found to have risen from 5 to 6% in 2004 to about 10% in 2014. Sobngwi et al. (2008) also reported a high prevalence of diabetes in patients displaying KS, which is endemic in sub-Saharan Africa. While we understand that the presented results are limited to a few either clinical-epidemiological or experimental studies, the inter-disciplinary importance of these works would surely rouse interest in these provocative hypotheses that could lead to several parallel investigations.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### ACKNOWLEDGMENTS

The authors thank Ms. Sally Davies for valuable help in the preparation and correction of the manuscript. This work was funded by the Department of Biomedical Sciences of Cagliari (FIR program).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Angius, Madeddu and Pompei. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Pro-inflammatory state in Monoclonal gammopathy of Undetermined significance and in Multiple Myeloma is characterized by low sialylation of Pathogenspecific and Other Monoclonal immunoglobulins

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Michael Diamantidis, University Hospital of Larissa, Greece Vittorio Bellotti, University College London, United Kingdom*

#### *\*Correspondence:*

*Jean Harb jean.harb@univ-nantes.fr*

*† These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 10 July 2017 Accepted: 03 October 2017 Published: 19 October 2017*

#### *Citation:*

*Bosseboeuf A, Allain-Maillet S, Mennesson N, Tallet A, Rossi C, Garderet L, Caillot D, Moreau P, Piver E, Girodon F, Perreault H, Brouard S, Nicot A, Bigot-Corbel E, Hermouet S and Harb J (2017) Pro-inflammatory State in Monoclonal Gammopathy of Undetermined Significance and in Multiple Myeloma Is Characterized by Low Sialylation of Pathogen-Specific and Other Monoclonal Immunoglobulins. Front. Immunol. 8:1347. doi: 10.3389/fimmu.2017.01347*

*Adrien Bosseboeuf <sup>1</sup> , Sophie Allain-Maillet <sup>1</sup> , Nicolas Mennesson1 , Anne Tallet <sup>2</sup> , Cédric Rossi <sup>3</sup> , Laurent Garderet 4,5,6, Denis Caillot <sup>3</sup> , Philippe Moreau7 , Eric Piver 2,8, François Girodon9 , Hélène Perreault 10, Sophie Brouard11, Arnaud Nicot11, Edith Bigot-Corbel 1,12,13, Sylvie Hermouet 1,14,15† and Jean Harb1,11,12,14\*†*

*1CRCINA, INSERM, Institut de Recherche en Santé 2 (IRS-2), Université de Nantes, Nantes, France, 2 Laboratoire de Biochimie, Centre Hospitalier Universitaire de Tours, Tours, France, 3Clinical Hematology, Centre Hospitalier Universitaire De Dijon, Dijon, France, 4UMRS938, INSERM Institut National de la Santé et de la Recherche Médicale, Paris, France, 5Département d'Hématologie et de Thérapie Cellulaire, Hôpital Saint Antoine, Paris, France, 6UPMC Université Paris 6, Sorbonne Universités, Paris, France, 7Hematology Department, Centre Hospitalier Universitaire (CHU) de Nantes, Nantes, France, 8UMR966, INSERM Institut National de la Santé et de la Recherche Médicale, Tours, France, 9 Laboratoire d'Hématologie, Centre Hospitalier Universitaire De Dijon, Dijon, France, 10Department of Chemistry, University of Manitoba, Winnipeg, MB, Canada, 11Centre de Recherche en Transplantation et Immunologie UMR1064, INSERM Institut National de la Santé et de la Recherche Médicale, Université de Nantes, Nantes, France, 12 Laboratoire de Biochimie, Centre Hospitalier Universitaire (CHU) de Nantes, Nantes, France, 13 Faculté de Pharmacie, Université de Nantes, Nantes, France, 14 Faculté de Médecine, Université de Nantes, Nantes, France, 15 Laboratoire d'Hématologie, Centre Hospitalier Universitaire (CHU) de Nantes, Nantes, France*

Multiple myeloma (MM) and its pre-cancerous stage monoclonal gammopathy of undetermined significance (MGUS) allow to study immune responses and the chronology of inflammation in the context of blood malignancies. Both diseases are characterized by the production of a monoclonal immunoglobulin (mc Ig) which for subsets of MGUS and MM patients targets pathogens known to cause latent infection, a major cause of inflammation. Inflammation may influence the structure of both polyclonal (pc) Ig and mc Ig produced by malignant plasma cells *via* the sialylation of Ig Fc fragment. Here, we characterized the sialylation of purified mc and pc IgGs from 148 MGUS and MM patients, in comparison to pc IgGs from 46 healthy volunteers. The inflammatory state of patients was assessed by the quantification in serum of 40 inflammation-linked cytokines, using Luminex technology. While pc IgGs from MGUS and MM patients showed heterogeneity in sialylation level, mc IgGs from both MGUS and MM patients exhibited a very low level of sialylation. Furthermore, mc IgGs from MM patients were less sialylated than mc IgGs from MGUS patients (*p* < 0.01), and mc IgGs found to target an infectious pathogen showed a lower level of sialylation than mc IgGs of undetermined specificity (*p* = 0.048). Regarding inflammation, 14 cytokines were similarly elevated with a *p* value < 0.0001 in MGUS and in MM compared to healthy controls. MM differed from MGUS by higher levels of HGF, IL-11, RANTES and SDF-1-α (*p* < 0.05). MGUS and MM patients presenting with hyposialylated pc IgGs had significantly higher levels of HGF, IL-6, tumor necrosis factor-α, TGF-β1, IL-17, and IL-33 compared to patients with hyper-sialylated pc IgGs (*p* < 0.05). In MGUS and in MM, the degree of sialylation of mc and pc IgGs and the levels of four cytokines important for the anti-microbial response were correlated, either positively (IFN-α2, IL-13) or negatively (IL-17, IL-33). Thus in MGUS as in MM, hyposialylation of mc IgGs is concomitant with increased levels of cytokines that play a major role in inflammation and anti-microbial response, which implies that infection, inflammation, and abnormal immune response contribute to the pathogenesis of MGUS and MM.

Keywords: myeloma, monoclonal gammopathy of undetermined significance, monoclonal immunoglobulin, immunoglobulin G sialylation, infection, inflammation, cytokines

#### INTRODUCTION

Infectious pathogens are implicated in various B-cell malignancies (Burkitt, Hodgkin, and non-Hodgkin lymphoma, chronic lymphocytic leukemia) *via* cell infection and direct transformation [Epstein–Barr virus (EBV), hepatitis C virus (HCV)], or *via* antigen (Ag)-driven stimulation and indirect cell transformation (*Helicobacter pylori*), or both (1–5). Chronic cancer-associated inflammation is established in hematological malignancies, especially in myeloma and chronic myeloproliferative neoplasms (MPNs). Myeloma is characterized by the accumulation of malignant, clonal, mature plasma cells, which produce a monoclonal immunoglobulin (mc Ig): Ig G, A, or more rarely, M, D, and E. In multiple myeloma (MM), the quantity of mc Ig is ≥30 g/L and, thus, represents the majority of Ig measured in blood serum, typically ≥90% of IgGs; thus, most patients still produce polyclonal (non-malignant) IgGs, at low levels. Myeloma derives from a chronic stage called monoclonal gammopathy of undetermined significance (MGUS) (6). In MGUS the quantity of mc Ig in blood is <30 g/L, and the mc Ig may represent 20–70% of all IgGs; thus in MGUS, the production of polyclonal (pc) IgG is maintained, though frequently reduced compared to healthy individuals. Most MGUS never evolve toward smoldering myeloma (SM) and MM: the risk of transformation of MGUS into SM and MM is estimated at 1% per year per patient, and involves the repeated acquisition of genetic alterations (7, 8). Clonal plasma cells also depend on certain inflammation cytokines for their growth [for instance, interleukin 6 (IL-6)]. Inflammation-linked cytokines produced at high levels by malignant hematopoietic cells in myeloma and in other blood malignancies include hepatocyte growth factor (HGF), IL-11, IL-6, and IL-8 (9). Clonal myeloma cells also secrete factors that inhibit the growth of normal hematopoietic progenitors and suppress the formation of polyclonal Ig [tumor growth factor β1 (TGF-β1) and stroma cell-derived factor 1α (SDF-1α)].

Inflammatory cytokines are produced in large quantity in chronic hematological malignancies but the reasons remain unclear, and are likely multiple. Some cytokines may be produced by malignant cells as a consequence of genetic alterations (IL-6), but there is strong evidence that cytokines are also produced independently from gene mutations or re-arrangements, both by clonal and non-clonal cells (9, 10). Latent infection is a plausible cause of chronic overproduction of inflammation cytokines by various cell types. A promising approach to understand hematological malignancy is that for subsets of patients, abnormal immune response to infection by lymphoid or myeloid cells leads to chronic Ag-driven cell proliferation, polyclonal at first, then oligoclonal, and finally monoclonal. Over time, the chronically stimulated lineage is at the origin of an increased risk of genetic alteration leading to clonality and malignant transformation. To support this pathogenic process in MGUS and myeloma, we recently reported that six infectious pathogens, including carcinogenic viruses [EBV, HCV, Herpes simplex virus (HSV)] and bacteria (*H. pylori*), are the targets of ~23% of purified mc IgG from MGUS, SM, and MM patients (11).

In MGUS and MM, chronic inflammation may directly influence the structure and function of the mc IgG produced by the clonal plasma cells. Moreover, IgG molecules can trigger pro- or anti-inflammatory responses mediated by their crystallizable (Fc) fragment domain. Numerous studies provided evidence that carbohydrates attached to the IgG Fc domain are essential for IgG function (12). In fact, IgG Fc contains a single, highly conserved asparagine 297 (N297) glycosylation site, and antiinflammatory activities of IgG have been associated with the presence of sialic acid. Thus, patients with autoimmune diseases such as rheumatoid arthritis show low levels of IgG Fc sialylation (13, 14). Inversely, IgG sialylation increases during pregnancy, and increased IgG sialylation is associated with the remission of rheumatoid arthritis (15). Moreover, the anti-inflammatory activity of intra-venous (i.v.) immunoglobulin (IVIg) injection in various murine models is due to their sialylated Fc fragments (16–19). Similarly, anti-gp120 antibodies (Abs) in HIV patients are less galactosylated and sialylated in long-term non progressors, who are infected but asymptomatic, compared to infected patients who shows disease symptoms (20).

The pro- or anti-inflammatory effector functions of IgG subclass Abs are mediated by their different affinities for activating FcγRs (FcγRI, RIIa, RIIIa, and RIIIb) and inhibiting FcγRIIb expressed by immune cells (21–23). Several studies demonstrated that the high level of sialylation on the IgG Fc fragment decreases their Ab-dependent cell-mediated cytotoxicity (ADCC) potential through less affinity for activating receptors (24). In the context of autoimmune diseases, Kaneko et al. demonstrated that the sialylated Fc fragment of IVIg was effective in treating arthritis in a mouse model (18). Similar results were obtained using four independent *in vivo* models under preventive as well as therapeutic conditions (25). The underlying mechanisms involve SIGN-R1 in mice and DC-SIGN lectin in humans; these molecules are expressed at the surface of regulatory macrophages and bind sialylated IgGs (19, 26). Subsequently, IL-33 and IL-4 cytokines modulate the inhibitory FcγRIIb receptors expressed by effector macrophages present at the site of inflammation, raising their activation threshold (26, 27). Recently, Quast et al. (28) demonstrated that tetra sialylation of IgG Fc domain impairs complement-dependent cytotoxicity (CDC) but has no impact on ADCC. This effect is due to decreased binding of C1q to Fc-galactosylated IgG. This tetra Fc-sialylated form has been demonstrated to enhance antiinflammatory activity up to 10-fold more than IVIg across different animal models (29). Conversely, bisecting *N*-acetylglucosamines are pro-inflammatory and enhance ADCC (30). The removal of core fucose residues selectively enhances the affinity of IgG for FcγIIIa receptors, leading to an increased ADCC and decreased CDC (31). Hence, Ab that cause fetal or neonatal alloimmune thrombocytopenia have a decreased IgG1-Fc fucosylation and an increased affinity for FcγRIIIa/b receptors (32). It was demonstrated in the context of allergy that the induction of tolerance for T cell-dependent (33) or T cell-independent Ag (34) produces regulatory sialylated IgGs in mice and in humans. Tolerance was linked to an increase in α2,6-sialyltransferase in plasma cells (33). However, Jones et al. (35) recently demonstrated that IgG sialylation was B-cell-independent and that sialylation also occur in the bloodstream due to the action of a liver-secreted α2,6-sialyltransferase and the presence of platelet α-granule-derived CMP-sialic acid. Thus, this model enables a rapid functional shift in existing IgG, independent of *de novo* synthesis or recycling.

In MGUS and MM, the glycosylation state of mc IgGs has rarely been studied. While Fleming et al. (36) showed higher sialylation of IgGs from MM patients compared to MGUS patients, Nishiura et al. (37) reported less galactosylated IgGs and, consequently, hyposialylated IgGs, in MM patients compared to MGUS patients and healthy volunteers (HVs). Similarly Mittermayr et al. (38) recently described a decrease of IgG sialylation in a few MM patients in comparison to MGUS patients. In the three studies, the authors studied the glycosylation of all IgG together, without separating mc IgG from polyclonal (pc) IgGs.

Here, we report on MGUS and MM with mc IgG, the most frequent type of mc Ig in both MGUS (70–75% cases) and MM (60% cases): after purification of mc IgGs and pc IgGs from patients, the degree of sialylation of each category of IgGs was determined, and the different levels of IgG sialylation were analyzed in relation with the inflammation status of patients and the infectious pathogen targeted by the purified mc IgG.

#### MATERIALS AND METHODS

#### Patients and Ethics Statement

The study was performed with the approval of the local ethics committee (# RC12 0085, University Hospital of Nantes) and the Commission Nationale de l'Informatique et des Libertés (CNIL # 912335). For technical reasons, only mc IgG could be purified, thus only patients presenting with a mc IgG were included in this study. Thus, we examined 148 patients with mc IgG: 68 MGUS, 6 SM, and 74 MM diagnosed at the French University Hospitals (CHU) of Dijon, Nantes, Paris (Saint-Antoine) and Tours. In this study, all MGUS patients had a mc IgG ≥4 g/L. Sera from 46 HVs and 40 patients diagnosed with MPN were also studied as controls. Written informed consents were obtained from patients in the relevant clinical departments, and in the blood bank for HVs enrolled by the Etablissement Français du Sang (EFS, Nantes, France). A convention has been signed between our laboratory (CRTI—INSERM UMR 1064) and the blood bank (EFS Pays de La Loire).

#### Purification of Monoclonal and Polyclonal IgG

After clotting, blood samples of patients were centrifuged at 2,200 × *g* for 15 min at 4°C, serum was collected, and aliquots were stored at −80°C or −20°C, depending on the collecting site. Total IgG concentration in serum was measured with an immunonephelemetric assay performed on a Beckman Immage Analyzer (Beckman Coulter, Villepinte, France). The concentration of the monoclonal (component) IgG is estimated by integrating the electrophoretic peak according to the orthogonal mode (the so-called "baseline method"). Purification of pc and mc IgGs and verification of their purity were performed as described (1, 3, 11, 39). Briefly, after separation using electric charge on agarose gel electrophoresis (SAS-MX high resolution, Helena Biosciences, Gateshead, UK), bands corresponding to mc IgG or gamma zone corresponding to pc IgGs were carefully cut and proteins were eluted from gels into PBS. Concentration of the purified pc and mc IgGs was determined using the Nanodrop Spectro-photometer ND-1000 with the IgG extinction coefficient (ε = 1.36 for a solution of 1 mg/mL). The recovered IgG amount after purification varied from 40 to 70% for both mc IgG and pc IgG, depending on experiments and the initial IgG concentration in serum. Purity of each IgG fraction was analyzed by isoelectrophoresis and immunoblotting (homemade isoelectrofocusing gel using a range of pH 3–10, blotting onto PVDF membrane and revelation using an HRP anti-human IgG gamma chain). Only highly purified mc IgG were used for sialylation studies (see Figure S1 in Supplementary Material).

#### Analysis of IgG Sialylation

An enzyme-linked lectin assay (ELLA) was developed and used for IgG sialylation detection, and an enzyme-linked immunosorbent assay (ELISA) was developed for total IgG detection, as described (40). Ninety-six well plates (NuncMaxiSorp™) were coated overnight at 4°C with 50 µL of affinipure donkey anti-human IgG, Fcγ-specific fragment Ab (Jackson ImmunoResearch, West Grove, PA, USA) diluted at 1/250 (5.2 µg/mL, ELLA) and 1/1,000 (1.3 µg/mL, ELISA) in 25 mM borate buffer pH9. After 3 washes with 200 µL PBS-Tween 0.05% (Sigma, St. Louis, MO, USA), 100 µL periodic acid (5 mM) per well were added for 10 min at room temperature (RT), protected from light. The plates were then saturated with 100 µL of B-grade bovine gelatin (Sigma, St. Louis, MO, USA) 0.25% in PBS-Tween 0.01%, at 37°C, for 2 h. After three washes, samples were diluted in PBS-Tween 0.1% and deposited in triplicates containing 1.25 ng/well for detection of total IgG, or 100 ng/ well for sialylation detection. Total IgG quantity was revealed by incubating the plates with 50 µL of peroxidase affinipure donkey anti-human IgG (H + L) diluted 1/1,000 (0.8 µg/mL, Jackson ImmunoResearch, West Grove, PA, USA) for 1 h. Sialic acid was revealed using 50 µL biotinylated *Sambucus nigra* agglutinin (SNA) diluted 1/750 (2 µg/mL, Glycodiag, Orleans, France) for 90 min and then 50 µL streptavidin HRP diluted 1/1,000 (1 µg/mL, Vector laboratories, Burlingame, CA, USA) for 1 h, at 37°C. Then 50 µL of TMB, the chromogenic substrate for HRP (Sigma-Aldrich, St. Louis, MO, USA) was added and the reaction was stopped by 50 µL sulfuric acid 0.5 M after 5 min for IgG detection, and after 15 min for sialic acid detection. Optical densities (OD) were measured using Spark 10 M multimode microplate reader (Tecan, Männedorf, Switzerland) at 450 nm. The relative sialylation was expressed as the sialic acid/IgG OD ratio. Control samples were used in all experimental settings, to assess reproducibility.

### Isoelectrophoretic Studies

A 1% agarose gel containing a 10% mixture of 3–10 and 8–10.5 ampholytes was prepared and pre-focalized with acetic acid (0.5 N) and sodium hydroxide (1 N) in order to establish a pH gradient. The pre-focalization was run for 30 min at 250 V and 30 mA for a total of 90 vH. Samples were then dropped at the anode side and the focalization was launched during 90 min at 1,200 V and 50 mA for a total of 900 vH. Sera proteins were then passively transferred on a pre-activated PVDF Immobilon-P membrane (Millipore, Billenca, CA, USA) for a few minutes. Finally, after a saturation step and several washes, the membrane was incubated with the rabbit anti-human IgG (H + L)-peroxydase (Dako, Santa Clara, CA, USA). Revelation was made using HRP revelation system (Sigma, St. Louis, MO, USA).

#### Mass Spectrometry

Purified IgGs were digested by trypsin and glycopeptides were isolated from peptides using two methods, i.e., reversed-phase high-performance liquid chromatography and a protocol involving the commercial ProteoExtract® Glycopeptide Enrichment Kit (EMD-Millipore, Etobicoke, ON, Canada). Fractions were concentrated for analysis by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-ToF-MS). The instrument used was an UltraFleXtreme™ (Bruker, Bremen, Germany) operated in positive ion, reflective mode.

#### Quantification of Inflammation Cytokines

Frozen aliquots of serum were used to quantify 40 cytokines and 2 soluble cytokine receptors linked to inflammation or/and infection using the Luminex technology (Bio-Plex 200) with Bio-Plex Pro Human Cytokine Panel kits (Bio-Rad, Hercules, CA, USA), following the manufacturer's instructions.

#### MIAA Assay

The Multiplexed Infectious Antigen microArray (MIAA) assay has been described previously (11, 39). The assay was developed to determine the infectious specificity of purified IgG using commercially available Ag or/and lysates from EBV, HCV, cytomegalovirus (CMV), Herpes simplex virus 1 (HSV-1), HSV-2, varicella zoster virus (VZV), *Helicobacter pylori* (*H. pylori*), *Toxoplasma gondii* (*T. gondii*), and *Borrelia burgdorferi* (*B. burgdorferi)*. Infectious Ag were purchased from Abcam (Cambridge, United Kingdom), Advanced Biotechnologies Inc. (Columbia, MD, USA) and ImmunoDiag (Hämeenlinna, Finland). Lysates were supplied by Advanced Biotechnologies Inc. (Columbia, MD, USA) and EastCoast Bio (North Berwick, ME, USA). The arrays consist of 8 × 8 matrices that included: (i) 13 Ag: 2 for EBV, 3 for HCV, 1 for *T. gondii,* 1 for *H. pylori,* 2 for HSV-1, 2 for HSV-2, and 2 for VZV; (ii) 5 lysates: CMV, *T. gondii*, *H. pylori*, HSV-1, and HSV-2; (iii) 2 mixes: one of 5 CMV Ags and one of 2 *B. burgdorferi* Ags; (iv) 2 negative controls: PBS, and PBS with 0.1% bovine serum albumin (BSA). For hybridization, IgG concentrations were adjusted to 400 µg/mL for serum and from 50 to 200 µg/mL for purified mc IgG. 80 µL of samples were incubated for 2 h at RT. After washing, slides were incubated with a labeled secondary Ab (0.2 µg/mL DylightTM 680 Labeled Goat antihuman IgG (H + L), from SeraCare, Milford, MA, USA; Ref. 5230-0342). Fluorescence signal, detected with the Odyssey infrared imaging system scanner at 21 µm resolution (LI-COR Biosciences, NE, USA) was quantified using the GenePix® Pro 4 Microarray Acquisition & Analysis Software (Molecular Devices, Sunnyvale, CA, USA).

#### Statistics

Data analysis was performed by GraphPad Prism 6.01 software. Patient parameters were expressed as medians and ranges, or/ and means ± SEM. The chi-2 test was used for categorical variables. For continuous variables (*n* ≥ 30) the Student *t*-test or the one-way ANOVA followed by Tukey's *post hoc* test were used. For continuous variables (*n* < 30), a normality test was systematically performed for each group. When parametric conditions were fulfilled, a Student's *t*-test or a one-way ANOVA followed by Tukey's *post hoc* test was performed. For non-parametric conditions, a Mann–Whitney *U* test or a Kruskal–Wallis test followed by Dunn's *post hoc* test was performed. The tests used are indicated in Figure and Table legends. A *P* value below 0.05 was considered statistically significant.

## RESULTS

## Purification of mc and pc IgGs

According to the patients' data, in the present cohort the mc IgG represented 40.8–87.7% (median: 70.8%) of total IgGs in MGUS, and 49.7–95.5% (median: 86.3%) in MM. These percentages represent estimations, since mc IgGs and total IgGs are not measured by the same techniques. However, the data indicated that almost all MGUS and MM patients in the cohort produced some level of pc IgGs, estimated to represent ~30 and ~14% (medians) of total IgGs in MGUS and in MM, respectively. After electrophoresis of the 148 sera on agarose gel, mc IgGs and pc IgGs from each patient were separated. The relevant bands were collected, and proteins were eluted from the gel (**Figure 1A**). The purity of IgGs was confirmed by immunoblotting after isoelectophoresis. Examples of the efficiency of this method are illustrated in **Figure 1B** and Figure S1 in Supplementary Material. The typical pattern of mc IgGs resolves into a series of sharp lines that are equidistant approximately 0.05 pH units apart. This difference in pH is due to the deamidation of glutaminyl and asparaginyl residues, yielding aspartic and glutamic acid, respectively (41, 42). The purified pc IgGs appear as a smear after isoelectrophoresis and are completely separated from mc IgG (**Figure 1B**). Only highly purified mc IgGs (*n* = 148) and pc IgGs (*n* = 142) were retained for further analysis (for 6 patients, pc IgGs could not be purified). In addition to deamidation, micro-heterogeneity of mc IgGs is due to carbohydrate differences, especially the sialylation level. Hence, immunofixation of the purified mc IgG, using biotinylated SNA, a lectin specific of sialic acid, shows that the lectin recognized some of the mc IgG bands, indicating that they are not fully sialylated (**Figure 1C**).

#### Characteristics of Patients

In this retrospective study, we analyzed 148 patients presenting with mc IgG and diagnosed with MGUS (*n* = 68), SM (*n* = 6) or MM (*n*= 74). Clinical data were available for 133/148 patients (59 MGUS, 6 SM, 68 MM). The biological and clinical characteristics of the 133 patients are shown in **Table 1**. The median age for MGUS, SM, and MM patients at the time of diagnosis was similar, 67.0, 67.1, and 63.6, respectively. The male/female ratio was 55.9% for MGUS, and higher for SM and MM patients: 83.3 and 60.3%, respectively (differences not significant, Chi-2 test). The International Staging System (ISS) and Durie–Salmon Staging (DSS) scores indicated that 32.7% of MM patients presented with ISS stage III at the time of diagnosis (DSS stage III: 44.6%). In MM, mc IgG were studied either at the time of diagnosis (17 patients; 23%), or during or after treatment (40 patients; 54%); for 17 patients (23%), information was not available. In addition, two categories of control sera without mc IgG were used in this study: first, a cohort of 46 HVs (with no chronic inflammation, and no hematological disease); second, a cohort of 40 patients at the time of diagnosis of MPN, a group of chronic hematological malignancies with strong associated inflammation. As expected, MPN patients showed higher leukocyte counts, a higher hemoglobin level, and higher platelet counts compared to MGUS and MM patients.

For MGUS, SM, and MM patients, the degree of sialylation of purified pc and mc IgGs was analyzed separately. For a significant number of patients in this cohort (35/128, or 25.7%: MGUS, *n* = 19; SM/MM, *n* = 16), our previous studies had revealed that the purified mc IgG specifically targeted an infectious pathogen (11). **Table 2** shows the serological status of patients, as determined by the MIAA assay; the results of these studies reflect the Ab function of pc IgG and mc IgG analyzed together. The MIAA used here allowed testing for panels of commercially available Ag and lysates from nine infectious pathogens: EBV (HHV-4), HCV, CMV (HHV-5), HSV-1 (HHV-1), HSV-2 (HHV-2), VZV, (HHV-3), *H. pylori*, *T. gondii*, and *B. burgdorferi* (39). Overall, the serological status of MGUS patients was similar to that of the general population, which is consistent with the persistence of pc IgGs for most patients. As previously reported, there were lower rates of positive serology for SM/MM patients compared to MGUS patients, likely explained by lower production of pc IgGs in advanced MM disease (11). The characteristics of the 35 patients (19 MGUS and 16 SM/MM) with infectious pathogen-specific mc IgG ("MIAA+ patients") as determined by the MIAA assay performed with purified mc IgG, are shown in **Table 3**. The purified mc IgG of 22 patients (12 MGUS and 10 SM/MM) specifically targeted EBV, and the EBNA-1 protein was the main target (20/22 patients) (11). The



*Nbr, number of patients; HV, healthy volunteers; NA, not available; ND, not determined; ISS, International Staging System; DSS, Durie–Salmon Staging.*

#### Table 2 | Serological status of patients.


*Sera of patients, which include both pc and mc IgG, were analyzed with the Multiplexed Infectious Antigen microArray (MIAA) assay. Results represent the number and % of patients with IgG in serum (pc IgG* + *mc IgG) that target the pathogens of the MIAA assay. Nbr, number.*

purified mc IgG from the 13 other patients specifically targeted HSV-1 (*n* = 6), CMV (*n* = 2), VZV (*n* = 2), and *H. pylori* (*n* = 3) (11). The sialylation results of mc and pc IgGs from the 35 MIAA+ patients were analyzed separately.

#### IgG Sialylation Level

The sialylation state of pc and mc IgGs is shown in **Figure 2A**. Analysis of the cohort of healthy controls showed that 95.6% of pc IgGs have a relative sialylation level between 0.5 and 1.5. The mean relative sialylation level of IgGs from healthy individuals was 1.023. Of note, 100% of pc IgGs samples from MPN patients had a similar relative sialylation level, between 0.5 and 1.5 (mean: 1.096). By contrast, pc IgGs from 142 MGUS, SM, and MM patients showed a large heterogeneity in relative sialylation level. For 52.1% of MGUS, SM, and MM patients, pc IgGs presented the same sialylation level as pc IgGs from healthy controls and MPN patients, but for 36.6% of MGUS, SM, and MM patients, pc IgGs were more sialylated, and for 11.3% of these patients, pc IgGs were less sialylated. Overall the degree of sialylation of pc IgGs from the MGUS/SM/MM group was significantly higher than for the group of healthy controls (mean = 1.703, *p* = 0.0007) or the group of MPN patients (*p* = 0.0062). Regarding mc IgG, Table 3 | Description of patients presenting with a pathogen-specific purified mc IgG, as assessed with the Multiplexed Infectious Antigen microArray (MIAA) assay (MIAA+ patients).


*Statistical analysis was performed using the chi-2 test for categorical variables and the Mann–Whitney U test for continuous variables. Significant differences are indicated. NS, not significant; ISS, International Staging System; DSS, Durie–Salmon Staging.*

for 75% of MGUS and SM/MM patients, the purified mc IgG presented with strongly reduced sialylation; only 4.1% of patients showed a high sialylation of purified mc IgG (**Figure 2A**). The sialylation level of purified mc IgGs from the MGUS/SM/MM group (mean = 0.443) was significantly lower than pc IgGs from the healthy controls (*p* = 0.0054), MPN patients (*p* = 0.0025), and MGUS/SM/MM patients (*p* < 0.0001) (**Figure 2A**). When MGUS and SM/MM were compared, the sialylation level of pc IgGs was not different, whereas that of mc IgGs was lower in SM/ MM (mean = 0.271) than in MGUS (mean = 0.645, *p* = 0.0048) (**Figure 2B**). Regarding SM/MM patients with mc IgGs specific for a pathogen (MIAA+ patients), the sialylation level of pc IgGs (MGUS or SM/MM) was not different from other pc IgGs (**Figure 2C**) but pathogen-specific mc IgG were significantly less sialylated (mean = 0.117) than all other mc IgGs (mean = 0.486, *p* = 0.048) (**Figure 2D**).

#### Relative Quantification of Sialylation Level by Mass Spectrometry

In a second series of experiments, we used HPLC-mass spectrometry to confirm the differences in IgG sialylation level observed between purified mc and pc IgGs. Mass spectrometry analysis was performed on 24 samples. Figure S2 in Supplementary Material shows the results obtained for two representative purified mc IgGs that differed in their sialylation level as assessed with ELLA and ELISA. Figure S2A in Supplementary Material shows a purified mc IgG1 that was found to be highly sialylated using the ELLA and ELISA techniques. Conversely, Figure S2B in Supplementary Material shows a purified mc IgG1 found to be poorly sialylated using the ELLA and ELISA techniques. In both cases, the sialylated forms corresponding to G1FS (1 galactose, 1 fucose, and 1 sialic acid) and G2FS (two galactoses, one fucose, and one sialic acid) with a *m/z* at 3,087 and 3,249, respectively, are indicated with red arrows. The respective percentages of IgG1 in G1FS and G2FS forms were 26.4% (Figure S2A in Supplementary Material) and 5.2% (Figure S2B in Supplementary Material), thus confirming the results obtained by ELLA and ELISA (**Figure 2**). We then compared the results obtained for six healthy individuals and eight patients (four MGUS and four MM) (**Figure 3**). The sialylation level of pc IgGs from healthy controls was not different from those of pc IgGs from MGUS/MM patients. Purified mc IgGs from MGUS/MM patients were less sialylated (mean = 7.35%) than pc IgGs from the same patients (mean = 10.21%; paired *t*-test: *p* = 0.03) or from healthy controls (mean = 14.05%; *p* = 0.02), thus confirming the results obtained by ELISA (**Figure 2**).

#### Pro-inflammatory Status of MGUS and MM Patients

For 64 patients (34 MGUS and 30 MM), we quantified in blood serum the level of 40 cytokines and 2 soluble cytokine receptors [IL-1 receptor α (IL-1Rα), IL-2Rα] linked to inflammation and, for certain molecules [interferon (IFN) α2, IFN-γ, eotaxin, IL-17, IL-22, IL-26, and IL-33] to anti-viral or anti-microbial immune responses (**Table 4**). Compared to the group of healthy controls,

IgGs from healthy volunteers (*n* = 46) and myeloproliferative neoplasms (*n* = 40), and from purified mc IgGs (*n* = 148) and pc IgGs (*n* = 142) from MGUS and SM/ MM patients (for 6 patients, pc IgGs could not be purified). Percentages indicate the % of patients whose IgGs present a low (<0.5), normal (0.5–1.5), of high (>1.5) level of sialylation. (B) Degree of sialylation of purified pc IgGs and mc IgGs from MGUS (*n* = 68) and SM/MM (*n* = 80) patients. Sialylation level of purified pc IgGs (C) and mc IgGs (D) from MM patients with pathogen-specific mc IgG (MIAA+ patients) were then compared with pc and mc IgGs from other MM patients. Bars indicate means ± SEM. Statistical analysis was performed using one-way ANOVA test followed by Tukey's Multiple Comparison Test for (A), unpaired *t*-test for (B) and (C), and unpaired *t*-test with Welch's correction for (D). \**p* < 0.05, \*\**p* < 0.01, \*\*\**p* < 0.001.

MGUS/SM/MM patients had significant increases in serum levels of 28 molecules: IL-1Rα, IL-2Rα, and 26 cytokines. Among those, 17 cytokines were elevated in the patients group with *p* < 0.001, compared to healthy controls: IL-6, IL-7, IL-10, IL-13, IL-15, IL-17, IL-33, IFN-α2, tumor necrosis factor (TNF)-α, granulocyte colony stimulating factor (G-CSF), granulocyte-macrophage CSF (GM-CSF), basic fibroblast growth factor (FGF), vascular endothelial growth factor (VEGF), monokine induced by IFN-γ (MIG, or CXCL9), monocyte chemotactic protein 1 (MCP-1 or CCL2), HGF, and leukemia inhibitory factor (LIF). Most of these cytokines are pro-inflammatory, which indicates a shift toward a pro-inflammatory environment in MGUS and SM/MM patients.

The comparison between the MGUS and SM/MM groups (**Figure 4**; Table S1 in Supplementary Material) revealed that only four molecules were slightly but significantly higher in SM/ MM patients: HGF, IL-11, SDF-1α, and RANTES. Interestingly, HGF is a survival factor which exerts a pro-tumoral and both pro- and anti-inflammatory action. TGF-β1, TGF-β2, and TGFβ3, known for their anti-proliferative and pro-differentiation effect on hematopoietic cells, were more expressed in MGUS than in SM/MM. Since high levels of HGF are considered of poor prognosis in MM, we explored eventual correlations between cytokine levels and β2-microglobulin, an important biomarker in the prognosis of MM. **Figure 5** shows the positive correlations found between β2-microglobulin concentration and serum levels of IL-9, IL-26, MIP-1β (pro-inflammatory molecules), and PDGF-BB. Of note, IL-26 is involved in antimicrobial immunity.

Figure 3 | Sialylation of IgGs as assessed by mass spectrometry: Purified pc IgGs from six healthy volunteers (HV) and pc IgGs and mc IgGs from eight patients [four monoclonal gammopathy of undetermined significance (MGUS) and four multiple myeloma (MM)] were analyzed by mass spectrometry. The percentage of sialylated glycoforms (G1FS + G2FS/total peaks) was determined in pc and mc IgG fractions. Bars indicate means ± SEM. Statistical analysis was performed using one-way ANOVA test after a Kolmogorov–Smirnov normality test followed by Tukey's comparison test ( \*1*p* < 0.05); and a paired *t*-test for pc IgG and mc IgG from MGUS and MM patients (\*2*p* < 0.05).

### IgG Hyposialylation, Pro-inflammatory Status, and Disease Severity

Hyposialylation of IgGs has been described as a hallmark of proinflammatory state in pathologic contexts other than MM. We analyzed cytokine levels in the serum of MGUS, SM, and MM patients according to the sialylation level of their pc IgGs. **Figure 6** shows that patients presenting with hyposialylated pc IgGs have significantly higher levels of major pro-inflammatory cytokines (IL-6, TNF-α, TGF-β1), and also have higher levels of HGF, IL-17, and IL-33, compared to patients with hyper-sialylated pc IgGs (*p*< 0.05). These results were confirmed when data were analyzed for potential correlations between cytokine levels and the degree of sialylation of pc IgGs or purified mc IgGs. In SM/MM group, the degree of sialylation of pc IgGs was negatively correlated with levels of IL-17 and IL-33, and also with the concentration of leptin (men only) (**Figure 7A**). The degree of sialylation of purified mc IgGs was also negatively correlated with leptin (men only) (**Figure 7B**), and positively correlated with the levels of IFN-α<sup>2</sup> and IL-13. Thus, in MM, sialylation of pc IgGs is inversely correlated with the levels of IL-17 and IL-33, two cytokines important for anti-microbial response. The sialylation level of purified mc IgGs increased with the level of 2 anti-inflammatory cytokines: IFN-α2 and IL-13.

In addition, we found that sialylation of purified mc IgGs was significantly lower for MM patients with advanced disease— Durie–Salmon staging (DSS) III—than for those with DSS I (**Figure 8**). Similarly, C-reactive protein (CRP) concentration, a common inflammation marker, was inversely correlated with the sialylation level of pc IgGs (MGUS, SM, and MM patients, Figure S3 in Supplementary Material), and the β2-microglobulin concentration was inversely correlated with the sialylation level of purified mc IgGs (MGUS, SM, and MM patients, Figure S3 in Supplementary Material). Altogether, these findings suggest that mc IgG hyposialylation could be a new marker of disease severity in MM.

## DISCUSSION

There is increasing evidence that chronic Ag stimulation, Ag-driven selection of plasma cell clones and subsequent mc IgG production, initiate MGUS and MM for subsets of patients. Nair et al. identified lyso-glucosylceramide (LGL1) as a potential target of mc Ig of MGUS and MM patients (43). We recently reported that six infectious pathogens, including carcinogenic viruses (EBV, HCV, and HSV) and bacteria (*H. pylori*), are the targets of ~23% of purified mc IgGs from MGUS, SM, and MM patients (11). Thus, for a significant percentage of patients, MGUS may result from chronic infectious Ag-driven clone proliferation and abnormal immune response, either to selfproteins or to infectious pathogens. Over time, the chronically stimulated lineage is at increased risk of genetic alteration and subsequent malignant transformation and overt MM. Several studies demonstrated that efficient anti-viral activity is associated with a dramatic shift in the IgG-glycosylation profile toward agalactosylated, afucosylated, and asialylated glycans (20, 44). Here, we demonstrate that the purified mc IgGs from MGUS, SM, and MM patients exhibit a very low level of sialylation in comparison to pc IgGs from HVs and pc IgGs from the same MGUS, SM, and MM patients. Furthermore, purified mc IgGs from MM patients were less sialylated than mc IgGs from MGUS patients, and purified mc IgGs targeting specifically an infectious pathogen had an even lower sialylation status than mc IgGs not directed against an infectious pathogen of the MIAA test. Moreover, hyposialylation of purified mc IgGs was concomitant with increased levels of cytokines that played a major role in inflammation and anti-microbial response. These results suggest that infection, inflammation, and an abnormal immune response are early events in a subset of MGUS and MM.

At least two main questions arise from these findings. The first concerns the molecular mechanisms that lead to hyposialylation of mc IgGs, and the link with inflammation. The second concerns the functional consequences of IgG hyposialylation on both the activation of macrophages and their subsequent production of pro- or anti-inflammatory cytokines, and on the Ab function of hyposialylated mc IgGs, especially when they target a pathogen.

Regarding the links between mc IgG hyposialylation and inflammation in chronic hematological malignancies, our results show that the inflammatory environment in MGUS and in MM is associated with the production of poorly sialylated mc IgG. Yet it is unlikely that alone, inflammation is sufficient to explain hyposialylation of mc IgGs in all cases, or all pc IgGs would be Table 4 | Cytokine profile of healthy volunteers (HVs), monoclonal gammopathy of undetermined significance (MGUS), smoldering myeloma (SM), and multiple myeloma (MM) patients.


*MGUS and MM* > *HV: cytokines elevated in MGUS and MM patients vs HV; statistically significant P values are indicated in bold. Statistical analysis was performed using the Mann– Whitney U test. Leptin results were analyzed according to sex.*

*ND, not detectable (below detection level).*

hyposialylated in MM. Hyposialylation of pc IgGs is observed in MM, but for only 11% of patients. Similarly, pc IgGs of MPN patients were normally sialylated, although chronic inflammation is more severe in MPNs than in MGUS and MM. Analysis of the expression and/or activity of the α2,6-sialyltransferase and sialidase enzymes in clonal plasma cells may be more informative to understand the potential mechanisms leading to hyposialylation of mc IgG. Indeed, Oefner et al. (33) showed that the stimulation with T-cell-dependent Ag under inflammatory conditions can result in the production of plasma cells that express low levels of α2,6-sialyltransferase and secrete desialylated IgGs. Fc sialylation is, thus, crucial for the differentiation between a tolerogenic immune status and a pathogenic immune status, the latter being directly correlated to the α2,6-sialyltransferase activity of plasma cells (33). Similar results were observed in T-cell acute lymphoblastic leukemia, where a decreased sialylation of membrane proteins was directly correlated with α2,6-sialyltransferase mRNA expression and activity (45). Furthermore, low IgG sialylation

and 2 cytokine receptors were quantified using the Biorad Luminex technology in the serum of 34 MGUS and 30 MM patients. The 6 cytokines found to be differently expressed in MGUS vs MM patients were: IL-11, HGF, SDF-1α, RANTES, TFG-β1, TFG-β2, and TGF-β3. Horizontal bars indicate median values ± ranges. Statistical analysis was performed using Student *t*-test. \**p* < 0.05 and \*\**p* < 0.01.

due to increased sialidase activity has been described in various cancers (46, 47). Unfortunately, we did not have access to patient plasma cells in the present study and the activity of these enzymes could not be investigated.

When inflammation was associated with specific viral Ag-specificity of the mc IgG, further decrease in the sialylation level of mc IgGs was observed in MM, especially for patients in DSS stage III (**Figure 8**). These observations are consistent with a deleterious effect of the hyposialylation of mc IgGs in MM disease evolution. It is established that the sialylation of IgGs dramatically changes their physiologic role, converting IgG from pro-inflammatory into anti-inflammatory (18, 26). The small fraction of sialylated IgGs is responsible for the immunosuppressive activity of IVIg (16). Recently, Barrios et al. (48) showed that the sialylation level of IgG structures decreased in patients with chronic kidney disease, thus demonstrating that sialylated glycans play an independent protective role in chronic kidney disease. Moreover, we recently demonstrated that the sialylation of the anti-donor

specific Ab against HLA class I are more sialylated in kidney transplant recipients who do not develop Ab-mediated rejection, than in patients who develop Ab-mediated rejection (40). Also, Quast et al. (28) demonstrated that the increase of the Fc IgG sialylation impairs the CDC due the inhibition to C1q binding. In fact, crystallographic and biophysical studies of sialylated and asialylated IgG Fc fragments showed that IgG Fc sialylation leads to conformational changes in the protein (49, 50).

The initial discovery that IgG sialylation plays a key role in the suppressive activity of IVIg in autoimmunity was a hallmark in the appreciation of the role of glycans in immune responses (16–18, 27). In murine models, sialylated IgGs bind to DC-SIGN, inducing the production of IL-33, an infection-linked cytokine that activates basophils to produce IL-4, leading to the upregulation of the inhibitory Fc receptor FcγRIIb (27, 51). In our study, the IL-33 and IL-4 levels were not different between MGUS and MM patients (**Figure 4**; Table S1 in Supplementary Material), thus in MM the IL-33 level may not be sufficient to induce IL-4 and expression of FcγRIIb. In fact, Musolino et al. (52) found that the IL-33 plasma levels were reduced in MM and were associated with more advanced disease. These observations support the hypothesis that latent infection and inflammation could be the early events for subsets of MGUS and MM, and *via* their pro-inflammatory effects, the hyposialylated IgGs (mc IgGs and in 11% of cases, pc IgGs also) contribute to the inflammatory environment and the progression to myeloma.

Our study shows that inflammation occurs early in myeloma pathogenesis since a very similar chronic state of inflammation was observed in MGUS and MM patients *vs* healthy controls: 35/42 molecules linked to inflammation were similarly increased in MGUS and MM. The similar inflammatory status observed in MGUS, considered a benign condition, and MM, a severe, overtly malignant and often invalidating disease, was unexpected. A such observation was observed by Zheng et al. (53) who showed that only IL-17 was highly increased in MM patients in comparison to MGUS patients but their cohort included 55 MM patients and only 8 MGUS patients. HGF was more strongly expressed in MM. HGF plays an important role in MM, inducing IL-11 and IL-6, two markers of disease activity and poor prognosis in MM (54–56). HGF and IL-11 are anti-inflammatory cytokines; in addition, IL-11 promotes bone destruction by osteoclasts and inhibits bone formation by osteoblasts, thus causing cancer-induced bone lesions (57). We also confirmed that TGF-β is less expressed in MM than in MGUS (58). This could be explained by the antiproliferative and pro-differentiation effects on hematopoiesis of TGF-β *in vivo* (59).

In the context of MM, we found positive correlations with β2-microglobulin concentration and expression levels of IL-9, IL-26, MIP-1β and PDGF-BB (**Figure 5**). β2-Microglobulin helps

patients with hyposialylated or hyper-sialylated pc IgGs. Forty cytokines and 2 cytokine receptors were quantified using the Biorad Luminex technology in the serum of 19 MGUS or SM/MM patients with hyposialylated pc IgGs (hypo), and 16 MGUS or SM/MM patients with hyper-sialylated pc IgGs (hyper). The 6 cytokines found to be more expressed in patients with hyposialylated pcIgGs were interleukin 6 (IL-6), hepatocyte growth factor (HGF), tumor necrosis factor (TNF)-α, TGF-β1, IL-17, and IL-33. Horizontal bars indicate median values ± ranges. Statistical analysis was performed using Mann–Whitney *U* test. \**p* < 0.05, \*\**p* < 0.01, and \*\*\**p* < 0.001. Normal values for IL-6: <9 pg/mL; HGF, median: 195 pg/mL, range: 63–1,283 pg/mL; TNF-α, median: 0 pg/mL, range: 6–98 pg/mL; TGF-β1, median: 47 pg/mL, range: 0–932 pg/mL; IL-17, median: 0 pg/mL, range: 0.22–31 pg/mL; and IL-33, not defined.

to characterize the severity and define the stage and prognosis of MM. Like MIP-1α, MIP-1β plays a role in hematopoiesis and osteoclast recruitment, and MIP-1α/β secretion correlates with lytic bone lesions in MM patients (60, 61). Several cytokines, including VEGF and PDGF-BB, are released by MM tumoral cells and also by endothelial cells, thereby contributing to the marked bone marrow micro-vessel density, a constant hallmark of active MM and of acquired refractoriness of MM plasma cells to conventional therapies (62, 63). Consistently, anti-tumor/vessel dasatinib, an inhibitor of PDGF receptor, significantly delays MM plasma cell growth and angiogenesis *in vivo* (64).

Since IgG hyposialylation is a hallmark of pro-inflammatory state, we investigated whether sialylation of IgG was linked to the secretion of specific pro-inflammatory cytokines. Such studies had never been done previously. We found that patients who have hyposialylated pc IgGs in addition to mc IgGs secrete several proinflammatory cytokines (IL-6, TNF-α, TGF-β, IL-17, and IL-33) or cytokines involved in MM progression (HGF) at significantly higher levels than patients with hyper-sialylated pc IgGs. The increased production of IL-17 may be of interest: in addition to its anti-microbial action, inhibition of Th1 response, and production of pc IgM and IgA, IL-17 has been shown to act directly on the expression of α2,6-sialyltransferase. Moreover, this enzyme is downregulated by IL-17 in autoimmune diseases, leading to decreased IgG sialylation (65). Accordingly, we found an inverse correlation between levels of IL-17 (and IL-33) and the degree of sialylation of pc IgGs in both MM and MGUS. Conversely, the levels of anti-inflammatory cytokines IFN-α2 and IL-13 were positively correlated with mc IgG sialylation. We also found an inverse correlation between the sialylation level of both pc and mc IgG and the leptin level in male patients. An increase in leptin level in serum has been observed in blood malignancies (66, 67) and particularly in newly diagnosed MM (68). This adipokine induces pro-inflammatory IL-1β as well as the expression of IL-6, TNF-α and many genes involved in the growth and metabolism of MM plasma cells (68, 69).

On the basis of our previous (11) and present findings, we propose a schematic model of the relation between chronic

mc IgG sialylation data). Statistical analysis was performed using the Spearman *t*-test.

inflammation/infection and the structure/function of mc IgG in myeloma (**Figure 9**). Over-expression of HGF is observed in MM vs MGUS, as well as IL-22, IL-26, and IL-33 in MM patients with a pathogen-specific mc IgG (MIAA+) vs MM patients with mc IgG of undetermined specificity, is consistent with the presence of a pro-inflammatory microenvironment. Under these conditions, MM plasma cells produce large quantities of hyposialylated mc IgGs, which activate macrophages *via* FcγRs. The resulting secretion by activated macrophages of TGF-β1, TNF-α, and IL-6 stimulates the production of Th17 cells, which secrete IL-17, IL-22, and IL-26. IL-22 is a marker of poor prognosis in MM (70), and IL-17 can induce the downregulation of sialyltransferase activity, thus maintaining the hyposialylation of mc IgG and to a lesser extent, of pc IgGs too (33).

## CONCLUSION

In MGUS, SM, and MM, mc IgGs are hyposialylated compared to pc IgGs from both healthy controls and MPN patients. Mc IgG hyposialylation was lowest in MM, particularly when the purified mc IgG targeted an infectious pathogen. Although the exact mechanisms of IgG hyposialylation in MM remain to be identified, IgG hyposialylation correlated with the overproduction of several cytokines (IL-17, IFN-α2, IL-33, and IL-13) that play a major role in inflammation and anti-microbial response. Altogether, the data suggest that infection, inflammation, and an abnormal immune response are early events for subsets of MM patients.

## ETHICS STATEMENT

The study was performed with the approval of the local ethics committee (# RC12 0085, University Hospital of Nantes) and the Commission Nationale de l'Informatique et des Libertés (CNIL # 912335).

#### AUTHOR CONTRIBUTIONS

JH, SH, AB, and EB-C designed the research, analyzed data, and wrote the paper. AB, JH, SA-M, NM, and HP performed experiments. AT, CR, DC, LG, PM, EP, and FG contributed patient samples and data, and critically read the manuscript. AN contributed to the statistical analysis of data. HP, AN, and SB critically read the manuscript. All authors gave final approval of the version to be submitted to publication and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved.

## ACKNOWLEDGMENTS

We thank all the colleagues from the Departments of Hematology or Internal Medicine of the University Hospitals of Dijon, Nantes, Paris Saint-Antoine and Tours, who contributed to the diagnosis and care of patients in this study. We also thank Emy Komatsu and Luka Markovic (Winnipeg, MB, Canada) for technical help with mass spectrometric measurements.

## FUNDING

This work was realized in the context of the IHU-Cesti project, which received funds from the French government *via* the Region Pays de la Loire. This work was realized in the context of the LabEX IGO program supported by the National Research Agency (ANR) *via* the "Investment for the future" program ANR-11-LABX-0016-01. The study was also supported by grants from the Comités Départementaux of Loire-Atlantique, Maine et Loire, Vendée and Finistère from the Ligue Nationale contre le Cancer, to EB-C (2013–2014); by a grant from the Cancéropôle Grand Ouest and Région Pays de la Loire, to SH and JH (HII-GO project, 2015–2017); and by a grant from the Cancéropôle Grand Ouest and Région Centre, to EP (2015–2017). The Cancéropôle Grand Ouest and Région Pays de la Loire financed the salary of AB (2015–2016).

### REFERENCES


## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at http://www.frontiersin.org/article/10.3389/fimmu.2017.01347/ full#supplementary-material.


high performance liquid chromatography: role of glycosyltransferases in the structures. *Cancer Res* (1990) 50:5345–50.


multiple myeloma during autologous stem cell transplantation. *Clin Cancer Res* (2014) 20:1366–74. doi:10.1158/1078-0432.CCR-13-2442


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflicts of interest.

*Copyright © 2017 Bosseboeuf, Allain-Maillet, Mennesson, Tallet, Rossi, Garderet, Caillot, Moreau, Piver, Girodon, Perreault, Brouard, Nicot, Bigot-Corbel, Hermouet and Harb. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Intestinal Microbiome Shifts, Dysbiosis, Inflammation, and Non-alcoholic Fatty Liver Disease

#### Emma T. Saltzman1,2 \*, Talia Palacios 1,2, Michael Thomsen1,2 and Luis Vitetta1,2 \*

*<sup>1</sup> Sydney Medical School, University of Sydney, Sydney, NSW, Australia, <sup>2</sup> Medlab Clinical, Sydney, NSW, Australia*

Adverse fluctuations in the distribution of the intestinal microbiome cohort has been associated with the onset of intra- and extra-intestinal inflammatory conditions, like the metabolic syndrome (MetS) and it's hepatic manifestation, non-alcoholic fatty liver disease (NAFLD). The intestinal microbial community of obese compared to lean subjects has been shown to undergo configurational shifts in various genera, including but not limited to increased abundances of *Prevotella, Escherichia, Peptoniphilus,* and *Parabacteroides* and decreased levels of *Bifidobacteria, Roseburia,* and *Eubacteria* genera. At the phylum level, decreased *Bacteroidetes* and increased *Firmicutes* have been reported. The intestinal microbiota therefore presents an important target for designing novel therapeutic modalities that target extra-intestinal inflammatory disorders, such as NAFLD. This review hypothesizes that disruption of the intestinal–mucosal macrophage interface is a key factor in intestinal-liver axis disturbances. Intestinal immune responses implicated in the manifestation, maintenance and progression of NAFLD provide insights into the dialogue between the intestinal microbiome, the epithelia and mucosal immunity. The pro-inflammatory activity and immune imbalances implicated in NAFLD pathophysiology are reported to stem from dysbiosis of the intestinal epithelia which can serve as a source of hepatoxic effects. We posit that the hepatotoxic consequences of intestinal dysbiosis are compounded through intestinal microbiota-mediated inflammation of the local mucosa that encourages mucosal immune dysfunction, thus contributing important plausible insight in NAFLD pathogenesis. The administration of probiotics and prebiotics as a cure-all remedy for all chronic diseases is not advocated, instead, the incorporation of evidence based probiotic/prebiotic formulations as adjunctive modalities may enhance lifestyle modification management strategies for the amelioration of NAFLD.

Keywords: intestinal microbiome, intestinal epithelial cell dysbiosis, dysbiosis, macrophage, inflammation, mucosal immunity, NAFLD

#### INTRODUCTION

NAFLD is a growing public health concern, laying claim to both a steadily rising prevalence as well as an increasingly young age at diagnosis (Welsh et al., 2013; Nobili et al., 2014). These trends reflect the increased rates of risk factors associated with an obesogenic lifestyle and the development of type 2 diabetes mellitus (T2DM) (Targher et al., 2007; Welsh et al., 2013). The prevalence of NAFLD in obese adults or those with T2DM has been reported to be 67.5 and 74%, respectively,

#### Edited by:

*Gayane Manukyan, Institute of Molecular Biology (NAS RA), Armenia*

#### Reviewed by:

*Michael Kogut, Agricultural Research Service (USDA), United States Mario M. D'Elios, University of Florence, Italy*

#### \*Correspondence:

*Emma T. Saltzman esal4025@uni.sydney.edu.au Luis Vitetta luis.vitetta@sydney.edu.au; luis\_vitetta@medlab.co*

#### Specialty section:

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology*

Received: *02 August 2017* Accepted: *10 January 2018* Published: *30 January 2018*

#### Citation:

*Saltzman ET, Palacios T, Thomsen M and Vitetta L (2018) Intestinal Microbiome Shifts, Dysbiosis, Inflammation, and Non-alcoholic Fatty Liver Disease. Front. Microbiol. 9:61. doi: 10.3389/fmicb.2018.00061* (Williams et al., 2011; Paquissi, 2016) compared to only 25% in the general adult population (Younossi et al., 2016). As a spectrum of diseases, NAFLD has been associated with significant morbidity and mortality, with advanced forms of the disease expressed as fibrosis, cirrhosis, and hepatocellular carcinoma (HCC).Additionally, insulin resistance (IR) and obesity have been identified as NAFLD risk factors (Gaggini et al., 2013), with NAFLD also reported to increase the risk of T2DM and cardiovascular disease, justly classifying NAFLD as the hepatic manifestation of the metabolic syndrome (MetS) (Adams et al., 2005; Targher et al., 2007, 2010; Dunn et al., 2008; Starley et al., 2010; Gregor and Hotamisligil, 2011; Lumeng and Saltiel, 2011; Dietrich and Hellerbrand, 2014; Paolella et al., 2014; Paquissi, 2016). NAFLD hence presents as a multi-systemic disease (Starley et al., 2010; Paolella et al., 2014). In light of the potential of NAFLD to progress to an ever-increasing prevalence, an appreciation of the molecular mechanisms that facilitate the manifestation of NAFLD disease chronicity and the pathways that trigger the transition across the spectrum of diseases is a pertinent aspect in designing novel effective therapeutic modalities. With the purported pathogenic mechanisms of NAFLD being intertwined with peripheral IR, increased liver lipolysis is reported to contribute to increased levels of hepatic free fatty acids (FFAs) (Paolella et al., 2014). IR simultaneously triggers increased gluconeogenesis and reduced glyconeogenesis, which further increases the production of circulating FFAs (Paolella et al., 2014).

The definition of NAFLD covers a spectrum of liver histologies, ranging from the benign and usually non-progressive simple steatosis (SS) (Tilg and Moschen, 2010; characterized by the accumulation of lipid droplets that exceed 5% of the total weight of the liver) to non-alcoholic steatohepatitis (NASH) which is found in 30% of NAFLD patients and is described as the beginning stages of inflammation and lobular ballooning which results from persistent hepatic injury (Jiang et al., 2015). Chronic inflammation can cause the liver to respond with compensatory tissue repair, a mechanism that initiates fibrosis and even cirrhosis through collagen deposition and scarring—which eventually forms the foundation of hepatocellular cancer (HCC) in extremely rare cases (Tilg and Moschen, 2010; Bieghs and Trautwein, 2014). Systems biology continues to probe the pathophysiology of this disease and the factors that establish it's manifestation and drives disease progression across the spectrum toward severe phenotypes.

The manifestation and progression of NAFLD is tentatively attributed to a range of factors as part of the multiple parallel hits hypothesis, which postulates that NAFLD is the result of inflammation in the liver induced by numerous intestinal-derived or adipose tissue-derived triggers (Tilg and Moschen, 2010). With the first hit said to be the onset and maintenance of SS, and the additional hits of gut-derived endotoxins and pro-inflammatory cytokines from adipose tissue reported to provide the impetus for NASH development and subsequent progression (Day and James, 1998). This hypothesis is flagging the importance of aberrant innate immunity as a central pathway for NAFLD progression (Miele et al., 2009; Tilg and Moschen, 2010) along with stress signaling networks and circulating adipocytokines and pro-inflammatory cytokines (Miele et al., 2009; Tilg and Moschen, 2010).

Although the molecular pathways that lead to the pathogenesis and progression of NAFLD remain poorly understood, it is accepted that inflammation is a major factor in hepatic injury (Bieghs and Trautwein, 2014). Currently, experimental data suggests that interactions of the innate immune system with the different resident liver cell types help perpetuate and maintain adverse inflammatory responses in the liver (Starley et al., 2010; Farrell et al., 2012; Bieghs and Trautwein, 2014). Serum markers of inflammation, including C-reactive protein (CRP), interleukins (ILs), and other general immunity markers are associated with the diagnosis and prognosis of NAFLD (Chiang et al., 2010; Harley et al., 2014), whilst at the cellular level, data has implicated an imbalance in T helper 17 (Th17) cells over regulatory T (Treg) cells, that occurs from an over-differentiation of T helper cells (Hammerich et al., 2011). A disturbance to the equilibrium between Th17 and Treg cells is a key event in the initiation of pro-inflammatory activity.

The innate immune system responds to cell damage or pathogenic invasion through pattern recognition receptors (PRRs), that are expressed intracellularly or on the surface of resident liver cells (Bieghs and Trautwein, 2014). These PRRs are programmed to detect damage-associated molecular patterns (DAMPs) that are released by injured cells or pathogen-associated molecular patterns (PAMPs), which are derived from intestinal bacterial metabolites (Pedra et al., 2009; Takeuchi and Akira, 2010). From a consideration of the consequences of intestinal epithelial dysbiosis, this review hypothesizes that the cascade of signals that activate adverse innate immune system and inflammatory activity implicated in NAFLD pathophysiology are triggered by the continuous release of endotoxins and other intestinal bacterial-derived products which can reach the liver through the gut-liver axis interface.

Resident liver cells have their own PRRs in toll-like receptors (TLRs). Activating these TLRs is an important step in the development of NAFLD as they are responsible for inducing gene transcription that facilitates responses of the innate immune system (Takeuchi and Akira, 2010). Kupffer cells, stellate cells, and hepatocytes amongst others, express TLRs and recognize a large array of PAMPs, which enable pro-inflammatory activity by activating different liver cells. PAMPs and DAMPs as well as irritants from the host's environment are themselves recognized by inflammasomes, receptors of the innate immune system. In response to sensing infectious microbes, inflammasomes are responsible for the pro-inflammatory activity observed in the initiation or manifestation of inflammatory diseases, like NAFLD. NLRP3 is an inflammasome that has been reported to be specifically and critically involved in NAFLD progression, with experimental and clinical data identifying higher levels of expression in NASH-affected subjects.

The intestinal microbiome is central to the narrative that NAFLD manifestation is largely a consequence of dysregulated innate immunity in response to persistent pro-inflammatory activity. The role of the intestinal microbiome is multi-factorial, functioning as an immunological, metabolic, and protective tool for optimal host health. When the intestinal microbiome is in dysbiosis, the health of the host is compromised as the microbiome is unable to maintain control of local homeostasis, increasing intestinal permeability. Disruption to intestinal epithelial homeostasis leads to hepatic exposure to exogenous and endogenous antigens that drives hepatoxic influences via the gut-liver axis interface (Glavan et al., 2016).

#### THE INTESTINAL MICROBIOME AND THE EPITHELIAL BARRIER

The microbial ecosystem of the gastrointestinal tract (GIT) comprises a metabolically and immunologically complex and active organ (Vitetta et al., 2013). Recognized as the most biodiverse and dense microbial site, the intestinal microbiome is estimated to harbor over 10<sup>14</sup> bacterial cells (Jiang et al., 2015). Serving as a key influence of host health, the GIT is the site that facilitates the exposure of environmental, dietary, and microbial antigens to the immune system (Vitetta et al., 2013). Existing in a state of symbiotic homeostasis, the intestinal microbiome and the immune system largely co-develop from birth (Nicholson et al., 2012). Subject to an array of complex interactions, dependent on host genetics, lifestyle, dietary and environmental cues, the microbiome and host share a bi-directional relationship, with both parties helping to shape each other's development and composition (Nicholson et al., 2012).

The intestinal epithelial barrier is a dynamic network made up of luminal and mucosal components (Paolella et al., 2014); with an epithelial cell layer interlaced with innate mucosal immunity and neuroendocrine elements, encasing the paracellular space that houses the intestinal microbiome diverse niches (Paolella et al., 2014; Ringel et al., 2015). The single layer of epithelial cells that line the intestines is bound by a tight junction protein (TJP) network, serving to form the physical barrier, which provides protection against potential pathogenic assaults and toxins that increases the risk of systemic sepsis. Intestinal epithelial cells have a rapid time of turnover of between 2 and 6 days (Ramachandran et al., 2000). The TJP network also regulates intestinal permeability, providing the pores and channels for passage of molecules (i.e., water, electrolytes, nutrients) by selective permeability (Paolella et al., 2014). The intestinal barrier is also involved in the co-ordination of responses of the innate immune system, with macrophage/dendritic cell activation contributing to host defenses against microbial-induced systemic infections (Paolella et al., 2014). The protrusions that characterize macrophages/dendritic cells allow for the sensation of potential pathogens that have breached the intestinal mucus layers, the mucosa, as well as those sensed in other parts of the intestinal lumen, resulting in the induction of responses such as phagocytosis and of the acquired immune system through B-cell activation (Kumar et al., 2011; Kinnebrew and Pamer, 2012). **Figure 1** details the maintenance of the homeostatic state of the intestinal barrier.

Intestinal microbial metabolites provide the substrate for the fermentation of complex dietary carbohydrates to produce short-chain fatty acids (SCFAs), as well as assist the host in harnessing maximal energy from dietary consumption (Jiang et al., 2015). The various metabolites exert varying effects on the host, from the beneficial production of signaling molecules (e.g., butyrate), to inducing mucus and other secretions, to provide the triggers that facilitate the innate mucosal system to maintain local homeostasis. In states of dysbiosis, the intestinal barrier increases in permeability as a result of a disruption to the regulation of the epithelial cell-to-cell tight junction protein network. A compromised intestinal barrier can be associated with bacterial translocation from the gut into the systemic circulation increasing the risk of sepsis. Lipopolysaccharides (LPS), a constituent of gram negative bacteria (Jiang et al., 2015), is found to be increased in the systemic circulation, indicative of dysbiosis (Boulangé et al., 2016). LPS has been associated with inducing apoptosis of lymphocytes under in vivo conditions (Norimatsu et al., 1995; Nielsen et al., 2012; Jiang et al., 2015) demonstrating an immune-modulatory effect. Studies have posited that a loss of lymphocytes in the intestinal mucosa is a consequence of intestinal epithelial dysbiosis and subsequent release of metabolic endotoxins (Jiang et al., 2015). LPS has also been implicated as an inductor of a pro-inflammatory environment which is conducive to MetS, IR and T2DM (Cani et al., 2008). Gram-negative bacteria containing LPS are therefore hypothesized to contribute to NAFLD development. Furthermore, dysbiosis and elevated systemic LPS can be envisaged as markers of intestinal toxicity (Nolan, 2010). Intestinal toxicity driven dysbiosis supports local mucosal inflammatory responses that is concomitant with an increase in intestinal permeability. This combined disruption of the intestinal barrier/mucosal immunity activity can promote and mediate NAFLD pathogenesis via the gut-liver axis (Littman and Pamer, 2011; Wieland et al., 2015).

## THE GUT-LIVER AXIS

The venous system of the portal circulation defines the gut-liver axis and highlights the close anatomical proximity and functional interactions of the gastrointestinal tract and the liver (Paolella et al., 2014; Brandl et al., 2017). The axis is described as a means of enhancing interactions between metabolites of the intestinal microbiome and receptors on the liver, which can trigger a cascade of events that culminates in IR, inflammation of the liver, and eventually the development of liver fibrosis (Paolella et al., 2014). The anatomical and functional link between the gut and liver delivers 70% of hepatic blood supply via the portal vein. The portal vein is the direct venous outflow from the intestines and thus when the intestinal mucosal barrier is compromised it exposes hepatic tissue to toxic factors derived from the intestines. Therefore, various metabolites produced by intestinal bacteria that reach the liver, have been linked to the manifestation of simple steatosis and NASH (Raman et al., 2013). Dynamic shifts in the gut-liver axis, contributed by either the physical barrier, the microbiome or the liver itself, are a result of alterations to the permeability of the intestinal epithelium and or microbial composition that have been implicated in NAFLD manifestation (Mehal, 2012). Experimental and clinical evidence increasingly implicate dysfunctions of the gut-liver axis in the development

2015; Robinson et al., 2015; Nakahashi-Oda et al., 2016) Adapted from Peterson and Artis (2014).

and progression of NAFLD through small intestinal bacterial overgrowth (SIBO) in conjunction with intestinal dysbiosis and increased permeability (Compare et al., 2012; Li et al., 2013; Miele et al., 2013; Vajro et al., 2013; Paolella et al., 2014). Recent experimental and clinical studies suggest that the gastrointestinal microbiome affects NAFLD pathogenesis through pathways that (i) facilitate metabolism and energy harvesting (Turnbaugh et al., 2006; Jiang et al., 2015), as described from high-fat fed mice models treated with high levels of pro-inflammatory cytokines that promote NAFLD development (Le Roy et al., 2013); (ii) dynamic interactions with the host's innate immune system where NAFLD is reported consequent to disrupted local immune cell functionality (Su et al., 2012).

#### INTESTINAL MICROBIAL COMPOSITION AND NAFLD

Widespread biodiversity exists in the microbial ecosystems of humans, particularly in the intestinal tract. However, despite the extensive variety of bacteria, four main phylum dominate in the intestines: Firmicutes, Bacteroidetes, Actinobacteria, and Proteobacteria (Mokhtari et al., 2017), with up to 90% of microbes estimated to belong to the Firmicutes and Bacteroidetes phyla (Eckburg et al., 2005). In both human and experimental model based studies, NAFLD has been associated with altered microbiome abundance, composition and dysbiosis (Wigg et al., 2001; Mouzaki et al., 2013; Zhu et al., 2013). Analyses have also correlated changes in microbiota composition with change in disease severity (Shavakhi et al., 2013; Eslamparast et al., 2014, 2015; Rahimlou et al., 2015; Yari et al., 2016).

In reviewing the literature, studies that have analyzed the intestinal microbial composition of NAFLD patients in comparison to healthy controls have reported identifying patterns or trends that can be associated with NAFLD development (Harris et al., 2012).

Studies that have profiled the intestinal microbiome of NAFLD patients report that specific configurational and compositional shifts are associated with intestinal epithelial

cell dysbiosis and the elicitation of pro-inflammatory immune responses, central to NAFLD manifestation and progression (Kirpich et al., 2015). Specific bacteria have been associated with NAFLD phenotypes, serving as both antagonists and protagonists in NAFLD pathogenesis. Studies which profiled the intestinal microbiome of healthy controls and NAFLD patients across the spectrum, were able to identify families, genera and phyla that differed significantly in their abundance between the healthy controls and those with a NAFLD diagnosis. Whilst data is inconsistent regarding the association between the intestinal microbiome profile and NAFLD, patterns are emerging which highlight potential relationships between bacterial types and host health (Lau et al., 2015). A study comparing subjects diagnosed with NAFLD compared to lean adults, reported that gram negative bacteria were significantly enriched (P =0 .009) and gram positive bacteria were markedly decreased (P = 0.001) in the NAFLD cohort (Wang et al., 2016). Findings from similar studies confirm the strong relationship between the composition and configuration of the intestinal microbiome and fatty liver histologies, suggesting that adverse shifts in intestinal microbiome profiles are related to the development of NAFLD (Wang et al., 2016). In a prospective cross sectional study, a 20% increase in the Bacteroidetes phylum (p = 0.005) and a 24% decrease in Firmicutes (p = 0.002) was found in healthy controls in comparison to NAFLD patients (Wang et al., 2016). Interestingly, among the species belonging to the Firmicutes phylum, SCFA-producing bacteria were significantly decreased. Specific microbiome signatures of intestinal bacteria that are reported to be associated with significant reductions in butyric acid (Consolandi et al., 2015) may comprise a significant marker for the depletion of intestinal bacterial species that are important for the maintenance of intestinal barrier integrity and innate mucosal immunity equilibrium. However, currently it is difficult to identify which microbiome differences are causal and which are coincidental in the development of intestinal barrier dysbiosis and moreover NAFLD. Variations between studies, which profile microbiome composition of NAFLD patients, may in part be accounted for by different analytical techniques that have been employed. Furthermore, differences in study design, including anthropometric measures, markers used for NAFLD diagnosis as well as ultra-sonographic vs. biopsy NAFLD diagnosis adds additional discrepancies. As such Mouzaki et al. (2013),reported an inverse relationship between NASH diagnosis and the proportion of the phyla Bacteroidetes detected. These results were contrasted by Zhu et al (Zhu et al., 2013) who reported that NASH diagnosis was accompanied by higher levels of alcohol-producing bacteria and endogenous levels of ethanol, a result that was supported by Wong et al. (2013).

The intestinal microbial variations were also observed at the genus level, with data purporting to Ruminococcus and Roseburia genera being shown to be inconclusive. Whilst Zhu et al. (2013) and Raman et al. (2013) reported a nonsignificant decrease in the abundance of Ruminococcus in NAFLD patients compared to a group of healthy controls, whereas Del Chierico et al. (2017) and Jiang et al. (2015) found an increase in the genus' abundance in NAFLD patients.

Despite the various limitations, preliminary studies highlight and support the hypothesis that configurational shifts in the intestinal microbiome composition may contribute to the development and progression of NAFLD. Whilst further studies on a larger scale with accurate measures and controlled variables are warranted, the potential contribution of intestinal microorganisms and their metabolites are implicated in the pathophysiology of NAFLD. Studies that administer probiotics/prebiotics, which posit to encourage the intestinal microbiome to re-establish intestinal-mucosal macrophage crosstalk homeostasis that then translates to reducing the progression of NALFD are very much warranted.

#### MECHANISMS LINKING THE INTESTINAL MICROBIOTA AND NAFLD

Research identifies several mechanisms by which the intestinal microbiome cohort-arrangement can affect NAFLD pathogenesis and maintenance. Increased intestinal permeability (Jandhyala et al., 2015), small intestine bacterial overgrowth (SIBO) (Zhu et al., 2013) and elevated serum endotoxin like lipopolysaccharide (LPS) (Brun et al., 2007; Soares et al., 2010), have been reported by studies with NAFLD patients, with varying disease severity and staging (Elshaghabee et al., 2016). LPS, a component of gramnegative bacteria, is elevated in cases of bacterial overgrowth and increased intestinal permeability, yielding hepatoxic effects through the activation of TLR4 and initiation of a cascade of proinflammatory innate immune responses (Wigg et al., 2001; Soares et al., 2010).

Experimental and clinical data suggests that SIBO and a disturbed intestinal epithelial barrier are involved in NAFLD pathogenesis (Bode et al., 1987; Purohit et al., 2008; Wan et al., 2016). Furthermore, investigations have shown that the serum from NAFLD patients have elevated levels of LPS-binding protein, TLR-4, and TNF-α in hepatic tissue (Ruiz et al., 2007; Wan et al., 2016).

The gut microbiota also contributes to NAFLD pathogenesis through enriched numbers of ethanol-producing bacteria, primarily Escherichia coli (Small et al., 2013; Zhu et al., 2013; Jiang et al., 2015). The alcohol produced by these bacteria are reported to be involved in compromising intestinal barrier integrity which instigates inflammatory activity and ultimately hepatoxic events. Ethanol is a common and dominant metabolite of numerous resident intestinal microbes. As a product of hetero-lactic organisms, endogenously produced ethanol is implicated as a pro-inflammatory hepatoxic factor in NAFLD pathogenesis (Cope et al., 2000; Baker et al., 2010). Linked with increased concentrations of serum ethanol, enrichment in the presence of alcohol-producing intestinal bacteria, like E. coli, has been demonstrated to increase intestinal epithelial permeability (Aron-Wisnewsky et al., 2013; Jiang et al., 2015).

A prelude to the liver's averseness to excessive accumulation of FFAs, is intestinal dysbiosis. Dysbiosis is the concept that describes compositional alterations away from the conventional symbiotic intestinal microbiota that may be associated with pathology within the host and is visually described in **Figure 2**. Herein we further posit that compositional alterations in the intestinal microbiota adversely affect the intestinal epithelial barrier exacerbating epithelial cell dysbiosis. It has been shown that intestinal epithelial cell disruption can directly adversely affect intestinal resident macrophages and act as critical effector cells in the initiation of inflammation in the pathogenesis of metabolic diseases (Chawla et al., 2011). Disturbances of the TJP network which link the epithelial cells to form the intestinal barrier is a central regulatory mechanism, facilitating selective permeability across the intestinal mucosa and limiting bacterial translocation. Examining the duodenum of NAFLD patients and healthy adults has revealed that in comparison to NAFLD patients, (Jiang et al., 2015) the TJP network was significantly more intact in the duodenum of a healthy adult, with regular alignment and extensive abundance of the microvilli(Jiang et al., 2015). These observations are in stark contrast to the significantly wider gaps and disrupted TJPs reported in NAFLD patients, suggesting a loss of barrier integrity with a consequent increase in bacterial translocation through increased intestinal permeability (Briskey et al., 2016). Additional assessment of the serum biomarkers of the TJPs, including occludin proteins, which are structural components of the tight junctions, have been reported with significantly higher levels in the intestinal mucosa of healthy adults compared to NAFLD patients (Jiang et al., 2015). Measuring the serum levels of the proteins that comprise the structural backbone of the intestinal TJP network lends further supportive data to the hypothesis that intestinal mucosal permeability is greater in NAFLD patients than lean subjects or controls, suggesting intestinal epithelial dysbiosis is a causal factor in NAFLD pathogenesis.

With the establishment of the intestinal microbiota as a participant in the onset and maintenance of low-grade systemic inflammation, the probing of the intestinal microbiome as a potential therapeutic target for extra-intestinal inflammatory conditions, such as NAFLD begun.

The supposition that the intestinal microbiome could indirectly and adversely influence the physiological function of an end–organ such as the liver, by contributing pro– inflammatory activity in the intestinal mucosa, is a novel concept with biological plausibility. As an example, the uptake of Shigatoxin from the pathogen enterohemorrhagic E. coli by M cells and the underlying macrophages in the Peyer's patches is a critical step that teaches about bacterial translocation. This step has been correlated to the efficiency of the infection by the pathogen (Etienne-Mesmin et al., 2011). Numerous bacterial pathogens and their products cross the epithelial barrier though M cell junctions that are then subsequently captured by intestinal resident macrophages (Alisi et al., 2017). Moreover, the LPS components of bacteria are ligands of TLR4 that are expressed on various immune cells, including intestinal macrophages (Vijayan et al., 2017). TLR4 in the M1 configuration is a mediator of inflammation that may imply that increased LPS/TLR4 signaling could be a driving factor in the accelerated inflammation process in patients with NAFLD (Alisi et al., 2017). LPS also induces activation of genes on macrophages such as the early growth response gene 1 (EGR1) as well as the LPS-induced expression of TNF-α, an action that is directly mediated through EGR1 and NF-kB (Xu et al., 2001). Therefore bacterial products that activate macrophages and other immune cells to produce proinflammatory mediators can trigger inflammation in an end organ such as the liver (Tateya et al., 2013). Signals released in response to microbial dangers are absorbed on a backgorund of increased and dysregulated intestinal barrier permeability. These signals are recognized by PRRs, including TLR-4 (Szabo et al., 2010; Wan et al., 2016), and when sensed by NLRP3, support the hypothesis that inflammasome-driven microflora are potential drivers of NAFLD onset (Wan et al., 2016).

Intestinal homeostasis is pivotal in optimal functionality of the innate immune system and hinges on macrophages eliminating pathogenic bacteria and their particles (Vitetta, 2016). Activated macrophages play a dual role within the innate immune system (Sansonetti, 2002; Vitetta, 2016). Firstly, they help to elicit appropriate immune responses to detected microbial proteins by facilitating the presentation of antigens to T lymphocytes (Sansonetti, 2002; Vitetta, 2016). Secondly, activated macrophages serve as a secretory source for an array of cytokines that regulate the activation of T cell lymphocytes, including IL-1, INF–α, and cytotoxic proteins (Sansonetti, 2002; Vitetta, 2016). The overall action of the macrophage within the immune responses rely on their ability to neutralize exogenous antigens, cellular debris, insoluble particles, and activated clotting factors via phagocytic activity (Tacke, 2017).

#### REPRISE

NAFLD's pathway to pathogenicity is characterized by the presence of ectopic fat within hepatocytes that results from an imbalance in the levels of lipogenesis and lipolysis (Machado and Cortez-Pinto, 2014). Triglycerides are synthesized from FFAs that are reported to accumulate in the liver. Therefore, it is envisaged that the concentration of FFAs function as a regulator of lipogenesis in the liver.

Previous studies have associated bacterial phyla, families, or even single genera with obese or lean phenotypes, with an increased lactobacilli count and decreased Bacteroidetes presence associated with leanness (Armougom et al., 2009). In support of this, an increased abundance of genus' of the Bifidobacterium animalis or Lactobacillus species were associated with weight management and a healthy body weight (Million et al., 2012). Lactobacillus reuteri has specifically been identified as being associated with an obese phenotype (Million et al., 2012). Assessing intestinal microorganisms for possible correlations with NAFLD is a biologically plausible next step in progressing an understanding of how much influence the intestinal microbiome as a metabolic and immunological organ may have on the development and progression of NAFLD.

Methodological and technological difficulties have largely prevented robust and definitive data from studies that specifically assess the intestinal microbiota of adults with NAFLD and the health of the intestinal mucosal barrier as well as a need for further knowledge into what constitutes a healthy microbiota. When the mucosal barrier of the intestine, which also serves as the largest immune area of the intestinal immune system, is impaired and disrupted, the liver is exposed to intestinal-derived bacterial factors, which are potentially hepatoxic through the gut-liver axis.

A majority of the current literature details the involvement of the innate branch of the immune system in pro-inflammatory pathways leading to NAFLD pathogenesis and progression. However, the adaptive immune system is also implicated in NAFLD development (Ganz and Szabo, 2013; Sutti et al., 2016). Linking the innate and adaptive branches, natural killer (NK) cells are abundant in hepatic tissue and have been reported to influence the development of liver injury and fibrotic deposition that spark NASH materialization (Ganz and Szabo, 2013). Studies in both animal and human models have found a decrease in circulating NK cells in obese subjects compared to lean counterparts (Ganz and Szabo, 2013), with further exploratory investigations suggesting a reduction in their levels and thus activity may in turn increase the sensitivity of obese patients to develop progressive forms of NAFLD, including cirrhosis (Radaeva et al., 2006). A subset of NK cells, the natural killer T (NKT) cells, also known for their exhibition of both innate and adaptive immunity features, serve to regulate hepatic immune responses by secreting both Th1 and Th2 cytokines (Godfrey et al., 2000). Experimental and clinical data indicates that a depletion in NKT cells can lead to the chronic pro-inflammatory environment that can accompanies hepatic steatosis (Li et al., 2005; Ronchi and Falcone, 2008).

With no effective medication having yet been tested for managing or treating NAFLD and the only universally accepted treatment strategy being lifestyle modifications that focus of weight loss, novel therapeutic agents are being pursued in an endeavor to address the rise of NAFLD as one of the most common non-communicable liver disease world-wide (Volynets et al., 2012).

Whilst lifestyle modification recommendations encourage weight loss, this approach requires significant commitment and efficiency decreases over the long term due to waning dedication. Intestinal microbial manipulation through the administration of probiotics presents as an attractive therapeutic adjunct. In an attempt to reduce intestinal epithelial inflammation, probiotics shift the intestinal microbial community toward beneficial bacterial communities such as the Parabacteroides, Prevotella, and Oscillibacter (Ohland and MacNaughton, 2010). These microbial communities are well known to produce antiinflammatory metabolites such as SCFAs (Schwiertz et al., 2010). SCFAs, such as butyrate, serve as important facilitators in the harvesting of energy and harnessing it for peripheral tissues and the intestinal epithelia (Elshaghabee et al., 2016). Oscillibacter and Parabacteroides are associated with T-cell differentiation by enhancing and maintaining the IL-10 producing Treg cells (Arpaia et al., 2013). Moreover, a probiotic formulation that attenuated hepatocellular carcinoma in a murine model offered further insight on how the intestinal microbiota influences the regulation of T-cell differentiation of mucosal immunity in the intestines and in turn, there is down-regulation of the level of pro-inflammatory cytokines (Li et al., 2016). A recently completed murine study by our group reported attenuation of steatosis by 60% in a high fat diet model of NAFLD (Briskey et al., 2016). As such the administration of probiotics to attenuate the progression of NAFLD is both clinically plausible and very much warranted.

Despite increasing focus and attention directed at identifying the mechanisms by which NAFLD develops and progresses, ambiguity remains surrounding the driving factors and molecular pathways that result in NAFLD onset. Immunological mechanisms, including the collaboration of the consequences of innate immunity, adaptive immunity, and TLR receptor

#### REFERENCES


signaling dysfunction with the gut-liver axis are each posited to contribute to disease pathogenesis and maintenance.

This narrative review has highlighted the requisite for the completion of a systematic literature review detailing the association between innate immune responses triggered by intestinal epithelial inflammation and dysbiosis and the onset of extra-intestinal pathologies, such as NAFLD. The intestinal microbiome is a significant environmental factor in NAFLD pathogenesis, specifically through effects on intestinal barrier integrity. A review of the literature which explores the correlation between changes in intestinal integrity, intestinal permeability, and therefore endotoxin translocation will help elucidate the specific patterns or profiles of intestinal microorganisms that are of interest in NAFLD manifestation and therefore relevant for therapeutic target purposes. Furthermore, such a review may also define bacterial species that elicit a hepatoprotective effect on the microbiome and extra-intestinal inflammation.

#### AUTHOR CONTRIBUTIONS

ES and LV: Conception and design of commentary, review; ES, LV, MT, and TP: Read, amended and approved final version of manuscript.

#### FUNDING

LV has received National Institute of Complementary Medicine and National Health and Medical Research Council of Australia competitive funding and Industry support for research into biocompounds and probiotics. ES, LV, and MT participate in research on probiotics in Medlab Clinical's research laboratory facility in Sydney, Australia.


endotoxin, and medical consequences: summary of a symposium. Alcohol 42, 349–361. doi: 10.1016/j.alcohol.2008.03.131


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest

Copyright © 2018 Saltzman, Palacios, Thomsen and Vitetta. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Fructose: A Dietary Sugar in Crosstalk with Microbiota Contributing to the Development and Progression of Non-Alcoholic Liver Disease

#### *Jessica Lambertz1 , Sabine Weiskirchen1 , Silvano Landert2 and Ralf Weiskirchen1 \**

*<sup>1</sup> Institute of Molecular Pathobiochemistry, Experimental Gene Therapy and Clinical Chemistry (IFMPEGKC), RWTH University Hospital Aachen, Aachen, Germany, 2Culture Collection of Switzerland AG (CCOS), Wädenswil, Switzerland*

#### *Edited by:*

*Wesley H. Brooks, University of South Florida, United States*

#### *Reviewed by:*

*Jieliang Li, Temple University, United States Ruchi Tiwari, Veterinary University (DUVASU), India Maryam Dadar, Razi Vaccine and Serum Research Institute, Iran Ashok Munjal, Barkatullah University, India*

> *\*Correspondence: Ralf Weiskirchen rweiskirchen@ukaachen.de*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 14 July 2017 Accepted: 01 September 2017 Published: 19 September 2017*

#### *Citation:*

*Lambertz J, Weiskirchen S, Landert S and Weiskirchen R (2017) Fructose: A Dietary Sugar in Crosstalk with Microbiota Contributing to the Development and Progression of Non-Alcoholic Liver Disease. Front. Immunol. 8:1159. doi: 10.3389/fimmu.2017.01159*

Fructose is one of the key dietary catalysts in the development of non-alcoholic fatty liver disease (NAFLD). NAFLD comprises a complex disease spectrum, including steatosis (fatty liver), non-alcoholic steatohepatitis, hepatocyte injury, inflammation, and fibrosis. It is also the hepatic manifestation of the metabolic syndrome, which covers abdominal obesity, insulin resistance, dyslipidemia, glucose intolerance, or type 2 diabetes mellitus. Commensal bacteria modulate the host immune system, protect against exogenous pathogens, and are gatekeepers in intestinal barrier function and maturation. Dysbalanced intestinal microbiota composition influences a variety of NAFLD-associated clinical conditions. Conversely, nutritional supplementation with probiotics and preobiotics impacting composition of gut microbiota can improve the outcome of NAFLD. In crosstalk with the host immune system, the gut microbiota is able to modulate inflammation, insulin resistance, and intestinal permeability. Moreover, the composition of microbiota of an individual is a kind of fingerprint highly influenced by diet. In addition, not only the microbiota itself but also its metabolites influence the metabolism and host immune system. The gut microbiota can produce vitamins and a variety of nutrients including short-chain fatty acids. Holding a healthy balance of the microbiota is therefore highly important. In the present review, we discuss the impact of long-term intake of fructose on the composition of the intestinal microbiota and its biological consequences in regard to liver homeostasis and disease. In particular, we will refer about fructose-induced alterations of the tight junction proteins affecting the gut permeability, leading to the translocation of bacteria and bacterial endotoxins into the blood circulation.

Keywords: fructose, gut-liver-axis, inulin, insulin resistance, microbiota, SCFA, probiotics, prebiotics

**Abbreviations:** AMPK, adenosine monophosphate kinase; BDL, bile duct ligation; DNL, *de novo* lipogenesis; ENL, enterolactone; F-1-P, fructose-1-phosphate; FOS, fructooligosaccharides; IBS, irritable bowel syndrome; IL-1β, interleukin-1β; IL-6, interleukin-6; LPS, lipopolysaccharide(s); MD, Mediterranean diet; NAFLD, non-alcoholic fatty liver disease; NASH, non-alcoholic steatohepatitis; PAMP(s), pathogen-associated molecule(s); SCFA, short-chain fatty acid(s); TLR(s), toll-like receptor(s); TNF-α, tumor necrosis factor α.

#### DIET INFLUENCES MICROBIOTA

Diet and connected nutritional status depict important, modifiable factors of human health. Major determinants are gut microbial community (microbiota) and its genes (microbiome) (1, 2). In turn, the gastrointestinal microbiota is highly influenced by the diet of the host. Turnbaugh et al. showed that a shift from a Mediterranean diet (MD) rich in polysaccharides to a Western diet, high in fat and sugar, low in fiber, is able to alter microbiota within a day (1). Similarly, diets high in sugar significantly decrease the microbial diversity in the gut after just 1 week (3).

Also another more recent study has shown that the consuming of a MD promotes a gut flora enriched in polysaccharidedegrading microbes and end products of polysaccharide fermentation, whereas in contrast, a Western diet leads to a community of microorganisms in the digestive tract that mostly contain proteolytic microbes and end products of protein and fat metabolism (4).

During metabolism, a part of the ingested food such as dietary fiber or resistant starch escapes digestion in the small intestine, reaches the colon and is fermented by gut microbes, which produce products such as short-chain fatty acids (SCFA), trimethylamine, ammonia, and hydrogen sulfide, which have beneficial impact on its host. With the diet also the economy of microbiota changes, because the different strains favor different environment and nutrients. Differences in microbial structure and function are reflected in the diversity of intestinal metabolites. For example, SCFA such as acetate, butyrate and propionate representing end products of fermentation of complex carbohydrates, were significant higher in samples taken from Egyptian teenagers consuming a MD than in teenagers from the United States absorbing a typical Western diet (4). SCFA are known to inhibit inflammation and obesity (5), while the end products resulting from lipid and protein degradation are associated with arteriosclerosis and colon cancer (6). Moreover, intestines of teenagers consuming a Western diet have elevated amino acid content and higher levels of lipid metabolism-associated compounds, and higher concentrations of 1-methylhistamine indicating allergic reactions and suggesting that the intestinal microbiota is majorly determined by the host diet (4). In line with these findings, the Western diet was found to be one of the causes for metabolic diseases such as non-alcoholic fatty liver disease (NAFLD). In combination with stress exposure, diets enriched in fat and fructose are even able to modulate brain immunity, and increase metabolic vulnerability to conditions associated with NAFLD, arteriosclerosis and cardiovascular disease (7). In contrast, the consumption of components of the MD enriched in olive oil, fish, nuts, whole grains, fruits, and vegetables is negatively correlated with the pathogenesis of NAFLD (8).

#### FRUCTOSE IN HUMAN DIETS

Fructose is an integral part of human diets. This monosaccharide appears naturally in ripe fruits, honey, and in small amounts in vegetables including carrot, onion, sweet potato, and paprika. In the 1960s, fructose has been shown to have positive effects in the treatment of diabetes because it does not need insulin to be metabolized. Fructose feeding had no influence on fasting blood glucose and its excretion into the urine (9, 10). This sugar was described as a "useful therapeutic agent" stabilizing blood glucose in diabetic patients to the normal fasting level, improving the overproduction of acetone, restoring the nitrogen balance, and decreasing the loss of water (9). All these changes were induced without any changes in insulin therapy, just by substitution of fructose for glucose (9). Similarly, intravenous administration of fructose was effective in the treatment of diabetic ketosis (11). In the 1960, high fructose corn syrup was inserted in the food industry as a substitute of sugar and the intake increased, while the clinical importance of this sugar is still discussed controversially and in the focus of research (12). In particular, fructose was identified as a sugar affecting lipid metabolism by rising plasma triglycerides and fasting plasma free fatty acids. Therefore, the usage of this sugar was challenged for treatment of diabetes (10). Today, the view on fructose has changed dramatically. It is now handled as a risk factor in the development of obesity and several metabolic disturbances. In addition, fructose worsens symptoms in irritable bowel syndrome (IBS), an inflammatory condition characterized by abdominal pain, diarrhea, and bloating (13). Human studies performed in patients suffering from IBS showed a high prevalence of fructose malabsorption up to 64% (14). Interestingly, also patients tested negative for disorders in fructose metabolism showed abdominal symptoms after fructose ingestion suggesting fructose intolerance as a highly common condition (13). In both studies, 25 g of fructose were orally administered in a fructose breath test to prove malabsorption. Noteworthy, beverages usually contain 10 g sugar/100 mL. On the other side, if beverages are sweetened with high-fructose corn syrup such as HFCS-55 containing 55% of fructose, consumers absorb 27.5 g fructose drinking a volume of 500 mL. This amount is more than tested in the mentioned human studies and again emphasizes the fact that fructose in beverages and food nutrients should be critically considered as risk factors for inflammatory diseases.

#### MICROBIOTA IMPACTS THE HOST IMMUNE SYSTEM

Bacteria in the gut are responsible for digestion and producing essential vitamins and minerals (2). In addition, they are important for host physiology, digestion of indigestible food materials, and production of bile acids (15). Moreover, the gut microbiota has an influence on several immune functions, protects against pathogens, and joins in the maturation of the gut barrier (16).

Our intestine is an individual immunological site where interaction between microbiota and its host takes place (17). If the homeostasis between microbiota and host is disturbed, inflammation and cancer can occur (17). The crosstalk of the intestinal microbiota with the immune system of the host can modulate insulin resistance and intestinal permeability (16). Furthermore, it has an impact on the body weight. This could be demonstrated by transplant experiments in which gut communities isolated from obese mice fed a Western diet were transplanted into germfree recipient mice. The colonization produced adiposity in the recipient mice within 2 weeks (1).

### SHORT-CHAIN FATTY ACIDS, METABOLITES OF MICROBIOTA AFFECT THE IMMUNE SYSTEM AND INFLUENCE DISEASE PROGRESSION

Not only the microbiota itself but also its products influence the metabolism, energy intake and immunity (18). The human digestive system is restricted of debranching enzymes necessary to digest fibers and higher non-digestible carbohydrates such as pectin, inulin, gums, and cellulose (15). When these nutrients reach the distal gut, they stimulate growth and activity of bacteria that can ferment these compounds (19). During this microbial fermentation, SCFAs are formed influencing gut health and affecting as signaling molecules metabolism and function of peripheral tissues (20). There are three main SCFA, namely acetate, propionate and butyrate, which are the most important gut-derived products, acting as signal transduction molecules with epigenetic impact (21). Almost 10% of the human energy requirement per day is provided by SCFA *via* the host colonic epithelial cells (22). In particular, butyrate is the main energy source for normal colonic epithelial cells, protecting against colorectal, cancer, and inflammation (23). Especially, the fermentation of butyrate is highly inducible by lactate through lactate-utilizing bacterial strains converting starch and fructooligosaccharides (FOS) into butyrate (24, 25). In line, high intake of non-digestible fibers caused an accumulation of lactate exhibiting a low pKa value and provoking metabolic acidosis (25). Therefore, the utilization of lactate by lactate-utilizing microbiota is important in the context of SCFAs formation. SCFAs are absorbed by both passive diffusion and *via* monocarboxylic acid transporters (26). However, the production and molecular effects of SCFAs is presently controversially discussed. On one hand, SCFAs provide energy for the body (17, 21), while on the other site they can lead to extra fat deposition in the body and obesity (27). Moreover, SCFAs can directly influence several different functions such as satiety and host metabolism, improve glucose homeostasis, and insulin sensitivity (20). Although SCFAs are a rich source of calories, their intestinal production is associated with lean body weight, reduced inflammation, and increased satiety (28). They are ligands for receptors regulating appetite, inhibiting gastric emptying, while at the same time stimulating insulin secretion. Thereby SCFAs influence eating habits and the metabolism of the host and prevent exaggerate energy intake and obesity (15). Moreover, butyrate is an important molecule in the lipid metabolism of the host used to synthesize cholesterol and palmitate, while propionate is the principal gluconeogenic substrate decreasing hepatic glucose production (22, 29). Furthermore, butyrate has direct anti-inflammatory potential in the gut and the brain that helps to maintain the gut-barrier integrity and protects against the influx of toxins (29, 30).

Acetate promotes antilipolytic activity and it may also have metabolically beneficial effects in white adipose tissue (17). Furthermore, it mediates effects in the central nervous system suppressing appetite (31). So it is obviously that SCFAs affect metabolism and energy homeostasis by impacting glucose homeostasis, insulin sensitivity, skeletal muscle and liver tissue functions, adipose tissue biology (20, 21). With regard to the current literature, there is no reported direct interaction between the monosaccharide fructose and SCFA. Fructose consumption influences the microbiota and therefore affects the composition of SCFA in the gut. Interestingly, cross-feeding studies between *Bifidobacterium* and two acetate-converting, butyrate-producing strains (i.e., *Anaerostipes caccae* DSM 14662 and *Roseburia intestinalis* DSM 14610) in which fructose was used as the sole energy source allowed oligofructose breakdown by the strains not able to degrade the substrate itself (32). In this coculture study, *Bifidobacterium* releases small amounts of free fructose and acetate during degradation of fructooligosacharides.

Noteworthy, all three SCFA were able to decrease tight junction permeability, protecting barrier properties during increased microvascular leakage, which is the reason for several disease conditions (33). However, the effective concentration of each SCFA necessary to mediate these effects is different. The most effective concentration of butyrate (0.5 mmol/L) to decrease paracellular permeability was lower than that of propionate (1 mmol/L) and acetate (32 mmol/L) suggesting that in particular butyrate producers in the gut strengthens the gut barrier (33). SCFAs can influence the enteric nerve system, stimulating motility and secretion activity or affecting immune cells thereby reducing inflammation and tumorigenesis (34). Similar to the effect of histone deacetylases, SCFAs have anti-inflammatory and immune-suppressive functions and act as modulators in immune homeostasis and cancerogenesis (17, 21).

Short-chain fatty acids are able to reduce the pH of the gut, altering the composition of microbiota. Changing the pH from 5.5 to 6.7 favors the population of gut microbiota that produces propionate, while reducing the pH to 5.5 favored bacteria producing butyrate (35). This process maintains the gut homeostasis and economy. As outlined above, butyrate is already more effective at lower concentrations than propionate and acetate, and it need to be considered that butyrate at higher concentrations may provide too much energy which can promote obesity.

Dietary fiber is known to promote weight loss and improve glycemic control. High fat diet enriched in SCFAs protected from obesity and improved glucose tolerance (36). However, studies on propionate have shown that the effects of SCFA are tissue specific. While propionate-dependent gluconeogenesis had a beneficial effect on metabolic health in the small intestine, it was detrimental in the liver (17).

The gut microbiota also produces diet-dependent many other metabolites such as secondary bile acids and amino acid derivatives that have essential functions in the body. An increased production of intestinal bile acid occurs in a high fat diet. Bile plays important roles in lipid and carbohydrate metabolism and also in mediating inflammatory responses (37). Diets rich in saturated fatty acids can change the composition of the bile acids and promote dysbiosis. The changed microbiota is then able to inhibit hepatic gluconeogenesis and glycolysis and impact insulin sensitivity (16). As a consequence, bile-sensitive bacteria, such as *Prevotella* will be less prevalent, and more bile-tolerant bacteria will be predominant. In contrast to these findings, the bile acids chenodeoxycholic acid and cholic acid have been shown to improve the conditions of fructose-induced NAFLD by regulating intestinal transepithelial permeability and prohibiting the translocation of bacterial endotoxin from the gut into the portal plasma, thereby diminishing the activation of hepatic Kupffer cells (38). Additionally, these bile acids protected against the loss of tight junctions proteins in the intestinal epithelium. *Vice versa* the composition of bile acids is mutual influenced by diet and microbiota modulating the immune system of the host and impacting the development of diseases such as NAFLD.

## HOW DOES FRUCTOSE CONTRIBUTE TO NAFLD AND NON-ALCOHOLIC STEATOHEPATITIS (NASH)

Non-alcoholic fatty liver disease is one of the most frequent hepatic disease (39). It incorporates a disease spectrum, which includes steatosis (fatty liver), NASH, and cirrhosis. While simple steatosis is reversible, NASH provokes hepatocyte injury, inflammation, and fibrosis that can aggravate to cirrhosis, liver failure, and even hepatocellular carcinoma (40). Moreover, NAFLD is associated with characteristics of the metabolic syndrome, which includes abdominal obesity, insulin resistance, glucose intolerance or type 2 diabetes mellitus, and dyslipidemia syndrome (41–43).

Gut microbiota has been recognized as the main environmental factor promoting metabolic diseases (28). Multiple studies reported fructose as a critical factor contributing to NAFLD progression by modulating intestinal microbiota (**Table 1**). It performs crosstalk with its host, maintaining the host's energy homeostasis and stimulating the host's immunity. Shifts in this composition can result in alterations of the symbiotic relationship, which can promote metabolic diseases (28). The key bacterial products involved in the pathogenesis of NAFLD are lipopolysaccharides (LPS) (44). LPS is derived from gram-negative bacteria and are known to critically contribute to inflammation-related processes and insulin resistance. It is able to cross the gastrointestinal mucosa *via* leaky tight junctions or infiltrating chylomicrons (28). Recent studies connected NAFLD to disturbances in the gut microbial environment. The microbial composition differed between healthy individuals and NAFLD patients (45). In line, a diet enriched in fructose not only induced NAFLD but also negatively affected the gut barrier and the microbiota, leading to dysbiosis (46).

In addition, the microbiota itself was shown to contribute to the progression of liver disease and injury and diet-induced NAFLD resulted in dysbiosis and a strong decrease in microbial diversity (66). Although the underlying mechanisms are not fully discovered yet, fructose is known to be highly involved in the development of NAFLD by altering gut metabolites (52, 60). Fructose increased the intestinal translocation of endotoxins and endotoxin levels in plasma (49, 64) contributing to inflammation and degrading of the mucosa barrier. Consequently, acute and chronic exposure to high fructose increased circulating endotoxin in patients with NAFLD, accompanied by markers of insulin resistance and inflammation (65). Interestingly, hepatic damage implicated by epitopic fat deposition occurred rapidly and significantly in non-human primates, even in the absence of weight gain (64) suggesting that arising periportal inflammation is the result of bacterial products overwhelming immune system and being presented to Kupffer cells (64). Altogether these studies indicate that fructose is able to induce inflammation promoting the development of NAFLD in two ways: it contributes to the formation of excessive fat triggering inflammation and it causes microbial dysbiosis promoting NAFLD pathogenesis.

In line, NAFLD and NASH patients were shown to consume more carbohydrates and particularly fructose (50). In a study, in which rats were fed a diet enriched in fructose, Wei et al. showed that the fecal metabolome profile was associated with the dietary fructose (60). Moreover, the chosen diet provoked formation of high quantities of SCFAs as well as C15:0 and C17:0 long chain fatty acids that are only produced by a special set of bacteria. This suggests that fructose-induced alterations in microbiota and not fructose itself are responsible for alterations in metabolome profile.

## FRUCTOSE SAVAGES THE INTESTINAL BARRIER

The intestinal barrier represents a physiological boundary protecting the host by preventing entry of intestinal microbiota and microbial products (**Figure 1**). In particular, tight junctions, adherens junctions, and desmosomes connect epithelial cells, which together form a selective permeable epithelial barrier limiting the penetration of potentially pathogen substances (47).

Chronic intake of fructose is associated with a loss of tight junction proteins in the duodenum, elevated translocation of endotoxin, and induction of toll-like receptors (TLRs) in the liver (50, 53, 54, 56, 58). TLRs can be activated by microbial pathogenassociated molecular patterns (PAMPs). LPS is the most common PAMP, occurring in the cell membrane of gram-negative bacteria. LPS binds to its receptor TLR4, inducing nuclear translocation of transcription factor nuclear factor kappa (NF-κB)-light-chain enhancer of activated B cells resulting in an increased expression of pro-inflammatory cytokines including tumor necrosis factor-α (TNF-α), interleukin-6 (IL-6), and interleukin-1β (IL-1β) (63). Especially the tight junction proteins occludin and claudin-1 have been shown to decrease during fructose consumption (53, 58, 67). This process causes mucosal inflammation and intestinal epithelial barrier disruption, increasing the translocation of microbial products (47). A recent study of Volynets et al. pointed out that a sucrose-rich diet and a fructose-rich diet affected the intestinal microbiome in different ways (55). While a Westerndiet high in sucrose primarily promoted weight gain, the intake of fructose, especially in combination with a Western diet, caused barrier dysfunction accompanied with loss of mucus thickness and endotoxin translocation (55).

Mice with a knockout in the *F11r* gene encoding the tight junction adhesion molecule A showed increased infiltration of intrahepatic macrophages, elevated recruitment of inflammatory monocytes to the liver, upregulation of TLRs and higher content Table 1 | Fructose in the crosstalk with microbiota in the pathogenesis of NAFLD.



Fructose in Crosstalk with Microbiota in the Pathogenesis of NAFLD

(*Continued*)


of inflammatory cytokines when fed a diet high in fructose and fat (47). High fructose consumption promotes gut inflammation accompanied by rising endotoxin release, epithelial dysfunction, and decline of tight junctions proteins (50, 55, 56) independent of the fat content in the diet and energy intake (55). These data again illustrate the high impact of nutritional fructose on the intestinal barrier function.

The hepatic portal system connects the liver and the intestine, commonly called the "gut-liver-axis" (44). Consequently, the liver is the first organ exposed to gut-derived exotoxins, receiving 70% of the blood supply from the intestine. Therefore, the liver acts as a first defense against bacterial pathogens possibly explaining the relation between the influx of endotoxin and the progression of NAFLD and other liver diseases (68). Additionally, the gut microflora is able to stimulate hepatic fat deposition contributing to NAFLD and NASH (68). Fructose-induced impairment of intestinal gut barrier function enables the entering of bacterial products into the portal vein system and the liver, leading to Kupffer cell-mediated activation of inflammasomes and inflam mation, increased formation of reactive oxygen species and proinflammatory cytokines such as TNFα that are major causes of insulin resistance and dyslipidemia (16, 50, 67). Long periods of feeding a high fructose and high fat cause a rise in serum LPS, liver TLR4 expression and circulating cytokines suggesting that LPS and TLR4 are key molecules in the pathogenesis of NAFLD (3, 44, 47). In combination with dysbiosis, impaired gut function can promote metabolic endotoxemia as it was observed in animals that received a high sugar diet ( 3).

In addition, endotoxins are able to damage hepatocytes, caus ing activation of Kupffer cells that produce and release inflam matory cytokines and oxygen radicals. These products further aggravate liver damage (44). In line, mice fed fructose, fat or a combination of both showed elevated endotoxin and TLR4 levels, which were associated with changing the polarization of Kupffer cells and infiltrating macrophages, contributing to NAFLD (56). Infiltration of lymphocytes in the liver was increased when mice were fed a high fat diet supplemented with high fructose compared to mice fed a high fat diet alone (54, 69). In the respective model, fructose consumption induced increased lymphocyte recruit ment that was accompanied by higher inflammation indicated by the elevated mRNA expression of TNFα (54).

#### FRUCTOS E CAUS ES DYSB IOS I S

The structure and biology of gut microbiota are highly individu ally so that individuals can be identified simply on the basis of their "microbiota fingerprint" (15). Mostly six different bacterial phyla colonize the healthy gut (**Figure 2**), namely *Firmicutes*, *Bacteriotedes*, *Proteobacteria*, *Actinobacteria*, *Fusobacteria*, and *Verrucomicrobia* (70). About 90% of total bacteria in the gut of an adult are represented by three major divisions, the *Firmicutes* (gram-positive), *Bacteroidetes* (gram-negative), and *Actinobacteria* (gram-positive) (15).

Microbial diversity significantly decreases when consuming a high sugar diet already after 1 week, independent of the fat content ( 3). Fructose was shown to induce a different pattern of dysbiosis than a high fat diet.

resulting in a higher permeability of the gut barrier. In addition, endotoxins enter the leaky barrier leading to epithelial disruption and increase penetration of pathogens into the blood stream. Reaching the liver, endotoxins increase inflammation by activation Kupffer cells through binding to TLR4 and formation of reactive

oxygen (ROS). The formed radicals induce hepatic damage and fibrosis. Furthermore, in the liver the fructokinase generate fructose-1-phosphate from fructose that is degraded into products providing substrate for *de novo lipogenesis* promoting steatosis.

Ferrere et al. found increased hepatic lymphocyte infiltration when mice were fed with a high fat and fructose diet compared to mice fed with high fat without fructose (57). Dysbiosis was observed in several studies when fructose was added to a normal diet (57). Mostly, a decrease of *Bacteriodes* and an increase of *Firmicutes* were observed when consuming a Western diet (27, 71–73). Independent studies comparing lean and diet-induced obese subjects suggested that *Firmicutes* promotes body fat accumulation (71, 72). Turnbaugh et al. reported that *Firmicutes* helps obese subjects to get more calories from the ingested food, which in turn leads to obesity (74). Also an increase in *Betaproteobacteria*, genera *Sutterella* was associated with enhanced hepatic fat deposition (3). Other studies demonstrated *Proteobacteria* as the main bacteria contributing to hepatic fibrosis and liver damage (47, 66). Furthermore, *Proteobacteria* is observed to be increased in several forms of dysbiosis and associated with NAFLD (75).

Rats subjected to different diets, marked changes in the microbiota already occurred already 1 week after the dietary switch to a diet enriched with high fat and high sugar (3). In a kind of circle, the increase in body fat mass correlated with shifts in the gut microbiota and gut-brain communication, possibly providing the basis for the pathogenesis of obesity (3).

Jegatheesan reported about alterations in colon mucosaassociated microbiota feeding a Western diet high in fat and fructose for 8 weeks (58). This was associated with lower levels of *Clostridium leptum*, *Bacteroides/Prevotella*, and *Lactobacillus/ Leuconostoc*, while the content of *Akkermansia muciniphila* was not affected (58). *A. muciniphila* is known to strengthen the epithelial barrier function and to impact the systemic health of the host (76). It has been shown that this cross-talk between the host and gut microbiota protects against high fat diet-induced LPS endotoxemia in obese mice (77). In the above mentioned study of Sen et al., a high fat/high sugar diet resulted in a specific

Figure 2 | Bacterial phyla colonizing the healthy or diseased gut. (A) *Bifidobacterium longum* is a gram-positive, rod-shaped, health-promoting lactic acid bacterium present in the human gastrointestinal tract contributing to the production of butyrate. (B) *Bacteroides thetaiotaomicron* is a gram-negative, anaerobic microbe which dominates the intestinal tract flora of most mammals and provides the host with metabolic capabilities. (C) *Enterobacter cloacae* is a gram-negative, facultative-anaerobic, rod-shaped bacterium of the normal gut flora helping to debranch organic substances for energy production. (D) *Citrobacter freundii* is a common component of the gut microbiome of healthy humans. It is a facultative anerobic, rod-shaped gram-negative bacteria. (E) *Escherichia coli* is a gramnegative, facultative anaerobic, rod-shaped, "coliform" bacterium, which is commonly found in the lower intestine of warm-blooded organisms. Although most *E. coli* strains are harmless and part of the normal gut flora, some serotypes are occasionally responsible for food contamination causing serious intoxication in their hosts. They have capacity to produce vitamin K2 and prevent colonization of the intestine with pathogenic bacteria. (F) *Enterococcus faecalis* is a gram-positive, commensal bacterium inhabiting the gastrointestinal tracts. They are often arranged in pairs or in chain form and have both an anaerobic and aerobic metabolism. (G) *Salmonella enterica* is a gram-negative bacterium, flagellated, facultative anaerobic with a rod-shaped phenotype. A number of *Salmonella* variations are serious human pathogens provoking (spontaneous healing) diarrhea. (H) *Clostridium difficile* is a gram-positive, anaerobic, spore-forming bacterium able to produce multiple toxins causing diarrhea and inflammation. It may become opportunistically established in the human colon during antibiotic therapy. (A–H) All images were taken from cultures deposited in the national Culture Collection of Switzerland AG (CCOS, Wädenswil, Switzerland, https://www.ccos.ch/). The respective CCOS strain numbers are: CCOS 606, CCOS 632, CCOS 668, CCOS 669, CCOS 684, CCOS 688, CCOS 739, and CCOS 958. All images were taken using a phase contrast microscope at 1,000×. Space bars, 10 µm.

increase in *Anaeroplasmatales,* which are an order of *Mollicutes* bacteria, a class of *Tenericutes* which is linked to diet-induced fat deposition and obesity (3). Adipose tissue inflammation plays a major role in the pathogenesis of NAFLD, intestinal dysbiosis might therefore be an additional factor leading to disease progression (78).

In line with this assumption, Raman et al. discovered a significant difference in the fecal microbiome in NAFLD patients and healthy subjects (75). In patients with NAFLD, the dysbiosis is mainly characterized by a decrease in *Bacterioides* (*Prevotella*) and an increase of *Clostridium coccoides* (45). Consistently, patients with IBS showed a similar dysbiosis characterized by an increase in *Clostridium* cluster XIVa and *Ruminococcaceae* with a concomitant decrease in *Bacteroidetes* and *Bifidobacteria* (79). In a more recent study, Xue et al. detected an increase of serum LPS and aerobic bacteria, such as *Escherichia coli* and *Enterococcus* and a decrease in the amounts of *Lactobacillus, Bifidobacteria*, and *Bacteriodes* in rats that were subjected to diets inducing experimental NAFLD (44).

De Minics et al. investigated the transformation of the microbiome in animal models of NAFLD, and compared high fat dietinduced NAFLD and bile duct ligation-induced liver damage. In the respective study, it was found that intestinal permeability and bacterial translocation are the key pathogenetic events triggering progression of liver damage. Moreover, the authors showed that the main effector in the degree of liver damage is related to microbiota changes (66). Similarly, fructose induced alterations in the microbiome of rats were associated with metabolic dysregulation and inflammation in gut, liver, and fat tissue that could be attenuated by antibiotic treatment or treatment with control fecal samples (59, 62). These observations again underpin the importance of microbiota in the development of metabolic diseases. Therefore, the authors of respective studies concluded that manipulation of the gut bacteria interferes with liver injury and progression of NAFLD.

To sum up, the human gut is colonized by several strains of microbiota, some with pro- and some with anti-inflammatory effects. A healthy gut is characterized by a homeostasis requiring a balance between the gut flora and the immune system of its host. Triggers that provoke intestinal leakage such as fructose cause a shift in this homeostasis favoring proinflammatory microbiota, suppression of anti-inflammatory microbiota, and reduction of their overall diversity.

#### FRUCTOSE-INDUCED PRODUCTION OF URIC ACID FURTHER PROVOKES LIVER DAMAGE

It was demonstrated that a Western diet leads to an increase in uric acid within the blood and that measured serum levels of uric acid directly correlate to the intake of fructose (80). The primary producers of uric acid are the hepatocytes and an anomalous metabolism of uric acid causes hepatocyte damage and produce oxidative stress (81). When fructose from dietary sources is absorbed through the fructose transporter GLUT5 within the intestinal epithelium and transported to the liver, it is rapidly phosphorylated in the liver by fructokinase, causing hepatic accumulation of fructose-1-phosphate (F-1-P) and a simultaneously increase in AMP (82) (**Figure 3**). Subsequently, the elevated hepatic concentration of F-1-P induces changes of several other metabolites such as glucose, lactate and uric acid (82). F-1-P is converted in triosephosphate providing substrates for *de novo* lipogenesis (DNL) (80, 82). Simultaneously, the decrease in intracellular phosphate stimulates AMP deaminase stimulating the generation of uric acid *via* xanthine oxidase and production of superoxide anion ( ) O2 <sup>−</sup> and hydrogen peroxidase (H2O2) (80, 83, 84). This process is reinforced under inflammatory conditions (80). Lanaspa et al. suggested that uric acid plays a pivotal role in the lipogenic ability of fructose. In the respective study, triglyceride accumulation was decreased by adding an inhibitor of xanthine oxidase (84). Likewise, Choi et al. showed that uric acid induced fat accumulation in HepG2 and in primary hepatocytes as a result of endoplasmatic reticulum stress induction, which could activate SREBP-1c and stimulate steatosis (85).

Mitochondrial oxidative stress inhibited aconitase resulting in accumulation of cytosolic citrate. This tricarboxylic acid promoted DNL by activating ATP-sensitive lipase converting citrate to acetyl-CoA, thereby inducing fatty acid synthesis (84). Simultaneously, oxidative stress in the liver mitochondria induced by uric acid leads to alterations of mitochondrial function and cell damage (81, 84). Even a single administration of fructose attenuates uric acid excretion in the ileum (83), and long-term consumption of fructose was shown to suppress renal uric excretion resulting in increased serum uric acid levels (86). Consequently, high uric acid impaired glucose tolerance, causing insulin resistance and inhibition of insulin signaling (87).

#### IMPROVEMENT OF DIET-INDUCED NAFLD

Currently, there is no real effective drug therapy for treatment of NAFLD. However, interventions in lifestyles, health-promoting diets, application of probiotics, nutritional supplementation with prebiotics inducing growth or activity of beneficial microorganisms, and exercise resulting in weight loss (88) have been shown to improve NAFLD (**Figure 4**). Some aspects of the beneficial effects of each intervention are discussed later.

### HEALTH-PROMOTING EFFECTS OF Mediterranean diets

The traditional MD is based on high intake in mono- and polyunsaturated fatty acids derived from olive oil or fish, vegetables, fruits, and nuts providing high fiber entry. MD has been shown to correlate negatively with NAFLD, and in combination with salt restriction, MD was shown to lower blood pressure, improve blood lipids, and improve steatosis and steatohepatitis, while omega-3 polyunsaturated fatty acids have been shown to reduce accumulation of lipid and liver enzymes, improve insulin sensitivity, and act as an anti-inflammatory compound (8). The high amount of fibers in the MD is accompanied by polyphenols, antioxidants and phytochemicals, plant metabolites capable to inhibit DNL, liver steatosis, and inflammation. Furthermore, fiber and phytochemicals enriched in whole grain are able to reduce energy

hepatocytes, thereby leading to a direct activation of genes such as SREBP-1c stimulating hepatic steatosis.

This results in increased cytosolic acetyl-CoA which is a substrate for *de novo* lipogenesis. In addition, uric acid can induce endoplasmatic reticulum stress in

function facilitating bacterial translocation. Bacterial-derived products [e.g., lipopolysaccharide (LPS)] are key driver of hepatic inflammation. Although effective pharmacological therapies for NAFLD are not available, lifestyle changes (modulation of diet), exercise, and weight loss have been shown to be beneficial on NAFLD outcome. Supplementation with probiotics and prebiotics restoring the microbial balance and changing the "*bad microbiota*" to "*good microbiota*" have healthpromoting effects by generation of short-chain fatty acids (e.g., acetate, propionate, and butyrate) interfering with NAFLD progression.

intake, promote SCFA-producing gut bacteria, and have significant prebiotic effects. Based on these positive health-promoting effects, it is obvious that changing to a MD is one therapeutically effective mean to improve the outcome or severity of NAFLD (8).

## SCFA INFLUENCE FAT METABOLISM *VIA* ACTIVATION OF ADENOSINE MONOPHOSPHATE KINASE (AMPK)

Adenosine monophosphate kinase is an important enzyme expressed mainly in the liver and skeletal muscles, playing a crucial role in cellular energy homeostasis (15). It is one of the central regulators of the body's metabolisms by promoting catabolic pathways to generate more ATP and inhibition of anabolic pathway. The enzyme is a heterotrimer composed of a catalytic α-subunit and two regulatory subunits (β and γ) (**Figure 5**). The activity of AMPK is strongly influenced by the gut microbiota (**Figure 6**). Drugs increasing expression of AMPK stimulate fatty acid oxidation in liver and muscle tissues, resulting in energy loss and prohibition of obesity (89). Specifically, certain food compounds and dietary strategies are suitable to activate AMPK and to mediate antidiabetic effects. Some intestinal bacteria are able

Figure 5 | Structure, regulation and functional aspects of adenosine monophosphate kinase (AMPK) biology. (A,B) The AMPK is a heterotrimeric protein kinase complex comprised of α-, β-, and γ-subunits. AMPK is activated by phosphorylation of a critical threonine residue located within the α-subunit that is triggered by binding of AMP and/or ADP to the γ-subunit. ATP competitively inhibits the binding of both AMP and ADP to the γ-subunit suggesting that AMPK is a critical sensor of AMP/ATP or ADP/ATP ratios (93). The space bar represents 10 Å. The CPK representation in (A) was generated with the interactive web-based tool NGL (94) and the ribbon drawing in (B) with Ribbons XP Version 3.0 (95) using the structure coordinates 5UFU deposited in the PDB Brookhaven database. More structural details of human AMPK are given elsewhere (96, 97).

to digest polyphenols and isoflavones from plants and convert it to enterolactone (ENL) or to equol, which both activates AMPK (63, 90). Furthermore, ENL ameliorates abnormal lipid metabolism, inhibits anabolic glycogen storage, and improves insulin sensitivity (63, 90). In the respective study it was shown that ENL dose dependently promoted glucose uptake under insulin absent condition, which was completely eliminated by an AMPK inhibitor suggesting that ENL is an antidiabetic substance. Conversely, the inhibition of AMPK prevented fatty acid oxidation in several organs and tissues, promoted the synthesis of cholesterol and triglycerides, and favored lipogenesis leading to excess fat storage and obesity (28). Therefore, food enriched in such pharmacological active substances including grapes, plums, strawberries, passion fruit, white tea, and soy are predicted to improve diet-induced NAFLD. Another study focusing on the beneficial impact of butyrate on the gut barrier reported that butyrate activates AMPK and induced a redistribution of the tight junction proteins zonula occudens-1 (ZO-1) and occludin (91). Additionally, butyrate activated AMPK induced SIRT1 phosphorylation that influences glucose homeostasis and insulin sensitivity in a positive way (92).

#### PROBIOTICS

There is now ample evidence showing the importance of probiotics in host health. As already discussed above, the main bacterial bioproduct, which is associated with the pathogenesis of NAFLD, is LPS. Xue et al. detected a positive correlation between serum LPS and aerobic bacteria, such as *Escherichia coli* and *Enterococcus* and a negative correlation between the amount of *Lactobacillus*, *Bifidobacteria*, and *Bacteriodes* in rats subjected to NAFLD inducing diets (44). Interestingly, it was demonstrated in the respective study that the amounts of anaerobic bacteria such as *Lactobaiullus* and *Bifidobacteria* could be restored by probiotics (44). The application of probiotics ameliorated the intestinal barrier in NAFLD and tight junctions were shown to be more complete in comparison to NAFLD rats without probiotics (44, 48). Furthermore, serum levels of LPS and TLR4 decreased significantly after administration of probiotics in a NAFLD rat model. In addition, the livers of the respective animals showed less hepatocyte swelling, less cell infiltration, lower degree of inflammation, and overall milder steatosis (44, 48). Along with that, serum levels of triglycerides, total cholesterol, LDL, and free fatty acids were diminished. Interestingly, the amount of high density lipoprotein (HDL) was increased compared to the control group and animals exhibited better glucose tolerance in comparison to NAFLD rats without probiotic treatment (44). Moreover, different *Lactobacillus* strains were identified that diminished hepatic fat accumulation, increased of serum alanine aminotransferase (ALT) levels, and improved intestinal gut barrier in a NASH model (48, 51). The combination of multiple

unit. (C) Representative fructooligosaccharides (FOS) are kestose (GF2), nystose (GF3), fructosylnystose (GF4) differing in number of fructose residues. The general molecular formula of a FOS is C6(*<sup>n</sup>*+1)H10(*<sup>n</sup>*+1)+2O5(*<sup>n</sup>*+1)+1 giving rise to a molecular mass of (*n* + 1) × 180.156 − *n* × 18.015 g/mol when *n* is the total number fructose residues.

*Lactobacillus* strains reduced plasma glucose and insulin levels, triglycerides, and oxidative stress. Moreover, high probiotic treatment reduced liver mass and liver cholesterol by increasing β-oxidation and lowering the expression of sterol regulatory element-binding transcription factor 1 (SREBP-1), stearoyl-CoA desaturase-1 (SCD-1), and fatty acid synthase (FAS) mRNA levels that are critically associated with fructose-induced metabolic syndrome (37). In similar studies, *Lactobacillus casei* was shown to decrease fat storage, oxidative stress, and hepatic inflammation in a NASH mouse model and to diminish the activation of the TLR4 signaling cascade (51, 98).

In a rat model, treatment with the butyrate-producing gram positive strain *Clostridium butyricum* prevented the progression of nutrient-induced NAFLD (92). When given as dietary supplement (i.e., MIYAIRI 588), this bacterial strain suppressed the diet-induced increase in endotoxin levels, and restored the expression of the tight junction proteins ZO-1 and occludin (92). It altered the intestinal flora and restored gut-barrier functions, reduced hepatic levels of cytokines as TNF-α and was able to regulate the activation of NF-κB (92).

Toll-like receptor 4 is the main receptor that detects gutderived endotoxins and regulates hepatic inflammation in NASH and probiotics were shown to decelerate the progress of NAFLD by inhibiting the LPS-TLR4-signaling pathway, improving intestinal flora dysbiosis, restoring normal gut homeostasis, and upregulating expression of tight junction proteins strengthening the gut barrier (44, 48). Patients with IBS exhibited a high activity of intestinal serine proteases, which could be biologically sequestered by *Bifidobacteria* acting as an antagonist of this endopeptidase (79). Therefore, probiotics are presently discussed as new therapeutics in clinical management of NAFLD (99).

#### PREBIOTICS

In addition, gut microbial modulation can also be achieved by intake of substances that induce the growth or activity of beneficial microorganisms. These prebiotics act as substrate for respective bacteria and can be supplied as a functional food component. Prototypically, the plant polyphenol Quercetin found in many fruits has been shown to modulate diet-induced dysbiosis in mice with NAFLD by increasing the production of SCFA, thereby improving the intestinal gut barrier and inhibition of diet-induced hepatic inflammasome activation (100). In addition, Quercetin supplementation counteracted the upregulation of lipogenic genes and reduced the amount of *Helicobacter*, which is more frequent in diet-induced NAFLD (100). In a similar study from Baldwin et al., the feeding of powdered grape extracts reduced fat gain and hepatic lipid accumulation in a high fat mouse model (101). Also fructooligosaccharides (FOS) were shown to act as prebiotics in a metabolic syndrome model (61). FOS are composed of relative short chains of less than 20 molecules of fructose that are linked to a glucose terminal residue and cannot be digested by humans (**Figure 7**). They can be extracted from many plants (e.g., blue Agarve) and naturally occur in high concentration in many fruits and vegetables such as bananas, artichokes, onions, chicory root, garlic, asparagus, and leeks. In combination with *Lactobacillus fermentum*, FOS were effective in preventing intestinal permeability, systemic inflammation, hepatic steatosis and insulin resistance without inducing an innate immune system reaction at the intestinal level during high fat diet (61). Similarly, the mentioned study of Baldwin et al. showed that FOS mediated an increase in *A. muciniphila*, which was accompanied by decreased metabolic endotoxemia and reduced expression of inflammatory markers in white adipose tissue (101). Increased abundance of this mucin degrader residing in the mucus layer correlated with improvement of high-fat diet-induced metabolic disorders, reduction of obesity, reduced metabolic endotoxemia, adipose tissue inflammation, and insulin resistance (77).

## PHYSICAL ACTIVITY

It is already well established that exercise is associated with the production and release of potent, pharmacologically active, anti-inflammatory mediators, which can principally counteract liver inflammation and chronic low-grade inflammation (102). In a recent study, Batacan et al. investigated the combined effect of exercise and diet treatments on intestinal microbiota in a rat model (103). The authors found that the development of the microbiota in response to physical activity depends on basic starting microbiota. Moreover, training could induce a microbiota composition which is able to break down carbohydrates more effectively, improving the performance through energy provision, and increasing the production of SCFAs. Therefore, the authors suggested exercise as a factor for remodeling microbiota and gut health which is negatively influenced by high

#### REFERENCES


fat diets preventing microbiota differentiation in response to exercise (103).

## CONCLUSION

Non-alcoholic fatty liver disease is a systemic disease induced and modulated by different metabolic components, in which the liver is the main affected organ. Fructose consumption causes dysbiosis in the microbiota, leading to an increased permeability of the gut barrier, hepatic inflammation, progressive development of metabolic syndrome, and insulin resistance (57, 62, 69). Microbiota is highly influenced by diet and lifestyle of its host and a major factor influencing the outcome of metabolic diseases. Changing gut flora by intake of dietary fiber (i.e., roughage), probiotics, prebiotics, as well as intensifying frequency and duration of physical activity, leads to improvement of hepatic inflammation and fibrosis. These dietary and lifestyle interventions affecting microbiota are presently the most effective treatment options to improve NAFLD. Therefore, these nutritional and behavioral therapies should be combined with diets that lack excessive NAFLD-promoting compounds such as fructose and fat that are majorly contributing to the pathogenesis of NAFLD.

#### AUTHOR CONTRIBUTIONS

JL and RW have written this review. SW prepared the final **Figures 1**, **3**, **4** and **6**. SL provided images of bacterial cultures depicted in **Figure 2**. RW prepared **Figures 5** and **7** and has made the final editorial assignments. All authors agreed to submit this review and to be accountable for the content of the work.

#### FUNDING

RW is supported by grants from the German Research Foundation (DFG, SFB/TRR 57 P13 and Q3) and the Interdisciplinary Centre for Clinical Research (IZKF) within the Faculty of Medicine at the RWTH Aachen University (E7-6). None of the funding sources exerted influence on content of this review or decision to submit this article for publication.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Lambertz, Weiskirchen, Landert and Weiskirchen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

, Laura Jordán-Martínez<sup>3</sup>

, Antonio J. Muñoz-Garcia<sup>3</sup>

,

,

# Role of Gut Microbiota on Cardio-Metabolic Parameters and Immunity in Coronary Artery Disease Patients with and without Type-2 Diabetes Mellitus

#### Edited by:

Lidia Sanchez-Alcoholado<sup>1</sup>†

Isabel Moreno-Indias1,2, Pilar Cardila-Cruz<sup>3</sup>

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Yuji Naito, Kyoto Prefectural University of Medicine, Japan Undurti Narasimha Das, UND Life Sciences LLC, United States Amanda Cox, Griffith University, Australia

#### \*Correspondence:

Maria I. Queipo-Ortuño maribelqo@gmail.com

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 25 July 2017 Accepted: 21 September 2017 Published: 05 October 2017

#### Citation:

Sanchez-Alcoholado L, Castellano-Castillo D, Jordán-Martínez L, Moreno-Indias I, Cardila-Cruz P, Elena D, Muñoz-Garcia AJ, Queipo-Ortuño MI and Jimenez-Navarro M (2017) Role of Gut Microbiota on Cardio-Metabolic Parameters and Immunity in Coronary Artery Disease Patients with and without Type-2 Diabetes Mellitus. Front. Microbiol. 8:1936. doi: 10.3389/fmicb.2017.01936 Maria I. Queipo-Ortuño1,2 \* and Manuel Jimenez-Navarro<sup>3</sup> <sup>1</sup> Unidad de Gestión Clínica de Endocrinología y Nutrición, Instituto de Investigación Biomédica de Málaga, Hospital Universitario Virgen de la Victoria, Universidad de Málaga, Málaga, Spain, <sup>2</sup> CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain, <sup>3</sup> Unidad de Gestión Clínica Área del Corazón, Hospital Universitario Virgen de la

, Daniel Elena<sup>3</sup>

, Daniel Castellano-Castillo1,2†

Victoria, Instituto de Investigación Biomédica de Málaga, Universidad de Málaga, CIBERCV Enfermedades Cardiovasculares, Málaga, Spain

Gut microbiota composition has been reported as a factor linking host metabolism with the development of cardiovascular diseases (CVD) and intestinal immunity. Such gut microbiota has been shown to aggravate CVD by contributing to the production of trimethylamine N-oxide (TMAO), which is a pro-atherogenic compound. Treg cells expressing the transcription factor Forkhead box protein P3 (FoxP3) play an essential role in the regulation of immune responses to commensal microbiota and have an atheroprotective role. However, the aim of this study was to analyze the role of gut microbiota on cardio-metabolic parameters and immunity in coronary artery disease (CAD) patients with and without type-2 diabetes mellitus (DM2). The study included 16 coronary CAD-DM2 patients, and 16 age, sex, and BMI matched CAD patients without DM2 (CAD-NDM2). Fecal bacterial DNA was extracted and analyzed by sequencing in a GS Junior 454 platform followed by a bioinformatic analysis (QIIME and PICRUSt). The present study indicated that the diversity and composition of gut microbiota were different between the CAD-DM2 and CAD-NDM2 patients. The abundance of phylum Bacteroidetes was lower, whereas the phyla Firmicutes and Proteobacteria were higher in CAD-DM2 patients than those in the CAD-NDM2 group. CAD-DM2 patients had significantly less beneficial or commensal bacteria (such as Faecalibacterium prausnitzii and Bacteroides fragilis) and more opportunistic pathogens (such as Enterobacteriaceae, Streptococcus, and Desulfovibrio). Additionally, CAD-DM2 patients had significantly higher levels of plasma zonulin, TMAO, and IL-1B and significantly lower levels of IL-10 and FOXP3 mRNA expression than CAD-NDM2. Moreover, in the CAD-MD2 group, the increase in Enterobacteriaceae and the decrease in Faecalibacterium prausnitzii were significantly associated with the increase in serum TMAO levels, while the decrease in the abundance of Bacteroides fragilis was associated with the reduction in the FOXP3 mRNA expression, implicated in the

**602**

development and function of Treg cells. These results suggest that the presence of DM2 is related to an impaired regulation of the immune system in CAD patients, mediated in part by the gut microbiota composition and functionality and the production and effects of their gut microbiota derived molecules.

Keywords: gut microbiota, coronary artery disease, zonulin levels, gut permeability increase, TMAO production, factor Forkhead box P3, type-2 diabetes mellitus, anti-inflammatory IL-10

#### INTRODUCTION

Previous studies have shown an important connection between metabolism, intestinal microbial composition and the development of cardiovascular risk (Tremaroli and Backhed, 2012; Tang and Hazen, 2014; Org et al., 2015). Several authors have strongly supported the contribution of intestinal microbiota in the conversion of L-carnitine and dietary phosphatidylcholine in trimethylamine (TMA), which is absorbed into the bloodstream and then oxidized to trimethylamine N-oxide (TMAO) (a pro-atherogenic compound) by the hepatic enzyme flavin-containing monooxygenase-3 (Ufnal et al., 2015; Stremmel et al., 2017). Independent of traditional CVD risk factors, this microbiota-dependent metabolite TMAO has been related with the development of clinical CVD in the general population (Koeth et al., 2013; Tang et al., 2013, 2014, 2015; Wang et al., 2014; Troseid et al., 2015). Furthermore, elevated TMAO plasma levels are associated with increased CVD (Wang et al., 2011; Tang et al., 2013, 2014; Koeth et al., 2014).

Nevertheless, the intestinal bacterial species responsible for phosphatidylcholine breakdown have not yet been completely described, however, some species of Bacteroidetes (such as, B. thetaiotaomicron and B. fragilis) may be involved in the formation of TMA since they express phospholipases which hydrolyse dietary phosphatidylcholine to choline (Sitaraman, 2013). In addition, it has been reported that bacteria of the Erysipelotrichia class (Firmicutes phylum) can produce TMA from choline whereby it imitates a choline deficiency (Serino et al., 2014), while several families of bacteria such as Deferribacteraceae, Anaeroplasmataceae, Prevotellaceae (Koeth et al., 2013), and Enterobacteriaceae (Craciun and Balskus, 2012; Zhu et al., 2014) have been also associated with TMA/TMAO production.

On the other hand, TMAO has also been suggested to be a strong candidate molecule to mediate the development of type-2 diabetes mellitus (DM2) in animal and human studies (Kim et al., 2015). Moreover, coronary artery disease (CAD) is an important determining factor of the long term prognosis among patients with DM2 and the prevalence of diabetes in patients with CAD is up to 50% in many countries. Patients with diabetes have a 2- to 4-fold greater risk of developing atherosclerotic CAD than non-diabetic patients (Aronson and Edelman, 2016; Hölscher et al., 2016). Accumulating evidence suggests that DM2 and cardiovascular diseases such as hypertension, atherosclerosis, or heart failure are associated with alterations in the integrity of the gut barrier and augmented gut permeability (Ufnal and Pham, 2017).

Finally, gut microbiota has been reported to be a factor that is able to link intestinal immunity and host metabolism. It has been suggested that the increase or decrease of the abundant quantities of some specific bacteria may result in an increased generation of Tregs or in the reduced differentiation of pathogenic T cells, which may prevent inflammatory diseases (Smith et al., 2013). Treg cells expressing the transcription factor Forkhead box protein P3 (FoxP3) play an essential role in the regulation of immune responses to commensal microbiota. Furthermore, metabolites of commensal bacteria might regulate Treg development (Pereira et al., 2017). The administration of Bifidobacterium infantis 35624 has been found to induce Foxp3 T regulatory cells in human peripheral blood (Konieczna et al., 2012). It has also been described that FOXP3 positive regulatory T cells play a crucial role in maintaining the immune balance having an atheroprotective role (Hasib et al., 2016).

Therefore, the aim of this study was to analyze the role of gut microbiota on cardio-metabolic parameters and immunity in CAD patients with and without DM2. Furthermore, we have also analyzed the association of the gut microbiota found in these patients with the glucose metabolism, inflammatory status, intestinal permeability, and serum levels of TMAO and FOXP3 gene expression in peripheral blood.

#### MATERIALS AND METHODS

#### Study Subjects

We included a total of 32 patients with CAD who were divided into two groups: 16 patients with DM2 (the CAD-DM2 group) and 16 patients without DM2 (the CAD-NDM2). The CAD group was defined by the presence of at least one coronary stenosis ≥50% of luminal diameter diagnosed by coronary angiogram. DM2 was defined by the history of diabetes diagnosed and/or the presence of treatment with medication (insulin or anti-diabetic drugs) and/or fasting blood glucose ≥126 mg/dl and/or glycated hemoglobin (HbA1c) ≥6.5%. Diabetic treatment of CAD patients with DM2 was the following: oral anti-diabetic (metformin n = 8 and glibenclamide n = 3) and insulin (n = 5).

All subjects underwent a complete physical examination, including measurements of height, weight, and blood pressure. The medical history of all patients was recorded, including the duration of diabetes, hypertension history, the history of taking anti-hypertensive and hypoglycemic drugs, and smoking history. The definition of systemic arterial hypertension was determined as a history of high blood pressure or treatment with antihypertensive medication. The definition of Dyslipidemia was determined as known but untreated dyslipidemia or currently

undergoing treatment with lipid lowering medication. The family history of coronary heart disease was determined by interviewing the patients. The definition of a positive smoking history was determined as current smokers or cessation of smoking within the previous 3 months before the start of the study.

The following were considered as exclusion criteria: acute inflammatory disease, severe infective disease and/or cancer, chronic kidney disease (GFR < 45 mL/min/1.73 m<sup>2</sup> ), hepatic impairment and autoimmune disease. This study was carried out in accordance with the recommendations of the Ethical Committee of the Virgen de la Victoria Hospital. Written informed consent was obtained in all cases in accordance with the Declaration of Helsinki. The protocol was approved by the Ethical Committee of the Virgen de la Victoria Hospital.

#### Biochemical Variable Analysis

Serum was separated from blood samples which were collected after an overnight fast, aliquoted and immediately frozen at −80◦C. Enzymatic methods (Randox Laboratories Ltd., United Kingdom) were employed to analyze the levels of serum cholesterol, triglycerides, HDL-cholesterol, glucose, and glycated hemoglobin (HbA1c) using a Dimension autoanalyzer (Dade Behring Inc., Deerfield, IL, United States). The Friedewald equation was used to calculate LDL-cholesterol. Standard enzymatic methods (Wako Bioproducts, Richmond, VA, United States) were used to measure gamma-glutamyl transpeptidase (GGT), glutamate-oxaloacetate transaminase (GOT), and glutamic pyruvic transaminase (GPT). RIA provided by BioSource S.A. (Nivelles, Belgium) was utilized for insulin quantification. The ELISA kit from BLK Diagnostics (Badalona, Spain) was employed to measure high sensitivity C-reactive protein (CRP) levels. To calculate the HOMA-IR we used the equation: HOMA-IR = fasting insulin (mIU/mL)/fasting glucose (mol/L)/22.5.

IL-10 and IL-1β cytokines were measured in serum samples using commercially available Novex <sup>R</sup> ELISA Kits (Life Technology), performed according to the manufacturer's instructions. The assay range was 7.8–500 pg/mL for IL-10 and 3.9–250 pg/mL for IL-1β.

#### Anthropometric Measures

Standardized procedures were used to measure body weight, height, waist, and hip circumferences (Lohman et al., 1988). BMI was calculated as weight (kilograms) divided by height (in meters) squared.

### PCR Amplification and Analysis of 16S rDNA Sequences

DNA was extracted from fecal samples using the QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany) following the manufacturer's protocol. Amplification of genomic DNA was performed using barcoded primers that targeted the V2 to V3 regions of the bacterial 16S rRNA gene. Amplification, sequencing, and basic analysis were performed by using a GS Junior 454 platform according to the manufacturer's protocols using a Titanium chemistry apparatus (Roche Applied Science, Indianapolis, IN, United States). For further details, see Moreno-Indias et al. (2015).

### Bioinformatic Analysis

Quantitative Insights into Microbial Ecology (QIIME) 1.8.0 software was used to analyze the 454-pyrosequencing data sets as previously described by our group (Moreno-Indias et al., 2015). Raw reads were filtered first following the 454 amplicon processing pipeline. The pyrosequencing reads were de-multiplexed and filtered further by means of the split\_library.py script of QIIME. Reads with an average quality score lower than 25, ambiguous base calls, primer mismatches or shorter than 100 bp were excluded from the analysis in order to increase the level of accuracy. After the quality filter, the pipeline analysis used to analyze the 16S gene reads was the following: sequences were denoised and singletons excluded. The operational taxonomic units (OTUs) were selected by clustering sequences with a similarity of >97% and the representative sequences, selected as the most abundant in each cluster, were passed on to the UCLUST to obtain the taxonomy assignment and the relative abundance of each OTU by means of using the Greengenes 16S rRNA gene database. QIIME was used to evaluate alpha and beta diversity, as described (De Filippis et al., 2013).

## PICRUSt Analysis

The PICRUSt analysis was used to predict metagenome function by picking OTUs against the Greengenes database.<sup>1</sup> As has been previously described by Bhute et al. (2016), "a closed reference based OTU picking approach was utilized to bin the amplicon sequences using latest Greengenes database 13.5 at 97% sequence similarity cut-off. The normalization for 16S rRNA gene copy number was carried out before prediction of the metagenome. This OTU table was used for predicting metagenome at different KEGG levels." The R packages "pheatmap" were used for data analysis and plotting. Statistical analysis was done in R 3.3.3. P-values were corrected for multiple comparisons using the Benjamini–Hochberg method. A corrected p < 0.05 was considered as statistically significant.

#### Intestinal Permeability

The plasma level of zonulin was determined by means of the enzyme linked immunosorbent assay (ELISA) using commercial kits (Immundiagnostik AG, Bensheim, Germany). Measurements were made in duplicate, and the mean values were used for the analysis. The detection limit for zonulin was established at 0.22 ng/mL. Intra- and inter-assay coefficients of variation were between 3 and 10%.

#### Total RNA Extraction, cDNA Synthesis, and Real-Time Quantitative PCR (qRT-PCR)

The total RNA was extracted from peripheral blood mononuclear cells using TRIPURE Reagent (Sigma–Aldrich, United States) according to the manufacturer's protocol. RNA was diluted

<sup>1</sup>http://huttenhower.sph.harvard.edu/galaxy/

in 20 µl RNase-free water and was subsequently quantified with a spectrophotometer (Nanodrop N-100, Thermo Scientific). Reverse transcriptions were performed using 1 µg of total RNA with the Transcriptor First Strand cDNA Synthesis Kit (Roche) and random hexamers in 20 µl reactions. The amplifications were performed by means of a MicroAmpH Optical 96-well reaction plate (Applied Biosystems, Foster City, CA, United States) on an ABI 7500 Fast Real-Time PCR System (Applied Biosystems, Foster City, CA, United States). Pre-validated and commercially available TaqMan <sup>R</sup> primer/probe sets were used as follows: 18S rRNA (4319413E) used as the endogenous control for the target gene in each reaction and FOXP3 gene (RefSeq NM\_001114377.1). A threshold cycle (Ct value) was obtained for each amplification curve and a 1Ct value was calculated first by subtracting the Ct value for human 18S rRNA cDNA from the Ct value for every sample and transcript. mRNA expression levels relative to 18S rRNA were calculated by means of the 2−1Ct method. All tests were performed in duplicate.

#### Quantification TMAO in Serum Samples

Trimethylamine N-oxide was quantified in serum samples using Nuclear Magnetic Resonance (NMR) (Embade et al., 2016). For NMR analysis, serum samples were thawed for 30 min and aliquots of 300 µL were mixed with 300 µL of phosphate buffer (pH 7.0) containing 5 mmol/L TSP (Trimethylsilyl propionate) and 5% v/v D2O. The final mixtures were gently shaken and transferred to NMR 96 rack tubes. NMR spectra were measured at 300 K on a Bruker Avance IVDr 600 MHz spectrometer (Bruker Biospin, Germany) and a Sample Jet Robot (Bruker Biospin, Germany) was used for automatization of the measurements. For each sample, three complementary NMR spectra were recorded. A standard <sup>1</sup>H spectrum with water suppression (NOESY) was assessed. We repeated the same experiment with an appended T2 relaxation filter implemented as a CPMG module to decrease broad signals from proteins and lipoproteins. Finally, a two dimensional <sup>1</sup>H,1H JRES was measured to help with the identification of the metabolites. All spectra were acquired and processed within the TopSpin program (Bruker Biospin, Germany) applying an automatic phase correction and referenced against internal TSP (δ = 0.00 ppm). To identify and quantify the desired metabolites (TMAO) different amounts of these compounds were added to the serum samples, were measured by NMR and the values of the intensity peaks were represented against concentration. The pure metabolite molecules used for referencing were all obtained from Sigma–Aldrich (St. Louis, MO, United States). For all the spectra we measured the intensity of the peaks corresponding to TMAO (singlets) and the concentrations in the samples were calculated with the power fitted calibration curves.

#### Statistical Analysis

Given that there are no previous studies assessing gut microbiota differences between patients with CAD with or without DM2, we have based our sample size estimation on a previous paper from Emoto et al. (2016). To calculate the sample size, we considered a difference in the mean relative abundance of Bacteroidetes between CAD patients and non-CAD controls with coronary risk factors of 0.084 and an estimated standard deviation between groups of 0.05, at least 9 patients in each study group were needed (90% power and a two-sided alpha level of 0.05). However, we increased the sample size to 16 samples per group.

The relative abundance of each OTU (taxa) were compared by a Wilcoxon test with a continuity correction using the Explicit software package specifically addressed to analyze microbiome data. α- and β-diversities were achieved by QIIME: α-diversity using a non-parametric t-test with a default number of Monte Carlo permutations of 999, and β-diversity using the ANOSIM statistical method with 99 permutations (Moreno-Indias et al., 2015). Continuous variables are summarized as mean ± SD. Discrete variables are presented as frequencies and percentages. Differences in clinical characteristics between two groups were analyzed using the Mann–Whitney U test. The Spearman correlation coefficients were calculated to estimate the correlations between variables. A lineal regression analysis was performed to identify individual bacteria as independent predictors for HOMA-IR, levels of inflammatory mediators, serum TMAO and plasma zonulin levels and relative expression of FOXP3 in the study groups. Statistical analyses were carried out with the statistical software package SPSS version 15.0 (SPSS Inc., Chicago, IL, United States). Values were considered to be statistically significant when the p < 0.05.

## RESULTS

The general characteristics of the patients in both study groups are summarized in **Table 1**. There were no significant differences in age, sex, and body mass index (BMI) between the two study groups. The number of patients with dyslipidemia and hypertension was significantly higher in CAD-DM2 patients when compared with the CAD-NM2 patients. Triglyceride, glucose, and HbA1C were significantly higher and HDL-cholesterol was significantly lower in CAD-DM2 patients when compared with CAD-NM2 (P < 0.05). There was no significant difference in serum total cholesterol and LDL-cholesterol between the two study groups (P > 0.05). CAD-DM2 patients had significantly higher levels of inflammatory markers IL-1B and significantly lower levels of anti-inflammatory IL-10 when compared with non-diabetic patients. TMAO plasma levels were also significantly increased in CAD-DM2 patients when compared with CAD-NDM2 (P < 0.05). Finally, FOXP3 mRNA expression in PBMC was significantly lower (2.9-fold decrease) in CAD-DM2 patients when compared with CAD-NDM2 (p < 0.001).

#### Analysis of the Diversity and Similarity of Microbial Communities in the Study Patients

A total of 249.815 good quality 16S rRNA gene sequences with an average of 11.895 ± 6527 sequences per sample passed the filters, which were applied by means of QIIME. In order to obtain a detailed structural overview of the microbiome of each subject who was enrolled in this study, an analysis of OTU was developed. The microbiota of all fecal samples after QIIME was

composed by 2701 OTUS with a relative abundance higher than 1% in at least two samples (97% similarity cut-off).

Before assessing alpha and beta diversity measures, samples were rarefied to 3119 sequences, which represents the lowest number of quality reads which were obtained from any individual sample in the data set. The CAD-DM2 group had a lower number of OTUs than the CAD no-diabetic group (mean = 925 versus 987, P < 0.05, respectively).

The Chao index and Shannon index were calculated to estimate the alpha diversity. The average of community richness (Chao 1) and Shannon index of each group suggested a significant decrease in bacterial richness and diversity in fecal samples of CAD-DM2 with respect to CAD-NDM2 (P = 0.036 and P < 0.001, respectively) (**Table 2**). The rarefaction curve of the OTUs observed started to plateau at approximately 100 reads which suggests that a greater number of reads per sample

TABLE 1 | Biochemical, clinical characteristics, TMAO and inflammatory mediators serum levels and PBMC relative expression of FOXP3 in both study groups.


BMI, body mass index; CVA, cerebrovascular accident; ACEI, angiotensin converting enzyme inhibitor; ARB, angiotensin II receptor blocker; LDL, low density lipoprotein; HDL, high density lipoprotein; HbA1c, glycated hemoglobin; GGT, gamma-glutamyl transferase; GOT, glutamic oxaloacetic transaminase; GPT, glutamic pyruvic transaminase, TMAO: trimethylamine N-oxide; L-1β, interleukin-1β; CRP, C-reactive protein; Foxp3, Forkhead box P3. Values are expressed as mean ± SD. <sup>∗</sup>P < 0.05.

would not have supplied a more extensive catalog of bacterial taxa.

Principal coordinate analysis (PCoA) plots based on unweighted UniFrac metrics was used to evaluate the beta diversity. The present study showed that apart from a few samples from both study groups closer together in the ordination, a separation between the samples from the diabetic group and the non-diabetic group could be observed from PC1 and PC2 scores that accounted for 22.08 and 24.39% of total variations. ANOSIM with permutations suggested that the intestinal microflora composition of these two groups is significantly different (P = 0.039) (**Figure 1**).

#### 16S rRNA Gene Pyrosequencing in Fecal Samples from CAD-DM2 and CAD-NDM2 Patients

In this study, variations in the composition of fecal microbiota of CAD-DM2 and CAD-NDM2 groups were observed at different bacterial levels. At the phyla level the majority of the OTUS were found to belong to Bacteroidetes (53.87% CAD-DM2 vs. 63.42% CAD-NDM2, P < 0.001), Firmicutes was the next most abundant (25.69% CAD-DM2 vs. 23.29% CAD-NDM2, P = 0.49) followed by Proteobacterias (3.67% CAD-DM2 vs. 2.05% CAD-NDM2, P = 0.04), but only Bacteroidetes and Proteobacterias exhibited significant differences between both, the CAD-DM2 and the CAD-NDM2 study groups. The remaining bacterial population belonged to the other four phyla (Fusobacteria, Actinobacteria, Verrucomicrobia, and Tenericutes) that had a relative abundance lower than 1% in at least two samples (**Figure 2**).

Twenty-two families were detected among all groups. We found a significant increase in Veillonellaceae (28.93% CAD-DM2 vs. 16.97% CAD-NDM2, P < 0.001), Enterobacteriaceae (19.53% CAD-DM2 vs. 4.73% CAD-NDM2, P = 0.016), Rikenellaceae (11.13% CAD-DM2 vs. 9.92% CAD-NDM2, P = 0.023, Streptococcaceae (6.64% CAD-DM2 vs. 2.55% CAD-NDM2, P < 0.001), and Desulfovibrionaceae (13.95% CAD-DM2 vs. 10.69% CAD-NDM2, P < 0.001) in CAD-DM2 group. Furthermore, a significant decrease in the abundance was found in CAD-DM2 when compared with CAD-NDM2 for Bacteroidaceae (49.02% CAD-DM2 vs. 60.48% CAD-NDM2, P < 0.001), Lachnospiraceae (14.67% CAD-DM2 vs. 20.56% CAD-NDM2, P < 0.001), Prevotellaceae (5.06% CAD-DM2 vs. 8.85% CAD-NDM2, P = 0.05), S24-7 (1.59% CAD-DM2 vs. 2.38% CAD-NDM2, P = 0.023, P < 0.001),

TABLE 2 | Estimate richness (Chao1) and diversity index (Shannon) indices among microbial communities obtained from fecal samples from CAD-DM2 and CAD-NDM2 patients.


The richness estimator Chao and diversity estimator Shannon were calculated at 3% distance. <sup>∗</sup>P < 0.05.

FIGURE 1 | Clustering of fecal bacterial communities according to the different study groups by principal coordinate analysis (PCoA) using unweighted UniFrac distances. Each point corresponds to a community coded according to the patient group. The percentage of variation explained by the plotted principal coordinates is indicated on the axes. CAD-DM2 (red dots) and CAD-NDM2 (blue dots).

Odoribacteraceae (1.90% CAD-DM2 vs. 3.57% CAD-NDM2, P = 0.02), and Paraprevotellaceae (0.79% CAD-DM2 vs. 1.81% CAD-NDM2, P < 0.05). In addition, no significant differences between the two study groups were found in other abundant families, such as Ruminococcaceae (37.87% CAD-DM2 vs. 40.91% CAD-NDM2, P > 0.05), Alcaligenaceae (48.10% CAD-DM2 vs. 63.63% CAD-NDM2, P > 0.05), Porphyromonadaceae (12.38% CAD-DM2 vs. 11.02% CAD-NDM2, P > 0.05), and Erysipelotrichaceae (2.63% CAD-DM2 vs. 1.75% CAD-NDM2, P > 0.05) (**Figure 3**).

Significant differences between the study groups were also found in the microbial composition at the genus level. A total of 40 genera were identified among the 32 fecal samples, with only

significant differences in eight genera between the CAD-DM2 and the CAD-NDM2 groups. Thus, the genera Enterobacter, Streptococcus, Dialister, Megasphaera, and Desulfovibrio were significantly higher in the CAD-DM2 group when compared with the CAD-NDM2 group. On the other hand, we found in the CAD-NDM2 group a significant rise in the abundance of Bacteroides, Faecalibacterium, and Prevotella when compared with the CAD-DM2 patients. No significant differences in the abundance of Sutterella, Phascolarctobacterium, Parabacteroides, Oscillospira, Roseburia, Ruminococcus, Odoribacter, Blautia, and other minority genera were found between the two study groups (**Table 3** and **Figure 4**).

At a specie levels, we found a significant decrease in the abundance of Bacteroides fragilis (0.43% CAD-DM2 vs. 1.38% CAD-NDM2, P < 0.05), Bacteroides ovatus (0.99% CAD-DM2 vs.

1.86% CAD-NDM2, P < 0.05), and Faecalibacterium prausnitzii (2.03% CAD-DM2 vs. 3.72% CAD-NDM2, P < 0.05) in the CAD-DM2 group when compared with the CAD-NDM2 patients.

#### Increased Circulating Zonulin Levels in CAD-DM2 Patients

The circulating zonulin levels were measured by means of the ELISA method. In the CAD-DM2 patients, circulating zonulin levels were significantly higher than in CAD-NDM2 patients (5.98 ± 1.4 vs. 3.34 ± 1.6 ng/ml, P < 0.001).

### Relationship between the Gut Microbiota Composition and Glucose Metabolism, Serum Levels of Inflammatory Mediators, Serum TMAO and Plasma Zonulin Levels and Relative Expression of FOXP3

In the CAD-DM2 patients, Firmicutes was positively correlated with HbA1c and fasting triglycerides (r = 0.849, P = 0.030, and r = 0.648, P = 0.043, respectively) and Proteobacteria was positively correlated with HOMA-IR (r = 0.754, P = 0.012), while Bacteroidetes was negatively associated with HOMA-IR (r = −0.849, P = 0.03). The abundance of Streptococcus was negatively associated with the serum HDL-cholesterol levels (r = −0.728, P = 0.017). We also confirmed a positive association between serum TMAO and the abundance of Enterobacteriaceae (r = 0.711, P = 0.021) and Desulfovibrio (r = 0.849, P = 0.03) and a negative association between TMAO and the abundance of the common gut commensals Faecalibacterium prausnitzii (r = −0.681, P = 0.030) and Ruminococcaceae (r = −0.699,

TABLE 3 | Complete list of genus level microbial abundance (OTUs) in both study groups.


Data are shown as a percentage of the total identified sequences per group. <sup>∗</sup>P < 0.05.

P = 0.04). Likewise, the abundance of Faecalibacterium prausnitzii was negatively associated with plasma zonulin (r = −0.642, P = 0.045) and was positively associated with serum IL-10 levels (r = 0.663, P = 0.037). Finally, FOXP3 mRNA expression in PBMC was positively associated with the abundance of Bacteroides fragilis (r = 0.742, P = 0.037) and Butyricimonas (r = 0.717, P = 0.020).

In CAD-NDM2 patients, plasma zonulin levels were positively associated with the abundance of Prevotella (r = 0.681, P = 0.030) and Rikenellaceae (r = 0.669, P = 0.035), while serum TMAO was positively associated with Bacteroides (r = 0.742, P = 0.014) and negatively related to Faecalibacterium prausnitzii (r = −0.669, P = 0.035). Furthermore, there was a positive association between IL-10 and Bifidobacterium (r = 0.794, P = 0.006).

Subsequent lineal regression analysis including all the bacterial groups showed that the increase in Proteobacterias (p = 0.029, β = 0.548, r <sup>2</sup> = 0.83) was associated with the higher HOMA-IR found in CAD-DM2. On the other hand, in the CAD-DM2 group the increase in Enterobacteriaceae (p = 0.005, β = 0.945, r <sup>2</sup> = 0.96) and the decrease in Faecalibacterium prausnitzii (p = 0.041, β = −0.911, r <sup>2</sup> = 0.91) were associated with the increase in serum TMAO levels, while the decrease in the abundance of Bacteroides fragilis (p = 0.039, β = 0.911, r <sup>2</sup> = 0.91) was associated with the reduction in the FOXP3 mRNA expression.

#### Functional Differences in Gut Microbiota between Both Study Groups

The PICRUSt analysis indicated that the genes for the metabolism of carbohydrates, lipids, amino acids, and energy were significantly over represented in CAD-NDM2 in comparision with CAD-DM2 (P < 0.05). Moreover, in the microbiota of CAD-NDM2 patients, we found inside the carbohydrate metabolism, significantly enriched for the proportion of genes related to glycolysis/gluconeogénesis (P = 0.036), pyruvate metabolism (P = 0.029), inositol phosphate metabolism (P = 0.027), pentose phosphate pathway (P = 0.016), propanoate metabolism (P = 0.020), butanoate metabolism (P = 0.023), and starch and sucrose metabolism (P = 0.016), additionally, in the lipid metabolism pathways we observed an increase in the proportions of genes for fatty acid biosynthesis (P = 0.013), fatty acid metabolism (P = 0.021), glycerophospholipid metabolism (P = 0.036), sphingolipid metabolism (P = 0.008), biosynthesis of unsaturated fatty acids (P = 0.020), and secondary bile acid metabolism (P = 0.005).

Furthermore, the metagenomic comparison between CAD-DM2 and CAD-NDM2 groups showed that gene families linked to amino acid metabolism [including the metabolism of glycine (P = 0.022), alanine (P = 0.043), and glutamine (P = 0.029)] and energy metabolism such as sulfur (P = 0.036) and nitrogen metabolism (P = 0.016) were significantly depleted in the CAD-DM2 group.

Finally, the microbiota from CAD-NDM2 patients were also significantly more enriched with genes for antigen processing and presentation (P = 0.006) and siderophore biosynthesis (P = 0.036) when compared with CAD-DM2 patients (**Figure 5**).

#### DISCUSSION

The present study in which we used 16S ribosomal RNA of gut microbiota for high throughput sequencing has demonstrated that the diversity and gut microbial composition were different between CAD-DM2 and CAD-NDM2. We also observed that the CAD-DM2 patients had significantly higher levels of zonulin, TMAO, and IL-1B and significantly lower levels of IL-10 and FOXP3 mRNA expression than CAD-NDM2.

Microbiota diversity is essential to maintain ecosystem stability and efficiency. The α-diversity analysis in this study showed that the Shannon and Chao 1 indices were significantly lower in the CAD-DM2 patient group when compared with the CAD-NDM2 group. These data revealed that bacterial communities in diabetic patients with CAD had lower taxa richness than those in the non-diabetic group. These results may suggest that the decrease in gut microbiota diversity in CAD-DM2 patients might be related to the presence of DM2. In addition, the unweighted UniFrac PCoA plot analysis indicated a clustering of the microbial populations of the CAD-DM2 patients away from that of the CAD-NDM2, confirming that the presence of DM2 in patients with CAD can significantly alter intestinal microbial populations.

On the other hand, the taxonomy based analysis of the assigned sequences showed that the presence of DM2 in patients from CAD is related to a decrease in the phylum Bacteroidetes and an increase in the phyla Firmicutes and Proteobacteria. Moreover, at genera level the CAD-DM2 patients had more opportunistic pathogens such as Enterobacteriaceae, Megasphaera, Streptococcus, Dialister, and Desulfovibrio, and lower beneficial or commensal bacteria including Bacteroides (Bacteroides fragilis), Faecalibacterium (Faecalibacterium prausnitzii), and Prevotella when compared with CAD-NDM2. Previous studies using terminal restriction fragment length polymorphism (T-RFLP) analysis demonstrated a decrease in the phylum Bacteroidetes and an increase in the phylum Firmicutes in CAD patients when compared with healthy subjects. Moreover, they observed that the order Lactobacillales, in which Streptococcus genus is included, was significantly increased in the CAD group when compared with healthy subjects (Emoto et al., 2016, 2017). Cui et al. (2017) revealed that the phyla Bacteroidetes and Proteobacteria were decreased, while the phyla Firmicutes and Fusobacteria were increased in coronary heart disease patients when compared with healthy controls by high throughput sequencing. In our study, the significant decrease in the phylum Bacteroidetes and the increase in the abundance of the phylum Proteobacteria in the CAD-DM2 group possibly reflected the fact of suffering from DM2, one of the major risk factors for CAD.

After using PICRUSt, we were able to predict that the gut microbiome of CAD-DM2 patients showed a depletion in genes involved in metabolic functions such as the metabolism of carbohydrates, lipids, amino acids, and energy as well as in genes of antigen processing and presentation related to inflammation and immune response and siderophore biosynthesis. The enrichment in this siderophore pathway could suggest a higher intra- or inter-species communication in CAD-NDM2 patients, due to the fact that siderophore act as quorum sensing molecules for gram-negative organisms (McHardy et al., 2013). These results indicate that the presence of DM2 in CAD patients may also influence the functional diversity.

Koren et al. (2011) compared the microbiota among the gut, oral cavities, and the atherosclerotic plaque in patients with aterosclerosis. This study suggested that the bacteria which is present in the atherosclerotic plaque may be derived from the gut or the oral cavities (Koren et al., 2011). Furthermore, the gut microbiota is essential to bioconvert cholesterol into bile acids, which are necessary for the excretion and the absorption of cholesterol (Fagerberg et al., 2010). Thus, the significantly

higher levels of Streptococcus (a common oral and gut taxón) found in CAD-DM2 patients when compared with CAD-NDM2 patients, which were negatively correlated with HDL cholesterol levels in these diabetic patients, could be influencing CAD evolution through the regulation of the lipid metabolism in the host.

The present study has demonstrated that CAD-DM2 patients had significantly higher plasma levels of TMAO when compared

with CAD-NDM2. Tang et al. (2015) also found that the presence of diabetes in those patients belonging to a heart failure cohort was associated with elevations in plasma TMAO which were statistically relevant. Moreover, Gao et al. (2014) described that dietary TMAO supplementation in mice increases impaired glucose tolerance, insulin signaling, and promotes the inflammation of adipose tissue. Additionally, we found that the presence of certain specific bacterial taxa in human feces was associated with the concentration of plasma TMAO in both study groups. We observed that the plasma TMAO concentrations were significantly and negatively associated with the abundance of Faecalibacterium prausnitzii in CAD-DM2 patients. Moreover, we found a significant positive correlation of plasma TMAO concentrations with Enterobacteriaceae and Desulfovibrio. According to our results, other human and animal studies have suggested that several families of bacteria are involved in the production of TMA/TMAO such as Prevotellaceae (Koeth et al., 2013) and Enterobacteriaceae (Craciun and Balskus, 2012; Zhu et al., 2014). Additionally, it has been previously reported that the increase in the conversion of choline to TMA can be caused by the expression of the cutC gene by bacteria such as Desulfovibrio (Craciun and Balskus, 2012).

On the other hand, in order to enter the bloodstream, the microbiota derived molecules TMAO need to pass the gut-blood barrier. Accordingly, we observed that plasma zonulin levels were significantly higher in CAD-DM2 patients. Recent research has revealed that circulating zonulin levels are significantly higher in patients with diabetes, polycystic ovary syndrome, obesity, non-alcoholic fatty liver disease, all of which are regarded as traditional risk factors of atherosclerosis (Li et al., 2016).

Therefore, in our study, CAD-DM2 patients presented an alteration in gut microbiota equilibrium (dysbiosis), a disruption of gut barrier function and an increase in gut permeability which altogether may result in aberrant production and absorption of microbe derived metabolites such as TMAO, which can exert their atherogenic effect through alterations in cholesterol and bile acid metabolism, activation of inflammatory pathways, and promotion of foam cells formation. This situation could contribute to an increased risk of major adverse cardiovascular events and death in the CAD-DM2 group.

The pathological basis of CAD is aterosclerosis. Furthermore, inflammation plays a decisive role in atherosclerotic plaque progression, plaque rupture, and thrombosis, which are the initial factors in acute coronary syndrome (Min et al., 2017). In addition, intestinal permeability may contribute to CAD through the production of inflammatory cytokines and the weakening of the stability of plaque. Thus, in this study the significantly higher serum IL-1B levels generated in the gut epithelium during the systematic or localized inflammation in CAD-DM2 patients could also raise intestinal permeability and facilitate intestinal translocation of microbial components and metabolites into the circulation. Moreover, IL-1β enhance cholesterol uptake from human macrophages by upregulation of their oxidized LDL receptors, resulting in foam cell formation and plaque growth (Ikonomidis et al., 1999). In a previous study, IL-1β levels were associated with a less favorable prognosis after acute coronary syndrome (Van Tassell et al., 2013).

Additionally, the significant decrease in the peripheral FOXP3 mRNA expression and serum IL-10 in the CAD-DM2 when compared with the CAD-NDM2 group was significatively associated with the decrease in the abundance of Bacteroides fragilis and Faecalibacterium prausnitzii, respectively. Bacteroides fragilis plays an important role in mucosal T cell homeostasis by means of regulating the function of T cells (Mazmanian et al., 2008). The immunomodulatory molecule polysaccharide A derived from the human commensal Bacteroides fragilis mediates the conversion of CD4+ T cells into Foxp3+ Treg cells that produce IL-10 during commensal colonization, stimulating immunological development within mammalian hosts (Round and Mazmanian, 2010). At the same time, IL-10 is a key cytokine in Treg-mediated suppression (Wang et al., 2001). It has been described as a decreased percentage of peripheral CD4 + CD25 + Foxp3 + Treg and serum IL-10 level in patients with DM2. Furthermore, the percentages of peripheral CD4 + CD25 + Foxp3 + Treg and serum IL-10 level were influenced by diabetic complications such as cardiovascular diseases (Qiao et al., 2016).

On the other hand, the commensal bacteria Faecalibacterium prausnitzii is an important supplier of butyrate to the colonic epithelium. Butyrate is a short chain fatty acid (SCFA) that promotes the integrity of gut epithelial tight junctions as well as increases the release of the anti-inflammatory and antiatherogenic cytokine IL-10 (Caligiuri et al., 2003) and promotes Foxp3 + Treg induction (Furusawa et al., 2013). So, the significant reduction in the abundance of Faecalibacterium prausnitzii and in the predicted proportion of genes related to the butanoate metabolism that we found in the CAD-DM2 patients when compared with CAD-NDM2 may partially explain the significantly lower levels of zonulin and serum IL-10 and the significantly higher levels of serum IL-1β and plasma TMAO in the CAD-DM2 patients.

Our study has certain limitations but also some important strengths. The limitations include the number of patients which was small and the possible influence of conflicting factors such as diabetes medication over gut microbiota. The strengths of our study lie in the well matched cohorts and the next generation sequencing of the microbiome.

## CONCLUSION

We have demonstrated that the presence of DM2 in CAD patients is associated with a gut microbiota change and reduced diversity together with functional metagenomic differences. CAD-DM2 patients had less beneficial or commensal bacteria (such as Faecalibacterium prausnitzii, and Bacteroides fragilis) and more opportunistic pathogens (such as Enterobacteriaceae, Streptococcus, and Desulfovibrio) that may impair intestinal barrier function (increasing zonulin levels), enhance the serum levels of TMAO, and may therefore contribute to inflammatory processes related to CAD by means of increasing the production of inflammatory cytokines (such as IL-1B). Moreover, the significant reduction in the abundance of Faecalibacterium prausnitzii and Bacteroides fragilis found in CAD-DM2 patients

was associated with the decrease in the release of the anti-inflammatory and anti-atherogenic cytokine IL-10 and the peripheral FOXP3 mRNA expression, respectively, both implicated in the development and function of Treg cells. These results suggest that the presence of DM2 is related to an impaired regulation of the immune system in CAD patients, mediated in part by the gut microbiota composition and functionality and the production and effects of their gut microbiota derived molecules. Thus, gut microbiota may represent a new target for therapeutic manipulation, treatment, and prevention of complex cardiometabolic diseases through regulation of the immune system.

## AUTHOR CONTRIBUTIONS

MQ-O, AM-G, and MJ-N: Conceived the study and developed the experimental design. LJ-M, PC-C, DE, AM-G, MQ-O, and MJ-N: Were responsible for the acquisition and selection of all samples utilized in this study. LS-A, DC-C, IM-I, and MQ-O: Performed all laboratory assays. LS-A, IM-I, DC-C, MJ-N, and MQ-O: Compiled the database and performed the statistical analysis and data interpretation. LS-A, DC-C,

#### REFERENCES


IM-I, LJ-M, PC-C, DE, MJ-N, and MQ-O: Wrote the paper. LJ-M, PC-C, AM-G, MJ-N, and MQ-O: Provided critical revision. All authors read and approved the final manuscript.

## FUNDING

This work was supported by the Fundación Andaluza de Cardiología and a grant from the Spanish Ministry of Health (FIS) (PI13/02542), co-founded by Fondo Europeo de Desarrollo Regional (FEDER). The research group belongs to the "Centros de Investigación en Red" CIBERobn "Fisiopatología de la Obesidad y Nutrición" (CB06/03/0018) and CIBERCV "Enfermedades Cardiovasculares" (CB16/11/00360) of the "Instituto de Salud Carlos III". Isabel Moreno-Indias and María Isabel Queipo-Ortuño acknowledges support from the "Miguel Servet Type I" program (CP16/00163) and (CP13/00065) respectively) from the Instituto de Salud Carlos III, Madrid, Spain, co-founded by Fondo Europeo de Desarrollo Regional (FEDER). Daniel Castellano-Castillo acknowledges support from the "FPU" program (FPU13/04211) from the Ministerio de Educacion, Cultura y Deporte.

human symptomatic carotid atherosclerotic plaques. J. Vasc. Res. 47, 221–230. doi: 10.1159/000255965



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Sanchez-Alcoholado, Castellano-Castillo, Jordán-Martínez, Moreno-Indias, Cardila-Cruz, Elena, Muñoz-Garcia, Queipo-Ortuño and Jimenez-Navarro. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Microbiome-Derived Lipopolysaccharide Enriched in the Perinuclear Region of Alzheimer's Disease Brain

#### *Yuhai Zhao1,2, Lin Cong1,3, Vivian Jaber1 and Walter J. Lukiw1,4,5\**

*1Neuroscience Center, Louisiana State University School of Medicine, Louisiana State University Health Sciences Center, New Orleans, LA, United States, 2Department of Anatomy and Cell Biology, Louisiana State University School of Medicine, Louisiana State University Health Sciences Center, New Orleans, LA, United States, 3Department of Neurology, Shengjing Hospital, China Medical University, Heping District, Shenyang, China, 4Department of Neurology, Louisiana State University School of Medicine, Louisiana State University Health Sciences Center, New Orleans, LA, United States, 5Department of Ophthalmology, Louisiana State University School of Medicine, Louisiana State University Health Sciences Center, New Orleans, LA, United States*

#### *Edited by:*

*Wesley H. Brooks, University of South Florida, United States*

#### *Reviewed by:*

*Ai-Ling Lin, University of Kentucky, United States Abdul Sadiq, University of Malakand, Pakistan*

> *\*Correspondence: Walter J. Lukiw wlukiw@lsuhsc.edu*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 25 July 2017 Accepted: 16 August 2017 Published: 04 September 2017*

#### *Citation:*

*Zhao Y, Cong L, Jaber V and Lukiw WJ (2017) Microbiome-Derived Lipopolysaccharide Enriched in the Perinuclear Region of Alzheimer's Disease Brain. Front. Immunol. 8:1064. doi: 10.3389/fimmu.2017.01064*

Abundant clinical, epidemiological, imaging, genetic, molecular, and pathophysiological data together indicate that there occur an unusual inflammatory reaction and a disruption of the innate-immune signaling system in Alzheimer's disease (AD) brain. Despite many years of intense study, the origin and molecular mechanics of these AD-relevant pathogenic signals are still not well understood. Here, we provide evidence that an intensely pro-inflammatory bacterial lipopolysaccharide (LPS), part of a complex mixture of pro-inflammatory neurotoxins arising from abundant Gramnegative bacilli of the human gastrointestinal (GI) tract, are abundant in AD-affected brain neocortex and hippocampus. For the first time, we provide evidence that LPS immunohistochemical signals appear to aggregate in clumps in the parenchyma in control brains, and in AD, about 75% of anti-LPS signals were clustered around the periphery of DAPI-stained nuclei. As LPS is an abundant secretory product of Gram-negative bacilli resident in the human GI-tract, these observations suggest (i) that a major source of pro-inflammatory signals in AD brain may originate from internally derived noxious exudates of the GI-tract microbiome; (ii) that due to aging, vascular deficits or degenerative disease these neurotoxic molecules may "leak" into the systemic circulation, cerebral vasculature, and on into the brain; and (iii) that this internal source of microbiome-derived neurotoxins may play a particularly strong role in shaping the human immune system and contributing to neural degeneration, particularly in the aging CNS. This "*Perspectives*" paper will further highlight some very recent developments that implicate GI-tract microbiome-derived LPS as an important contributor to inflammatory-neurodegeneration in the AD brain.

Keywords: Alzheimer's disease, inflammatory degeneration, lipopolysaccharide, microbiome, microRNA, small non-coding RNAs

## INTRODUCTION—INFLAMMATORY SIGNALING IN THE ALZHEIMER'S DISEASE (AD) BRAIN

Multiple aspects of increased inflammatory signaling and an altered innate-immune system are consistent features of AD neuropathology; however, it is not well understood where these pathogenic signals originate or how they progressively contribute to the AD process (1–5). AD is characterized by the appearance of complex networks of many different kinds of chemokines and cytokines including, prominently, interleukin 1β (IL-1β) and tumor necrosis factor (TNFα), 40 and 42 amino acid amyloid beta (Aβ40, Aβ42) peptides, and adhesion molecules, in addition to the progressive deposition of these Aβ peptide containing amyloid plaques and neurofibrillary tangles (NFT) in the parenchyma of AD brain (6, 7). Activated microglia, astrocytes, or neurons appear to mediate the release of these pro-inflammatory molecules and cellular immune components (6, 8–12). Indeed, chemokines, cytokines, the insoluble Aβ42-enriched peptide deposits, NFTs, apoptotic, damaged and vanishing neurons, and activated microglia, and other related pro-inflammatory signals are potent neuropathological stimulants that appear to maintain the AD brain in a "*chronic state of self-reinforcing inflammation"* (2, 7, 10–13). Very recent studies that evaluated the pro-inflammatory potential of several different chemokines, cytokines, Aβ peptides, and lipopolysaccharides (LPS), either alone or in combination, have indicated that when compared, bacterial LPSs exhibit the strongest induction of pro-inflammatory signaling in human neuronal–glial cells in primary coculture of any single inducer, and different LPS extracts from different gastrointestinal (GI)-tract resident Gram-negative bacteria appeared to have different pro-inflammatory potential (12, 14–16). For example, exposure of LPS from the Gram-negative GI-tract abundant *Bacteroides fragilis* to primary human neuronal–glial cells in coculture was found to be an exceptionally powerful inducer of the NF-κB p50/p65 dimer, a known pro-inflammatory transcription factor complex that triggers the expression of pathogenic pathways involved in neurodegenerative inflammation (15, 16). In both neocortex and hippocampus, LPS has been detected to range from a ~7- to ~21-fold increase abundance in AD brain (**Figures 1A–D**). Along with an avalanche of very recent work from independent laboratories, these observations prompted us to further examine the presence and anatomical location of LPS in AD brains versus age- and gender-matched controls (12, 17, 18).

## INTERNALLY DERIVED NOXIOUS EXUDATES OF THE GI-TRACT MICROBIOME

Major Gram-negative bacilli of the human GI-tract, such as the abundant *B. fragilis* and *Escherichia coli* (*E. coli*), are capable of discharging a remarkably complex assortment of pro-inflammatory neurotoxins. These consist of four major components: (i) bacterial amyloids (10, 21); (ii) endotoxins and exotoxins (5, 12); (iii) LPS (12, 18); and (iv) small non-coding RNAs (sncRNAs) [(22–25), unpublished observations]. Either

Figure 1 | Continued

#### Figure 1 | Continued

(A–D) Western and (E–F) immunohistochemical analysis of lipopolysaccharide (LPS) (~37 kDa) signals in human brain temporal lobe neocortex [*N* = 4 control and 4 sporadic Alzheimer's disease (AD) cases; quantified in (B)]; and (C) hippocampus [*N* = 3 control and *N* = 3 sporadic AD cases; quantified in (D)] were compared against β-actin (~42 kDa) abundance in the same sample (using anti-*Escherichia coli* LPS; cat# ab35654 from Abcam, Cambridge UK and anti-β-actin cat# 3700, Cell Signaling, Danvers, MA, USA). All Western methodologies have been previously described in detail (12, 19). Densitometric readings of immunereactive bands were obtained using ImageQuantTL [GE Healthcare (12, 19, 20)]; all control and AD tissues were age- and gender-matched; there were no significant differences between the age (control 82.5 ± 8.1 years, AD 81.3 ± 8.8 years), gender (all female), postmortem interval (PMI) (all tissues 3.8 h or less), RNA quality, or RNA yield between each of the two groups; in these samples, LPS abundance was found to be on average greater than sevenfold as abundant in AD when compared to control neocortex; LPS was found to be on average >21-fold as abundant in AD when compared to control hippocampus; in (B,D) a dashed horizontal line at 100 is included for ease of comparison; \**p* < 0.01 (ANOVA); (E,F) for immunohisto-chemistry control and AD neocortex and/or hippocampal brain tissues were embedded, sectioned (10 µm), fixed, and incubated with primary antibodies (1:1,000; 1× PBS with 2% BSA, 2% goat or donkey serum, and 0.1% TX-100) overnight at 4°C, washed with PBS, and then incubated with Alexa Fluor-conjugated species-specific secondary antibodies (LPS; red fluorescence λmax ~ 650 nm); sections were next counter-stained with DAPI (blue fluorescence; λmax ~ 470 nm) for nuclei (E), and/or Aβ peptide (green fluorescence; λmax ~ 510 nm) (F) and imaged with Zeiss LSM 700 Confocal Laser microscope system (Richmond, VA, USA); note perinuclear staining of LPS in AD; while there appears to be random association of LPS with Aβ deposits in controls, >75% of all LPS signals were found to be associated with brain cell nuclei in AD; the significance of this is not currently known; the association of LPS with the major cellular repository for genetic material suggests that the significance of this association may be genetic; white arrows highlight LPS-nuclear envelope association; a total of 26 control and AD brains (PMI 3.8 h or less) were examined and yielded highly similar results; (E,F) magnification 50×.

alone or in various combinations, these neurotoxins are intensely pro-inflammatory toward primary human brain cells (12, 15, 16). As integral components of the outer leaflet of the outer membrane of Gram-negative bacteria, LPS shed into the local environment have historically been thought to play some host–pathogen immune-evasion strategy useful to bacterial survival while eliciting strong immune and inflammatory responses within the host. Interestingly, secreted LPS along with proteolytic endotoxins and amyloid monomers are generally soluble as monomers. However, over time, they aggregate into highly insoluble fibrous lipoprotein lesions that associate with the progressive degenerative neuropathology of several common, age-related disorders of the human systemic circulation, and CNS including systemic inflammatory response syndrome, multiple sclerosis, prion disease, and AD (12, 20, 26). LPS, the major molecular component of the outer membrane of Gram-negative bacteria normally serves as a physical barrier providing the bacteria protection from its surroundings. LPS is also recognized by the immune system as a marker for the detection of bacterial pathogen invasion and responsible for the development of inflammatory response is perhaps the most potent stimulator and trigger of inflammation known (27). LPS activates toll-like receptors (TLRs), membrane-spanning protein receptors expressed in microglial cells of the innate-immune system, which recognize common damage- or pathogen-associated molecular-patterns [DAMPS, PAMPs (2, 28)]. Interestingly, of the 13 currently characterized TLRs, the microglial TLR2 and TLR4 are activated by amyloid, LPS, lipoglycans, and/ or other microbial triggers that subsequently induce cytokine production, inflammation, phagocytosis, and innate-immune defense responses that directly induce the development of CNS pathology. In addition to the TLR2 and TLR4 receptors, at least one additional microglial transmembrane LPS receptor—CD14 mediates phagocytosis of both bacterial components and Aβ42 peptides, hence expanding roles for microglia and microglial LPS receptors in AD pathophysiology (12, 29).

To cite other recent examples, a secreted, highly proinflammatory zinc metalloprotease metalloproteinase *B. fragilis* endotoxin called fragilysin (BFT) derived from enterotoxigenic strains of *B. fragilis* have been recently shown to contribute to: (i) anaerobic bacteremia, sepsis and systemic inflammatory distress, diarrheal disease; (ii) systemic inflammation, GI-tract, and colorectal cancers; (iii) inflammatory neurodegeneration in part *via* the disruption of epithelial cell-based GI-tract barriers *via* cleavage of the synaptic adhesion zonula adherens protein E-cadherin; and (iv) enterotoxigenic microbes specifically impact microglial-mediated innate-immune responses, detoxifying and phagocytic mechanisms, and amyloidogenesis characteristic of inflammatory aspects of neurodegeneration (12, 15, 16, 30–34). Prokaryotic sncRNAs play essential roles in the regulation of many bacteriological processes including the expression of exotoxins and endotoxins and the regulation of bacterial virulence (22). In eukaryotes, microRNAs (miRNAs) also function as key regulators in many biological processes through posttranscriptional suppression of mRNAs and the downregulation of gene expression. Typical trans-acting microRNA-size sncRNAs are abundant in all prokaryotic cells including bacteria and fungi, but their production, release, and leakage from the confines of a healthy GI-tract into systemic and cerebral circulation and downstream effects along the gut microbiome–brain axis are a highly novel and largely unexplored research area (12, 22, 25). There is considerable speculation that, as for other bacterial exudates, such RNA-based neurotoxins may be pathogenic and highly detrimental to the homeostatic function of the neuronal, glial, endothelial, and other brain cells that comprise the CNS (23, 24).

#### LEAKAGE OF NEUROTOXIC MOLECULES INTO THE SYSTEMIC CIRCULATION AND THE CNS

Gram-negative bacterial exudates of the human GI-tract are not only the primary source of a remarkable array of neurotoxic pro-inflammatory amyloids, endo- and exotoxins, LPSs, and sncRNAs but also serve as potent sources of membranedisrupting agents (12, 15, 16, 35, 36). As aforementioned, BFT can alone induce the disruption of epithelial cell-based GI-tract membrane barriers *via* presenilin 1-dependent cleavage of the zonula adherens protein E-cadherin, thus leading to progressive functional decline in membrane integrity (12, 15, 16, 30–34). Other recent reports suggest that intestinal dysbiosis and "*leaky*  *gut syndrome*" constitutes a key pathophysiological link for transport of microbiome-derived toxins across GI-tract and blood–brain biological barriers that result in a progression from systemic to CNS inflammation (12, 21). The progressive failure of major physiological barriers is reminiscent of the activation of the thanatomicrobiome (*the "death"-associated microbiome*) and the deactivation of protective biological barriers that occurs at the time of death when normal endothelial cell structures and signaling: (i) becomes increasingly inoperative and "*leaky*" (1, 12, 37); and (ii) progressively unable to support normal homeostatic brain functions that are accompanied by a progressive and insidious functional decline (12, 28, 37). These recent findings indicate that AD-affected brains have remarkably large loads of bacterialderived toxins compared to controls. The transfer of noxious, pro-inflammatory molecules from the GI-tract microbiome to the CNS may be increasingly important during the course of aging when both the GI-tract and blood–brain barriers become significantly more permeable (12, 28, 38).

#### PERINUCLEAR LOCALIZATION OF LPS IN AD BRAINS

While other recent studies have reported an LPS-mediated stimulation of chronic inflammation, beta-amyloid accumulation, and episodic memory decline in murine models of AD (39, 40) and a biophysical association of LPS with amyloid deposits and blood vessels in human AD patients (18), here, we provide the first evidence of a perinuclear association of LPS with AD brain cell nuclei (**Figures 1E,F**). Strong adherence of LPS to the nuclear periphery has recently been shown to inhibit nuclear maturation and function that may impair or block export of mRNA signals from brain cell nuclei, a highly active organelle with extremely high rates of transcription, mRNA processing, and export into the cytoplasm [(41–43), unpublished observations]. This may in part be responsible for the widely observed, generalized downregulation of global gene expression in AD, independently reported by several AD gene expression research laboratories, through the biophysical blockage of mRNA trafficking through nuclear pores (41, 42, 44, 45). LPS may be further injurious to the nuclear membrane just as LPS contributes to cerebrovascular endothelial cell membrane injury (12, 18, 40). Lastly, evidence is accumulating that neurotoxic exudates from other GI-tract microbiota may contribute to dysfunction in additional, ultimately fatal neuropsychiatric illnesses that involve progressive inflammatory neurodegeneration (8, 12). New opportunities to modulate existing gut microbiota and their exudates using probiotics and/or modifications through soluble or insoluble dietary fiber intake could provide novel targets for more effective clinical intervention [**Figure 2** (18, 46, 47); unpublished observations]. Interestingly, the high intake of dietary fiber is a strong inhibitor of *B. fragilis* abundance and proliferation in the intact human GI-tract and as such is a potent inhibitor of the neurotoxic *B. fragilis*-derived amyloids, LPS, enterotoxins, and sncRNAs. Hence, dietary fibermediated suppression of *B. fragilis* abundance may turn out to be beneficial for *both* the human GI-tract microbiome and CNS health (34, 38, 46).

Figure 2 | The human gastrointestinal (GI)-tract microbiome as a source of strong pro-inflammatory exudates—highly schematicized depiction of anaerobic, Gram-negative bacilli (such as *Escherichia coli* and *Bacteroides fragilis*) of the human GI-tract microbiome and their potentially pathogenic, immunogenic, and pro-inflammatory neurotoxins [amyloids, endotoxins and exotoxins, lipopolysaccharide (LPS), and small non-coding RNAs (sncRNAs)] that may contribute to systemic and CNS inflammation and neuro-immune disruption; two major sources of these complex mixtures are *E. coli* and *B. fragilis*; major anaerobic Gram-negative bacilli of the human middle and lower GI-tract, respectively; the *B. fragilis* toxin (BFT) fragilysin is one of the most potent pro-inflammatory molecules known (12, 15, 16, 30, 37, 38); these intensely pro-inflammatory LPS species may be able to "leak" through at least two major biophysiological barriers—the GI-tract barrier and the blood–brain barrier—to access brain compartments [see Ref. (2, 12, 28, 30, 31, 34)]. Neurotoxic mixtures secreted by multiple GI-tract microbes or other microbial species may have considerable potential to support inflammatory signaling within the CNS (2, 12, 21, 28, 30, 31, 34); *B. fragilis* proliferation and (BFT) fragilysin levels may be kept in check by increased intake of soluble and insoluble dietary fiber (34, 38, 46); interestingly, BFT-derived fragilysin may exert neurotoxic activities *via* multiple mechanisms: (i) by increasing the permeability or "leakiness" of the intestinal epithelium *via* the dissolution of tight junctions in epithelial cells (28, 30); and (ii) by promoting amyloid peptide aggregation and progressive amyloidogenesis (15, 16, 18, 37, 38); Figure 2 modified and updated from Lukiw (15, 16).

## CONCLUDING REMARKS

It is not generally appreciated that, in the human body, microbial genes outnumber human genes by about 100 to 1, and the impact of bacterial genetics on human health and disease may have been vastly underestimated (8, 12, 15–17, 48). The assumption of the privileged immunological status of the CNS has also been recently questioned in multiple investigations, particularly in terms of inflammatory neurodegenerative diseases such as AD, as both microbial-derived nucleic acid sequences and/or noxious exudates representative of GI-tract Gram-negative bacteria are showing up within CNS compartments, including, prominently, anatomical regions of the CNS involved in inflammatory and pathological signaling and neuro-immune disruptions that characterize the AD process (9, 12, 15, 16, 18, 49). For example, LPS has been recently localized to the same anatomical regions involved in AD-type neuropathology to levels of greater than sevenfold over control in the temporal lobe neocortex and >21-fold over control in the hippocampus. This suggests that GI-tract microbiome-derived LPS may be an important initiator and/or significant contributor to inflammatory degeneration in the AD CNS (**Figures 1** and **2**). An alternative, yet, highly speculative view is that the human CNS may have its own microbiome, which could also explain the presence of Gram-negative bacterial secretory components in the brain as well as multiple forms of microbial-derived nucleic acid sequences (12, 49).

#### AUTHOR CONTRIBUTIONS

YZ, LC, VJ, and WL conceived and discussed the experimental design; YZ, LC, VJ, and WL performed the experiments; YZ and WL performed bioinformatics and contributed to the medical artwork; WL reviewed the results and further researched and wrote this paper.

## FUNDING

The work in this paper was presented in part at the Vavilov Institute Autumn 2016 Seminar Series (Институт Вавилова Осень 2016 Семинар Серии) in Moscow, Russia October 2016, at the

#### REFERENCES


Society for Neuroscience (SFN) Annual Meeting, San Diego, CA, USA November 2016, and will be presented in Abstract-Special Symposium format at the Society for Neuroscience (SFN) Annual Meeting, Washington, DC, USA November 2017. Sincere thanks are extended to Drs P. N. Alexandrov, J. G. Cui, F. Culicchia, W. Poon, K. Navel, C. Hebel, C. Eicken, and the late Dr. J. M. Hill for helpful discussions in this research area, for short postmortem interval (PMI) human brain tissues or extracts, for initial bioinformatics and data interpretation, and to D. Guillot and A. I. Pogue for expert technical assistance and medical artwork. Thanks are also extended to the University of California at Irvine Brain Bank and the many neuropathologists, physicians, and researchers of the US and Canada who have provided high quality, short post-mortem interval (PMI) human CNS or extracted tissue fractions for scientific study. Research on the microRNAs, pro-inflammatory, and pathogenic signaling in the Lukiw laboratory involving the innate-immune response, neuroinflammation, and amyloidogenesis in AD, prion, and in other neurological diseases was supported through an unrestricted grant to the LSU Eye Center from Research to Prevent Blindness (RPB); the Louisiana Biotechnology Research Network (LBRN) and NIH grants NEI EY006311, NIA AG18031, and NIA AG038834 (WL).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Zhao, Cong, Jaber and Lukiw. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Gut Dysbiosis and Adaptive Immune Response in Diet-induced Obesity vs. Systemic Inflammation

Jana Pindjakova<sup>1</sup> , Claudio Sartini <sup>2</sup> , Oriana Lo Re<sup>1</sup> , Francesca Rappa<sup>3</sup> , Berengere Coupe<sup>4</sup> , Benjamin Lelouvier <sup>4</sup> , Valerio Pazienza<sup>5</sup> and Manlio Vinciguerra1, 6 \*

<sup>1</sup> Center for Translational Medicine, International Clinical Research Center, St. Anne's University Hospital, Brno, Czechia, <sup>2</sup> Department of Primary Care and Population Health, University College London, London, United Kingdom, <sup>3</sup> Section of Human Anatomy, Department of Experimental Biomedicine and Clinical Neurosciences, University of Palermo, Palermo, Italy, <sup>4</sup> Vaiomer, Labège, France, <sup>5</sup> Gastroenterology Unit, IRCCS "Casa Sollievo della Sofferenza" Hospital, San Giovanni Rotondo, Italy, <sup>6</sup> Division of Medicine, Institute for Liver and Digestive Health, University College London, London, United Kingdom

A mutual interplay exists between adaptive immune system and gut microbiota. Altered gut microbial ecosystems are associated with the metabolic syndrome, occurring in most obese individuals. However, it is unknown why 10–25% of obese individuals are metabolically healthy, while normal weight individuals can develop inflammation and atherosclerosis. We modeled these specific metabolic conditions in mice fed with a chow diet, an obesogenic but not inflammatory diet—mimicking healthy obesity, or Paigen diet—mimicking inflammation in the lean subjects. We analyzed a range of markers and cytokines in the aorta, heart, abdominal fat, liver and spleen, and metagenomics analyses were performed on stool samples. T lymphocytes infiltration was found in the aorta and in the liver upon both diets, however a significant increase in CD4+ and CD8+ cells was found only in the heart of Paigen-fed animals, paralleled by increased expression of IL-1, IL-4, IL-6, IL-17, and IFN-γ. Bacteroidia, Deltaproteobacteria, and Verrucomicrobia dominated in mice fed Paigen diet, while Gammaproteobacteria, Delataproteobacteria, and Erysipelotrichia were more abundant in obese mice. Mice reproducing human metabolic exceptions displayed gut microbiota phylogenetically distinct from normal diet-fed mice, and correlated with specific adaptive immune responses. Diet composition thus has a pervasive role in co-regulating adaptive immunity and the diversity of microbiota.

## Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Paul Cordero Sanchez, University College London, United Kingdom Fabio Galvano, University of Catania, Italy

\*Correspondence: Manlio Vinciguerra manliovinciguerra@gmail.com

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 28 April 2017 Accepted: 07 June 2017 Published: 22 June 2017

#### Citation:

Pindjakova J, Sartini C, Lo Re O, Rappa F, Coupe B, Lelouvier B, Pazienza V and Vinciguerra M (2017) Gut Dysbiosis and Adaptive Immune Response in Diet-induced Obesity vs. Systemic Inflammation. Front. Microbiol. 8:1157. doi: 10.3389/fmicb.2017.01157 Keywords: obesity, inflammation, gut microbiota, adaptive immune system

#### INTRODUCTION

The main feature of obesity is an excess of adipose tissue, which is the result of an imbalance existing between the intake and the expenditure of energy. The causes of obesity are both genetic and environmental; the diseases often comes along with the establishment of several chronic co-morbidities, such as high fasting hyperglycaemia, hypertriglyceridemia, dyslipidaemia, and hypertension (Alberti et al., 2005). Clinical diagnosis of metabolic syndrome is defined by the copresence of at least three of the above criteria (Alberti et al., 2005). Metabolic syndrome enhances the odds of having type 2 diabetes and of developing diseases of the cardiovascular system. The majority of people with the metabolic syndrome are in obese, suggesting that the excess mass of adipose tissue may play a causative role in this cluster of diseases (Despres et al., 2008). However, this hypothesis has been strongly debated because several epidemiological analyses have evidenced people with a normal body mass index (BMI) who nevertheless display markers of inflammation and metabolic diseases [here termed metabolic syndrome leans (MSL)], such as high levels of triglycerides and accumulation of fat in the liver (Alberti et al., 2005); in fact, independently of BMI, and with variability linked to race and geographical areas, approximately 1 adult in every 4 or 5 had metabolic syndrome (Alberti et al., 2005). Conversely, a lack of clinical consistency for several or all metabolic syndrome components is found in some individuals with long-established and morbid obesity, which is actually recognized as healthy despite a high BMI. These subjects are referred to as metabolically healthy obese (MHO), and their prevalence has been estimated to be between 10 and 40% of the obese population, notwithstanding design differences between studies, such as age, ethnicity, geography, sample size, and the lack of a standardization (Munoz-Garach et al., 2016). As the prevalence of obesity and metabolic syndrome rises continuously with enormous

economic and social costs, innovative countermeasures on the biological mechanisms, beyond prevention and lifestyle interventions, are required. In particular, the biological and disease mechanisms underlying the pathology of MSL and the health of MHO are not understood. Inflammation has been persistently associated with both obesity-associated diseases and the metabolic syndrome, indicating that low-grade inflammation is a potential and modifiable risk factor (Cox et al., 2015). The gut microbiota can be considered a distinct organ with endocrine properties; gut microbiota is involved, through a tight molecular interplay with the host organism, in the homeostasis of host organism energy and in stimulating of its immune system (Clarke et al., 2014). It has been proposed that gut microbiota participates to the establishment of metabolic diseases via the onset of low-grade inflammatory processes (Zupancic et al., 2012; Marchesi et al., 2016), and its composition is rapidly and heavily modulated by the diet (David et al., 2014). However, under healthy conditions commensal bacteria colonizing the gut interplay with the host immunity to maintain a state of

panels) and aorta sections (lower panels) in C57/BL6 mice fed with ND, HD, or PD. (D) Steatosis, lobular inflammation, and ballooning were scored semi quantitatively (0–4). \*p < 0.05; \*\*p < 0.01; \*\*\*p < 0.001 vs. ND.

homeostasis. In this respect, an immune system–gut microbiota cooperation which operates at optimal levels is instrumental for setting protective mechanisms against pathogenic agents and, at the same time, for keeping in check the regulatory pathways implicated in the avoidance of triggering immune responses to harmless antigens (Belkaid and Hand, 2014). This reciprocal interaction involves both innate (Thaiss et al., 2014) and adaptive immunity (Kato et al., 2014; Zhang and Luo, 2015). In this respect, signals from gut microbiome play crucial role in maturation (or differentiation) of IL-17 expressing Th17 cells as well as IFN-γ expressing Th1 cells (Ivanov et al., 2008; Gaboriau-Routhiau et al., 2009). Although, it has been suggested that dysbiosis can cause immune dysfunctions by activating B and T cells regardless of their distance from the location of their induction (Honda and Littman, 2016), there is scarce knowledge on the relationship between distinct immune cell populations, more in particular those belonging to the adaptive immunity, and the heterogeneity of digestive system-residing and symbiotic bacteria.

Here we modeled the metabolic and clinical features of MSL and MHO humans in C57/BL6 mice fed for 20 weeks with a chow diet, a high fat obesogenic but not inflammatory diet (mimicking healthy obesity) or a hypercholesteraemic, pro-atherogenic, low fat diet (Paigen diet, mimicking systemic inflammation, and fatty liver in the lean subjects; Getz and Reardon, 2006), under the same housing environment. We then analyzed possible interactions among adaptive immune system in multiple tissues, and gut microbiota. Mice fed these distinct "unhealthy" diets reproducing human metabolic exceptions, MSL and MHO, had a gut microbiota with phylogenetic characteristics significantly divergent from normal diet-fed littermates, and displayed specific intra-tissue adaptive immune responses.

## MATERIALS AND METHODS

#### Dietary Mice Models

Four week old male C57BL/6 (B6) mice were purchased from Velaz, Ltd. (Prague, Czech Republic). Animals (n = 10 per experimental group) were housed in specific pathogen-free facilities and fed ad libitum with basal (normal) or specific (Paigen or Western) diet for 20 weeks with fresh, clean water available at all time. Diet compositions were as following: normal diet (ND): proteins % 18.6, fat % 10 (Linoleic Acid, % 3.34, Linolenic Acid, % 0.07 Arachidonic Acid, % 0.01, Omega-3 FA, % 0.07, Saturated FA % 2.72, Monounsaturated FA, % 3.31, Polyunsaturated FA, % 3.42), carbohydrates % 60.6, cholesterol % 0, choline chloride % 0; High fat diet (HD): proteins % 17.3, fat % 21.2 (Linoleic Acid, % 1.70, Linolenic Acid, % 0.16

FIGURE 3 | (A) Cell suspension obtained from tissues were surface-stained for T lymphocyte markers with antibody combination CD45, CD4, and CD8 (A) and gated for CD45+ CD4+ T lymphocytes and CD45+ CD8+ T lymphocytes. (B) For myeloid cell subsets, the cell suspensions were surface-stained with antibody against CD45, CD11b, CD11c, Ly6G, and F4/80 (B) and gated for CD45+ CD11b+ Ly6G+ neutrophils, CD45+ CD11b+ CD11c+ dendritic cells, CD45+ CD11b+ CD11c-F4/80+ macrophages.

Arachidonic Acid, % 0.02 Omega-3 FA, % 0.23, Saturated FA, % 7.92 Monounsaturated FA, 6.28, Polyunsaturated FA, % 3.14), carbohydrate % 48.5, cholesterol % 0, choline chloride % 0.2); Paigen diet (PD): proteins % 20.8, fat % 15 (Linoleic Acid, % 1.70, Linolenic Acid, % 0.14, Arachidonic Acid, % 0.01, Omega-3 FA, % 0.16, Saturated FA % 7.18 Monounsaturated FA, % 5.34, Polyunsaturated FA, % 1.91), carbohydrate % 61, cholesterol % 1.25, choline chloride % 0.5. The experiments were performed in accordance with the law governing the protection of animals and the principles derived from the requirements of the Act No. 359/2012 Sb., on the protection of animals against cruelty and the decree 419/2012 Sb. Ministry of Agriculture of Czech Republic on the protection of experimental animals (including relevant EU regulations). The experiments were approved by the local Animal Ethics Committee on the Welfare of Experimental Animals and by the Ministry of Education of Czech Republic (MSMT-2582/2016-14)—project number 66-2015. Serum levels of fasting glucose, fasting insulin, triglycerides, and cholesterol were measured as we have previously described (Cederroth et al., 2008; Pazienza et al., 2016).

## Histology

Samples of liver, aorta, and heart from each mouse and were fixed in formalin and embedded in paraffin for histological analysis. Sections with a thickness of 4 µm were obtained from paraffin blocks and stained with hematoxylin and eosin for histological examinations (Benegiamo et al., 2013). Histological classification of NAFLD was performed by applying a semiquantitative scoring system grouping histological traits into broad classes (steatosis, fibrosis, portal inflammation, hepatocellular injury, and miscellaneous features; Kleiner et al., 2005).

## Tissue Digestion and Single Cell Suspension Preparation

To prepare single cell suspension from solid tissue (aorta, heart, and abdominal fat), required digestion, the tissue was minced with a sterile scissors and placed in 1 ml DMEM containing: for heart and aorta—2.5 mg/ml Collagenase type XI, 0.25 mg/ml Hyaluronidase type I-s, 0.25 mg/ml DNase I, 2.5 mg Collagenase type I; for abdominal fat −1 mg/ml Collagenase IV of 3% DMEM. Tissues were incubated in water bath for 1 h with vortex every

15 min and washed by cold PBS. Erythrocytes were removed by RBC lysis buffer (Biolegend), cells were washed by PBS and transferred to fresh tubes through 70 mm nylon mesh. Finally, the cell suspension was resuspended in 1 ml PBS per sample. Spleen and liver were cut into small pieces and passed through tissue grinder to Petri dish, and then the cell suspension was passed through the 70 µm cell strainers and processed as mentioned above. Peripheral blood was collected into heparinized syringe, resuspend in PBS and spin down. Erythrocytes were removed by RBC lysis buffer and cells passed through the 70 µm cell strainer. Single cell suspensions were used for flow cytometry or PCR.

## Flow Cytometry

Cells in single cell suspensions were stained in 100 µl aliquots of FACS buffer (2% FBS in PBS) after incubation with fluorochrome-labeled antibodies for 30 min at 4◦C followed by washing in FACS buffer. Combination of surface markers for Tlymphocytes was CD45, CD4, and CD8 and myeloid cell subsets were stained for CD45, CD11b, CD11c, F4/80, and Ly6G using specific antibodies (Biolegend). Analysis was performed using a BD Biosciences FACSCanto <sup>R</sup> flow cytometer and FlowJo <sup>R</sup> software (TreeStar Inc., Olten, Switzerland).

#### Gene Expression

Total RNA was isolated from cell suspensions using Trizol LS Reagent (Life Technologies). RNA was converted to cDNA using gb Reverse Transcription Kit (Generi-Biotech, Czech Republic). Equal amounts of cDNA were analyzed by Real-Time quantitative PCR using gb Elite PCR Master Mix (Generi-Biotech, Czech Republic) on a LightCycler <sup>R</sup> 480 Real Time PCR System (Roche). Relative quantifications were performed using the comparative CT method with normalization to GAPDH and results expressed as fold difference relative to a relevant control sample. Primers and probes were from Qiagen: GAPDH (Mm99999915\_g1), IL-17A (Mm00439618\_m1), IFN-γ (Mm01168134\_m1), IL-4 (Mm00445259\_m1), TGFβ (Mm01178820\_m1), IL-6 (Mm00446190\_m1), IL-12 p35 (Mm00434165\_m1).

### Metagenomics Profiling

The microbial population present in the fecal samples from mice was determined using next generation high throughput sequencing of variable regions of the 16S rRNA bacterial gene. The workflow performed at VAIOMER (France) includes the steps of (i) Library construction and sequencing; (ii)

PCR amplification was performed using 16S universal primers targeting the V3–V4 region of the bacterial 16S ribosomal gene (Vaiomer universal 16S primers). The joint pair length was set to encompass 476 base pairs amplicon thanks to 2 × 300 paired-end MiSeq kit V3. For each sample, a sequencing library was generated by addition of sequencing adapters. The

FIGURE 6 | (A) Alpha diversity using Shannon index of the fecal microbiota for each groups. (B) Relative abundance of major Phylum (Bacteroidetes and Firmicutes) for each group. (C) Relative abundance of most significant species, using RDP v11.4 databank in fecal samples of ND, HD, or PD mice. Graphs are displayed as mean ± SEM. \*\*p < 0.01; \*\*\*p < 0.001, One-Way Anova followed by Kruskal–Wallis test.

detection of the sequencing fragments was performed using MiSeq Illumina <sup>R</sup> technology; (iii) Bioinformatics pipeline, The targeted metagenomic sequences from microbiota were analyzed using the bioinformatics pipeline established by Vaiomer from the FROGS v1.3.0 guidelines. Briefly, after demultiplexing of the bar coded Illumina paired reads, single read sequences are cleaned and paired for each sample independently into longer fragments. Operational taxonomic units (OTU) are produced via single-linkage clustering and taxonomic assignment is performed in order to determine community profiles. PhyloSeq v1.14.0 R package was used to provide a set of classes and tools to facilitate the import, storage, analysis, and graphical display of microbiome census data. The samples with <5,000 sequences after FROGS processing were not included in the statistics (rarefaction analysis, alpha diversity, beta diversitymultidimensional scaling). The raw sequencing data are available upon request.

#### LEfSe Method

The OTU files generated were uploaded and formatted for LEfSe analysis using the per sample normalization of sum values option. The linear discriminant analysis effect size was determined using default values (alpha value of 0.5 for both the factorial Kruskal– Wallis test among classes and the pairwise Wilcoxon test between subclasses, threshold of 2.0 for the logarithmic LDA score for discriminative features) and the strategy for multi-class analysis set to "allagainst-all." LEfSe cladograms from the LDS effect size data were generated with Bacteria as the tree root. Differential features detected as biomarkers from the raw data used to generate the cladograms were plotted as abundance histograms with class and subclass information.

## Statistical Methods

The parametric Student's t-test (2-sample t-test) was used to compare the difference in mean of immune cells by type of diet (HD vs. ND), and difference in mean of cytokines by type of diet (HD vs. ND). The non-parametric Mann–Whitney U-test was also used to check if the results were basically similar to the ttest using GraphPad Prism Software (version 5.00 for Windows, San Diego, CA, USA): a p < 0.05 was considered significant. To explore the association of gut microbiota with immune cells and cytokines levels, analyses were carried out using STATA/SE software. As preliminary analysis, mean and standard deviation (SD) of each gut microbiota type and adaptive immune system parameters measured in the aorta, heart, liver, spleen, and fat were calculated. The Pearson's correlations between each gut microbiota and adaptive immune system parameters were also examined. In the final analysis, associations between gut microbiota and adaptive immune system parameters levels were explored by using linear regression models. In each of the models, the associations between each bacterial taxa and adaptive immune system parameters were reported as absolute difference (β), with 95% Confidence Intervals (CI), in immune system parameters levels by % of increase in the proportion of the bacterial taxa. Coefficient of determination (R 2 ) was also reported.


TABLE 1 | Associations from linear models between gut microbiota and alterations of adaptive immune system parameters measured in the Aorta.

(Continued)


In each of the models, the associations are reported as absolute difference (β), with 95% Confidence Intervals (CI), in immune system parameters levels by 1 unit increase in gut microbiota. In bold are reported the statistically significant results (p < 0.05). Per each of the models, the Coefficient of determination (R<sup>2</sup> ) is also reported.

## RESULTS

## Modeling Healthy Obesity and Metabolic Syndrome during Leanness in Mice

To model diets able to mimic MSL and MHO conditions in humans, three groups (N = 10) of 4 weeks old C57/BL6 mice were fed different dietary regimens: (i) a control normal diet (ND, 21.2% kcal from proteins, 58% kcal from carbohydrate, and 17% from fat); (ii) high fat diet, rich in fatty acids (HD, 21.2% kcal from proteins, 24% kcal from carbohydrate, and 58% from fat) and with 0.1% cholesterol, and (iii) atherogenic/inflammatory Paigen diet (PD), containing similar composition of the normal diet with in addition 1.25% cholesterol and 0.5% sodium cholate (**Figure 1A**). C57/BL6 mice had similar baseline weight before starting being fed the diets (mean = ∼21 ± 0.4 g). After 15 weeks of dietary regimens, body weight was unchanged in mice on the control ND or the PD, which both increased body weight during growth by ∼33% (ND = 28.3 ± 0.7 g and PD = 28.8 ± 0.77 g, respectively, **Figure 1B**). In contrast, mice on the HD gained ∼65% in weight (HD = 35.9 ± 0.6 g), compared to their baseline, indicating that only HD diet was obesogenic (p < 0.001 vs. ND and vs. PD). We then examined glucose and insulin levels at the end of the dietary treatment. Basal insulin and glucose fasting levels were considerably higher in PD vs. ND and HD (**Figures 2A,B**). A similar trend was observed for serum triglycerides and cholesterol levels, which were highest in the PD group vs. ND and HD (**Figures 2C,D**). Obesogenic HD regimen triggered lipid accumulation in the liver under the form of simple steatosis, whereas atherogenic/inflammatory PD regimen induced NAFLD/NASH at the end of its pathologic spectrum, characterized by lipid accumulation, ballooning, fibrosis, and inflammatory infiltrates, as quantified by NAFLD/NASH score (**Figure 1C** upper panels, **Figure 1D**), consistent with previous finding that the cholesterol and cholate components of Paigen diet induces genes involved in inflammation and fibrosis, respectively, in the liver (Vergnes et al., 2003). Cross-sectional analysis of aortas walls suggested an increased infiltration of inflammatory cells in the PD-fed mice, in comparison to ND or HD fed mice (**Figure 1C**, lower panels). Altogether, these data indicate that PD triggers prominent features of metabolic syndrome and inflammation in mice in absence of weight gain, compared to obesogenic HD.

#### Dissecting Diet-Dependent Intra-Tissue Adaptive Immune Changes

Cells of the innate immune system, in particular macrophages, mediate chronic inflammation (Sell et al., 2012). Moreover, B and T lymphocytes of the adaptive immune system have been recently recognized as important modulators of glucose homeostasis, indicating that antigen-driven immune responses could influence insulin resistance. Like macrophages, lymphocytes can be divided into populations with primarily proinflammatory functions (including CD8+ cytotoxic T cells, Th1, Th17) or primarily regulatory functions (including Treg or Th2) and the skewing of the adaptive immune milieu toward a proinflammatory phenotype can exacerbate the metabolic disturbances associated to obesity (Nishimura et al., 2009; Winer et al., 2009; Shen et al., 2015). Moreover, it is known since 1980 that T lymphocyte subsets and related cytokines are present in atherosclerotic lesions and affect their development (Lichtman, 2013). Here, to analyze the changes of adaptive immunity between MSL and MHO, we analyzed T cell populations from the blood and the spleen (secondary lymphoid organ), from the heart and the aorta (cardiovascular tissue), from the liver and from the adipose tissue (metabolic and nutrient hubs) of ND, PD, and HD-fed mice, using the flow cytometry gating strategy depicted in **Figure 2**. Briefly, CD4+ and CD8+ T cells were identified from the CD45+ lymphocyte populations, whereas to study the myeloid lineage, CD11b+ subpopulation was further analyzed for Ly6G+ cells (neutrophils), CD11b+CD11c+ cells


TABLE 2 | Associations from linear regression models between gut microbiota and alterations of adaptive immune system parameters measured in the heart.

(Continued)


In each of the models, the associations are reported as absolute difference (β), with 95% Confidence Intervals (CI), in immune system parameters levels by 1 unit increase in gut microbiota. In bold are reported the statistically significant results (p < 0.05). Per each of the models, the Coefficient of determination (R<sup>2</sup> ) is also reported.

(conventional dendritic cells), and CD11b+CD11c-F4/80+ cells (macrophages; **Figure 3**). In parallel, we measured by qPCR the intra-tissue expression levels of the following panel of cytokines that play a major role in the adaptive immune system being secreted by helper CD4+ T cells (Th1, Th2, Th17, and Treg) and stimulating several cell types: IL-1α, IL-4, IL-6, IL-12, IL-17 and IFN-γ. IL-1α, IL-12, IL-17, and IFN-γ are generally regarded as pro-inflammatory and pro-atherogenic, while IL-4 and IL-6 display pro- and anti-inflammatory properties which are context-dependent (Hunter and Jones, 2015; Zarzycka et al., 2015). Our analyses showed a great enrichment in myeloid cells, CD45+, CD11b+, CD11c, F480+, Ly6G+ upon PD- in the spleen compared to ND- and HD-feeding (**Figure 4A**), a massive lymphocyte infiltration in the aorta and in the liver upon both PD and HD compared to ND diet (**Figure 4B**), a significant increase in CD4+ and CD8+ positive cells percentage exclusively in the aorta and in the hearts of PD-fed animals compared to ND and HD (**Figure 4C**); no changes were observed in abdominal fat tissues (data not shown). At the cytokine level, IL-17 was greatly increased in the aorta, heart and fat only in PD-fed mice compared to HD and ND (**Figure 5A**). IL-1α, IFN-γ, and IL-4 levels were augmented in the aorta and/or in the heart only in PD-fed mice compared to HD and ND (**Figures 5B,C**). Finally, we report a trend in increased IL-6 and IL-12 mRNA levels in adipose tissue of PD mice compared to HD- and ND-fed mice (**Figure 5D**). Collectively, our data surprisingly indicate activation of several components of the adaptive immune system in the metabolic syndrome lean PD mouse model compared to an established mouse model of diet-induced obesity.

### Gut Microbiota Profiling By Metagenomic Sequencing

The reciprocal interaction between the gut microbiota and the adaptive immunity contributes to the insurgence of metabolic diseases and of inflammation (Kato et al., 2014; Zhang and Luo, 2015; Marchesi et al., 2016). It is however unknown how this interplay adapts to the MSL or MHO clinical features. To this aim we identified bacterial populations contained in fecal samples from ND-, HD-, and PD-fed mice using next generation high throughput sequencing of variable regions (V3–V4) of the 16S rDNA bacterial gene (Lluch et al., 2015; Paisse et al., 2016). Alpha diversity analyses, representing the mean of species diversity in each sample showed that ND-fed mice had a higher taxonomic diversity than the HD-fed mice, which in turn have a higher taxonomic diversity than the PD-fed mice (**Figure 6A**). Feces microbial composition after 20 weeks of different diet is highly different between the three groups, as shown by beta diversity metrics based multi-dimensional scaling Unifrac analysis (**Figure 7A**) and by hierarchical clustering (**Figure 7B**). The community structures observed in the different groups were significantly different. At the phylum level, Firmicutes and Bacteroidetes dominated the fecal microbiota in all groups (**Figure 7C**). No differences of Firmicutes and Bacteroidetes relative abundance were observed between ND and HD groups (**Figure 7C**). However, an increase in Bacteroidetes and a decrease in Firmicutes were observed in PD groups (**Figure 6B**). At the family level, the fecal microbiota was dominated by Porphyromonadaceae in all groups (**Figure 7D**) and are significantly higher in Paigen Diet mice compared to HD and ND groups (**Figure 7F**). Focusing on diet effect between the three groups of mice, broad population changes were seen from phylum to genus level (**Figure 7E**), significantly enriched taxa for all groups are identified using LDA Effect Size (LEfSe) analysis. Clostridia class are significantly enriched in ND and HD mice compared to PD mice (**Figures 7E,F**). Actinobacteria and Deltaproteobacteria are enriched in HD group compared to ND and PD groups (**Figures 7E,F**). Bacteroidia and Verrucomicrobia are enriched in PD group compared to ND and HD groups (**Figures 7E,F**). Interestingly we have identified (with databank


TABLE 3 | Associations from linear regression models between gut microbiota and alterations of adaptive immune system parameters measured in the adipose tissue.

#### TABLE 3 | Continued


In each of the models, the associations are reported as absolute difference (β), with 95% Confidence Intervals (CI), in immune system parameters levels by 1 unit increase in gut microbiota. In bold are reported the statistically significant results (p < 0.05). Per each of the models, the Coefficient of determination (R<sup>2</sup> ) is also reported.

RDP v11.4) an increase in Akkermansia muciniphila and Bacteroides dorei in PD groups compared to HD and ND groups (**Figure 6C**). Therefore, the most striking result of our metagenomic analyses in gut microbiota composition between the MSL and MHO mimicking diets (PD and HD, respectively) is the preponderance of Bacteroidia and Verrucomicrobia in PD compared to HD and control ND.

#### Association of Gut Microbiota Profile with Adaptive Immune Factors

We then sought to explore correlation between changes in gut microbiota composition with the over-responses of the adaptive immune system in mice, irrespective of the diet administered, using linear regression. The associations from regression analyses between gut microbiota classes and cytokines or immune cell types are shown in full in **Tables 1**–**5**. Some statistically significant associations were observed in each of the organs analyzed (aorta, heart, adipose tissue, liver, and spleen). For example (1) a linear increase in Bacteroidia and a decrease in Clostridiae were associated to an increase in IL-17, IFN-γ, IL-4, and CD8+ cells in the aorta. A decrease in Mollicutes and an increase in Verrucomicrobia was associated to increased infiltration of leukocytes (CD45+) as well as CD4+ and in CD8+ T cells and to an increase in IL-17, in the aorta (**Table 1**); (2) A linear increase in Bacteroidia and a decrease in Clostridiae was associated to an increase in IFN-γ, IL-6, lymphocytes, and CD4+ cells in the heart. An increase in Verrucomicrobia was associated to increased CD45+ cells and lymphocytes and to an increase in IL-1a, IFN-γ, and in CD4+ cells in the heart (**Table 2**). (3) A linear decrease in Actinobacteria and in Betaproteobacteria was associate to an increase in IL-6 and/or in IL-12 in the adipose tissue (**Table 3**). An increase in Verrucomicrobia was associated to increased IL-17 in the adipose tissue (**Table 3**). (4)


TABLE 4 | Associations from linear regression models between gut microbiota and alterations of adaptive immune system parameters measured in the liver.

(Continued)


In each of the models, the associations are reported as absolute difference (β), with 95% Confidence Intervals (CI), in immune system parameters levels by 1 unit increase in gut microbiota. In bold are reported the statistically significant results (p < 0.05). Per each of the models, the Coefficient of determination (R<sup>2</sup> ) is also reported.

A linear increase in Bacteroidia and a decrease in Clostridiae were associated to increased myeloid cells and CD11b+ cells in the liver. Also a decrease in Mollicutes and an increase in Verrucomicrobia was associated to increased myeloid cells in the liver (**Table 4**). (5) A linear increase in Bacteroidia and in Verrucomicrobia, and a decrease in Clostridiae, were associated to increased myeloid cell markers CD11b+, CD11c+, F4/90+, and Ly6G+ cells in the spleen (**Table 5**).

#### DISCUSSION

The results of our study in mice suggest that diet composition might have a pervasive role in co-regulating adaptive immunity and gut microbiota's profile in healthy obese subjects and in atherogenesis/inflammation in subjects with normal BMI. There has been recently a great focus on a particular subset of overweight and obese individuals having normal metabolic profile despite highly increased adipose mass (MHO = metabolic healthy obese; Karelis, 2008; Flegal et al., 2013). Individuals with adverse metabolic status despite a normal BMI have also been described (MSL = metabolic syndrome lean; Karelis, 2008; Flegal et al., 2013). It is currently unclear whether metabolic dysfunctions affects the higher morbidity and mortality observed in individuals with higher BMI: the concept of "benign obesity" has been challenged by some meta-analyses (Kramer et al., 2013) but not by others (Dhana et al., 2016), suggesting that MetS and not elevated BMI is an unequivocal risk factor for cardiovascular diseases (CVD). As it was previously reported, we confirmed that mice in a C57/BL6 genetic background fed a Paigen diet (PD) developed features of MetS, including hyperinsulinemia, hyperglycaemia, steatohepatitis, and inflammatory infiltration into the aorta, without increase in body weight (Getz and Reardon, 2006). Although, atherosclerosis is not observed without ApoE−/<sup>−</sup> mutation in mice, this study reports for the first time a systemic activation of the immune system upon an atherogenic diet, with high tissue infiltration of myeloid cell subsets CD45+CD11b+CD11c, CD45+ F4/80+, CD45+ CD11b+Ly6G+ in the spleen, a massive lymphocyte infiltration in the aorta and in the liver, a significant increase in CD4+ and CD8+ positive cells in the aorta and in the hearts, paralleled by increased IL-17, IL-1, IFN-γ, and IL-4 levels in the aorta and heart. The elevated level of gene expression for IL-1α detected in the heart might indicate the activation of myeloid cell types, probably macrophages, which have been described as the main proinflammatory cell population in atherosclerotic plaques (Jonasson et al., 1986), and also play a crucial role in the development of heart failure (Heidt et al., 2014). In general, IL-1 critically orchestrates the inflammatory events that are considered building blocks for the formation atherosclerotic plaques, precursors and risk factor for CVD such as myocardial infarction (Van Tassell et al., 2013; Gallego-Colon et al., 2015; Taleb et al., 2015). In this study, mice in a C57/BL6 genetic background fed a high fat diet (HD) developed obesity, increased body weight and fatty liver without systemic inflammation and activation of the adaptive immune system. We took advantage of these two phenotypically characterized mice models of MSL and MHO, to scrutinize the composition of gut microbiota in the stool of these metabolic exceptions. High-throughput 16S targeted sequencing showed a dominion of Bacteroidia, Deltaproteobacteria and Verrucomicrobia and under-representation of Clostridia in MSL PD-fed mice. Generally, Proteobacteria and Verrucomicrobia are not abundant in the healthy gut, but abundant in the gut dysbiosis of patients with type 2 diabetes or with inflammatory bowel disease (IBD) (Larsen et al., 2010; Qin et al., 2012; Shin et al., 2015). More abundant bacteria such as Bacteroidiales and Clostridiales are more and less represented, respectively, in type 2 diabetes compared to obesity (Larsen et al., 2010; Qin et al., 2012). Bacteroidiales are associated to weight loss (Million et al., 2013). Our results are completely in line with the above reports


TABLE 5 | Associations from linear regression models between gut microbiota and alterations of adaptive immune system parameters measured in the spleen.

(Continued)


In each of the models, the associations are reported as absolute difference (β), with 95% Confidence Intervals (CI), in immune system parameters levels by 1 unit increase in gut microbiota. In bold are reported the statistically significant results (p < 0.05). Per each of the models, the Coefficient of determination (R<sup>2</sup> ) is also reported.

and with the strong pro-inflammatory and pro-MetS role of PD compared to ND and HD. To our knowledge, a limited number of studies described clearly a role for gut microbiota in the onset of the MHO phenotype. It has been shown that separate cohorts of mice belonging to the same genetic background (C57/BL6) became either diabetic or resistant to diabetes and related metabolic dysfunctions despite being eating the same high-fat diet triggering obesity (Serino et al., 2012). The gut microbiota of the diabetes-resistant mice displayed a 20% decrease in the abundance of Firmicutes that were replaced by a parallel increase in Bacteriodetes (Serino et al., 2012). Moreover, the microbiota of diabetes-resistant mice presented with less bacteria of the helicobacter genus compared to the diabetic mice; instead, actinobacteria levels were unchanged (Serino et al., 2012). Our and these published studies suggest that the gut microbiota might reflect faithfully the metabolic phenotype irrespective of variability in the genetic background and diets of the host. Results of a recent study performed in the brown bear (Ursus arctos) are consistent with this (Sommer et al., 2016). The bear is a mammal accumulating enormous quantities of adipose fat in a seasonal manner (summer); by doing so, bears develop hyperlipidemia while maintaining metabolic health and being resistant to the development of atherosclerosis (Arinell et al., 2012). In fact, during summer season, the gut of bears harbored a different composition of microbiota than during winter season. In summer it was shown that gut microbiota was richer in Actinobacteria, Firmicutes, Proteobacteria, and poorer in Bacteroidetes (Sommer et al., 2016).

Several groups have provided data supporting a role for gut microbiota in the establishment of the MSL. If intestinal microbiota is suppressed in atherosclerosis-prone mice, an inhibition of dietary-choline-dependent atherosclerosis is observed (Wang et al., 2011). Patients with symptomatic atherosclerosis and normal body weight showed enrichment of the genus Collinsella of Actinobacteria in the gut (Karlsson et al., 2012). Generally, gut microbiota can affect atherosclerosis even in absence obesity or high fat feeding by different pathways: (i) infection activating the immune system and causing an inflammatory and proatherogenic response at distant sites; (ii) alteration of the levels of serum triglycerides and cholesterol, and of the metabolism of bile acids; (iii) dietary components (such as choline) and microbial metabolites [such as trimethylamine N-oxide (TMAO) generated from microbial metabolism of phosphatidylcholine, which is common in red meat and shellfish] lead to the production of both beneficial and harmful molecules (Jonsson and Backhed, 2017). For this reasons, gut microbiome is sometimes described as "endocrine" organ contributing to organism homeostasis [47].

Here, observed pro-inflammatory cytokines levels and tissue infiltrates correlating with decreased numbers of Clostridia corroborate previous findings on their regulatory functions. Clostridia strains presented in colon environment synergise to induce Tregs development [48, 49]. Tregs are fundamental to maintain mucosal homeostasis; therefore their insufficient development has pathological potential. Atherosclerosis develops upon stimulation of dendritic cells with oxidized low-density lipoproteins, the pathology is orchestrated by Th17 produced IL-17 [50], pro-autoimmune role of Th17 in atherosclerosis as well as association to HD induced chronic inflammation is welldescribed. Bacteroidetes has been associated to "healthy" nonobese homeostatic microbiome and with immunomodulation [51]. Interestingly addition of short chain fatty acids to the diet can result in Bacteroidetes abundance also during HD [52]. Bacteroides fragilis polysaccharide A (PSA) promotes T cells development [53], furthermore dysbiosis is frequently described as reduced Firmicutes/Bacteroidetes ratio, and this change was associated with IL-17 production and Th17 responses [54, 55].

Our result shows that increase proportion of Verrucomicrobiae correlates with higher percentage of myeloid cells markers. Interestingly, Roopchand et al. has shown that presence of A. muciniphila from Verrucomicrobiae together with presence of Bacteroidetes has significant role in protection to diet-induced obesity and metabolic dysbiosis in mice fed with HD [56]. Furthermore, in support to our data Ganesh et al. showed that presence of A. muciniphila increased levels of IL-17 in Salmonella infected mice [57].

It has been already demonstrated that diet shapes gut microbiome composition (De Filippo et al., 2010), and it is also now recognized that commensal microorganisms impact host gene expression not only in the gastrointestinal tract but also in other systems (Levy et al., 2015). Moreover, microbial cell components and secreted intermediate metabolites appear to be implicated in the response of the host to microbial colonization at the level of gene expression, which in turn reciprocally influence the disease progression. Noteworthy, both immunosuppressive drugs and probiotics affect the balance between microbiota and the immune system (Bartman et al., 2015). Particularly, probiotics supplementation have been shown to be effective in restoring and/or renovating the microbiota changes stimulating a number of health benefits, nevertheless whether modulation of gastrointestinal microbiota composition could have an effect on the amelioration of metabolic syndrome in obese or lean subjects, remains to be further investigated.

Supporting the link between metagenomics and immunogenomics, our data underline that understanding the reciprocal cross-talk between host immunity and microbiota

#### REFERENCES


will pave the way to the development of new therapeutic strategies against microbiome-driven common diseases, such as the metabolic syndrome.

#### AUTHOR CONTRIBUTIONS

All authors made substantial contributions to the conception of the work; to the acquisition, analysis, or interpretation of data for the work; to the drafting the work and revising it critically for important intellectual content; and they finally approved the version to be published. All authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of the work.

#### FUNDING

This work was supported by the European Social Fund and European Regional Development Fund—Project MAGNET (No. CZ.02.1.01/0.0/0.0/15\_003/0000492) to MV, and by "Italian Ministry of Health" from grant RC1603GA31 (to VP).

#### ACKNOWLEDGMENTS

We thank all the members of CTM-ICRC, St'Agata, and the technical personnel of the University of Veterinary and Pharmaceutical Sciences (Brno, Czech Republic) for help and assistance.


in steady-state and after myocardial infarction. Circ. Res. 115, 284–295. doi: 10.1161/CIRCRESAHA.115.303567


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer PCS declared a shared affiliation, though no other collaboration, with the authors CS and MV to the handling Editor, who ensured that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Pindjakova, Sartini, Lo Re, Rappa, Coupe, Lelouvier, Pazienza and Vinciguerra. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# **Investigation of the Cross-talk Mechanism in Caco-2 Cells during** *Clostridium difficile* **Infection through Genetic-and-Epigenetic Interspecies Networks: Big Data Mining and Genome-Wide Identification**

#### *Cheng-Wei Li† , Ming-He Su† and Bor-Sen Chen\**

*Laboratory of Control and Systems Biology, National Tsing Hua University, Hsinchu, Taiwan*

#### *Edited by:*

*Marina I. Arleevskaya, Kazan State Medical Academy, Russia*

#### *Reviewed by:*

*Alexander Oksche, Mundipharma Research, United Kingdom Alejandro Ramirez, Iowa State University, United States*

> *\*Correspondence: Bor-Sen Chen bschen@ee.nthu.edu.tw †Co-first authors.*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

> *Received: 06 April 2017 Accepted: 13 July 2017 Published: 02 August 2017*

#### *Citation:*

*Li C-W, Su M-H and Chen B-S (2017) Investigation of the Cross-talk Mechanism in Caco-2 Cells during Clostridium difficile Infection through Genetic-and-Epigenetic Interspecies Networks: Big Data Mining and Genome-Wide Identification. Front. Immunol. 8:901. doi: 10.3389/fimmu.2017.00901* *Clostridium difficile* is the leading cause of nosocomial antibiotic-associated diarrhea and the major etiologic agent of pseudomembranous colitis. In severe cases, *C. difficile* infection (CDI) can cause toxic megacolon, intestinal perforation, and death. The intestinal epithelium is the first tissue encountered in the adhesion and colonization of *C. difficile*, and serves as a physical defense barrier against infection. Despite the well-characterized cytotoxicity, few studies have investigated the genome-wide interplay between host cells and *C. difficile*. The aim of this study is to investigate the genetic-and-epigenetic molecular mechanisms between human colorectal epithelial Caco-2 cells and *C. difficile* during the early (0–60 min) and late stages (30–120 min) of infection. To investigate the cross-talk mechanisms during the progression of infection, we introduced a systems biology approach using big data mining, dynamic network modeling, a genome-wide data identification method, system order detection scheme, and principal network projection method (PNP). We focused on the construction of genome-wide genetic-andepigenetic interspecies networks (GEINs) and subsequent extraction of host–pathogen core networks (HPNs) to investigate the progression of underlying host/pathogen geneticand-epigenetic mechanisms from the early to late stages of CDI. Based on our results, we suggest that the cell-wall proteins CD2787 and CD0237, which both play an important role in cell adhesion and pathogen defense mechanisms, can be considered as potential drug targets. In addition, the crucial proteins employed by *C. difficile* for sporulation, including CD1214, CD2629, and CD2643, can also be considered as potential drug targets since spore-mediated re-infection is a critical issue.

**Keywords:** *Clostridium difficile***, Caco-2 cells,** *Clostridium difficile* **infection, genetic-and-epigenetic interspecies network, cross-talk molecular mechanism, reactive oxygen species, endoplasmic reticulum stress**

## **INTRODUCTION**

*Clostridium difficile* (*C. difficile*) is characterized as the major infectious cause of antibiotic-associated diarrhea and is the etiologic agent of pseudomembranous colitis. It was first identified by Hall and O'Toole in 1935, but no further studies linked the bacterium to human disease until 1978 (1, 2). *C. difficile* infection (CDI) usually occurs after a disturbance of the normal gut microbiome following antibiotic treatment. After the disruption of the microbiota, *C. difficile* can colonize to intestinal epithelial cells and produce pathogenic factors to breach the barrier. The major toxins of *C. difficile* are enterotoxin CD0663 (TcdA) and cytotoxin CD0660 (TcdB). Both toxins enter host cells *via* receptor-mediated endocytosis and are cytotoxic to host tissue by inactivating small Rho GTPases (RAC1, RHOA, and CDC42) (3). The glucosylationdependent inactivation of Rho GTPases results in actin cytoskeleton depolymerization and tight junction breakdown. However, *C. difficile* has not been fully investigated due to difficulties in its genetic manipulation, which makes it hard to generate isogenic strains for further study. In a previous study, the glucosylation of RHOA was shown to achieve saturation at 60 min postinfection, and all GTPases (CDC42, RAC1, and RHOA) also lose enzyme activity at this time point (4, 5). In addition, the MTT-dependent cell viability assays presented in our data source study (6) reveal that significant cell death also occurs at 60 min postinfection. Based on these observations, we define the early (0–60 min) and later stages (30–120 min) of host cells in CDI to investigate the progression of molecular mechanisms between two species. The 30-min overlap allows us to observe the causality and coherence of cross-talk molecular mechanisms.

Another mechanism regulating colonic gene expression is the microRNA (miRNA) system. miRNAs are small non-coding RNA molecules (~22 nucleotides) that bind to mRNA *via* complementary base pairing, resulting in mRNA silencing in human cells. Interestingly, a recent study indicates that the host utilizes miRNA silencing to shape the gut microbiota, including the Clostridium genus (7), suggesting that miRNAs play a crucial role in not only host gene repression but also microbiota shaping.

Unlike miRNAs, long non-coding RNAs (lncRNAs) are too large (~200–1,000 nucleotides) to pass through the bacterium cell wall and cell membrane for pathogen-gene regulation. In human cells, lncRNAs participate in gene regulation in a similar but more complex manner than miRNA (8), controlling various cellular responses. Furthermore, other epigenetic regulations, such as DNA methylation and histone modification, confer rapid and strong cellular responses to bacterial invasion. In order to investigate the progression of host–pathogen cross-talk mechanisms, as well as how these epigenetic activities contribute to progression during CDI, we identified the genome-wide geneticand-epigenetic interspecies networks (GEINs) in both the host and pathogen during the early and late stages of CDI. We then extracted the host–pathogen core networks (HPNs) from the GEINs to investigate the core pathways involved in the cellular responses of the host and pathogen during the early and late stages of CDI. In addition, we also discussed the offensive and defense mechanisms employed by the host and pathogen, respectively.

#### **RESULTS**

## **The Identified GEINs at the Early and Late Stage of CDI**

By applying a system identification method and system order detection scheme to two-sided microarray data (see materials and methods, supplementary methods and the flowchart in **Figure 1**), we identified the early-stage and late-stage GEINs of three biological replicates using the network visualizing software Cytoscape (9) (**Figure 2**). The numbers of identified nodes and edges are also shown in **Tables 1** and **2**, respectively. Among all three replicates, the node number of host transcription factors (TFs) at the early stage is higher than that at the late stage. In addition, the identified edges in **Table 2** show significant differences in host-TFs to host-genes regulation between the two stages. These results reveal that the activities of host TFs are more abundant during the early stage.

To further characterize genes in Caco-2 cells according to their functional groups, we performed an enrichment analysis *via* the function annotation tool DAVID (10) on the conserved target genes among all three replicates based on the biological process categories of GO database and the protein information resources of the Swiss-Prot database (**Table 3**). The early stage of CDI was characterized by the disturbance of cell shape and epithelial cell barrier, as well as immune activation and metal binding, which plays an important role in the scramble for metallic nutrients between the host and pathogen. At the late stage, the results including inflammatory-related functions and molecule secretion/transport suggests that a strong inflammatory response is triggered to eliminate the pathogen.

Since GEINs are very complex, it is difficult to investigate the precise host–pathogen interaction process from these networks directly. We, therefore, performed the principal network projection method (PNP) method to extract the core nodes with high projection values, which compose the corresponding HPNs from early-stage and late-stage GEINs of Caco-2 cells during CDI.

#### **The HPNs during the Infection of** *C. difficile* Construction of HPNs to Investigate the Epigenetic Activities in Host Core Networks of CDI

Applying the PNP method to GEINs in **Figure 2**, host/pathogen proteins with top 2,000 projection values based on intraspecies ranking in all three replicates and their connected genes/miRNAs/lncRNAs/complex were selected as core nodes of GEINs of each stage. Since the identified GEINs in **Figure 2** belong to three biological replicates from the same cell line, the identified differential interactions and regulations can be viewed as the adaptability of cells while facing stress and stimulus at different replicates. For more complete information, the combinations of these interactions/regulations in three replicates are considered real GEINs in the early and late stages as shown in Figures S1 and S2 in Supplementary Material, respectively. Next, we extracted core nodes from the real GEINs in Figures S1 and S2 in Supplementary Material using the PNP method to consist HPNs as shown in Figures S3 and S4 in Supplementary Material at the early and the late stages, respectively.

In order to adapt to CDI, some post-translation epigenetic modifications in host cells can also be found in HPNs. These epigenetic modifications can be detected by the basal level κ *H i* in the host protein expression dynamic Eq. 1 in the Section "MATE-RIALS AND METHODS." During the early stage of CDI (Figure S3 in Supplementary Material), host MAPK pathway members (UBA52 and HSPA5) can be regulated by the deacetylase protein (HDAC11) and the ubiquitin protein (UBE2D3). In addition, the

**FIGURE 1** | Flowchart of the systems biology approach used to construct genetic-and-epigenetic interspecies networks (GEINs) for host–pathogen core networks (HPNs) and to investigate the cross-talk mechanisms during *C. difficile* infection (CDI) for drug targets. The blue gray blocks represent the external information utilized in this study, including the big data mining for constructing candidate genetic-and-epigenetic interspecies network (GEIN), microarray data identification, and the surveyed literature for drug design; the rounded rectangular blocks denote the schemes and methods utilized to construct the candidate GEIN and real cross-talk GEINs at the early and late stages of CDI, and then extract the HPNs of each stage; and the light yellow blocks are the identified real GEINs and the cross-talk HPNs at the early and late stages of infection, as well as the inferred drug targets.

interaction; the red lines denote the transcriptional regulation; and the green lines signify the microRNA (miRNA) repression.

host proteins (DUSP6 and EGFR) involved in the interleukin-3, -5 signaling pathway, can also be regulated by the methyltransferase protein (PRDM14), the deubiquitinase protein (OTUB1), and the ubiquitin protein (UCHL5). These activities of the MAPK pathway and interleukin-related pathway participate in the immune response of host cells in response to the invasive bacterium. This finding can be supported by the functional enrichment analysis of early stage GEIN in **Table 3**. Furthermore, in Figure S3 in Supplementary Material, the small GTPase (CDC42) and downstream effectors (GRB2 and KBTBD7) that participate in the GTPase

**TABLE 1** | The number of identified nodes in the early stage and late stage of each replicate.


*E\_R1 and L\_R1 represent the early stage and late stage of replicate 1, respectively. Similarly, E\_R2 and L\_R2 denote the early stage and late stage of replicate 2, and E\_R3 and L\_R3 are the early stage and late stage of replicate 3. HP denotes host cytosolic protein [excluding host receptor and host transcription factor (TF)]; HR, HT, HM, HL, and HC represent host receptor, host TF, host microRNA, host long non-coding RNA and host complex, respectively; PP means pathogen protein excluding pathogen TF; and PT denotes pathogen TF.*

**TABLE 2** | The number of identified edges in the early stage and late stage of each replicate.


*The arrow lines in first column represent transcriptional/post-transcriptional regulations; the solid lines in the first column denote protein–protein interactions; HG signify host genes; and PG are pathogen genes.*

signaling pathway could be regulated by the deacetylase protein (HDAC4) and ubiquitin proteins (UBA52, USP43). In addition to post-translation modification, our results show that host heat shock protein (HSP) HSPA5 is DNA methylated in the HPN at the early stage of CDI. Considering that host–pathogen interspecies interactions play a central role in bacterial invasion, we then aim to investigate the cross talk between *C. difficile* proteins and host plasma membrane proteins.

#### Cross-talk Networks among Host–Pathogen Interactions and Their Validations

The cross talk between host and pathogen has been extensively investigated. However, the epigenetic modulation and interspecies protein–protein interactions of CDI are still largely unknown. To further investigate the offensive and defense mechanisms between the host and pathogen, we rearranged the HPNs in Figures S3 and S4 in Supplementary Material from the perspective of signal transduction pathways. The rearranged core signal transduction **TABLE 3** | The functional enrichment analysis of the conserved target genes among three replicates based on GO terms and protein information resources of Swiss-Prot.


*The early stage of C. difficile infection is characterized by the change of cell shape and tight junction and this could result from the activities of GTPases and pathogen toxins. The metal-binding ability is crucial for both host and pathogen cells due to its important role in the scramble of metallic nutrients and the transport of toxic molecule, including ROS. Signaling in the case of the early stage of infection was increased for immune response, while signaling in the case of the late stage of infection was increased for the abundant cellular processes, including macrophage activation, phagocytosis, and inflammatory response. The cofactor transport and secretion-related process also contribute to the cytokine production and secretion.*

pathways of HPNs in Figures S3 and S4 in Supplementary Material at the early and late stage of CDI are shown in **Figures 3** and **4**, respectively.

For the validation of our identified host–pathogen protein– protein interactions in **Figures 3** and **4**, we surveyed the existing literature for studies reporting recognized host–pathogen interactions during CDI. The interaction [CD2787, fibronectin 1 (FN1)] in both **Figures 3** and **4** has been experimentally verified (11). CD2787 is involved in cell adhesion by binding to FN1, and degrading host extracellular matrix proteins. Meanwhile, the interaction (CD0660, HSP90B1) in **Figure 3** and (CD0660, HSP90B2P), (CD0663, HSP90B1), and (CD0663, HSP90B2P) in **Figure 4** can be verified by Chaves-Olarte et al. (12) and Na et al. (13), which revealed that *C. difficile* toxins can enter host cells through GP96. Similarly, the suggested cytotoxic interactions (CD0660, CDC42), (CD0663, CDC42), (CD0660, RHOA), and (CD0663, RHOA) in **Figure 3**, and (CD0663, RAC1) in **Figure 4** can also be verified in Just et al. (4, 14) and Chen et al. (15).

While in the genetic-and-epigenetic core pathways at the late stage of CDI shown in **Figure 4**, a new pathogen protein CD1466 was secreted out by *C. difficile*. CD1466 encoding an ATP-binding protein belongs to the ABC-type transport system, which is the same as CD0478 in the early stage. The results show that an ATPbinding cassette transporter CD1466 interacts with host receptors SNW1 and EGFR, causing various signal cascades in the host cell. The interaction between ATP-binding cassette transporters and SNW1 has been reported by mass spectrometry (16). In addition, the suggested interaction between CD1466 and EGFR during CDI could provide *C. difficile* with a fast and efficient way to change transporter function (17). SNW1 in **Figure 4** was observed to enter the cytoplasm and then interact with TFs *via* acetylation to induce immunity-related processes. In addition,

The upper clump represents the pathogen core pathways and the lower clump signifies the host core pathways at the early stage of CDI. The gray solid lines denote the protein–protein interaction; the red arrow lines are transcriptional regulation; the green dot lines signify protein translation; the gray blue dash lines represent protein secretion; and the purple clump with arrow and dash lines indicate the activity of reactive oxygen species (ROS). The pathogenic factors (CD0660, CD0663, and CD0478) of *C. difficile* trigger ROS production and dysfunction of protein folding of Caco-2 cells. Therefore, host cells employ autophagy and DNA damage response to remove induced cellular injuries and activate the immune response to eliminate pathogen. In response, *C. difficile* generate various antioxidative proteins to counteract the host-produced ROS.

EGFR is participating in several cellular responses, including GTPase activity, immune response, inflammation, and apoptosis. In **Figure 4**, the downstream TFs of EGFR-mediated pathways include NFIC and NFKB1, involved in inflammatory response. The inflammatory response triggered by CD1466 could be used by host cells to remove invasive pathogens. However, severe inflammation could also induce tissue damage of host cells. The tradeoff of inflammation in CDI will be discussed in later sections. In addition to CD1466, CD0663 and CD0660 are also secreted in the late stage of CDI (**Figure 4**). The result suggests that CD0663 can bind to CD46, HSP90B1, HSP90B2P, and RAC1 (**Figure 4**). The suggested interaction between CD0663 and RAC1 in the result (**Figure 4**) can be supported by the previous report (14). By mass spectrometry, the interaction between CD0663 and HSP90B1 has also been reported (18). Therefore, according to sequence-homology and the result in **Figure 4**, we suggested that

**FIGURE 4** | Core pathways rearranged from the host–pathogen core network in Figure S4 in Supplementary Material at the late stage of *C. difficile* infection (CDI). The upper clump represents the pathogen core pathways, and the lower clump signifies the host core pathways at the late stage of CDI. The gray solid lines denote the protein–protein interaction; the red arrow lines are transcriptional regulation; the green dot lines signify protein translation; the gray blue dash lines represent protein secretion; and the purple clump with arrow and dash lines indicate the activity of reactive oxygen species (ROS). The abundant activities of CD0663 and the acetylation of fibronectin 1 (FN1) result in enhanced ROS production and a strong inflammatory response, while these counter mechanisms in turn increase the cellular stress of host cells. The endoplasmic reticulum (ER) stress response reflects that the accumulated cellular stress could risk the host cell. Therefore, the tissue damage caused by severe inflammation and the accumulated cellular stress eventually triggers the apoptosis process of host cells. In addition, *C. difficile* utilize DNA damage response and antioxidative proteins against human-produced ROS, and reduce the toxins production and cell growth rate for sporulation to transform to endospore.

the interspecies PPIs between pathogen CD0663 and host CD46 and HSP90B2P could occur during CDI.

## **Drug Targets Prediction for Treating CDI**

Considering the crucial role of CD2787 (cwp84) in pathogenesis at the early stage of CDI, we recommend CD2787 as a potential drug target in the prevention of CDI. Moreover, other cell-wall proteins participating in the infectious process, such as CD0440, CD0237, CD0266, and CD1987 are also potential drug targets since the inhibition of these cell surface proteins may reduce the occurrence of cell adhesion. In **Figures 3** and **4**, four pathogen genes (*CD0130*, *CD3256*, *CD1275*, and *CD2781*) were previously predicted as essential survival/growth genes by applying flux balance analysis (FBA) and synthetic accessibility (SA) to a curated *C. difficile* metabolic network, as well as transposon-directed insertion site sequencing (TraDIS) to the *C. difficile* transposon mutant library (19, 20). However, CD0130 and CD3256 have important humanhomologs, MAT2A and VARS, respectively (19). The inhibition of these proteins may result in unpredictable dysfunction of host cells. Therefore, CD1275 and CD2781 are more conservative choices for further drug design. Similarly, proteins involved in the defense mechanisms employed by *C. difficile* against reactive oxygen species (ROS), including CD0141 and CD2115 in the early stage of CDI, and proteins responsible for redox-state homeostasis (CD1690 and CD1631) in the late stage of CDI, have humanhomologs (19). Therefore, the three remaining anti-ROS enzymes (CD2356, CD0171, and CD0179) are recommended as potential targets to repress the defense mechanisms of *C. difficile*. The heavy economic burden of CDI results from its high recurrence rate, suggesting that the inhibition of spore formation of *C. difficile* is a feasible therapeutic way to reduce the recurrence rate of CDI. In our results, the key sporulation pathway members included CD1214, CD2629, and CD2643. These proteins could become targets of further drug design.

In addition to the inhibition of pathogen mechanisms, another strategy of drug design includes the promotion of host defense mechanisms. It has been reported that low doses (0.2 ng/ml) of CD0660 is sufficient to induce significant cell rounding *in vitro* (21), suggesting that amplification of host defense appear to be required. In our results, the major pathogenic effects induced by *C. difficile* pathogenic factors include tight junction/cytoskeleton breakdown and the accumulation of endoplasmic reticulum (ER) stress, which result from the inactivation of Rho GTPases (RHOA, CDC42, and RAC1) and the dysfunction of chaperone proteins (HSP90B1, HSPA5, and HSP90B2P), respectively. Therefore, we aim to increase the expression of these genes to maintain the activities of these dysfunction components. The severe inflammation and apoptosis process, induced by NFKB1, REL, and IL-8, are responsible for cell death at the late stage of CDI. Though the activities of immune response and inflammation are important counter mechanisms against bacteria, the overexpression of these related genes cause tissue damage and apoptosis of host cells. Therefore, we suggest repressing the expression of NFKB1, REL, and IL-8 to relieve these activities.

Our results show that CD2356, CD0171, and CD0179 participate in the defense mechanisms of *C. difficile* against oxidative stress. The co-operation among these proteins provides a well-designed protection against human-produced ROS. The inhibition of these antioxidative proteins facilitates the host eliminating ability against pathogens, and the scattered ROS can induce the rapid necrosis of pathogen cells. Therefore, we recommend that CD2356, CD0171, CD1064, and CD0179 are potential drug targets for further drug design (see supplementary methods and Table S2 in Supplementary Material and Figure S5 in Supplementary Material).

## **DISCUSSION AND CONCLUSION**

## **Comparison of the Pathogen Core Network with Previously Proposed Essential Genes or Proteins in CDI**

During the pathogenesis of the bacterium, essential genes are required for the survival and growth of the pathogen. The absence of essential genes results in a decreased growth rate or death of the organism. The pathogen core networks extracted from GEINs display not only essential proteins required for the survival of the pathogen but also important enzymes that contribute to the pathogenesis and defense mechanism of the pathogen against various stress conditions. Here, we compare HPNs with the existing predicted essential proteins of *C. difficile* to show the difference and advantage of HPNs.

Recently, by applying FBA and SA to a curated *C. difficile* metabolic network, Larocque et al. predicted 76 essential *C. difficile* genes (19). Seven *C. difficile* proteins (CD2664, CD2335, CD3550, CD0198, CD1225, CD0130, and CD0123) in the HPN of the early stage of CDI (Figure S3 in Supplementary Material) and three *C. difficile* proteins (CD2588, CD1816, and CD0130) in the HPN of the late stage of CDI (Figure S4 in Supplementary Material), encoded by *C. difficile* genes appear to be required for the survival of the pathogen, have been identified in this study. Moreover, applying transposon-directed insertion site sequencing (TraDIS) to the *C. difficile* transposon mutant library led to the identification of 404 genes with no transposon insertion in the library, which can be considered as essential genes for *C. difficile* growth (20). Thirteen *C. difficile* proteins (CD2664, CD2335, CD0067, CD3550, CD3540, CD0198, CD1255, CD2714, CD3256, CD1316, CD0095, CD2739, and CD0123) in the HPN of the early stage of CDI (Figure S3 in Supplementary Material) and 15 *C. difficile* proteins (CD3170, CD2588, CD2744, CD2771, CD2462, CD3304, CD2781, CD3540, CD1145, CD2461, CD2793, CD0059, CD1275, CD0052, and CD1767) in the HPN of the late stage have been identified in this study as the products of likely essential genes for the growth of *C. difficile*. Overall, 15 pathogen genes in the HPN of the early stage of CDI, and 17 pathogen genes in the HPN of the late stage, were previously identified as likely essential genes of *C. difficile* (Table S1 in Supplementary Material).

Besides these previously identified essential proteins, HPNs also provide numerous crucial *C. difficile* enzymes that participate in the offensive and defensive mechanism utilized by the pathogen, such as well-known toxins (CD0660 and CD0663), toxin-regulators (CD0659, CD0661, and CD0664), cell-wall proteins (CD2787, CD1987, CD0237, and CD0440) that provide celladhesion abilities, defensive proteins (CD0141, CD0171, CD2115, CD1631, and CD1690) against ROS, and sporulation-related proteins (CD1214, CD2643, CD2629, and CD1511). The presence of these proteins in HPNs reveals that there are various crosstalk activities between the host and pathogen in the offensive and defensive mechanism.

## **The Induced Homeostatic, Apoptotic, Intestinal Inflammatory Responses in CDI Patients Mediated by Acetyltransferase and Methyltransferase Proteins**

The interactions between the small GTPase (CDC42) and downstream effectors (GRB2 and KBTBD7) suggest that the dynamic of cytoskeleton homeostasis can be affected by *C. difficile* toxins and host epigenetic activities to influence the change of cell shape displayed in **Table 3**. Similarly, at the late stage of CDI (Figure S4 in Supplementary Material), the subunits (NFKB1 and REL) of the NF-κB complex are regulated by acetyltransferase proteins (GCNT2 and B3GNT6) and the methyltransferase protein (PRDM14), resulting in the assembly of the NF-κB complex and the induction of apoptosis in host cells. The change in the methylation level of HSPA5 has been proposed in Hesson et al. (22) *via* NOME-Seq analysis of HSPA5 in intestinal disease, suggesting that altered nucleosome positioning induces differences in accessibility. In addition, in the late-stage HPN shown in Figure S4 in Supplementary Material, the significant difference in the basal level of the NF-κB subunit NFKB1 gene reflects the change of methylation level, which is consistent with a previous study showing that NFκB1 is hypomethylated in intestinal inflammation (23). These identified HPNs reveal the ability of acetylation and DNA methylation likely to alter cellular functions in order to adapt to bacterium infection.

## **Cross-talk Mechanisms in CDI Patients Mediated by Ubiquitin and Acetyltransferase Proteins**

Once the bacterium is attached to the surface of Caco-2 cells in the early stage of infection (**Figure 3**), the cell surface-associated cysteine protease CD2787 (cwp84) of bacterium interacts with FN1 on the host membrane. This negative interaction (CD2787, FN1) and the MIB2-induced ubiquitination of FN1 suggest that CD2787 could provide a degrading activity on FN1, which can be supported by the biological experiment (11). Since FN1 is involved in the maintenance of cell shape, it can bind cell surface and various compounds, including actin. We suggested that the degradation of FN1 results in the morphological change of Caco-2 cell or the disturbance of the actin cytoskeleton. The results also suggest that three pathogen proteins (CD0660, CD0663, and CD0478) secreted by *C. difficile* interact with four host receptors [FPR1, SCARA3, HSP90B1, and Arrestin Beta 2 (ARRB2)]. FPR1 has been reported as the receptor of CD0660 (TcdB) (24). Owing to its interaction with toxins (CD0660 and CD0663) and ubiquitination modified by ARIH2, we suggested that *C. difficile* toxins enter the host-cell cytoplasm *via* receptor-mediated endocytosis. For *C. difficile* toxins, another identified human cell surface receptor HSP90B1, which encodes a member of HSP 90 kDa family, plays a role in protein folding. It has been reported that the ER chaperones

HSP90B1 and HSPA5 could be significantly downregulated by HDAC inhibition at the protein level (25). HSP90B1 has been identified as a cell surface receptor for *C. difficile* proteins, but HSP90B1 did not respond to CD0660 (13). The observation can be supported by the result in the GEIN that HDAC inhibition impairs the binding affinity of HSP90B1 for CD0660.

The interaction between CD0663 and CD46 (**Figure 4**) indicates that pathogen CD0663 could be the extracellular stimuli of CD46, and potentially responsible for CDI to recruit phagocytic cells such as neutrophils to clear toxins and whole microbes in the early stage. The result in **Figure 4** suggests acetylation and interaction with toxins (CD0660 and CD0663) of HSP90B2P, which participate in impairing its chaperone-ability. Interestingly, we noticed that there are more host proteins influenced by CD0663 than CD0660 in the late stage of infection. By contrast, CD0660 binds more host proteins than CD0663 in the early stage (**Figure 3**). These results provide an explanation for the perennial argument about the cytotoxic responsibility of CD0660 (TcdB) and CD0663 (TcdA), suggesting that both toxins are likely essential for pathogenesis. The results suggested that this role change may result from CD2973-induced acetylation of CD0663 in the late stage (**Figure 4**). The rise of CD0663 activity could be the turning point for the progression of CDI. We observed that another host protein, FN1, could also change its role from the early stage to the late stage of CDI. FN1 is repressed by ubiquitination and CD2787-induced degradation in the early stage of CDI (**Figure 3**), but the CD46-triggered activation and NAT8Linduced acetylation of FN1 increased the corresponding expression levels (*p*-value *<sup>&</sup>lt;* <sup>8</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) despite the influence of CD2787, thus allowing FN1 to trigger downstream activities in the late stage of CDI in **Figure 4**. Therefore, we suggested that CD0660 functions prior to CD0663 and triggers rapid responses in the early stage of infection and CD0663 works actively in the late stage.

To further understand this, we separate the rearranged core pathways in **Figures 3** and **4** based on different cellular responses and their corresponding pathways for further discussion. For the early stage of CDI, the genetic-and-epigenetic scheme of the core pathways (**Figure 3**) can be separated into three parts: the offensive mechanism of the pathogen and the corresponding pathogenesis of host cells (**Figure 5A**); the remedial actions employed by host cells in response to pathogen-induced injuries (**Figure 5B**); and the counter mechanisms of Caco-2 cells and the defense mechanisms of *C. difficile* (**Figure 5C**). During the late stage of CDI, the core pathways (shown in **Figure 4**) can be separated into two parts: the strong cellular responses triggered by epigenetic acetylation in host cells for depleting the tenacious pathogen at the late stage of CDI (**Figure 6A**) and the severe inflammation and apoptosis processes of Caco-2 cells as well as the endospore formation of *C. difficile* (**Figure 6B**).

## **A Precise View of Pathogenic Effects and Host Responses at the Early Stage of** *C. difficile* **Invasion**

#### Pathogenic Factors Utilized by *C. difficile* and the Resulting Pathogenesis in Caco-2 Cells

During CDI, the first event that initiates pathogenesis is cell adhesion of the two organisms. We identified that CD2787 plays

#### **FIGURE 5** | Continued

The host/pathogen cross-talk mechanism through core pathways at the early stage of *C. difficile* infection (CDI). **(A)** The secreted pathogenic factors of *Clostridium difficile* (*C. difficile*) and their induced cytopathic effect in Caco-2 cells; **(B)** the remedial schemes employed by host cells in response to *C. difficile* toxins; **(C)** offensive mechanisms *via* host-secreted reactive oxygen species (ROS), microRNA (miRNAs), and immune response, and thus the defense mechanism of *C. difficile* at the early stage of CDI. The red arrow lines denote transcriptional regulation; the gray solid lines indicate the protein–protein interaction; the green dot lines signify protein translation; the gray blue dash lines indicate protein secretion; and the purple clump with arrow and dash lines represent the activity of ROS. In **(A)**, *C. difficile* toxins hijack a RAC1-related pathway to promote ROS production in host cells. The significant change of acetylation level of heat shock proteins (HSP90B1 and HSPA5) and their interactions with toxins impair their chaperone-activity, resulting in the formation of misfolded protein and the induction of autophagy. As the remedial schemes for these injuries, in **(B)**, Caco-2 cells employ the DNA damage response and autophagy to deplete the damage caused by pathogens, and innate immune response to remove the invasive bacterium. The alternative routes of these counter mechanisms and the miRNA silencing used by host cells to interfere with the activities of pathogen are also displayed in **(C)**. Furthermore, the scheme in **(C)** shows the multiple pathways of *C. difficile* against the oxidative stress presented by Caco-2 cells.

**FIGURE 6** | The host/pathogen cross-talk mechanism through core pathways at the late stage of *C. difficile* infection. **(A)** The enhanced reactive oxygen species (ROS) production and cellular stress of host cells, and the anti-ROS mechanism of *Clostridium difficile* (*C. difficile*); **(B)** the stress-induced apoptosis of Caco-2 cells and the leaving of *C. difficile*. The red arrow lines denote transcriptional regulation; the gray solid lines indicate the protein–protein interaction; the green dot lines signify protein translation; the gray blue dash lines indicate protein secretion; and the purple clump with arrow and dash lines represent the activity of ROS. The definitions of node symbols are the same as those defined in **Figure 5**. In **(A)**, the abundant activity of CD0663 and the acetylation of fibronectin 1 increase the production of ROS and trigger the formation and transport of cytokine, inducing a strong inflammatory response. However, the presence of endoplasmic reticulum (ER) stress response reflects that the accumulated cellular stresses also risk the survival of the host cell. Meanwhile, *C. difficile* employ DNA damage response and antioxidative proteins to counteract the oxidative stress. In **(B)**, the severe inflammation, accumulated oxidative stress and ER stress, and the activity of NF-κB complex trigger the apoptosis of Caco-2 cells. In addition, *C. difficile* actively transform to endospore to lie dormant and avoid the risking environment.

an important role in cell adhesion by binding FN1 (**Figure 5A**), resulting in the degradation of the extracellular matrix proteins of host cells (11). On the pathogen side, CD2787 also interacts with the cytotoxin CD0660 (TcdB) to promote the secretion of toxins, and signals *via* the toxin regulator CD0664 (TcdC) to enhance the production of CD0660 (**Figure 5A**). On the host side, the enterotoxin CD0663 (TcdA) and cytotoxin CD0660 (TcdB) are the two major *C. difficile* pathogenic factors and the main causes of clinical symptoms of CDI. In **Figure 5A**, the result suggests that CD0660 can activate CD0663 *via* CD1625, a histidine kinase. Another route for CD0660 to control toxin production is through interaction with the pathogen transcriptional regulator CD1064 (**Figure 5A**). This protein regulates a wide range of genes, including *CD0660* and *CD0663* (**Figure 5A**). Similarly, CD0663 (TcdA) can enhance toxin production *via* CD1064 and CD0664 (**Figure 5A**). It should be noted that the toxin activator, CD0659, regulates CD0664 (**Figure 5A**). Therefore, the results suggest that higher expressions of CD0660 and CD0663 in the early stage of CDI (CD0660: *<sup>p</sup>*-value *<sup>&</sup>lt;*<sup>5</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> , CD0663: *p*-value *<sup>&</sup>lt;*<sup>4</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) reflects the higher activities of toxins at this stage. Moreover, the result also suggested that CD2787 could activate an ATP-binding cassette transporter CD0478 through histidine kinase CD1352. CD0478 is involved in the antibiotic resistance of *C. difficile* (26). On the host side, we suggested that this protein, which interacts with human cell receptors, is responsible for the subsequent response of human cells in CDI, suggesting its potential immunogenicity.

A G protein-coupled receptor FPR1, which interacts with *C. difficile* toxins (**Figure 5A**), is involved in the activation of immunerelated phagocytic cells. On the host side, the result in **Figure 5A** suggests that FPR1 can negatively interact with its downstream protein PTEN, a tumor suppressor protein. PTEN inhibition has been reported to activate the small GTPase RAC1 (27), which leads to the decreased PTEN expression (*p*-value *<sup>&</sup>lt;*<sup>3</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and the increased expression of RAC1 (*p*-value *<sup>&</sup>lt;*<sup>2</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) during CDI infection. The activated RAC1 is characterized by its participation in assembly of the NADPH complex and, thus, the production of ROS in CD0660-infected Caco-2 cells (28). Interestingly, the results show that another glycosylation-independent pathway exists for RAC1 regulation of ROS production. The result shows that RAC1 activation of RPS3 mediates ETS1 translocation to the nucleus for transcription (**Figure 5A**). The SYK gene is regulated by ETS1 and is also involved in ROS production. The highly expressed SYK in the early stage of CDI (*p*-value *<sup>&</sup>lt;*<sup>3</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) reflects an induced oxidative stress caused by ROS. Oxidative stress reflects the imbalance of the redox-state in a biological system. The overproduction of ROS or insufficient reducing agents in the system could lead to oxidative stress. Oxidative stress is usually activated by the host to kill pathogen. However, the simultaneous production of ROS and free radicals causes severe damage of cellular components, including lipids, proteins, and DNA. In the case of CDI, toxins hijack the RAC1-dependent pathway to generate ROS, resulting in oxidative-stress-mediated necrosis of host cells.

Heat shock proteins are usually activated to assist with the folding and refolding of cellular proteins in response to environmental stress. The result suggests that the deacetylation of HSPA5 can switch its folding ability into a degrading ability (29). Moreover, UBE2D3-induced ubiquitination of HSPA5 results in its repression in the early stage of CDI (*p*-value *<sup>&</sup>lt;*<sup>2</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ). The result shows that *HSPA5* gene is also inhibited by the activated miR155HG and its DNA methylation in the early stage of CDI. Therefore, the result suggests that the chaperone-ability and degrading activity of HSPA5 are repressed in the early stage of CDI.

Furthermore, the result suggests that the deacetylation of HSPA5 can activate the expression of Serine/Threonine Kinase 11 (STK11) *via* GATA1. STK11 is responsible for autophagy activation in response to the aggregation of misfolded proteins. However, the activated miR7-3HG directly inhibits STK11, which results in dysregulated autophagy activation.

#### Caco-2 Cells Adopt Autophagy, DNA Damage Response, and the Activation of PAK1 and GRB2 As Remedial Schemes in Response to Pathogen-Induced Damage

SCARA3 and LRRK2, which encode a ROS scavenger protein and a leucine-rich repeat kinase, respectively, are employed to counteract environmental stresses, including overproduction and secretion of ROS, and to be responsible for induce autophagy initiation, respectively. The result suggests that LRRK2 can enhance the expression of YY1 by repressing ETS1 (**Figure 5B**). The activated LRRK2 and YY1 (*p*-value *<sup>&</sup>lt;*<sup>2</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> and *<sup>&</sup>lt;*<sup>1</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> , respectively) in the early stage of CDI also support this relationship. These identified regulatory interactions suggest that host cells recruit numerous proteins and genes to maintain the redoxstate balance and protein homeostasis.

Over the past decade, *C. difficile* toxins (CD0660 and CD0663) have been characterized by their capacity for glycosyltransferasedependent inactivation of host Rho family GTPases (RHOA, CDC42, and RAC1). It has been shown that these modifications result in cytoskeleton rearrangement, severe inflammation, and subsequent apoptosis in human cells (3). In **Figure 5B**, the result suggests that CDC42, which interacts with CD0660 and CD0663 and is modified by acetylation, impairs its association to the CDC42-activated kinase 1 (PAK1) and subsequent GRB2. The inhibited PAK1 and GRB2 (PAK1: *<sup>p</sup>*-value *<sup>&</sup>lt;*<sup>2</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> , GRB2: *<sup>p</sup>*-value *<sup>&</sup>lt;*<sup>4</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) also support this relationship. PAK1 and GRB2 are involved in activation of the immune response such as the interleukin-23 signaling pathway, which contributes to the regulations of cell adhesion, inflammation, and apoptosis in intestinal epithelial cells (30). Interestingly, the result suggests that PRRX2 functions as a double-edge sword (**Figure 5B**). The repression of PRRX2, resulting from CD2787-mediated degradation of FN1, not only decreases the efficiency of chaperone functions as previously mentioned but also unlocks the repression of *PAK1* and *GRB2* genes. This process can restore cytoskeleton homeostasis and initiate the interleukin-23 signaling pathway.

#### The Offensive Mechanisms of Caco-2 Cells and the Defense Mechanisms of *C. difficile* at the Early Stage of CDI

The evolutionary war between microbes and humans has lasted for thousands of years. In addition to the defense mechanisms developed against microbial infection, host cells also evolved strategies to kill pathogens. However, lipids and proteins from human cells cannot easily pass through bacterium cell walls and membranes. These offensive interactions exist but still remain largely unknown and generally species specific. The major means employed by host cells include the production of tiny and toxic molecules (ROS), recruitment of phagocytes (neutrophil, macrophages), and newly detected miRNA regulation (7). The production of ROS and immune response initiation has been discussed in previous sections. The alternative routes of these offensive processes are also shown in **Figure 5C**.

Through its interaction with FN1 and CD0478 (**Figure 5C**), ARRB2 can signal through the COP9 signalosome subunit COPS5 to positively activate GRB2. GRB2 can participate in cytoskeleton homeostasis and the innate immune response. The result in **Figure 5C** shows GRB2 activates ETS1 to upregulate *SYK* expression and, thus, ROS production. The co-operation of ROS and innate immune processes form an offensive mechanism to deplete pathogens. ARRB2 can also transmit stimulation signals to androgen receptor (AR) (**Figure 5C**). Once stimulated, AR dissociates from accessory proteins, translocates to the nucleus, and then induces the transcription of *PAK1* and *miR155HG* (**Figure 5C**). Both PAK1 and miR155HG are involved in the innate immune response. The feature of this pathway is the activation of miR155HG (**Figure 5C**). The high expression level of this miRNA increases the probability of passing through pathogen cell barriers (cell walls and cell membranes) to silence the pathogen gene *CD3256*. The relationship between host miRNAs and pathogen genes has recently been reported (7). This seminal observation revealed that the host can regulate and shape gut microbiota through miRNAs. The result suggests that miR155HG and miR1 can silence *CD3256* and *CD0130*, respectively (**Figure 5C**). CD3256 and CD0130 function in pathogen pathways that counteract host-produced ROS. This indicates that host cells can interfere with pathogen defense mechanisms and enhance the host's ability to eliminate *C. difficile*.

Oxidative stress is usually employed by host cells to kill pathogens, but*C. difficile* toxins enhance ROS production through a RAC1-mediated pathway, causing injury to the host and DNA damage, to avoid or defend against ROS (**Figure 5A**). When ROS is released at the interface between the host and pathogen, pathogen cell-wall proteins are the first to be encountered in CDI. In **Figure 5C**, on the pathogen side, the result suggests that the cell-wall proteins CD2787 (cwp84) and CD0440 (cwp27) can positively activate CD0812, which is a universal stress protein. Universal stress proteins can promote endurance under stress conditions, such as heat, nutrient starvation, chemical agents, and oxidative stress. The homoserine kinase CD2119, which participates in several essential metabolic pathways, is also employed by CD0440 to activate CD2356 (**Figure 5C**). CD2356 is a thioredoxin reductase that can remove superoxide radicals and balance the redox state of pathogens. A high expression level of CD2356 in this stage (*p*-value *<sup>&</sup>lt;*<sup>6</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>4</sup> ) could function to resolve oxidative stress. In addition to performing its own activity, the result also suggests that CD2356 can urge CD0130 to interact with CD0141 (**Figure 5C**). CD0130 encodes an S-adenosylmethionine synthase and has been identified as an essential survival gene of *C. difficile*, and the absence of this gene will result in gaps in metabolic networks and biomass decrease (19). Its metal ion binding ability can help CD0141 with exporting copper ions. Copper (Cu) is an essential element for most species, including bacteria. However, overexposure to Cu is toxic. In fact, copper and its alloys are natural antimicrobial materials that have long been used as bactericides before antibiotics were discovered. The toxicity of copper ion is primarily due to its reaction with human-produced H2O<sup>2</sup> (a member of ROS) to generate the hydroxyl radical (*·*OH), which damages pathogen lipids, proteins, and DNA. To reduce this risk, the copper homeostasis protein CD0141 is activated to export copper from pathogen (**Figure 5C**). Interestingly, we identified that *CD0130* is silenced by the host miRNA miR1 (**Figure 5C**). This regulation reduces the probability of CD0130- CD0141 co-operation, thus interfering with pathogen defense mechanisms. Unfortunately, the expression of CD0130 remains high in the early stage of CDI (*p*-value *<sup>&</sup>lt;*<sup>3</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ), and this may be due to the lack of miR1 upregulation.

CD0237 (FliD) encodes a flagellar protein that together with CD2787 participates in *C. difficile* adhesion. In **Figure 5C**, the result suggests that this cell surface protein responds to ROS by activating CD1185 and CD2753. CD1185 is a diguanylate kinase protein that participates in the formation of c-di-GMP, a ubiquitous second messenger involved in bacteria biofilm formation and pathogen aggregation (31). The aggregated pathogen cells also promote the formation of biofilm. CD2753 is also a c-di-GMP-signaling component that is specific to *C. difficile* 630. The assembly of these second messenger subunits induces various cellular responses. One downstream TF, CD2643, which regulates the sporulation process, is activated to transcriptionally regulate *CD1185*, creating a self-activation loop. CD1185 is methylated, but its high expression (*p*-value *<sup>&</sup>lt;*<sup>2</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) at the early stage and the establishment of the feedback loop reveals the high activity of c-di-GMP and, thus, the formation of biofilm, which is a powerful scheme against stress conditions, including oxidative stress. CD2115, a copper-transporting ATPase, is activated by CD2753 (**Figure 5C**), playing a similar role as CD0141. The existence of similar pathways modulating copper homeostasis indicates the important role of Cu ion in oxidative stress by highlighting the redundancy employed by pathogens to counteract oxidative stress.

Furthermore, in **Figure 5C**, the result suggests that CD0237 (FliD) can transmit signals *via* its downstream flagellum subunit CD0745, a putative chemotaxis protein, to trigger CD1128 mediated DNA replication. CD1128 is the DNA polymerase 1 (PolA) of *C. difficile*, and the initiation of DNA replication triggers bacterial reproduction. The increased amount of pathogen promotes the concentration of toxins and the resistance against stress conditions. CD1128 is methylated by CD2726 (**Figure 5C**), but its high expression level (*p*-value *<sup>&</sup>lt;*<sup>5</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and the positive interaction with CD0745 suggest that pathogen still recruits this protein for DNA replication. CD1128 then positively activates the transcriptional regulator CD1064 to regulate *CD2356* (**Figure 5C**). Another CD1128-mediated anti-ROS pathway includes CD3256 and CD0171. CD3256 (valS) is a valinetRNA ligase and has been predicted to be an essential gene for the growth of *C. difficile* (20). Its activation ability to CD0171 is important for pathogen since CD0171 encodes a key redoxsensing regulator. CD0171 is only active as a repressor when the intracellular NADH/NAD<sup>+</sup> ratio is low, thus regulating the redoxstate of the pathogen (32). Interestingly, the result suggests that the host could interfere with this antioxidative defense process through miRNA silencing (**Figure 5C**). miR155HG is upregulated in the host cell (*p*-value *<sup>&</sup>lt;*<sup>4</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and identified as a repressor of CD3256. The inhibition of CD3256 by miR155HG results in not only redox-state disturbance but also pathogen-biomass decrease since CD3256 was predicted as an essential component for the growth of *C. difficile* (20).

## **The Strong Cellular Activities of Caco-2 Cells and the Infection Results of Host and Pathogen at the Late Stage of Infection**

The Emphasized ROS Production and

#### Stress-Accumulation of Host Cells, and the Failure of Antioxidative Defense Mechanisms in *C. difficile*

When CDI proceeds to the late stage, pathogen toxins exist in host cells, and the bacteria continue to secrete pathogenic factors into the interface between the two species. In this situation, the scattered CD0663 binds the host membrane protein CD46 (**Figure 6A**), triggering various cellular responses and the complement system. As discussed above, CD2973-induced acetylation of CD0663 could promote its activity in the late stage. This in turn increases the probability of CD0663–CD46 interaction and the following responses. CD46 activates these processes through FN1 (**Figure 6A**). For example, FN1 can trigger ROS production by activating NOX5. NOX5 is a key member of the NADPH oxidase complex and is also responsible for superoxide generation. Unlike the indirect transcriptional regulation utilized in the early stage, ROS production *via* NOX5 in the late stage provides a much rapid and efficient antimicrobial effect since it is directly activated by host cell surface FN1 and immediately released into the interface (**Figure 6A**). RAB33B encodes a small GTP-binding protein and plays a role in vesicular transport in protein secretion, such as cytokine release. RAB33B is repressed by methylation (**Figure 6A**), but its high expression level (*p*-value *<sup>&</sup>lt;*<sup>1</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and positive interaction with FN1 indicate that host cells enhance protein transport through RAB33B (**Figure 6A**). In addition to direct interaction, FN1 can transmit signals through the nuclear factor NFIC to upregulate the expression of *RAB33B* (**Figure 6A**).

Another important downstream protein of FN1 is C22orf28 (**Figure 6A**). Referred to as RTCB, C22orf28 was recently reported as an essential component in the ER stress response (33). The ER stress response is a cellular stress response resulting from the accumulation of unfolded proteins in the ER. The major aim of the ER stress response is to refold or degrade misfolded proteins, thus relieving the ER stress. If this objective cannot be achieved in a certain time span, cells then activate apoptosis. The presence of C22orf28 reveals that the prolonged accumulation of misfolded proteins from the early stage induces ER stress and risks the survival of host cells. To relieve the stress, host cells employ lncRNA HOTAIR to activate *C22orf28* (**Figure 6A**), but the decreased level of HOTAIR (*p*-value *<sup>&</sup>lt;*<sup>3</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) may in turn limit the activity of C22orf28. Overall, FN1 can be activated by CD46 and modified by NAT8L-triggered acetylation, which enhance the expression of FN1 and, therefore, promote cellular processes such as ROS production, ER stress response, and vesicular transport (**Figure 6A**). Therefore, the result suggested the presence of CD46 and the acetylation of FN1 as critical events during the progression from the early stage to the late stage of CDI.

Similar to the early stage, the chaperone function of HSP90B1 is impaired by the acetylation and interaction with *C. difficile* toxin CD0663 (**Figure 6A**), which results in misfolded protein formation. This result suggests that HSP90B1 is a conserved receptor for *C. difficile* toxins. To improve the impaired chaperone-ability, HSP90B1 urges the TF GATA2 to regulate its own expression. The acetylation of GATA2 also enhances its transcription ability to regulate the transcription of both *NOX5* to improve ROS production and *RAB33B* to strengthen cytokine release (**Figure 6A**).

In addition to HSP90B1, another HSP 90 kDa member, HSP90B2P, is influenced by *C. difficile* toxins (**Figure 6A**). *C. difficile* toxins (CD0663 and CD0660) can directly interact with HSP90B2P to repress the protein-folding ability of HSP90B2P in the host cells (**Figure 6A**). The acetylation of HSP90B2P also impairs its own chaperone function. In that case, HSP90B2P shows a similar response as HSP90B1. It can upregulate the expression of *NOX5* through AR, whose acetylation could promote its regulatory ability. HSP90B2P also activates the nuclear factor NFIC *via* the signal transduction protein VCAM1 (**Figure 6A**). Activated NFIC then upregulates the expression of *RAB33B* (**Figure 6A**). These same responses presented by HSP90B1 and HSP90B2P reveal that *C. difficile* toxins, whether intra- or extracellular, could affect the function of HSP90 proteins in a similar manner. The third manner in which toxins affect HSP90 proteins is through the inactivation of RAC1 (**Figure 6A**). The toxininduced glycosylation of RAC1 leads to the production of UDP, a putative "danger signal" and the ligand for the receptor P2RY6 (34). The secreted UDP can alarm neighbor cells by binding to P2RY6 and, therefore, activate HSP90B2P-related processes (**Figure 6A**).

Host miRNAs, as mentioned above, can pass through pathogen cell walls and membranes, enter the cytosol, and then silence pathogenic genes. In the late stage of infection, the result suggests that host miRNAs, miR4500HG, and miR16, can inhibit the pathogen genes *CD1631* and *CD1690*, respectively (**Figure 6A**). Both silenced genes encode protective proteins against oxidative stress (CD1631: superoxide dismutase, CD1690: thioredoxin). This result is similar to the previous stage, suggesting that host miRNAs may target anti-ROS genes in *C. difficile*.

In response to the enhanced ROS production and the activities of the complement system, *C. difficile* performs multiple methods to counteract host mechanisms. In **Figure 6A**, the flagellar sigma factor CD0266 (FliA) replaces CD2787 to control toxin production and induce defense mechanisms against ROS. It can directly interact with CD0660 (TcdB) or activate another toxin, CD0663 (TcdA), *via* the transmembrane protein CD2337 (**Figure 6A**). The downstream target of CD0663 is a GTP-sensing transcriptional repressor CD1275 (CodY) (**Figure 6A**). This protein plays an important role in pathogen growth repression and its corresponding gene has been predicted as an essential gene for the growth of *C. difficile* (20). The presence of this protein indicates that pathogens transform from rapid growth to the stationary phase and reflect the risk situation presented by the host cell. Interestingly, the high expression of CD1275 (*p*-value *<sup>&</sup>lt;*<sup>1</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and the decreased expression of CD0660 (*p*-value *<sup>&</sup>lt;*<sup>3</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) reveal that CD1275 in turn inhibits the expression of toxins, which is consistent with a previous study (35). CD2337 can also activate the metabolic dehydratase CD2341 and consequently its downstream essential protein CD0130 (**Figure 6A**). CD0130 plays a central role in this pathway to trigger various responses. However, a recent study indicated that the acetylation of the CD0130-homolog in *E. coli* impairs the corresponding enzymatic activity (36). The result in **Figure 6A** shows that CD0130 can activate the release of pathogenic factor CD1466 and signal *via* CD1931 (LexA) to regulate the toxin *CD0663*. CD1931 is a transcriptional repressor that can repress toxins and a number of genes involved in the DNA damage response (37). The impaired enzymatic activity of CD0130 results in low activity of CD1931 (*p*-value *<sup>&</sup>lt;*<sup>1</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and, thus, initiates the DNA damage response to remove ROSinduced DNA damage. The result suggests that CD0130 also employs alternative proteins to counteract ROS, such as superoxide dismutase CD1631, a conserved and powerful antioxidative protein, and CD1287, a fur family transcriptional regulator (**Figure 6A**). CD1287 can enhance the expression of *CD1631* and *CD0179* (**Figure 6A**), another defensive protein against H2O2, to eliminate ROS (38). However, these protective activities are limited by the acetylation of CD0130. Furthermore, DNA methylation of *CD0179* and host miR4500HG-induced silencing of *CD1631* (**Figure 6A**) also repress antioxidative defense mechanisms, resulting in DNA damage and component damage of the pathogen.

Since the function of CD1412 is currently unknown, the results suggest that it may participate in the defense mechanism triggered by CD0266 (**Figure 6A**). Once activated by CD0266, CD1412 then triggers CD0179 and the downstream CD1690 protein (**Figure 6A**). CD0179 encodes a glutamate dehydrogenase and is secreted by the pathogen, forming a protective layer against H2O<sup>2</sup> (38). Its downstream target, thioredoxin CD1690, can co-operate with CD0179 to remove ROS around *C. difficile*. Activation of the CD1412-mediated pathway and the high expression of CD1287 indicate that *C. difficile* recruits these proteins against oxidative stress. However, the DNA methylation of *CD0179* and host miRNA-induced silencing of *CD1631* and *CD1690* confer limited efficiency.

#### The Apoptosis Process Triggered by Severe Inflammation, Accumulated Cellular Stresses of Caco-2 Cells, and the Leaving of *C. difficile via* Endospore Formation

Stimulated by CD1466, SNW1 participates in the NOTCH signaling pathway to modulate NF-κB activity. The result in **Figure 6B** suggests that SNW1 can enhance the transcription ability of GATA2 and, therefore, improve RAB33B-meidated protein secretion. In addition, SNW1 can activate NFKB1 through the diacylglycerol kinase (**Figure 6B**), which is involved in several immunerelated pathways, including the interleukin-3, -5 signaling pathways. The activation of NFKB1 is an important event of CDI, and the acetylation of this NF-κB subunit could enhance its transcription ability to upregulate IL-8, a major cytokine for neutrophil recruitment (**Figure 6B**). Tight junction breakdown occurs at the early stage, allowing recruited neutrophils to enter the lumen and then clear pathogens by phagocytosis. However, neutrophil infiltration can also cause severe tissue damage in the host cell. In fact, neutrophil infiltration is an important clinical feature induced by CDI. Another important property of NF-κB is its ability to regulate apoptosis. In **Figure 6B**, the chaperone-ability of HSP90B2P is impaired by acetylation and the interaction with toxins. The prolonged ER and oxidative stress, initiated in the early stage, eventually trigger apoptosis *via* the signaling proteins VCAM1 and N4BP3 (**Figure 6B**). NFKB1 can upregulate its own expression and together with REL, another NF-κB subunit activated by HSP90B2P, promote NF-κB complex activity, thus initiating apoptosis. *NFKB1* and *REL* are methylated and silenced by miR143HG and miR155HG, respectively (**Figure 6B**). However, their high expression level (*p*-value *<sup>&</sup>lt;*<sup>1</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) indicates that host cells undergo apoptosis in the final step of infection. Moreover, HSP90B2P utilizes ETS1 to enhance the expression of the lncRNA MIAT (**Figure 6B**), which has also been reported to participate in apoptosis in epithelial cells (39).

High levels of ROS and scattered neutrophils threaten the survival of *C. difficile*, forcing pathogens to leave the infection site. The decreased expression of the cell surface protein CD2787 (*p*value *<sup>&</sup>lt;*<sup>4</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) represses the adhesion ability of *C. difficile*. The result suggests that another cell-wall protein, CD1987, utilizes the transmembrane protein CD2781 to initiate the CD1128-related DNA replication and the following sporulation (**Figure 6B**). CD1128 encodes DNA polymerase 1 (PolA) of *C. difficile*, and the acetylation of this protein can affect replication timing. DNA replication in bacterium triggers either pathogen reproduction or sporulation. Endospores are special survival structures of some bacteria, including the *Clostridium* genus. They are highly resistant to ultraviolet radiation, heat, desiccation, chemical agents, and oxidative stress, allowing bacteria to lie dormant for even centuries. The result in **Figure 6B** also suggests that *C. difficile* can initiate the sporulation process *via* CD1214. CD1214 is the stage 0 sporulation protein that helps pathogens transform to an endospore. CD1214 is methylated during infection but is activated by CD1128 (**Figure 6B**) and, thus, upregulates its own expression to enhance its activity.

In addition, the result suggests that CD2787 triggers another sporulation pathway *via* CD2247 (**Figure 6B**). CD2247 encodes a serine peptidase, which plays a crucial role in *C. difficile* spore germination. CD2247 also controls the sporulation process (**Figure 6B**). For instance, it can activate the biosynthesis of spore coat protein CD1511 and improve the activity of CD2629, which is an ATPase required for spore coat formation and proper localization. CD2247 also activates a crucial sporulation regulator CD2643 (SigE) (40), and this sigma factor can upregulate the expression of *CD2629* (*p*-value *<sup>&</sup>lt;* <sup>8</sup> *<sup>×</sup>* <sup>10</sup>*<sup>−</sup>*<sup>3</sup> ) and co-operate with CD1214. These activities suggest that pathogens actively transform to an endospore form in the late stage of infection.

## **Pathogenic Model of CDI**

Using a systems biology approach to construct genome-wide GEINs *via* big data mining and two-sided genome-wide data identification, we investigated the pathogenic model of CDI (**Figure 7**). The result suggests that the acetylation of HSP90B1 and the deacetylation of HSPA5, and the interaction of these HSPs with pathogen toxins (CD0660 and CD0663), impair the chaperoneactivity of host cell, resulting in the formation and accumulation of misfolded proteins. In addition, *C. difficile* toxins (CD0660 and CD0663) display their cytotoxicity by inactivating Rho GTPases (RHOA and CDC42) *via* glucosylation, causing the disturbance of actin cytoskeleton homeostasis and tight junction breakdown (3). Another central epigenetic activity is ubiquitination. The ubiquitination-mediated endocytosis of FPR1 allows *C. difficile* toxins enter host cells and then CD0660 and CD0663 hijack a

#### **FIGURE 7** | Continued

Overview of the pathogenic model of *C. difficile* infection (CDI). The upper panel (**Figure 7A**) displays the dominant role of epigenetic activity and the cross-talk interplay mechanisms between Caco-2 cells and *C. difficile* during the early stage of infection. The lower panel (**Figure 7B**) shows the molecular and epigenetic activities during the late stage of CDI and the outcome of both organisms. The cross marks signify the dysfunction or the breakdown of proteins and structures accordingly; the solid arrow lines represent the protein-interaction/cellular-response with literature support; the dash arrow lines denote the identified/predicted response; the red lines denote crucial epigenetic activities; the red dots in the endoplasmic reticulum of Caco-2 cell signify misfolded proteins; and the definitions of epigenetic activities are the same as those defined in **Figure 5**. In **(A)**, the epigenetic activities dominate the initialization of host responses toward bacterial invasion. The acetylation/deacetylation of heat shock proteins (HSPs) lead to the formation of misfolded proteins, the ubiquitination of FPR1 allows pathogen toxins enter Caco-2 cells *via* endocytosis and then trigger reactive oxygen species (ROS) overproduction, and the glucosylation of GTPases impair the homeostasis of cytoskeleton. In this situation, Caco-2 cells utilize DNA damage response and autophagy as defense mechanisms, and eliminate pathogens *via* ROS with the promotion of microRNA silencing. Between two stages, the acetylation of CD0663 and fibronectin 1, and the presence of CD46 and CD1466 serve as the turning points of the progression of infection, resulting in enhanced ROS production and severe inflammation as shown in **(B)**. The acetylation activity results in the dysfunction of HSPs and the assembly of NF-κB complex, inducing not only the production of IL8, which recruits neutrophils, but also the apoptosis process of host cells. To avoid the recruited phagocytic neutrophils and human-produced ROS, *C. difficile* cells activate the sporulation pathway to transform to endospore.

RAC1-mediated pathway to activate SYK-dependent ROS production (24, 28). ROS production is a primary response of the immune system, but the toxin-activated overproduction of ROS and free radicals could damage host lipids, proteins, and DNA. To repair the damaged DNA, SCARA3 and YY1 are activated to initiate the DNA damage response. The dysfunction of HSPs together with YY1 trigger the autophagy process, removing misfolded proteins, and ROS-induced damage (41). As a countermeasure, host miRNA can pass through the pathogen cell wall and then repress pathogen genes, including the antioxidative proteins CD3256 and CD0130 to promote the eliminating ability of host cells.

When the infection proceeds to the late stage (**Figure 7B**), several events together result in the turning point of CDI. The result suggests that the acetylation of CD0663 promotes the activity of this toxin and, thus, increases the probability of CD0663 to bind to CD46. CD46 can activate the complement system and enhance ROS production *via* NOX5. The complement system guides neutrophils, which can be recruited by interleukin-8 to remove toxins and pathogens (42). In addition, FN1 can be activated by CD46 and modified by acetylation, and these interactions could enhance the activity of FN1 and its downstream inflammatory responses. Furthermore, the newly produced CD1466 displays its immunogenicity by interacting with SNW1 and EGFR, thus inducing cytokine production and inflammation. Taken together, the presence of CD1466 and CD46, and the acetylation of CD0663 and FN1, result in enhanced oxidative stress and a severe inflammatory response. Meanwhile, the acetylation of HSPs (HSP90B1 and HSP90B2P) and their interactions with toxins (CD0663 and CD0660) impair the chaperone-activity, aggravating the accumulated ER stress (43). These activities in turn increase the cellular stress of the host cell. For example, neutrophil infiltration and the increased NADPH oxidase could cause tissue damage of the host cell, and the presence of C22orf28 reflects high ER stress (33). On the other hand, these processes are assuredly life threatening to *C. difficile*. To counteract these threatening stresses, *C. difficile* utilize numerous redox-related proteins, including superoxide dismutase CD1631, extracellular glutamate dehydrogenase CD0179, and thioredoxin CD1690 in the defense against oxidative stress. Finally, the accumulated oxidative and ER stress trigger the apoptosis process *via* the NF-κB complex. For *C. difficile*, the high levels of ROS and scattered neutrophil risk the survival of pathogens. Therefore, *C. difficile* transform to endospores by activating the sporulation pathway, and then lie dormant for the next infection.

As discussed above, CDI is characterized by cytoskeleton dysfunction, severe inflammation, and subsequent apoptosis. The actin cytoskeleton breakdown is mainly induced by CD0663 and CD0660, but the correlation between cytoskeleton dysfunction and apoptosis remains unclear. Here, we suggest that the main cause of apoptosis during CDI is the accumulation of cellular stress, including oxidative and ER stress. The emergence of oxidative and ER stress has been reported in studies using CD0660 infection models (21, 44) but generates few attention and further researches. We suggested that Caco-2 cells activate the DNA damage response and autophagy to counteract these stressors in the early stage of CDI. Unfortunately, the accumulated stress and tissue damage caused by severe inflammation induce host cells to undergo apoptosis in the late stage of infection.

There has been a long-lasted argument about whether CD0660 or CD0663 is responsible for the cytotoxicity in host cells. Some early studies reported that only CD0660 is essential for the virulence of *C. difficile* and a number of patients with CDI are caused by CD0663-negative and CD0660-positive strains (45). However, a later study using toxin knockdown technology on *C. difficile* 630 indicated that both CD0660 and CD0663 are responsible for disease (46). Our identified results also support this conclusion. During the early stage of CDI, CD0660 functions prior to CD0663 and triggers numerous host responses, such as ROS production and chaperone dysfunction. In the late stage, acetylation enhances the activity of CD0663, which takes the place of CD0660, inducing a severe inflammatory response and aggravating ER stress. The CD0663-activating ability of CD0660 identified in our toxin-regulation pathways is also consistent with this temporal relationship. Without CD0660, CD0663 may be not sufficient to initiate pathogenesis in host cells in the early stage of CDI and the low expression level of CD0660 cannot trigger the subsequent apoptosis in the absence of CD0663 during the late stage of infection.

By contrast, unlike pathogenic effects on the host, the intraspecies interactions and cellular mechanisms inside *C. difficile* are largely unknown. We found that *C. difficile* could generate redox-balancing proteins against oxidative stress and utilize toxin production and bacterial reproduction as offensive mechanisms at the early stage of CDI. During the late stage of infection, the result suggests that *C. difficile* utilized anti-ROS proteins, including CD0179, CD1631, and CD1690, as well as the DNA damage response to counteract the oxidative stress presented by host cells. The decreased activities of toxin production and bacterium-cell growth, and endospore formation also reveal that *C. difficile* actively transform to endospores to leave the infection site. Finally, the molecular mechanisms of progression from the early stage to the late stage of CDI are investigated, and the potential drug targets are also proposed for further drug design.

The crucial events of progression in CDI, such as the acetylation of CD0663 and FN1, the chaperone dysfunction of HSPs, and the pathogen silencing induced by host miRNA, are mainly caused by epigenetic regulations identified in our systems biology method. These results suggest that epigenetic regulation plays an important role in the progression of infection since these cellular activities could change the cellular functions of host cells in a more rapid and efficient manner than the adjustment of gene regulation.

This study has three main limitations. (1) The reported results are modeling outcomes of two-sided genome-wide expression data and are speculations and not proven. (2) Several new potential interactions, especially interspecies PPIs, based on sequencehomology between *C. difficile* and *E. coli*, and the interspecies PPIs between *C. difficile* and *homo sapiens*, need for further experimental evidence. (3) The findings relate to only on strain of *C. difficile* and only once specific cell line, Caco-2 cells. The true complexity of an entire gut significantly complicates interactions with other cells/bacteria/host. This is great and valuable work but certainly has limitations. To the best of our knowledge, there are few studies focusing on the epigenetic modulation on pathogenic and offensive mechanisms in the host cell infected by *C. difficile*. In addition, no existing whole-epigenomic data of the host cells can be considered as the basis of such studies. We show that epigenetic regulation will play a more important role in host/pathogen cross-talk mechanisms in CDI, which provides a novel direction for deeper studies on molecular mechanisms for drug target predictions and multiple drug discovery. With the identification of GEINs and HPNs, our knowledge of the bioinformatics of *C. difficile* and the core proteins is bound to increase, driving further experimental hypotheses and investigation directions for cross-talk mechanisms, including the protein–protein interaction network (PPIN) and gene regulation network (GRN) between host and *C. difficile* during infection. These further studies will complementarily complete our genetic-and-epigenetic network, providing a novel basis for developing the whole-genome cellular network of CDI.

### **MATERIALS AND METHODS**

#### **Overview of the Construction of GEINs and HPNs in Caco-2 Cells during the Early and Late Stage of CDI**

A flowchart showing HPNs of the host and pathogen at the early and late stages of infection by big data mining, model construction, and network identification for investigating the cross-talk molecular mechanisms and inferring potential drug targets is shown in **Figure 1**. These processes can be divided into four steps: (1) big data mining and data preprocessing of host/pathogen gene/miRNA expression data (see supplementary methods); (2) construction of candidate GEIN, which consists of candidate host/pathogen intraspecies PPINs, candidate interspecies PPINs between host and pathogen, candidate host/pathogen gene/miRNA regulation networks, candidate miRNA regulation networks of host-miRNAs on host/pathogen-genes, and candidate lncRNAs regulation networks of host-lncRNAs on host-genes (see supplementary methods and Figure S6 in Supplementary Material); (3) identifying real GEINs of each stage from the candidate GEIN *via* system identification method and system order detection scheme, using the genome-wide microarray data of Caco-2 cells and *C. difficile* during infection; and (4) extraction of HPNs from the real GEINs using the PNP (see supplementary methods). We then investigated the crucial molecular mechanisms that contribute to the progression of CDI and inferred potential drug targets for further drug design.

#### **Genome-Wide Microarray Data of Caco-2 Cells and** *C. difficile* **during Infection**

The two-sided microarray raw data proposed by Janvilisri et al. (6) has two parts. The first one contains the mRNA/miRNA expression profiles of three biological replicates of the Caco-2 cell line at 0, 30, 60, and 120 min postinfection with *C. difficile* 630. Each biological replicate contains two technical replicates (GEO accession number GSE18407). The Caco-2 cell line was cultured in Dulbecco's modified Eagle's medium at 37°C prior to infection. The second part contains the mRNA expression profiles of three biological replicates of *C. difficile* 630 in Caco-2 cells at 0, 30, 60, and 120 min postinfection (GEO accession number GSE18407; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE18407). The platforms used in the host and pathogen were Phalanx Human OneArray and *C. difficile* 630/QCD32g58 array, respectively, which include 39,200 and 13,824 probes, respectively. The microarray data were validated using qRT-PCR. We took the average of the microarray data of the two technical replicates for further network identification. We applied one-way analysis of variance to calculate *p*-value of a gene using microarray data between early and late infection stages.

#### **Dynamic Models of GEINs for Caco-2 Cells and** *C. difficile* **during Infection**

Since the candidate GEIN (see supplementary methods) was constructed by big data from numerous databases, experimental datasets, and literature, they contain some inevitable false-positive information. To address this, we built the dynamic model to characterize the molecular mechanisms of GEINs and to prune the false-positives in candidate GEIN, thus producing the real GEINs for Caco-2 cells and*C. difficile* during the infection process. For the PPIN of host proteins in candidate GEIN, the dynamic interaction model of the ith host protein can be described by the following dynamic equation:

$$\begin{split} p\_i^H(t+1) &= p\_i^H(t) + \sum\_{f=1}^{F\_i} a\_{if}^H p\_i^H(t) p\_f^H(t) + \sum\_{q=1}^{Q\_l} c\_{iq}^H p\_i^H(t) p\_q^P(t) \\ &+ \mathfrak{a}\_i^H \mathfrak{g}\_i^H(t) - \gamma\_i^H p\_i^H(t) + \mathfrak{e}\_i^H + \mathfrak{w}\_i^H(t), \\ &\quad \text{for } i = 1, 2, \dots, I, \ \mathfrak{a}\_i^H \ge 0 \text{ and } -\gamma\_i^H \le 0 \end{split} \tag{1}$$

where *p H <sup>i</sup>* (*t*), *p H f* (*t*), *g H <sup>i</sup>* (*t*) and *p P <sup>q</sup>* (*t*)represent the expression levels of the *i*th host protein, the *f* th host protein, the *i*th host gene and the *q*th pathogen protein at time *t*, respectively; *a H if* and *c H iq* denote the interactive ability between the *i*th and *f* th host protein and between the *i*th host protein and *q*th pathogen protein, respectively; *F*<sup>i</sup> and *Q*<sup>i</sup> signify the number of host proteins and pathogen proteins that interact with the *i*th host protein; α *H i* , *−*γ *H i* , and κ *H i* indicate the translation rate from the corresponding mRNA, the degradation rate, and the basal level of the *i*th host protein, respectively. In general, the basal level κ *H i* in Eq. 1 represents an unknown activity affecting the expression of the *i*th host protein other than those mentioned above, such as epigenetic acetylation and ubiquitination. ϖ *H <sup>i</sup>* (*t*) denotes the stochastic noise of the *i*th host protein at time *t*. The biological meaning of Eq. 1 is that the expression level of the *i*th host protein can be affected by various molecular mechanisms including the host intraspecies PPIs ∑*<sup>F</sup><sup>i</sup> f*=1 *a H if p H <sup>i</sup>* (*t*)*p H f* (*t*), interspecies PPIs ∑*<sup>Q</sup><sup>i</sup> q*=1 *c H iqp H <sup>i</sup>* (*t*)*p P <sup>q</sup>* (*t*), protein translation α *H i g H <sup>i</sup>* (*t*), protein degradation *−*γ *H i p H <sup>i</sup>* (*t*), basal level κ *H i* , and the corruption of stochastic noise ϖ *H <sup>i</sup>* (*t*)In addition, the translation rate should be constrained to be non-negative and the protein degradation rate should be constrained to be non-positive in real PPIs.

For the GRN of host genes in the candidate GEIN, the dynamic model of the *j*th host gene can be described as follows:

$$\begin{split} \boldsymbol{g}\_{l}^{H}\left(t+1\right) &= \boldsymbol{g}\_{l}^{H}\left(t\right) + \sum\_{i=1}^{l\_{j}} \boldsymbol{b}\_{ji}^{H}\boldsymbol{p}\_{i}^{H}\left(t\right) + \sum\_{n=1}^{N\_{j}} \boldsymbol{\varepsilon}\_{jn}^{H}\boldsymbol{l}\_{n}^{H}\left(t\right) \\ &+ \sum\_{i'=1}^{l\_{j}'} \sum\_{i''=1}^{l\_{j}''} \boldsymbol{\varkappa}\_{j\left(l\_{i'}^{H}\left(t^{-1}\right) + \boldsymbol{i}^{\nu}\right)}^{H} \boldsymbol{p}\_{i'}^{H}\left(t\right) \boldsymbol{p}\_{i'}^{H}\left(t\right) \\ &- \sum\_{k\_{l}}^{K\_{l}} d\_{jk}^{H}\boldsymbol{g}\_{l}^{H}\left(t\right) \boldsymbol{m}\_{k}^{H}\left(t\right) - \boldsymbol{\lambda}\_{j}^{H}\boldsymbol{g}\_{j}^{H}\left(t\right) + \boldsymbol{\delta}\_{j}^{H} + \boldsymbol{\varepsilon}\_{j}^{H}\left(t\right), \\ &\text{ for } j = 1, 2, \dots, J, -d\_{jk}^{H} \leq 0 \text{ and } -\boldsymbol{\lambda}\_{j}^{H} \leq 0 \end{split} \tag{2}$$

where *g H <sup>j</sup>* (*t*), *p H <sup>i</sup>* (*t*), *m H k* (*t*), and *l H <sup>n</sup>* (*t*) indicate the expression levels of the *j*th host gene, the *i*th host TF, the *k*th host miRNA, and the *n*th host lncRNA at time *t*, respectively; *b H ji* , *−d H jk* and *e H jn* represent the regulation ability of the *i*th host TF, the *k*th host miRNA, and the *n*th host lncRNA on the *j*th host gene, respectively; *Ij*, *Kj*, and *N<sup>j</sup>* denote the number of host TFs, host miRNAs and host lncRNAs, respectively, which regulate the expression level of the *j*th host gene; *p H i ′* (*t*)*p H i ′′* (*t*) imply the *i*th host complex where *p H i ′* (*t*) and *p H i ′′* (*t*) represent the subunit 1 and 2 of the *i*th host complex, respectively; *I ′ <sup>j</sup>* and *I ′′ <sup>j</sup>* are the same, to denote the number of host complex, which regulate the *j*th host gene in the candidate GRN; *x H j*(*I ′′ j* (*i ′−*1)+*i ′′*) signifies the regulation ability of the *i*th host complex on the *j*th host gene; *−*λ *H <sup>j</sup>* and δ *H j* indicate the degradation rate and the basal level of the *j*th host gene, respectively. In general, the basal level δ *H j* in Eq. 2 denotes an unknown regulation other than those mentioned above such as DNA methylation. *ε H <sup>j</sup>* (*t*) represents the stochastic noise due to modeling residue at time *t*. Notably, in the case of the regulation ability *x H j*(*I ′′ j* (*i ′−*1)+*i ′′*) of the *i*th host complex on the *j*th host gene, the index *I ′′ j* ( *i ′ −* 1 ) + *i ′′* guarantees the proper coordinate of the regulation ability *x H j*(*I ′′ j* (*i ′−*1)+*i ′′*) of the *i*th host complex *p H i ′* (*t*)*p H i ′′* (*t*)in the host GRN system matrix of the *j*th host gene, i.e., the regulation abilities of host complexes on the *j*th host gene can be aligned to an one row matrix of *x H j*1 *, . . . , x H j*(*I ′′ j* ) *, x H j*(*I ′′ <sup>j</sup>* <sup>+</sup>1) *, . . . , x H <sup>j</sup>*(2*<sup>I</sup> ′′ j* ) *, x H <sup>j</sup>*(2*<sup>I</sup> ′′ <sup>j</sup>* <sup>+</sup>1) *, . . . , x H j*(*I ′′ j* (*i ′−*1)+*i ′′*) *, . . . , x H j*(*I ′′ j I ′ j*) .

Therefore, the biological meaning of Eq. 2 is that the expression level of the *j*th host gene can be regulated by numerous molecular mechanisms, including the host TF regulations ∑*<sup>I</sup><sup>j</sup> i*=1 *b H ji p H <sup>i</sup>* (*t*) host lncRNA regulations <sup>∑</sup>*<sup>N</sup><sup>j</sup> n*=1 *e H jnl H <sup>n</sup>* (*t*), host complex regulations *I ′* ∑*j i ′*=1 *I ′′* ∑*j i ′′*=1 *x H j*(*I ′′ j* (*i ′−*1)+*i ′′*) *p H i ′* (*t*)*p H i ′′* (*t*), host miRNA repressions *−* ∑*<sup>K</sup><sup>j</sup> k*=1 *d H jkg H <sup>j</sup>* (*t*)*m H k* (*t*), mRNA degradation effect *−*λ *H j g H <sup>j</sup>* (*t*), basal level δ *H j* , and the corruption of stochastic noise *ε H <sup>j</sup>* (*t*). Similar to protein model, the miRNA repression ability and gene degradation rate should be constrained to be non-positive. Since DNA methylation can directly influence the binding affinities of RNA polymerase to target genes (47), we assumed that the regulation by methyltransferase could cause the significant change of the basal level δ *H <sup>j</sup>* and the change of δ *H <sup>j</sup>* between the early stage and late stage of CDI in the dynamic model (2) implies the occurrence of methylation at the *j*th host gene in the infection process.

In Eq. 2, the expression of the *k*th host miRNA *m H k* (*t*) and the *n*th host lncRNA *l H <sup>n</sup>* (*t*) at time *t* can also be regulated by other regulators. Therefore, the dynamic regulatory equation of the *k*th host miRNA was modeled as follows:

$$m\_k^H(t+1) = m\_k^H(t) + \sum\_{i=1}^{l\_k} \mathcal{y}\_{ki}^H \boldsymbol{\uprho}\_i^H(t) - \boldsymbol{\upmu}\_k^H \boldsymbol{m}\_k^H(t) + \boldsymbol{\uprho}\_k^H + \boldsymbol{\upxi}\_k^H(t),$$
 
$$\text{for } k = 1, 2, \dots, K \text{ and } -\boldsymbol{\upmu}\_k^H \le \mathbf{0} \tag{3}$$

where *m H k* (*t*) and *p H <sup>i</sup>* (*t*) denote the expression levels of the *k*th host miRNA and the *i*th host TF at time *t*, respectively; *y H ki* represents the regulatory ability of the *i*th host TF on the *k*th host miRNA; *I<sup>k</sup>* signifies the number of host TFs that regulate the expression level of the *k*th host miRNA; *−*μ *H k* and ϕ *H k* indicate the miRNA degradation rate and the basal level of the *k*th host miRNA, respectively; and ς *H k* (*t*) implies the stochastic noise at time *t*. The dynamic model of host miRNAs in Eq. 3 characterizes molecular regulatory mechanisms, including the transcription regulations ∑*Ik i*=1 *y H kip H <sup>i</sup>* (*t*), miRNA degradation effect *−*μ *H <sup>k</sup> m H k* (*t*), basal level ϕ *H k* , and the corruption of stochastic noise ς *H k* (*t*). In addition, the degradation rate should be constrained to be non-positive.

Similarly, the dynamic model of the *n*th host lncRNA in candidate GEIN can be described by the dynamic equation as follows:

$$l\_n^H(t+1) = l\_n^H(t) + \sum\_{i=1}^{l\_n} z\_{ni}^H p\_i^H(t) - \boldsymbol{\chi}\_n^H l\_n^H(t) + \boldsymbol{\uprho}\_n^H + \boldsymbol{\uprho}\_n^H(t),$$
 
$$\text{for } n = 1, 2, \dots, N \text{and } -\boldsymbol{\upchi}\_n^H \le \mathbf{0} \tag{4}$$

where *l H <sup>n</sup>* (*t*) and *p H <sup>i</sup>* (*t*) represent the expression levels of the *n*th host lncRNA and the *i*th host TF at time *t*, respectively; *z H ni* denotes the regulation ability of the *i*th host TF on the *n*th host lncRNA; I<sup>n</sup> signifies the number of host TFs regulating the expression level of the *n*th host lncRNA; *−*χ *H n* and ρ *H n* indicate the degradation rate and the basal level of the *n*th host lncRNA, respectively; and ϑ *H <sup>n</sup>* (*t*) is the stochastic noise due to the modeling residue. The dynamic model of host lncRNAs in Eq. 4 characterizes molecular regulatory mechanisms, including the transcription regulations ∑*In i*=1 *z H nip H <sup>i</sup>* (*t*), degradation effect *−*χ *H n l H <sup>n</sup>* (*t*), basal level ρ *H n* , and the corruption of stochastic noise ϑ *H <sup>n</sup>* (*t*). Furthermore, the constraint of this model is that the lncRNA degradation rate should be non-positive.

For the PPIN of pathogen proteins in candidate GEIN, the dynamic model of the *q*th pathogen protein can be described by the following equation:

$$\begin{aligned} p\_q^P(t+1) &= p\_q^P(t) + \sum\_{o=1}^{O\_q} a\_{qo}^P p\_q^P(t) \, p\_o^P(t) + \sum\_{i=1}^{I\_q} c\_{qi}^P p\_q^P(t) \, p\_i^H(t) \\ &+ \alpha\_{qq}^P \boldsymbol{\xi}\_q^P(t) - \boldsymbol{\gamma}\_q^P \boldsymbol{p}\_q^P(t) + \boldsymbol{\kappa}\_q^P + \boldsymbol{\varpi}\_q^P(t) \, , \\ &\text{for } q = 1, 2, \dots, Q, \ \boldsymbol{\alpha}\_q^P \ge 0 \, \text{and } -\boldsymbol{\gamma}\_q^P \le 0 \end{aligned} \tag{5}$$

where *p P <sup>q</sup>* (*t*), *p P <sup>o</sup>* (*t*), *g P <sup>q</sup>* (*t*), and *p H <sup>i</sup>* (*t*) represent the expression level of the *q*th pathogen protein, the *o*th pathogen protein, the *q*th pathogen gene, and the *i*th host protein at time *t*, respectively; *a P qo* and *c P qi* denote the interactive ability between the *q*th pathogen protein and *o*th pathogen protein, and between the *q*th pathogen protein and *i*th host protein, respectively; *O<sup>q</sup>* and *I<sup>q</sup>* signify the number of pathogen proteins and host proteins that interact with the *q*th pathogen protein, respectively; α *P q* , *−*γ *P q* and κ *P q* indicate the translation rate, the degradation effect and the basal level of the *q*th pathogen protein, respectively; and ϖ *P <sup>q</sup>* (*t*) denotes the stochastic noise of the *q*th pathogen protein at time *t*. The biological meaning of the Eq. 5 is that the expression level of the *q*th pathogen protein can be affected by various molecular interactive mechanisms, including the intraspecies PPIs ∑*Oq o*=1 *a P qop P <sup>q</sup>* (*t*)*p P <sup>o</sup>* (*t*), the interspecies PPIs ∑*Iq i*=1 *c P qip P <sup>q</sup>* (*t*)*p H <sup>i</sup>* (*t*), protein translation α *P q g P <sup>q</sup>* (*t*), protein degradation *−*γ *P q p P <sup>q</sup>* (*t*), basal level κ *P q* , and the corruption of stochastic noise ϖ *P <sup>q</sup>* (*t*). Similar to the host protein dynamic model, the translation rate should be constrained to be nonnegative and the protein degradation rate should be constrained to be non-positive.

For the GRN of pathogen genes in candidate GEIN, the dynamic model of the *h*th pathogen gene can be described as follows:

$$\begin{aligned} \mathcal{g}\_h^P(t+1) &= \mathcal{g}\_h^P(t) + \sum\_{q=1}^{Q\_h} b\_{hq}^P \mathcal{p}\_q^P(t) - \sum\_{k=1}^{K\_h} d\_{hk}^P \mathcal{g}\_h^P(t) \, m\_k^H(t) \\ &- \lambda\_h^P \mathcal{g}\_h^P(t) + \mathfrak{G}\_h^P + \varepsilon\_h^P(t) \,, \\ &\text{for } h = 1, 2, \dots, H, \ -d\_{hk}^P \le 0 \text{ and } -\lambda\_h^P \le 0 \text{ (6)} \end{aligned}$$

where *g P h* (*t*), *p P <sup>q</sup>* (*t*), and *m H k* (*t*) indicate the expression levels of the *h*th pathogen gene, the *q*th pathogen TF, and the *k*th host miRNA at time *t*, respectively; *b P hq* and*−d P hk* represent the regulatory ability of the *q*th pathogen TF and the *k*th host miRNA on the *h*th pathogen gene, respectively; *Q<sup>h</sup>* and *K<sup>h</sup>* denote the number of pathogen TFs and host miRNAs that regulate the *q*th pathogen gene; *−*λ *P <sup>h</sup>* and δ *P h* indicate the degradation rate and the basal level of the *q*th pathogen gene, respectively; and *ε P h* (*t*) represent the stochastic noise due to the modeling residue at time *t*. The biological meaning of Eq. 6 is that the expression level of the *h*th pathogen gene can be regulated by various molecular mechanisms, including the pathogen TF regulations ∑*Qh q*=1 *b P hqp P <sup>q</sup>* (*t*), host miRNA repressions *−* ∑*Kh k*=1 *d P hkg P h* (*t*)*m H k* (*t*), mRNA degradation *−*λ *P hg P h* (*t*), basal level δ *P h* , and the corruption of stochastic noise *ε P h* (*t*). In addition, the host-miRNA repression rate and degradation rate should be constrained to be non-positive. By comparing to the well-proposed models in the cross-talk genomewide GEINs (48), this study considered additional molecular mechanisms in the GEIN, including transcription regulations of host lncRNA and host protein complex in Eq. 2 and the dynamic models of host miRNA in Eq. 3 and host lncRNA in Eq. 4, to characterize molecular interaction during cell infection in more detail.

#### **Parameter Estimation of the Dynamic Models of Candidate GEIN** *via* **the System Identification Method**

In order to identify the precise parameters, we applied a system identification method to the dynamic genetic-and-epigenetic Eqs 1–6 in the candidate GEIN. We rewrote the host PPIN dynamic Eq. 1 as the linear regression form below,

$$\begin{aligned} \boldsymbol{p}\_{i}^{H}\left(t+1\right) &= \begin{bmatrix} \boldsymbol{p}\_{i}^{H}\left(t\right) \boldsymbol{p}\_{1}^{H}\left(t\right) & \cdots & \boldsymbol{p}\_{i}^{H}\left(t\right) \boldsymbol{p}\_{i}^{H}\left(t\right) & \boldsymbol{p}\_{i}^{H}\left(t\right) \boldsymbol{p}\_{1}^{P}\left(t\right) \\\\ \vdots & & \vdots \\\\ \boldsymbol{\gamma} &= \boldsymbol{p}\_{i}^{H}\left(t\right) \boldsymbol{p}\_{Q\_{i}}^{P}\left(t\right) & \boldsymbol{g}\_{i}^{H}\left(t\right) & \boldsymbol{p}\_{i}^{H}\left(t\right) & 1 \end{bmatrix} \begin{bmatrix} \boldsymbol{d}\_{1}^{H} \\ \vdots \\ \boldsymbol{d}\_{H\_{i}}^{H} \\ \boldsymbol{d}\_{H\_{i}}^{H} \\ \vdots \\ \boldsymbol{d}\_{i}^{H} \\ \boldsymbol{d}\_{iQ\_{i}}^{H} \\ \vdots \\ \boldsymbol{d}\_{i}^{H} \\ 1 - \boldsymbol{\gamma}\_{i}^{H} \end{bmatrix} + \boldsymbol{\mathfrak{m}}\_{i}^{H}\left(t\right) \\ \boldsymbol{\end{aligned}$$

where φ *HP i* (*t*)represents the regression vector that can be obtained from the microarray expression data and θ *HP i* is the unknown parameter vector to be estimated for the *i*th host protein in host PPIN.

Equation 7 of the *i*th host protein can be augmented for *T<sup>i</sup>* data points as the following form:

$$\begin{bmatrix} \mathbf{p}\_{l}^{H}(t\_{2})\\\mathbf{p}\_{l}^{H}(t\_{3})\\\vdots\\\mathbf{p}\_{l}^{H}(t\_{T\_{l}}+1) \end{bmatrix} = \begin{bmatrix} \boldsymbol{\Phi}\_{l}^{HP}(t\_{1})\\\boldsymbol{\Phi}\_{l}^{HP}(t\_{2})\\\vdots\\\boldsymbol{\Phi}\_{l}^{HP}(t\_{T\_{l}}) \end{bmatrix} \boldsymbol{\Phi}\_{l}^{HP} + \begin{bmatrix} \boldsymbol{\varpi}\_{l}^{H}(t\_{1})\\\boldsymbol{\varpi}\_{l}^{H}(t\_{2})\\\vdots\\\boldsymbol{\varpi}\_{l}^{H}(t\_{T\_{l}}) \end{bmatrix}, \text{ for } i = 1,2,\dots,I \tag{8}$$

which could be simply represented as follows:

$$P\_i^H = \Phi\_i^{HP} \Phi\_i^{HP} + \Gamma\_i^{HP}, \text{for } i = 1, 2, \dots I \tag{9}$$

$$\text{where } P\_{l}^{H} = \begin{bmatrix} p\_{l}^{H}\left(t\_{2}\right) \\ p\_{l}^{H}\left(t\_{3}\right) \\ \vdots \\ p\_{l}^{H}\left(t\_{T\_{l}} + 1\right) \end{bmatrix}, \Phi\_{l}^{HP} = \begin{bmatrix} \Phi\_{l}^{HP}\left(t\_{1}\right) \\ \Phi\_{l}^{HP}\left(t\_{2}\right) \\ \vdots \\ \Phi\_{l}^{HP}\left(t\_{T\_{l}}\right) \end{bmatrix}, \Gamma\_{l}^{HP} = \begin{bmatrix} \mathfrak{D}\_{l}^{H}\left(t\_{1}\right) \\ \mathfrak{D}\_{l}^{H}\left(t\_{2}\right) \\ \vdots \\ \mathfrak{D}\_{l}^{H}\left(t\_{T\_{l}}\right) \end{bmatrix}.$$

Therefore, the parameters in the vector θ *HP <sup>i</sup>* can be estimated by applying the following constrained least-squares estimation problem,

$$\begin{aligned} \min\_{\boldsymbol{\Theta}\_{i}^{HP}} \quad & \left\| \boldsymbol{\Phi}\_{i}^{HP} \boldsymbol{\Theta}\_{i}^{HP} - \boldsymbol{P}\_{i}^{H} \right\|\_{2}^{2} \\ \text{subject to } \begin{bmatrix} \mathbf{0} & \cdots & \mathbf{0} & \mathbf{0} & \cdots & \mathbf{0} & -1 & \mathbf{0} & \mathbf{0} \\ \mathbf{0} & \cdots & \mathbf{0} & \mathbf{0} & \cdots & \mathbf{0} & \mathbf{0} & 1 & \mathbf{0} \end{bmatrix} \boldsymbol{\Theta}\_{i}^{HP} \leq \begin{bmatrix} \mathbf{0} \\ \mathbf{1} \end{bmatrix} \tag{10} \end{aligned} \tag{10}$$

The parameters in the host PPIN dynamic Eq. 1 can be estimated by solving the constrained least-squares problem (10) *via* the help of *lsqlin* function in MATLAB optimization toolbox, and simultaneously the host protein translation rate α *H i* is guaranteed to be non-negative and the host protein degradation *−*γ *H i* is guaranteed to be non-positive, i.e., α *H <sup>i</sup> ≥* 0 and *−*γ *H <sup>i</sup> ≤* 0. Similarly, the constrained least-squares estimation problems of Eqs 2–6 are shown in supplementary methods.

As mentioned above, to avoid the overfitting problem in the parameter identification process, we have applied cubic spline to interpolate extra data points (five times number of the parameters in the parameter vector to be estimated, i.e., θ *HP I* in host PPIN, θ *HG j* in host GRN, θ *HM k* in host miRNA dynamic model, θ *HL <sup>n</sup>* in host lncRNA dynamic model, θ *PP q* in pathogen PPIN, and θ *PG h* in pathogen GRN). Therefore, with the microarray expression data, we could solve the constrained least-squares estimation problems in Eq. 10 and Eqs S3, S6, S9, S12, and S15 in Supplementary

#### **REFERENCES**


Material and identify the precise parameters in GEINs gene by gene (or protein by protein) *via* the *lsqlin* function of MAT-LAB optimization toolbox. Since the measurement technology of genome-wide protein expression of Caco-2 cells and*C. difficile* has not yet been realized, and about 73% variance of protein abundance can be explained by the corresponding mRNA abundance (49), the microarray data of gene expressions can replace protein expressions, providing sufficient information for solving above constrained least-squares parameter estimation problems.

#### **AUTHOR CONTRIBUTIONS**

C-WL and M-HS are co-first author. C-WL and M-HS: data analysis and interpretation, manuscript writing, methodology development, conception and design, data analysis and interpretation, manuscript writing. C-WL and B-SC: methodology development, conception and design, data analysis and interpretation, manuscript writing. C-WL and B-SC reviewed the paper, prepared figures, wrote and improved the scientific quality of the manuscript. All authors read and approved the final manuscript.

#### **ACKNOWLEDGMENTS**

The work was supported by the Ministry of Science and Technology of Taiwan under grant No. MOST 104-2221-E-007 -124 -MY3 and 105-2811-E-007-041.

#### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu. 2017.00901/full#supplementary-material.


production and intestinal epithelial barrier dysfunction. *PLoS One* (2013) 8(11):e81491. doi:10.1371/journal.pone.0081491


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Li, Su and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Real-Time qPCR as a Method for Detection of Antibody-Neutralized Phage Particles

Anna Kłopot<sup>1</sup> , Adriana Zakrzewska<sup>1</sup> , Dorota Lecion<sup>1</sup> , Joanna M. Majewska<sup>1</sup> , Marek A. Harhala<sup>1</sup> , Karolina Lahutta<sup>1</sup> , Zuzanna Ka ´zmierczak<sup>1</sup> , Łukasz Łaczmanski ´ 1,2 , Marlena Kłak<sup>2</sup> and Krystyna D ˛abrowska1,2 \*

<sup>1</sup> Bacteriophage Laboratory, Institute of Immunology and Experimental Therapy, Polish Academy of Sciences, Wrocław, Poland, <sup>2</sup> Research and Development Center, Regional Specialist Hospital, Wrocław, Poland

The most common method for phage quantitation is the plaque assay, which relies on phage ability to infect bacteria. However, non-infective phage particles may preserve other biological properties; specifically, they may enter interactions with the immune system of animals and humans. Here, we demonstrate real-time quantitative polymerase chain reaction (qPCR) detection of bacteriophages as an alternative to the plaque assay. The closely related staphylococcal bacteriophages A3R and 676Z and the coliphage T4 were used as model phages. They were tested in vivo in mice, ex vivo in human sera, and on plastic surfaces designed for ELISAs. T4 phage was injected intravenously into pre-immunized mice. The phage was completely neutralized by specific antibodies within 5 h (0 pfu/ml of serum, as determined by the plaque assay), but it was still detected by qPCR in the amount of approximately 10<sup>7</sup> pfu/ml of serum. This demonstrates a substantial timelapse between "microbiological disappearance" and true clearance of phage particles from the circulation. In human sera ex vivo, qPCR was also able to detect neutralized phage particles that were not detected by the standard plaque assay. The investigated bacteriophages differed considerably in their ability to immobilize on plastic surfaces: this difference was greater than one order of magnitude, as shown by qPCR of phage recovered from plastic plates. The ELISA did not detect differences in phage binding to plates. Major limitations of qPCR are possible inhibitors of the PCR reaction or free phage DNA, which need to be considered in procedures of phage sample preparation for qPCR testing. We propose that phage pharmacokinetic and pharmacodynamic studies should not rely merely on detection of antibacterial activity of a phage. Real-time qPCR can be an alternative for phage detection, especially in immunological studies of bacteriophages. It can also be useful for studies of phage-based drug nanocarriers or biosensors.

Keywords: qPCR, plaque assay, phage, quantitation, antibody, neutralizing, humoral response, immune response

## INTRODUCTION

Bacteriophages can be used in multiple medical applications (Mie¸dzybrodzki et al., 2012; Kutter et al., 2015; Vandenheuvel et al., 2015; Karimi et al., 2016; Saeed et al., 2017), veterinary (Grant et al., 2016), biotechnology (O´slizło et al., 2011; Lee et al., 2012; O'Sullivan et al., 2016), agriculture (Zaczek et al., 2015 ˙ ), or food processing (Endersen et al., 2014). Development of all these phage

#### Edited by:

Yves Renaudineau, University of Western Brittany, France

#### Reviewed by:

Dirk Dittmer, University of North Carolina at Chapel Hill, United States Ruchi Tiwari, Veterinary University (DUVASU), India

> \*Correspondence: Krystyna D ˛abrowska dabrok@iitd.pan.wroc.pl

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 28 July 2017 Accepted: 23 October 2017 Published: 06 November 2017

#### Citation:

Kłopot A, Zakrzewska A, Lecion D, Majewska JM, Harhala MA, Lahutta K, Ka ´zmierczak Z, Łaczmanski Ł, Kłak M and ´ D ˛abrowska K (2017) Real-Time qPCR as a Method for Detection of Antibody-Neutralized Phage Particles. Front. Microbiol. 8:2170. doi: 10.3389/fmicb.2017.02170

applications relies on experimental testing with the accurate detection of phage particles in various conditions. Specifically, phage quantitation is the key step in comparisons between different phage strains, different health status of animals or humans, and different experimental design. The most common and prevalent method for phage quantitation is the plaque assay. It simply employs a microbiological culture, where a sensitive bacterial host is cultured with a phage in a double layer plate, eventually allowing for direct counting of plaques as soon as bacteria become visible in the plate (Adams, 1959). There are important limitations of this method. The most important one is the fact that microbiological detection of phage activity does not meet the real count of bacteriophage particles in samples; it in fact allows for testing how many phages were able to infect their host effectively. Phage capability of infection, in turn, may be dependent on a myriad of factors: from simple ion content in the environment and presence of organic compounds or detergents, to the presence of specific antibodies, complement system elements or competitive phages of other types (Ishiguro et al., 1983; Matsushita et al., 2011; Refardt, 2011; Cheng et al., 2013, 2016; Hodyra-Stefaniak et al., 2015; Szermer-Olearnik et al., 2017). Once inactivated, phage cannot be detected by the plaque assay, which makes quantitation of real phage particle content very difficult or even impossible.

Non-infective phage particles may preserve other biological or technological properties that are not related to their ability to infect bacteria. Phage immunoreactivity, i.e., the potency of phage to interact with antibodies and other elements of the immune system, is a phenomenon independent on phage infectivity (Kirsch et al., 2008; Samoylova et al., 2012; D ˛abrowska et al., 2014). Most bacteriophages are complex, multi-protein structures where the infection apparatus is only a fraction of the whole particle. Other structural proteins can interact with mammalian system (D ˛abrowska et al., 2006, 2007; Barr et al., 2013) regardless phage infectivity. Therefore, phage pharmacokinetics/pharmacodynamics studies should not rely merely on detection of antibacterial activity of a phage.

Techniques that rely on nucleic acid amplification and detection are among the most valuable tools in biological research. Real-time quantitative polymerase chain reaction (qPCR) detection of eukaryotic viruses in environmental and human or animal samples is a standard and commercialized method (Watzinger et al., 2004; Hmaied et al., 2015). Bacteriophages have been quantitatively analyzed and discriminated by real-time qPCR directly in microbiological cultures, and the authors found this method to be a good alternative to the plaque assay (Edelman and Barletta, 2003; Clokie, 2009; Anderson et al., 2011; Refardt, 2012; Dieterle et al., 2016). Real-time PCR has been further demonstrated as applicable for a rapid screening allowing phage detection in food (milk, fruits, vegetables, seafood, meat) (Imamovic and Muniesa, 2011; Flannery et al., 2014; Perrin et al., 2015; Parente et al., 2016; Hartard et al., 2017) and water samples (Farkas et al., 2015; Kunze et al., 2015; Unnithan et al., 2015; Mankiewicz-Boczek et al., 2016) or in feces (Imamovic et al., 2010; Chehoud et al., 2016). However, potential applicability of real-time qPCR for detection of inactivated (non-infective) but still biologically active (e.g., immunoreactive) phage has never been investigated.

Here we propose real-time qPCR as a quantitative method for phage detection in immunological studies of bacteriophages. Specifically, it was tested in comparisons between different phage strains, different immunological statuses of animals or humans, and different experimental designs. Two closely related staphylococcal bacteriophages, A3R and 676Z, together with the coliphage T4, were used as the model phage strains. We assessed and validated a real-time qPCR-based method for immunological studies of samples derived from animals and humans (in vivo and ex vivo experiments), as well as for optimization of comparative ELISAs of different phages.

## MATERIALS AND METHODS

### Preparation of Phages T4, A3R, and 676Z and Determination of Phage Titers Using Plaque Assay

T4 phage was purchased from American Type Culture Collection (ATCC, Manassas, VA, United States) and phages A3R and 676Z are part of the IIET Microorganisms Collection (Institute of Immunology and Experimental Therapy, Polish Academy of Sciences, Wroclaw, Poland). Enriched broth cultures of phages were purified by filtration through polysulfone membranes and by fast protein liquid chromatography: gel filtration on Sepharose 4B (Sigma–Aldrich, Poland). The final preparation was dialyzed using 1000 kDa-pore membranes against PBS to remove the bacterial residuals and lipopolysaccharide (LPS), and filtered through 0.22 µm PVDF filters (Millipore, Europe). Prior to dialysis of T4 phage we used LPS-affinity chromatography EndoTrap HD according to the manufacturer's instructions (Hyglos GmbH, Bernried, Germany) in order to further remove LPS. LPS removal was done by three successive incubations of the preparation with the slurry followed by centrifugations. Each purified phage T4 preparation was tested for phage concentration by determining phage titer after serial dilution with PBS (dilutions from 10−<sup>1</sup> to 10−<sup>9</sup> ), 25 µl of each dilution was spotted on a culture plate pre-covered with susceptible bacteria, two spots for each dilution. The plate was incubated overnight at 37◦C to obtain visible plaques. The plaques were counted, mean values of two spots were calculated, and the phage concentration was calculated per milliliter with regard to the dilution and spot volume. For phages A3R and 676Z we used double layer agar plates according to Adams in order to determine phage titer (Adams, 1959).

## Isolation of Genomic DNA from T4, A3R, and 676Z Phages and Preparation of DNA Standards

Phage genomic DNA was isolated using GenElute Mammalian Genomic DNA Miniprep (Sigma–Aldrich, Poznan, Poland). For this procedure we used phage lysates each containing 10<sup>8</sup> pfu. After incubating samples with DNase and RNase (Sigma–Aldrich, Poland) (50 µg of each, 10 min, 37◦C), 40 µg

of proteinase K (Sigma–Aldrich, Poland) as well as 100 µl of Resuspension Buffer was added to each sample. Samples were then incubated for 5 min at 70◦C. Next, 200 µl of Lysis Solution C was added to the samples and samples were incubated for 10 min at 70◦C. Afterward 200 µl of 96% ethanol (Sigma–Aldrich, Poland) was added to each sample. Samples were applied to the columns provided with the kit and prepared in advance using Column Preparation Solution. Following column centrifugation (6500 × g, 1 min, room temperature) 500 µl of wash buffer was applied to the columns and the columns were centrifuged (12,000 × g, 1 min, room temperature). The washing step was repeated twice. Finally phage genomic DNA was eluted using 60 µl of DNase-free water. DNA was quantified using NanoDrop (Wilmington, DE, United States). Based on this we prepared 10 ng/µl stocks of phage genomic DNA. In order to prepare a standard curve for each phage genomic DNA we used 10 ng/µl stock solutions to prepare the following dilutions of phage genomic DNA: 1, 0.1, and 0.01 ng/µl. For the phages A3R, 676Z, and T4 the highest standard equals approximately 6.5 × 10<sup>7</sup> , 8.6 × 10<sup>7</sup> , and 5 × 10<sup>7</sup> phage particles per microliter, respectively. Numbers of virus particles were calculated as genomic equivalents. In order to calculate genomic equivalents based on the amount of nanograms of DNA in qPCR samples we used an online calculator (Stothard, 2000) to calculate the molecular weight of a single phage genome from the exact genomic sequences of the bacteriophages (accession numbers for A3R, 676Z, and T4, are JX080301, JX080302, and NC\_000866, respectively). The following molecular weights were calculated for single phage DNA molecules: 87108105.24 Da for A3R, 91769200.84 Da for 676Z, and 104340909.87 Da for T4 phage. These molecular weights of phage genomes were used to calculate numbers of single genomic DNA molecules in investigated samples (1 Da weighs approximately 1.67 × 10−<sup>24</sup> g), resulting in the values called genomic equivalents. The standard solutions were used in duplicate to prepare the standard curve for qPCR. For the purpose of absolute quantifications, standard curves were created by plotting quantification cycle (Cq) values against the number of genomic equivalents.

#### Primer Design and qPCR Reaction

The genome sequences of T4, A3R, and 676Z phages were obtained from the GenBank (accession numbers for T4, A3R, and 676Z are NC\_000866, JX080301, and JX080302, respectively). Based on the genome sequence, qPCR primers were designed using the Primer-BLAST software at the National Center for Biotechnology Information (National Center for Biotechnology Information, 2017).

To detect T4 phage we used forward primer 5<sup>0</sup> -ACT GGC CAG GTA TTC GCA-3<sup>0</sup> and reverse primer 5<sup>0</sup> -ATG CTT CTT TAG CAC CGG CA-3<sup>0</sup> . To detect A3R and 676Z phages the following primers were used: forward primer 5<sup>0</sup> -TGA AGA AGA CCG TGC AGG ATT-3<sup>0</sup> and reverse primer 5<sup>0</sup> -TCA GAA GGA GCT GAT TGA GCG-3<sup>0</sup> .

The amount of genomic DNA in each test sample was determined using 5× HOT FIREPolEvaGreen qPCR Mix Plus (Solis BioDyne, Tartu, Estonia) following the manufacturer's recommendations. Briefly, each PCR reaction contained 1 µl of DNA template and 15 pM of each primer as well as 2 µl of 5× HOT FIREPolEvaGreen qPCR Mix Plus in a final volume of 10 µl. Cycling conditions were as described by 5× HOT FIREPolEvaGreen qPCR Mix Plus's manufacturer. The amount of phage genomic DNA in test samples was calculated based on the standard curve generated using standard solutions prepared as outlined above. qPCR normalization was performed according to MIQE Guidelines (Bustin et al., 2009).

## ELISA

A MaxiSorp flat-bottom 96-well plate (Nunc, Thermo Scientific, Poznan, Poland) was covered with purified phage preparations ´ obtained by chromatography as described above (100 µl per well, as indicated in the figures) sterilely, at 4◦C, overnight. Plates were washed five times with PBS and blocked with five times diluted Superblock (Thermo Fisher Scientific Inc., Rockford, IL, United States) for 1 h (100 µl per well) at room temperature. Blocking solution was removed and the plate was washed five times with PBS with 0.05% Tween 20 (Serva, Heidelberg, Germany). Serum diluted 1:100 was applied to the wells in 100 µl per well. Each sample was processed in duplicate. The plate was incubated at 37◦C for 2 h. Plates were washed five times with PBS with 0.05% Tween 20 (Sigma–Aldrich, Poland). Diluted detection antibody was added in the amount of 100 µl per well: peroxidase-conjugated AffiniPure goat anti-mouse IgM (Jackson ImmunoResearch Laboratories, West Grove, PA, United States) or peroxidase-conjugated AffiniPure goat anti-mouse IgG (Jackson ImmunoResearch Laboratories, West Grove, PA, United States) at a final dilution of 200,000. The detection antibody was incubated in the wells for 1 h at room temperature in the dark, removed, and the plate was washed five times with PBS with 0.05% Tween 20. TMB X-Treme (Nordic BioSite AB, Sweden) substrate reagent for peroxidase was used (50 µl) according to the manufacturer's instructions (ImmunO4, Westminster, MD, United States). Twenty-five microliters of 2 N H2SO<sup>4</sup> (Sigma–Aldrich, Poland) was added to each well without substrate removal, then absorbance was measured at 450 (main reading) and 550 nm (background). The background values were subtracted from the main readings and the average value of each duplicate was calculated. In the case of relative increases of antibody levels OD values are presented.

## Experimental Design

#### Experiment 1: Comparison of qPCR and Plaque Assay Method for T4 Phage Detection in Human Sera ex Vivo

In this experiment equal volumes of T4 phage (50 µl, titer 10<sup>7</sup> pfu/ml) were mixed with equal volumes of human serum samples. We specifically used serum samples from three healthy donors that contained high amounts of antibodies against T4 phage as well as samples from three healthy donors that contained low amounts of such antibodies. The serum samples mixed with T4 phage were incubated for 1 h at 37◦C in order to allow for antibody-mediated destruction of T4 phage followed by detection of the amount of live phage in serum samples using the plaque

method. In addition we used the qPCR method described above to detect genomic DNA.

#### Experiment 2: Comparison of qPCR and Plaque Assay Method for T4 Phage Detection in Serum of Mice Immunized to T4 Phage

Three BALB/c female mice were injected intraperitoneally on day 0 with 1 × 10<sup>11</sup> pfu T4 phage/mouse and three BALB/c female mice were injected with an equal volume of sterile PBS. Eleven days after the challenge animals were bled from the tail and serum was separated from the blood by double centrifugation at 2250 × g. The serum samples were subjected to the ELISA test in order to determine the amount of antibodies specific for T4 phage belonging to classes IgM and IgG. Thirteen days after the challenge mice were injected i.v. with T4 phage 1 × 10<sup>11</sup> pfu/mouse determined by the plaque method and bled from the tail 0.5 and 5 h later. Serum was separated as outlined above. A portion of each serum was used to detect the amount of T4 phage via the plaque method, whereas a 50 µl portion of serum was subjected to isolation of viral nucleic acid using a viral RNA and DNA kit (Macherey Nagel, Duren, Germany) according to the protocol outlined by the manufacturer. Obtained DNA was used to detect the amount of T4 phage via the qPCR method.

#### Experiment 3: Quantitation of Phage Particles on Plastic Surfaces

We used dilutions of each phage (T4, A3R, and 676Z) that contained 1 × 10<sup>9</sup> , 5 × 10<sup>9</sup> , and 1 × 10<sup>10</sup> pfu/ml as determined by the plaque assay, six replicates each. Hundred microliters of each dilution for each phage was used to coat a 96-well Maxisorp plate (Nunc). As a negative control we used PBS solution that was also used as a diluent for phage solutions. The plate was incubated overnight at 4◦C. The next day the plate was washed five times with 150 µl of PBS per well and nonspecific binding was blocked with fivefold diluted Superblock (Thermo Fisher Scientific Inc., Rockford, IL, United States) for 45 min at room temperature. Then, 10 µg of proteinase K (A&A Biotechnology, Gdynia, Poland) diluted in Tris-EDTA buffer pH 8.0 (final volume 100 µl) was added to each well and the plate was incubated for 4 h at 50◦C. After incubation the content of each well was transferred to a single tube. Tubes were incubated for 20 min at 85◦C in order to inactivate proteinase K. Following brief centrifugation samples were frozen at −80◦C before further use for qPCR reactions as described above.

#### Experiment 4: Comparison of the Amount of Phage Detectable by qPCR and the Relative Signal from the ELISA

In this experiment A3R and 676Z phages at 1 × 10<sup>9</sup> , 5 × 10<sup>9</sup> , and 1 × 10<sup>10</sup> pfu/ml dilutions as determined by the plaque assay as well as PBS were used in triplicate to coat Maxisorp plates. One plate was processed for qPCR analysis of adhered phage particles as described above (see Experiment 1), and the other identical plate was processed for ELISA. In the ELISA experiment we used mouse serum specific to major capsid protein (AFN38316) as the first antibody source and goat anti-mouse IgG as the detection antibody. The ELISA test was conducted as outlined above.

## Ethics Statement

The female 6- to 8-week-old BALB/c mice were purchased from Mossakowski Medical Research Centre, Polish Academy of Sciences, Warsaw, Poland and kept in specific pathogen-free (SPF) conditions in the Animal Breeding Centre of the IIET. The experiments were approved by the First Local Committee for Experiments with the Use of Laboratory Animals, Wroclaw, Poland (no. 92/2016). The Bioethics Committee of Wrocław Medical University approved obtaining blood samples from healthy donors. All study subjects gave written informed consent for participation in the study.

## Data Analysis and Statistics

Each experiment was repeated at least twice and the results of one experiment are shown. The data are presented as mean ± SE. Statistical analysis was done by one-way ANOVA followed by Tukey's multiple comparison test with a significance level of p = 0.05 or two-tailed unpaired t-test with a significance level of p = 0.05. The graphs and statistical analysis were performed using GraphPad Prism software.

## RESULTS

## Characterization of qPCR Reaction

We determined efficiency and specificity of qPCR reactions performed with primers specifically recognizing genomic DNA of staphylococcal phages A3R and 676Z and coliphage T4. In all investigated phages, primers were specific to genes coding for major head proteins. It should be noted that these genes are 100% homologous in A3R and 676Z phages [overall homology of the genomes is 94.6% (Łobocka et al., 2012)]; thus, the same set of primers was used in both phages. Sample standard curves obtained for phages A3R, 676Z, and T4 are presented in **Figure 1**. These curves together with corresponding correlation coefficients (**Table 1**) show a strong correlation between the Cq value and the genomic equivalents representing numbers of phage particles. The efficiency of amplification reaction for any phage genomic DNA was higher than 75%.

The sensitivity of phage detection by qPCR reaction was determined by Cq values of the PCR reaction performed with the phage concentration range from 0 to 10<sup>11</sup> pfu/ml, for all three investigated phages. We propose the range from 10<sup>3</sup> to 10<sup>11</sup> pfu/ml as a reliable correlation between phage concentration and Cq value; in this range we observed a linear relationship between Cq values and the number of each phage particles (**Figure 2**). For the lower phage titers Cq values were characterized by high deviations: in a range between 35.36 ± 0.3 and 35.57 ± 1.32 for A3R phage, between 33.4 ± 0.85 and 34 ± 0.923 for 676Z phage, and 31.32 ± 0.69 and 32.57 ± 0.064 for T4 phage.

## Comparison of qPCR and Plaque Assay Method for T4 Phage Detection in Serum of Mice Immunized to T4 Phage

Optimized qPCR detection of phage (see above) was used to detect the phage in in vivo model with animals eliciting phageneutralizing antibodies. Mice were challenged with T4 to develop a specific antibody response as previously described (D ˛abrowska et al., 2014); a high antibody level was confirmed by ELISA (**Supplementary Figure S1**). The mice specifically immunized to T4 phage and control mice were injected intravenously with the phage; phage concentration in blood serum was assessed 0.5 and 5 h after injection by the plaque assay or by qPCR. The amounts of phage detected in murine sera are compared in **Figure 3**. Both methods showed that the presence of specific antibodies results in a decrease of phage concentration in animals' sera, but there was a striking difference in the demonstrated levels of the decrease. Half an hour after phage administration to mice, the difference between phage concentration in phageimmunized mice and control mice as detected by the plaque assay was more than five orders of magnitude, but only two orders of magnitude by qPCR. Five hours after phage administration to mice, this difference as detected by the plaque assay was more than 10 orders of magnitude (no active phage was detected in immunized mice) but less than three orders of magnitude by qPCR (still approximately 10<sup>7</sup> genomic equivalents/ml of phage particles detected). This demonstrates that the phage, even when neutralized in terms of antibacterial activity, can still be present in the circulation in high amounts and it can be detected and quantitatively assessed by qPCR.

## Comparison of qPCR and Plaque Assay Method for T4 Phage Detection in Human Sera ex Vivo

The potency of qPCR to detect immunologically inactivated phage was further verified in human sera ex vivo. T4

TABLE 1 | Correlation coefficients calculated for qPCR reaction (corresponding to Figure 1).


FIGURE 2 | Correlation between Cq values of qPCR reaction and the amount of phage particles used as a template in qPCR reaction. Three independent serial dilutions (N = 3) containing from 0 to 10<sup>11</sup> pfu/ml of 676Z phage, A3R phage, or T4 phage were used as templates in qPCR reactions. The X-axis presents the logarithm of phage titers used as template DNA in the qPCR reaction and the Y-axis presents Cq values of the qPCR reactions.

bacteriophage was incubated with human sera, both with those containing a high concentration of phage-specific antibodies (positive sera) and with those with a low content of phagespecific IgG (negative sera); a previously described collection of positive and negative sera was used (D ˛abrowska et al., 2014). After incubation, the phage was detected either by plaque assay or by qPCR and the results were compared. The plaque assay revealed statistically significantly lower (70% decrease) activity of phage after incubation with positive sera than that after incubation with negative sera (p = 0.037). In the qPCR method the amount of phage detected in both groups was not statistically different (p = 0.39) (**Figure 4**). This is in line with the proposition of qPCR as a method for detection of phage whose antibacterial activity has been neutralized.

#### Quantitation of Phage Particles on Plastic Surfaces

Due to the devastating effect that antibodies may exert on phage activity, probably the major area of immunological studies of bacteriophages is identification of phage-specific

FIGURE 3 | Comparison of qPCR and plaque assay method for T4 phage detection in serum of mice immunized to T4 phage. Control and pre-immunized BALB/c mice were injected i.v. with T4 phage 10<sup>11</sup> pfu/mouse, murine sera were collected 0.5 and 5 h after and phage concentration in the sera was tested by (A) plaque assay method or (B) qPCR method.

antibodies. Comparative studies of bacteriophages as different antigens by ELISA need to include quantitation of phage particles immobilized on ELISA plates. Here, quantitation of T4, A3R, and 676Z phage particles adhering to 96-well plates was conducted by qPCR, using the developed standard curves (see section Characterization of qPCR Reaction).

In general, all investigated bacteriophages were detectable by qPCR when immobilized on ELISA plates. We further observed that the amount of immobilized phage correlated positively with concentration of the phage that was used for incubation with the plate (**Figure 5**). Comparison of the three investigated phages revealed that their individual affinity to plastic surfaces was markedly different. Incubation with the highest phage concentration (1 × 10<sup>10</sup> pfu/ml) allowed for the following approximated amounts of recovered bacteriophages (as genome equivalents): 1 × 10<sup>8</sup> for T4, 3 × 10<sup>9</sup> for A3R, and 6 × 10<sup>9</sup> for 676Z (**Figure 5**).

We further compared the amount of phage detectable by qPCR and the relative signal from the ELISA, to assess the reliability of ELISA for comparisons between phages. ELISA plates covered with two very similar bacteriophages, A3R and 676Z, were processed either for adhering phage quantitation by qPCR or for immunodetection of phage by major capsid proteinspecific antibody (ELISA). Although the qPCR method revealed approximately two times more 676Z phage adhering to the ELISA plate (in comparison to A3R phage), no differences were detected by ELISA (**Figure 6**).

676Z phage. The plates were processed either for (A,B) immunodetection of phage by ELISA with major capsid protein-specific antibodies, or (C,D) for qPCR detection of phage immobilized on the plate, with PCR primers specific to the genes coding for major capsid proteins. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Estimation of the efficiency of phage immobilization in the well was calculated from the total amount of phage DNA in the solution that was applied to the wells and from the amount of phage DNA recovered from the wells. It revealed that when 1 × 10<sup>10</sup> pfu/ml solutions of phages A3R and 676Z were applied to the wells we recovered 4.5 and 9.67%, respectively.

## DISCUSSION

In this work we investigated qPCR as a useful alternative for phage detection, in comparison to standard methods based on phage cultures with bacteria. We focused on immunological studies of bacteriophages, since antibodies are probably the most efficient biological factors able to neutralize phages.

We demonstrated that the phage, even when neutralized in terms of antibacterial activity, can still be present in the circulation in high amounts and it can be detected and quantitatively assessed by qPCR (**Figure 3**). This implies that the phage may exert biological effects much longer than it can be detected by standard microbiological methods. Phage effects that are independent of its antibacterial activity may apply to phage use as drug nanocarriers (Karimi et al., 2016; Bashari et al., 2017), vaccine platforms (Aghebati-Maleki et al., 2016; Pires et al., 2016; Tao et al., 2016), as well as to possible immunomodulation exerted by the phage (Barr, 2017; Górski et al., 2017). Our observations demonstrate that the time lapse between "microbiological disappearance" and true clearance of phage particles from the circulation can be substantial. This particularly applies to common phages that may run up against a frequent and strong anti-phage immune response in the population, such as T4 (D ˛abrowska et al., 2014).

A further consequence of the strong influence of antibodies on the results of microbiological detection of phages is the fact that standard attempts to detect phages in human and animal sera can be unsuccessful even when the phages are highly represented there. Phage translocation and "phagoviremia" has been postulated as an important physiological phenomenon (Górski et al., 2006), although not demonstrated yet. There are still not enough data to assess levels of immunization of humans and animals to phages naturally belonging to their gut- (or other) microbiota. However, a model phage was demonstrated as able to induce a specific systemic response and a high serum level of specific IgG (Majewska et al., 2015). Assuming that phages of the microbiome eventually induce antibody production, it may be impossible to detect true phage translocation to the circulation only by microbiological methods. We propose the qPCR method as an appropriate and reliable method for studies of phage translocation to blood. This method can also be applied in other experiments that require detection of a "neutralized" phage, which is merely a phage that is not active against bacteria. Notably, qPCR has already been applied for phage detection in other (than phage neutralizing) conditions: in phage cultures (Edelman and Barletta, 2003; Clokie, 2009; Anderson et al., 2011; Refardt, 2012; Dieterle et al., 2016), food (Imamovic and Muniesa, 2011; Flannery et al., 2014; Perrin et al., 2015; Parente et al., 2016; Hartard et al., 2017; Muhammed et al., 2017), environmental samples (Farkas et al., 2015; Kunze et al., 2015; Unnithan et al., 2015; Mankiewicz-Boczek et al., 2016), and in feces (Imamovic et al., 2010; Chehoud et al., 2016), where some interference from antibodies cannot be excluded (Majewska et al., 2015).

We have also shown that different bacteriophages may differ strongly in their ability to immobilize on plastic surfaces. Here, we report more than one order of magnitude difference between phage recovery from plastic that was covered by different phages in the same conditions (**Figure 5**). This should be considered in any comparative studies of bacteriophages that include phage immobilization on various surfaces, e.g., in biosensor constructing or in the ELISA assay for phage immune-reactivity. As demonstrated by comparison of the ELISA signal to qPCR phage detection, differences in phage ability to adhere to a plastic surface can be missed by immunodetection (**Figure 5**). Notably, the two staphylococcal bacteriophages investigated in this work are highly similar. They have identical genes coding for major capsid proteins (the ones detected by qPCR here) (Łobocka et al., 2012), but minor structural elements of their virions differ. Eventually, their adherence to plastic also differs as much as twofold (**Figures 5**, **6**).

Quantitative analysis of phage by real-time PCR as a method has a few limitations that should be considered when planning its use. First, although qPCR can detect even those phage particles that are not able to infect bacteria, it is sensitive to inhibitors of the PCR reaction. Thus, in conditions where these inhibitors cannot be removed, phage detection may be more efficient in the plaque assay that in qPCR. For example, our post hoc analysis indicated that the animals in this study were challenged with 5.8 × 10<sup>11</sup> pfu/ml of T4 as measured by plaque assay and 1.56 × 10<sup>11</sup> of T4 as measured by qPCR. We also observed an inhibitory effect of raw human sera when applied into the composition of the qPCR reaction (data not shown). Thus, we isolated phage DNA from murine sera by a standard isolation kit to compare phage circulation between immunized and non-immunized mice. This approach turned out to be appropriate to overcome the problem of PCR reaction inhibitors in serum samples. The second considerable requirement for experiments employing qPCR is to use phage preparations without free phage DNA. In the qPCR method, free phage DNA present in phage solution can easily produce a false positive signal mimicking phage particles. It means that production of phage preparations, including often complex procedures of phage purification, need to be gentle enough not to destroy phage particles releasing phage nucleic acids.

#### Concluding Remarks

This study demonstrates a substantial time lapse between inactivation of antibacterial activity and true clearance of phage particles from the circulation of pre-immunized individuals. qPCR allows for detection of inactivated bacteriophages that cannot be detected by plaque assay. Further, qPCR demonstrated marked differences in the ability of investigated bacteriophages to immobilize on plastic surfaces, while the ELISA did not detect differences in phage binding to plates. We propose that phage pharmacokinetic and pharmacodynamic studies should not rely merely on detection of antibacterial activity of a phage; real-time qPCR can be an extension for phage detection methods.

## AUTHOR CONTRIBUTIONS

AK performed most of the experiments, analyzed the results, and participated in writing the manuscript. AZ, DL, JM, MH, KL, MK, and ZK participated in experimental work. ŁŁ participated in experimental design, data analysis, and reviewed the manuscript. KD conceived and designed the experiments, participated in experimental work and in data analysis, and wrote the paper.

#### FUNDING

This work was supported by the National Science Centre in Poland grant UMO-2012/05/E/NZ6/03314, and by Wroclaw Centre of Biotechnology, program The Leading National Research Centre (KNOW) for years 2014–2018.

#### ACKNOWLEDGMENTS

fmicb-08-02170 November 3, 2017 Time: 18:28 # 9

We are grateful to Professor Andrzej Gorski, Bacteriophage Laboratory, IIET, for valuable scientific discussions and for

#### REFERENCES


his continuous, valuable support for immunological studies of phages.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02170/full#supplementary-material

FIGURE S1 | T4-specific antibody detection in mice pre-immunized with T4. Significant induction of serum IgM and IgG specific for T4 phage was detected by ELISA (IgM level: T4-challenged mice vs. PBS control, p = 0.005; IgG level: T4-challenged mice vs. PBS control, p < 0.0001).

Dit/Tal modules in the genome of Lactobacillus casei BL23. Appl. Microbiol. Biotechnol. 100, 9201–9215. doi: 10.1007/s00253-016-7727-x



raspberries. J. Virol. Methods 224, 95–101. doi: 10.1016/j.jviromet.2015. 08.005


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Kłopot, Zakrzewska, Lecion, Majewska, Harhala, Lahutta, Kazmierczak, Łaczma ´ nski, Kłak and D ˛abrowska. This is an open-access article ´ distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Unveiling and Characterizing Early Bilateral Interactions between Biofilm and the Mouse Innate Immune System

Christiane Forestier<sup>1</sup> , Elisabeth Billard<sup>2</sup> , Geneviève Milon<sup>3</sup> and Pascale Gueirard<sup>1</sup> \*

<sup>1</sup> CNRS UMR 6023, Laboratoire Microorganismes: Génome et Environnement, Université Clermont-Auvergne, Clermont-Ferrand, France, <sup>2</sup> INRA USC 2018, Inserm U1071, Laboratoire Microbes Intestin Inflammation et Susceptibilité de l'Hôte, Université Clermont-Auvergne, Clermont-Ferrand, France, <sup>3</sup> Institut Pasteur, Paris, France

#### Edited by:

Marina I. Arleevskaya, Kazan State Medical Academy, Russia

#### Reviewed by:

Livia Visai, Universita degli Studi di Pavia and Istituti Clinici Scientifici Maugeri di Pavia, Italy Maria Cristina Cerquetti, Universidad de Buenos Aires, Argentina François J. M. A. Meurens, INRA UMR703 Ecole Nationale Vétérinaire, Agroalimentaire et de l'Alimentation de Nantes-Atlantique, France

> \*Correspondence: Pascale Gueirard pascale.gueirard@uca.fr

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 24 August 2017 Accepted: 08 November 2017 Published: 21 November 2017

#### Citation:

Forestier C, Billard E, Milon G and Gueirard P (2017) Unveiling and Characterizing Early Bilateral Interactions between Biofilm and the Mouse Innate Immune System. Front. Microbiol. 8:2309. doi: 10.3389/fmicb.2017.02309 A very substantial progress has been made in our understanding of infectious diseases caused by invasive bacteria. Under their planktonic forms, bacteria transiently reside in the otherwise sterile mammal body tissues, as the physiological inflammation insures both their clearance and repair of any tissue damage. Yet, the bacteria prone to experience planktonic to biofilm developmental transition still need to be studied. Of note, sessile bacteria not only persist but also concur preventing the effectors and regulators of the physiological inflammation to operate. Thus, it is urgent to design biologically sound experimental approaches aimed to extract, at the earliest stage, immune signatures of mono-bacteria planktonic to biofilm developmental transition in vivo and ex vivo. Indeed, the transition is often the first event to which succeeds the "chronicization" process whereby classical bacteriatargeting therapies are no more efficacious. An in vivo model of micro-injection of Staphylococcus aureus planktonic or biofilm cells in the ear pinna dermis of laboratory transgenic mice with fluorescent immune cells is proposed. It allows visualizing, in real time, the range of the early interactions between the S. aureus and myeloid cell subsets- the resident macrophages and dendritic cells, the recruited neutrophil granulocytes/polymorphonuclear neutrophils, monocytes otherwise known to differentiate as macrophages or dendritic cells. One main objective is to extract contrasting immune signatures of the modulation of the physiological inflammation with respect to the two bacterial lifestyles.

Keywords: bacteria, biofilm, intravital imaging, macrophage/monocyte, mouse, polymorphonuclear neutrophil

#### INTRODUCTION

Most invasive bacteria display two different lifestyles: whereas the free-floating planktonic bacteria' life style dominates, in some clinical settings, bacteria sensing hostile conditions adhere to biotic or abiotic surfaces and form biofilms (Moormeier and Bayles, 2017). The planktonic to biofilm/sessile lifestyle transition is associated with important metabolic changes and selfproduction of proteins-lipids-exopolysaccharides-rich as well as extracellular DNA-containing extracellular matrix (Costerton et al., 1999).

According to the National Institutes of Health, biofilms have an enormous impact on human medicine, accounting for over 80% of infectious processes in otherwise sterile tissue (s). Whereas the physiological inflammation is able to both clear invasive planktonic bacteria and to repair tissue damages insuring the return to tissue structural and functional homeostasis, this physiological inflammation does not operate in tissues experiencing sustained colonization by bacterial biofilms. Moreover, contrasting with planktonic bacteria that are cleared by commonly used antibiotics, provided that they do not harbor genetic resistance determinants, the majority of bacteria within the biofilms are resistant to these antibiotics (Lebeaux et al., 2014).

In this review, our present understanding of the professional phagocyte sensors of microbial agonists expected to operate over the in vivo planktonic to biofilm lifestyle switch is briefly introduced. Transiently invasive planktonic bacteria are usually cleared by myeloid cells of either the neutrophil granulocyte lineage or/and by mononuclear phagocytes. However, the bilateral interactions engaged or not between sessile bacteria and the myeloid cells are still poorly studied. Until now, most experimentalists have conducted in vitro studies with sessile bacteria, exposing them to either one or the other myeloid cell lineage or both (Watters et al., 2016). This review focuses on recent developments obtained in rodent models to characterize inflammatory responses against Staphylococcus aureus or Pseudomonas aeruginosa sessile bacteria. A new experimental approach combining the mouse ear pinna model and the intravital imaging approach is proposed to analyze these innate immune responses at the dynamic level.

## THE PROFESSIONAL PHAGOCYTE SENSORS OF MICROBIAL AGONISTS EXPECTED TO OPERATE OVER THE IN VIVO PLANKTONIC TO BIOFILM LIFESTYLE SWITCH

The in vivo developmental transition from planktonic to sessile bacteria reflects a range of bilateral cross talks in the fluctuating dynamic tissue milieu colonized by the bacteria under study. At the earliest stage of this developmental transition, communications between key bacterial messengers as well as interactions between bacterial agonists and sensors displayed by the resident and recruited myeloid cells such as the professional phagocytes are initiated and renewed.

## The Nucleotide-Based Second Messengers

The cyclic dinucleotides (c-di-NMPs) are recognized as Microbial Associated Molecular Patterns/MAMPs and induce a host type I interferon immune response prolonged by IFNγ production (Valle et al., 2013; Snyder et al., 2017). Moreover, the c-di-NMPs play a central role in many bacterial species during the lifestyle transition (Valle et al., 2013). Using the P. aeruginosa model organism, Valentini and Filloux (2016) showed that the bacteria use c-di-GMP as a checkpoint during the different steps of biofilm development. There is indeed a direct correlation between high levels of c-di-GMP in the bacteria and biofilm formation, and between low levels of c-di-GMP and motility (planktonic phenotype). The c-di-GMP second messenger is used by Escherichia coli and Salmonella enterica serovar Typhimurium over their planktonic to biofilm developmental transition (Allewell, 2016; Valentini and Filloux, 2016) whereas the c-di-AMP is used as a second messenger by other bacteria such as S. aureus (Corrigan et al., 2011).

The Guanosine tetraphosphate (ppGpp) and pentaphosphate (pppGpp), also called (p)ppGpp or alarmones are synthetized when bacteria are exposed to cells such as phagocytes in the infected tissue. Considered as bacterial signature of a so called stringent response, these alarmones represent intracellular signaling molecules known to participate to intracellular bacteria survival: Geiger et al. (2012) demonstrated that the stringent response is induced, in vitro, after S. aureus phagocytosis by neutrophil granulocytes or polymorphonuclears (PMN). The rapid (p)ppGpp synthesis leads to increased psm transcription and to participation of synthetized phenol soluble modulins (PSMs) concurring to bacteria survival after phagocytosis (Geiger et al., 2012). Depicted as pro-inflammatory agents, the PSMs also account for the bacteria escape from the transient intracellular niche, followed either by bacteria survival inside the cytosol or by lysis of the cell, all these rapid processes contributing to damages of the S. aureus- hosting tissues (Geiger et al., 2012; Peschel and Otto, 2013).

## The Quorum Sensing Circuit and Its Additional Regulators

Quorum sensing (QS) is a cell-to-cell signaling process that allows bacteria to sense and process high cell densities. It involves the synthesis, release and accumulation of signaling molecules called auto-inducers (AIs) (Papenfort and Bassler, 2016). At high concentrations, AIs induce cellular signaling cascades that notably control biofilm formation. The QS system of S. aureus called accessory gene regulator (Agr) has been extensively studied (Paharik and Horswill, 2016). With other regulators, it constitutes a complex regulatory network that, at any moment, either modifies AGR activity itself, or its downstream signaling or metabolic pathways. At the early stage of S. aureus developmental transition from planktonic to biofilm lifestyle, the low Agr concentration allows the production of inter-bacteria/intercellular adhesins whereas toxins' expression is repressed (Balasubramanian et al., 2016). Later, high levels of Agr induce biofilm structuration and dispersion by up-regulating the expression of the pro-inflammatory PSM molecules (Otto, 2008; Periasamy et al., 2012; Kavanaugh and Horswill, 2016; Paharik and Horswill, 2016). Environmental factors also modulate the Agr function in S. aureus, with a well-described inhibitory effect of the reactive oxygen species (ROS) produced by innate immune cells (Kavanaugh and Horswill, 2016).

Among other regulators that interact with the agr system at later stage of the biofilm development and maturation, the S. aureus exoprotein (Sae) two-component system promotes the synthesis and secretion of leucocidins and other virulence

determinants that actively participate to S. aureus survival after contact with PMN in vitro (Paharik and Horswill, 2016). Sae mutants have indeed a decreased capacity to resist to the PMN- clearing functions after contact (Yarwood and Schlievert, 2003; Voyich et al., 2009; Paharik and Horswill, 2016).

## RODENT MODELS TO STUDY INFLAMMATORY RESPONSES TO BIOFILMS IN VIVO

In rodent laboratory models, the most rapidly recruited cells are phagocytes, namely the PMN. Over any local disruption of mouse tissue homeostasis, PMN stored in the bone marrow egress into the blood vascular bed. Through chemoattractants and cytokines produced by tissue resident mast cells and macrophages (Teng et al., 2017), the PMN rapidly cross the microvessels endothelial cells, reaching the extracellular compartment colonized by invasive bacteria. They represent the first wave of innate immune cells to be recruited from the blood circulation. Most often, PMN efficiently kill and degrade invasive planktonic bacteria by using different antimicrobial strategies: phagocytosis, production of ROS and antimicrobial peptides, as well as cytotoxic components released from their subcellular distinct granules. More recently, neutrophil extracellular traps (NETs) have been observed and included as an additional strategy (Brinkmann et al., 2004). Among the in vivo models that allow characterization of the bilateral interactions co-engaged by PMN and sessile bacteria, we selected those relying on either P. aeruginosa or S. aureus (**Table 1**).

Of note, the features of the biofilm-colonized devices condition the interaction profiles: whereas in devices of hollow type (catheter, silicon splint) bacterial biofilms are protected from the immune cells, on solid devices (K-wire, pin), the bacterial biofilms are directly exposed to the different waves of immune cells. Depending on the model used, many other parameters operate (listed in **Table 1**). In most models, an intense and rapid accumulation of PMN is observed at the proximity of the biofilms (Wagner et al., 2003; Prabhakara et al., 2011a; Torre et al., 2015; Moser et al., 2017; Wang et al., 2017), the PMN representing the most abundant population of recruited cells. Using a non- invasive bioluminescence-based approach, Bernthal et al. (2011) showed that this recruitment is IL-1β dependent, as a 50% decrease in PMNs numbers was observed in the bacteria colonized knee joints of IL-1β deficient mice, as compared to wild-type mice. This PMN infiltration is increased when diabetic mice are treated with insulin in a bacteria-hosting wound, the latter incorporating actin and DNA from lysed PMN, which therefore contributes to the building of biofilms (Watters et al., 2014). Of note, a low PMN recruitment in the target tissues (Thurlow et al., 2011; Scherr et al., 2014; Secor et al., 2017) can be also operating: in particular was noticed an association between the reduced PMN recruitment and the production of Filamentous Pf1 like bacteriophage (Pf phage) by P. aeruginosa (Secor et al., 2017). Taken globally, whatever the experimental conditions depicted in **Table 1**, there is a need to capture more comprehensive information documenting, at least at the earliest stages, the in vivo dynamic interactions between bacteria and PMN.

Could mature biofilms' matrix components protect bacteria from activated PMN? PMN were indeed shown to be activated, in vitro, by bacterial DNA and polysaccharides components such as alginate, the measured outcome being an increase of their respiratory burst (Pedersen et al., 1990; Alvarez et al., 2006; Fuxman Bass et al., 2008; Jensen et al., 2010). In vivo, P. aeruginosa biofilm interactions with PMN lead to up regulation of the QS-dependent effectors such as rhamnolipids which cause PMN lysis (Alhede et al., 2014). This shielding specific property of rhamnolipids is described in mouse models relying upon the intraperitoneal implant of pre-colonized silicone device (Van Gennip et al., 2012) or on biofilm development in the respiratory tract (Bjarnsholt et al., 2005; Jakobsen et al., 2012). In these models, QS mutants are unable to produce rhamnolipids and are rapidly phagocytosed and cleared by PMN.

Monocytes/macrophages are other key actors recruited and sensing both the other inflammatory cells -e.g., PMN -as well as bacteria agonists. Under homeostatic conditions, circulating macrophage/monocytes qualified as classical/intermediate ones are continuously recruited from the blood and either mature into macrophages or remain as monocytes within tissues (Sprangers et al., 2016; Jakubzick et al., 2017). In the skin, long- lived resident macrophages are also present and maintain their population by self-renewal (Sprangers et al., 2016; Jakubzick et al., 2017). As for PMN, microbe-specific molecules participate to the rapid emigration of classical/intermediate monocytes in the extravascular space which contribute to the resolution of inflammation by recognizing and phagocytosing bacteria and dying cells, and by producing ROS and reactive nitrogen species (RNI) (Sprangers et al., 2016; Jakubzick et al., 2017; Kashem et al., 2017). Once PMNs experience apoptotic death, a second wave of monocytes qualified as non- classical monocytes contribute to the resolution of inflammation and repair of the disrupted tissue (Sprangers et al., 2016).

Only a few studies (Prabhakara et al., 2011a; Thurlow et al., 2011; Hanke et al., 2012, 2013; Scherr et al., 2015; Corrado et al., 2016; Silva-Santana et al., 2016; see **Table 1**) assess the complex recruitment waves and networks of both PMN and monocyte subsets. Of note, an early wave of monocyte recruitment is sometimes predominant, as compared to PMN recruitment (Thurlow et al., 2011). Two main studies were conducted to monitor the abundance and functional features of monocyte/macrophage recruitment in vivo at specific time points after contact with biofilms. The biofilms' matrix components clearly modulate the functional properties of recruited cells. In a S. aureus pre-colonized catheter-based model, mobilized cells are mainly distant from the catheter and a non-significant proportion of F4/80<sup>+</sup> macrophages are able to invade the biofilms. Cells present deeply into the biofilm rapidly die, therefore preventing phagocytosis of biofilm bacteria to be otherwise exerted by


TABLE 1


 responses to biofilms

in vivo.

macrophages (Thurlow et al., 2011). A significant reduction of pro-inflammatory cytokines (IL-1β, TNF. . .) and chemokines (CXCL2, CCL2) production at the boundary of the biofilmcolonized tissue is also observed, associated with a reduced Nitric Oxide synthase (iNOS) induction and an increased arginase-1 expression. The authors conclude to a macrophage polarization toward a counter-inflammatory activated M2 phenotype and to fibrosis with MyD88- signaling as a major effector pathway regulating these two phenomena (Thurlow et al., 2011; Hanke and Kielian, 2012; Hanke et al., 2012). In a lung experiencing colonization with phage Pf- producing P. aeruginosa, Secor et al. (2017) also documented a M2 polarization profile. Taken globally, it appears that the macrophage M2 polarization and fibrosis do concur prolonging bacteria biofilm persistence, although the universal character of such responses remains to be assessed.

## THE MOUSE EAR PINNA DERMIS IMAGED BY INTRA-VITAL CONFOCAL MICROSCOPY AT STEADY AND NOT STEADY STATE

Intra-vital microscopy is increasingly used in different biomedical research fields to study dynamic processes at the cellular level in their specific tissue environment. Compared to classical methods such as ex vivo histology or flow cytometry, intra-vital confocal microscopy live imaging allows dynamic interactions to be captured once fluorescent reporters are expressed by both the microbes and the laboratory mouse cell lineages with which are engaged, more or less durably, dynamic interactions.

Several skin-related models were described in the literature to study the immunobiology of biofilm infections by using classical approaches (**Table 1**). In these models, the biofilmloaded cutaneous sites were the back or the flank of animals, which represent unsuitable sites for intravital microscopy of cutaneous innate immune responses.

The ear pinnae is one of the most studied appendage in which are delivered microorganisms, enabling the observation of the early and either transient or prolonged dynamic interactions with resident or recruited myeloid cells (Amino et al., 2006, 2007; Peters et al., 2008; Ng et al., 2011; Sumaria et al., 2011; Jain and Weninger, 2013; Tavares et al., 2013; Carneiro et al., 2017). As an imaging site, the ear presents several technical advantages such as the accessibility of the tissue, easy and fast protocols of preparation to perform imaging experiments and obtain reproducible results, and the possibility of imaging for long periods of time in blocks of 20–40 min (Li et al., 2012).

The mouse ear pinna appendage harbors a thin epidermis and an underlying dermis, respectively avascularized and highly vascularized, which contain a broad range of lympho-myeloid cells (Jain and Weninger, 2013). Using a multiphoton microscopy approach and quantitative flow cytometry, Tong et al. (2015) elaborated a 3D immune cell atlas of mouse skin and compared the ear pinnae, dorsal back, footpad, and tail skin. Langerhans cells and dendritic epidermal T cells are present in the epidermis, whereas the dermis mainly harbors in the upper dermis myeloid cells such as dendritic cells, mast cells as well as lymphoid cells (αβT cells, γδT cells, and group 2 Innate Lymphoid Cells), and mainly resident macrophages in the deeper dermis (Tay et al., 2014; Tong et al., 2015). All these dermis-located immune cells are included in a collagen- and elastin- rich and more or less hyaluronan- rich extracellular matrix. In term of cell numbers, the specificities of the ear pinna cutaneous site are the following ones: the total leukocyte density is high, with around 4800 cells per mm<sup>2</sup> and a majority of cells present in the dermis. Macrophages, dermal dendritic cells and mast cells represent the majority of ear dermal leukocytes with respectively around 6000 macrophages per mm<sup>3</sup> and more than 2000 dermal dendritic cells and mast cells per mm<sup>3</sup> (Tong et al., 2015). Of note, the ear pinna dermis presents two specificities regarding the presence of mast cells: their high prevalence and their perivascular localization, in close association to blood vessels, in contrast to dermal dendritic cells (Tong et al., 2015). By using intravital multi-photon microscopy, Ng et al. (2011) showed that PMNs are present in the ear dermis of naive mice, but in small numbers, and patrol, as do dermal dendritic cells. Both cell lineages are likely surveying the presence of either epidermisrestricted microbiota derived agonists that reach the dermis or endogenous agonists through the sensors displayed at the plasma membrane or within the macropinocytosis/endosomal machinery.

### THE MOUSE EAR SKIN MODEL TO STUDY THE DYNAMICS OF INNATE IMMUNE RESPONSES AGAINST PLANKTONIC OR SESSILE Staphylococcus aureus

By combining intra-vital confocal microscopy approach and the mouse ear pinna infection model, inflammatory responses against biofilms could be analyzed for the first time at the dynamic level in the tissue environment. The **Figure 1** shows the schematic workflow of methods proposed to characterize and compare the innate immune responses against S. aureus planktonic and sessile bacteria. Transgenic mice with fluorescent immune cells visible in the skin such as Lysozyme-EGFP (circulating PMNs and monocytes, dermal macrophages), CD11c-EYFP (dermal dendritic cells, epidermal langherans cells) and Mcpt5-Cre+R26Y+ (dermal mast cells) mouse strains have been selected (Faust et al., 2000; Lindquist et al., 2004; Dudeck et al., 2011). The conditions to prepare bacteria inoculum will be set up. The planktonic inoculum will be obtained from an overnight culture of fluorescent S. aureus Lyo-S2 strain in Trypticase Soja broth (Marquès et al., 2015). Biofilms will be generated from a planktonic culture incubated at 37◦C under static conditions. After 24 h of incubation, biofilms will be gently collected (A). The first series of experiments will be performed

with Lysozyme-EGFP mice inoculated into the ear tissue with the same number of CFU of either planktonic or sessile bacteria (B). Inoculum will be micro-injected in two injection points in the dermis of the ear pinna tissue with a nanofil syringe (Mac-Daniel et al., 2016). At early time-points, mice will be anesthetized and the cellular recruitment will be analyzed by confocal microscopy at the injection points (C1a and C1b). The behavior of recruited or resident innate immune cells (moving speed, trajectory, distance covered) will be analyzed, as well as their specific interactions with bacteria. In parallel, additional groups of mice will be inoculated with planktonic or biofilm bacteria to perform quantitative analyzes over 14 days in the ear tissue and the cutaneous draining lymph node (auricular lymph node), i.e., determination of the phenotype of recruited cells (C2), of the cytokine levels (C3), and counting of bacteria in the target tissues (C4). The inflammation visible to the eye and macroscopic cutaneous lesions (skin necrosis) will be observed and compared.

## CONCLUSION

The objective of the review was to highlight the requirement of developing new in vivo models to analyze, at the early stage of infection, the dynamics of innate immune responses against S. aureus planktonic or sessile bacteria. A new experimental approach in the field combining the mouse ear pinna model and the intravital imaging approach is proposed. The long term objectives are to use this model as a pre-clinical model to test necessary new therapeutic approaches targeting the host immune system, as proposed initially by Hanke et al. (2013).

#### AUTHOR CONTRIBUTIONS

fmicb-08-02309 November 18, 2017 Time: 15:47 # 7

CF: intellectual contribution to the work; EB: intellectual contribution to the work; GM: substantial, direct and intellectual contribution to the work; PG: substantial,

#### REFERENCES


direct and major intellectual contribution to the workcorresponding author. All authors listed approved the work for publication.

#### ACKNOWLEDGMENT

The authors wish to acknowledge the contribution of Caroline Vachias at the Confocal Microscopy Facility ICCF (Imagerie Confocale Clermont-Ferrand) at Université Clermont Auvergne.


during early innate response to Pseudomonas aeruginosa lung infection. Clin. Exp. Immunol. 137, 478–485. doi: 10.1111/j.1365-2249.2004.02554.x



cells recruited into the site of infection. Shock 20, 503–510. doi: 10.1097/01.shk. 0000093542.78705.e3


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Forestier, Billard, Milon and Gueirard. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Aaron Lerner1,2\*, Torsten Matthias2 and Rustam Aminov3,4*

*1B. Rappaport School of Medicine, Technion-Israel Institute of Technology, Haifa, Israel, 2AESKU.KIPP Institute, Wendelsheim, Germany, 3 Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, 4School of Medicine & Dentistry, University of Aberdeen, Aberdeen, United Kingdom*

Many essential functions of the human body are dependent on the symbiotic microbiota, which is present at especially high numbers and diversity in the gut. This intricate host–microbe relationship is a result of the long-term coevolution between the two. While the inheritance of mutational changes in the host evolution is almost exclusively vertical, the main mechanism of bacterial evolution is horizontal gene exchange. The gut conditions, with stable temperature, continuous food supply, constant physicochemical conditions, extremely high concentration of microbial cells and phages, and plenty of opportunities for conjugation on the surfaces of food particles and host tissues, represent one of the most favorable ecological niches for horizontal gene exchange. Thus, the gut microbial system genetically is very dynamic and capable of rapid response, at the genetic level, to selection, for example, by antibiotics. There are many other factors to which the microbiota may dynamically respond including lifestyle, therapy, diet, refined food, food additives, consumption of pre- and probiotics, and many others. The impact of the changing selective pressures on gut microbiota, however, is poorly understood. Presumably, the gut microbiome responds to these changes by genetic restructuring of gut populations, driven mainly *via* horizontal gene exchange. Thus, our main goal is to reveal the role played by horizontal gene exchange in the changing landscape of the gastrointestinal microbiome and potential effect of these changes on human health in general and autoimmune diseases in particular.

#### *Edited by:*

*Juarez Antonio Simões Quaresma, Universidade Federal do Pará, Brazil*

#### *Reviewed by:*

*Sheryl S. Justice, The Ohio State University Columbus, United States John Peter Konhilas, University of Arizona, United States*

#### *\*Correspondence:*

*Aaron Lerner aaronlerner1948@gmail.com*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 27 July 2017 Accepted: 09 November 2017 Published: 27 November 2017*

#### *Citation:*

*Lerner A, Matthias T and Aminov R (2017) Potential Effects of Horizontal Gene Exchange in the Human Gut. Front. Immunol. 8:1630. doi: 10.3389/fimmu.2017.01630*

Keywords: probiotics, microbiome, dysbiome, horizontal gene transfer, intestine, gut, biofilm, environment

#### INTRODUCTION

Single-celled microorganisms have shaped our planet for several billions of years before the arrival of multicellular organisms. The appearance of the humankind is a relatively recent evolutionary event but, within a very short time on the evolutionary scale, it has become a very powerful force on the planetary scale. It extensively changes the biosphere, consumes copious natural resources, affects global climate, and engages in wars that are driven by competition for natural resources, while the human population experiences an explosive growth, destroying natural ecosystems to sustain this growth. Interaction between humans and commensal microbiota shaped both of them, in a mutual and bidirectional way, thus benefiting microbes and us (1). Given the extensive influence of microorganisms across the entire biosphere, and the microbiota on human health, the gut–microbiome integrity is of prime significance to human health, disease prevention, and survival. In this regard, our body contains a "second genome," allowing to cohabit with the human one to form a steady equilibrium for the two counterparts and for long-term mutual survival.

**Abbreviations:** HGT, horizontal gene transfer; MGEs, mobile genetic elements.

In parallel with the evolutionary selected adaptation mechanisms on both side of interaction, the new developments are entering this interaction to support and provide food for the ever-growing human population. These are the contemporary intensive agricultural systems with the widespread use of herbicides, insecticides, fungicides, fumigants, desiccants, harvest aids, antimicrobials, growth regulators, and many other substances. Other developments include the use of genetically manipulated microorganisms, plants, and animals as well as new nutrients, new food technologies, engineered microbial delivery systems, and various food additives.

Due to the close relationship and intimate cross talks between the human cells and gut microbiota, the effects of the latter on human health are substantial. Thus, the aim of the present manuscript is to update on potential outcomes of the microbiome, which is changing as a result of changing lifestyle, on human health in general and on chronic diseases in particular. It is not the aim of the present review to discuss horizontal gene transfer (HGT) in pathogens leading to the exchange of virulence genes that could contribute to the emergence of new "superbugs" (2). This review deals with the potential effects of our changing lifestyle on the evolutionary equilibrium with the "normal" microbiota, which had been established during the previous coevolution and cospeciation (3). The focus of this review is on horizontal gene exchange, which has been demonstrated for within the three domains of life, eukaryotes, bacteria, and archaea (4), with a particular emphasis on human host–microbe interaction. Finally, a potential role of HGT in the gut ecosystem in regard to human health and chronic disease induction, is discussed. While there are many other excellent reviews elsewhere discussing HGT in the gut, the main focus of these reviews is limited to the HGT events among the intestinal microbiota. In this review, broader implications of HGT are discussed that go beyond the gut microbiota boundaries and involve the host side as well.

#### HGT: DEFINITION, CHARACTERISTICS, AND MECHANISMS

Horizontal gene transfer is the lateral exchange of genes between unicellular and/or multicellular organisms. In contrast to the vertical gene transfer, i.e., between generations, HGT enables the transfer of genetic sequences between remote species mediated usually by transformation, transduction, and conjugal transfer or with specific gene transfer agents (5, 6).

Horizontal gene transfer has been demonstrated for almost all phylogenetic groups in all three domains of life as a crucial factor in the evolution of various organisms (7–9), including plants (10, 11), viruses (12), archaea (13), fungi (14, 15), and animals (16). While evolution of the Eukaryotes is largely driven by vertical inheritance, the predominant form of evolution among the bacteria and archaea is HGT, the rate of which is comparable to the point mutation's rate, surpassing the gene duplication rate (17). Thus, for these two domains, HGT is an essential way for genome diversification and novel function procurement to survive under the pressure of natural selection and to reproduce. Therefore, HGT is a main driver of microbial evolution and ecology. Bacteria have developed multiple natural genetic tools to exchange genetic material between strains, species, genera, and even higher taxa. While the phenomenon of HGT can be encountered in virtually any ecosystem, the focus of the present review is HGT in the human gut. This is complemented by a potential contribution of changing dietary and lifestyle habits to HGT.

One of the well-documented HGT events that the mankind has experienced in its history is the consequence of the rampant usage of antibiotics, which resulted in the widespread dissemination of antibiotic resistance genes among many bacteria (18). These mechanisms allowed them to acquire protection against the pressure of antibiotic selection to afford successful survival and proliferation. In general, HGT drives the enlargement of protein families in bacteria and archaea to deal with the environmental and anthropogenic impacts imposed upon them (19).

There are many mechanisms of HGT reviewed elsewhere (5, 6, 20–22). In brief, these mechanisms include the following.

**Transformation:** Genetic modification of the cell due to the uptake of foreign genetic material. Natural competence and transformation are relatively common in bacteria and, to some extent, in archaea, but not in eukaryotes. Artificial transformation is often used in recombinant technology for research, industrial or medical purposes.

**Transduction:** When microbial DNA is transferred from one bacterium to another one by a viral intermediate.

**Conjugation:** The transfer of DNA *via* a conjugative element [plasmid, transposon, ICE, or other mobile genetic elements (MGEs)] from a donor to a recipient cell during cell-to-cell touch (23, 24).

**Gene Transfer Agents:** Virus-like elements encoded by the host found in Alphaproteobacteria, order Rhodobacterales, and some other bacteria.

It is generally accepted that HGT *via* transduction, conjugation, and gene transfer agents is more efficient than transformation by naked DNA because in the former mechanisms DNA is protected from degradation. By means of HGT, entire genes and functional sequences can be incorporated into the genome of a recipient. Although the best-known examples are among the bacteria and archaea, they can be found in the Eukarya as well, including primates and humans (25). Thus, HGT had occurred and continues to occur, on a considerable scale, in all three domains of life, and is likely to be one of the important factors that have contributed to the diversification during evolution. Unlike evolution *via* gene duplications and mutations, which are slow and progressive processes, HGT allows a rapid acquisition of a new function important for species to pass through the natural selection barrier and successfully reproduce.

#### HGT IN THE GUT

Since the gut is a niche mainly colonized by a plethora of microbial species, it is logical to presume that the genomes of methanogenic archaea in the intestine have acquired their ability to survive and proliferate in this environment through interdomain HGT from the microbial counterpart that dominate this niche. Recently, contribution of HGT to the gene repertoire in a gut-adapted commensal methanogen *Methanobrevibacter smithii*, which is the richest archaeon in the human gross intestine, has been evaluated (26). A phylogenetic tree-based genome-wide survey of putative genes, presumably acquired as a result of HGT, established that over 15% of the coding sequences in *M. smithii* could be inferred as of bacterial origin. Laterally acquired genes largely contribute to surface functions and encode glycosyltransferases and adhesin-like proteins, which also can act as virulence factors in pathogens. In addition, several important ABC transporters, especially metal transporters are potentially of microbial origin. Metals such as zinc, for example, are important for bacterial growth, and there is a strong competition for it among intestinal microbiota as well (27). Thus, the microbial genes acquired by this archaeon contributed to the host adaptation by permitting an extended variety of surface structures and enhancing the efficiency of metal ion uptake in the competitive gut niche. Taken together, adaptation of *M. smithii* to the niche involved the acquisition of bacterial genes into its genome to adjust its lifestyle.

A comparative study of fecal samples from mono and dizygotic twins revealed that the pan-genome of *M. smithii* "contains 987 genes conserved in all strains, and 1,860 variably represented genes" (28). Strains from monozygotic and dizygotic twins had a comparable degree of shared genes and SNPs and were significantly more similar than strains isolated from mothers or members of their families. The 101 adhesin-like proteins in the pan-genome (45 ± 6 per strain) exhibited strain-specific differences in expression and responsiveness to format. The authors hypothesized that *M. smithii* strains use their different repertoires of adhesin-like proteins to create diversity in their metabolic niches, by permitting them to create syntrophic relationships with bacterial partners with differential metabolic capacities and patterns of co-occurrence. It is generally accepted that the core genome genes are less prone to HGT than that of the auxiliary genome. Thus, the majority of genes in the pangenome of *M. smithii* are laterally circulating among the strains of this species.

More information on the magnitude of HGT operating in the human intestine came from the study of Zaneveld et al. (29). They revealed that enteric-adapted genomes are more comparable in gene content at a given evolutionary distance than non-gut genomes. Thus, common functional needs or magnified HGT causes similarities in genes within the gut compartment. Notably, niche specialization at short phylogenetic distances is also important in the mammalian intestine. More recently, a hypothesis that the animal gut is a melting pot for HGT has been summarized in two mini-reviews that surveyed HGT events in the mammalian gut and the role of HGT in the long-term adaptation of microbes to the intestinal milieu (30, 31). The authors concluded that the mammalian intestine is "a melting pot of genetic exchange, resulting in the large extent of HGT occurrence" (30).

Most of our knowledge of HGT has been obtained from the *in vitro* studies. Where the features are substantially different, especially in the gut, due to extremely dense and diverse microbiota. The plethora of functions include the suppression of settlement by pathobionts, degradation of dietary and *in situ*produced components, production of nutritional factors, modulating and maintaining a functional mucosal immunity, supporting inter epithelial tight junction integrity, and contribution to intestinal epithelial homeostasis. Recently, evidence of increased DNA exchange among the Bacteroidales species within the human intestine has been reported (32). Genes that are extensively exchanged among these species encode proteins involved in fitness and multiple cycles-like alterations of gene expression. Fimbriae components may enhance attachment, utilization of new substrates increase the nutritional base, and secretion of antimicrobial molecules may confer a competitive advantage within the ecological niche. The genetic content of the "transferome" suggests that the gene transfer from the successfully adapted members of an ecosystem confers useful properties to the recipients, increasing their fitness and conferring them a competitive edge within the gut microbial ecosystem.

Humans and cultivated life stoke are the main consumers of antibiotics, thus enhancing HGT processes (24, 33–36). In fact, the gut microbiota represents a major habitat of antibiotic resistance genes (37–40), thus also called "gut resistome," in the frame of the "gut mobilome" (41–43). Interestingly and somewhat alarmingly, high frequencies of HGT in infants' meconium and early fecal samples have been recently reported (44). Antibiotic-susceptible commensal bacteria may acquire resistance to antibiotics *via* mutations in target genes or the acquirement of resistance genes by HGT, mainly by the transfer mediated by MGEs. MGE-mediated transfer of genetic cargo from one organism to another greatly contributes to the dissemination of antibiotic resistance genes, because it can take place between closely or non-related species and in diversified environments, including the animal and human gut (45–50).

Notably, the taxonomically distant prokaryotes of intestinal microbiome may share reservoir of closely related antimicrobial resistance genes (51). The role of bacteriophages, plasmids, conjugative transposons, and integrons in transfer of genes by pathogenic human enteric pathobionts and subsequent acquisition of pathogenicity also pointed to HGT as an important step in the expansion of virulence traits and antibiotic resistance (52–54). More so, existence of ancient and possibly, recent transfers of antibiotic resistance genes from antibioticproducing actinobacteria to pathogenic proteobacteria was lately described (55).

The *in vivo* gut luminal milieu may encourage the transfer incidence and enhance the steady inheritance of MGEs, even in the lack of the antibiotic influence (56, 57). The soil microcosm exploration, represented by *Escherichia coli* as a donor with a genetically large conjugative plasmid RP4luc and the presence or absence of earthworms, supplied direct evidence that the gut passage is a prerequisite for a plasmid transfer to soil microbiota (58). Surprisingly, the plasmid transfer rates were even higher than can be achieved in filter mating experiments, suggesting that the HGT rhythm in nature could be higher than the laboratory assessed. If MGEs from soil can penetrate the earthworm gut, then they can lodge in the creature's intestine that is next in the food chain, for example, moles and birds.

Additional example of the enhanced HGT under the gut environment has been demonstrated in ciliates (59). Ciliates are common in many aquatic ecosystems and in the rumen. Their food vacuoles are formed by phagocytosis and follow their path through the cell, thus, imitating a primitive enteric tract. Ciliates may enhance the rhythm of conjugal transfer between *E. coli* strains by two orders of magnitude, and the suggested pathway involved is the accretion of bacteria in vesicles allowing extensive HGT (59). This avenue may enhance the diffusion of antibiotic resistance in bacterial communities (60). The insect intestine can also be considered as a hot spot for HGT. For example, "the rates of conjugative plasmid transfer between *Salmonella enterica* Newport and *E. coli* in the intestine of the lesser mealworm beetle are by two orders of magnitude higher compared with filter mating" (61).

Conditions in the intestinal tract are highly expedient to HGT. The continuous flow of food, high density of microbiota, stable and optimal temperature, the formation of biofilm structures, and the vast diversity of enteric microbiota provide ideal conditions for HGT to occur in the human gut (62). Moreover, recent studies have disclosed the mechanisms of host– microbe molecular crosstalk that may enhance to the magnitude of HGT in the gut. It appears that the bacteria can perceive and react to host signals. For example, microbial sensing and responding to the level of host stress hormones is extensively reported (63–66). Even in the host–bacteria stress cross talks, a genetic material can be involved, expressed by the increased conjugative gene transfer between gut bacteria (67). It was shown *in vitro*, that "the physiological concentrations of norepinephrine enhanced the transfer of a conjugative plasmid from a clinical strain of *Salmonella* sp. to an *E. coli* recipient. Notably, the adrenergic receptor antagonists deprived the stimulatory effect of norepinephrine on conjugation. These signals of host stress may possibly affect HGT under the *in vivo* conditions as well" (5). In our recent work involving mice mono-associated with a human gut symbiont *Roseburia hominis*, we have discovered that a number of genes involved in HGT are upregulated in the bacterium in response to the intestinal environment (68). Finally, viral communities are abundant in the human intestine and viral sequences are transferred from bacterial cells to eukaryotic hosts (69). Not only phages are moving between microbes during antibiotic therapy but also whole phage communities are exchanged between human subjects during fecal microbial transplantation (69).

#### HGT IN BIOFILMS: POTENTIAL APPLICABILITY TO HUMAN GUT

Biofilm formation represents one of the main basic bacterial strategies for growth and survival in nature and disease (70). Biofilms are communities of microbes embedded in matrices composed of extracellular polymeric substance, and they were implicated for both the healthy and disease states of the host. Biofilm habitats are common in many biological ecosystems. The majority of microbiota found in natural, clinical, and industrial settings persist in association with surfaces and not in the planktonic state. They are usually found in many ecosystems including the teeth of humans and animals, and in the intestinal lumen (71).

The immune system recognizes many different bacterial patterns, but these components can be camouflaged in the biofilm mode of life. Transition from the planktonic to the biofilm-associated state induces bacterial production of small molecules, which can increase inflammation, induce cell death and necrosis and may potentially enhance posttranslational modification of naïve proteins to immunogenic ones, thus provoking undesirable immune reactions (72, 73). While planktonic cells are readily cleared, cells in biofilms are much less susceptible to clearance by neutrophils and macrophages. Moreover, in the presence of these host cells, biofilm formation is enhanced, and the components of the host immune system can be incorporated into the extracellular polymeric substance matrix (71). In particular, biofilm formation in the gut is facilitated by human secretory immunoglobulin A molecules (74). And the biofilm environment is well known for the HGTpromoting properties (5).

Cooperative phenotypes are important for the functioning of bacterial communities in many contexts, including syntrophy, linking *via* quorum sensing, biofilm formation, exchange of antibiotic resistance, and progress of polymicrobial infections. The human gut accommodates a dense and diverse microbial population critical to health, yet, cooperation within this important ecosystem, which has evolved over a long coevolutionary process, is poorly understood (75). Based on the above mutual, bidirectional, cooperative, and fine-tuned equilibrium between us and the microbiota, several questions may arise, such as, whether there is a risk associated with perturbation of this equilibrium, or, whether probiotics and the use of recombinant enzymes in foods may affect HGT in the human gut, or, if the changed microbial/microbial product profile could affect human health.

#### PROBIOTICS

The first person to bring the idea of colonization of the intestinal tract by advantageous microbes was Élie Metchnikoff, who observed, more than 100 years ago, that the largest percentage of centenarians live in Bulgaria. He related the phenomenon to the increased consumption of milk fermentation products, in particular Bulgarian yogurt. He forwarded the hypothesis that aging is caused by toxic bacteria in the gut, and he encouraged the use of Bulgarian yogurt and its principal component, *Lactobacillus delbrueckii* subsp. *bulgaricus*, to prevent this toxicity (76). The theory of aging did not hold but the encouraged use of lactic acid producing bacteria as helpful for health has grown enormously and, in fact, the global probiotics market extended to USD 27.9 billion in 2011 and is anticipated to reach USD 44.9 billion in 2018 (77).

The rostral definition of probiotics introduced by Lilly and Stillwell (78), however, had another meaning than that of the Lerner et al. HGT in the Gut

Metchnikoff 's, and it has been allocated to the "protozoa, in particular to the growth promotion of *Tetrahymena pyriformis* in response to a factor produced by *Colpidium campylum*. After several refinements, the definition has been assembled with the original idea of Metchnikoff," and the actual definition of probiotics is "live microbial feed supplement which beneficially affects the host animal by improving its intestinal microbial balance" (79).

Frequently used probiotics include lactic acid producing bacteria, particularly lactobacilli, bifidobacteria, lactococci, and streptococci. Less often used are yeasts, bacilli, and nonpathogenic *E. coli* strains. The majority of probiotics, therefore, are facultative anaerobes, and their main effects are mediated through the secretion of lactate, and other short chain fatty acids that may inhibit the pathobionts and affect the communities of the enteric microbiome. Also, it is presumed that the intake of probiotics may modulate the immune system.

#### HGT POTENTIAL OF PROBIOTICS

Probiotics and starter cultures have a generally regarded as safe status. The stature, however, had been acquired well before the recent safety concerns such as the carriage of antibiotic resistance genes on MGEs have been raised. From the safety standpoint, it is necessary to distinguish between the intrinsic resistance, which may constitute a normal physiology and metabolism (for example, inability of an antibiotic to enter the cell because of a particular cell wall structure or formation of a thick capsule) and the transferable antibiotic resistance genes (38). Phenotypic screening of probiotics in dietary supplements, for example, revealed a substantial level of antibiotic resistance, in particular, toward vancomycin, streptomycin, aztreonam, gentamicin, and/or ciprofloxacin antibiotics (80). Suggestions for risk assessment of antibiotic resistance in probiotic supplements obviously include broader genetic screening as well as the use of computational simulations, dynamic imaging, and functional genetics (81). Genes encoding virulence factors are also of concern, especially if they are located on MGEs and can be exchanged between the probiotic and commensal strains (82).

There is a growing body of literature on interspecies genetic exchange, underlining the importance of gene acquisition/loss within or between various probiotic strains. Examples of HGT among the probiotic strains have been reported for *Lactobacillus rhamnosus* (83), *Lactobacillus gasseri* (84), *Lactobacillus paracasei* (85), *Lactobacillus reuteri* (86–88), *Lactobacillus plantarum* (88), and some other probiotics. As mentioned earlier, the market for probiotics (and of course the consumption of them) is growing rapidly. Also, the gastrointestinal tract is the hot spot for the HGT events. Would the ingestion of probiotic cultures, which may act as donors or recipients, therefore increase the antibiotic resistance gene pool in the enteric ecosystem? Some authors suggest that there is a gene flux from Gram-positive cocci such as enterococci or streptococci to Gram-negative bacteria, with genes encoding for streptogramin resistance as an example (89). Being lactic acid producing bacteria, they contain plasmids containing genes conferring resistance to tetracycline, erythromycin, chloramphenicol, lincosamide, macrolides, streptomycin, and streptogramins (90).

*Leuconostoc* and *Pediococcus* species can serve as recipients for the broad host range antibiotic resistance plasmids from *Lactococcus* species (91). Conjugation transfer from enterococci to lactobacilli and lactococci can take place in the gastrointestinal tract of animals as well as *in vitro*; the transfer frequencies to lactobacilli, however, are pretty low (92). A recent systemic review has concluded that there is not enough evidence for the impact of probiotics on the stool microbiota composition in healthy human being (93). However, while the bacterial composition is not affected, the gene pool exchange *via* HGT between the probiotic and endogenous strains could happen. For example, the transfer of a tetracycline resistance gene from probiotic *L. reuteri* to bacteria in the human gut has been reported (94). To deal with this problem, Rosander et al. (95) removed antibiotic resistance gene-carrying plasmids from the commercial strain of *L. reuteri* ATCC 55730. Besides the antibiotic resistance genes, the danger of amplification of which is well recognized, other factors of concern could be toxins and virulence factors. Most recently, antibiotic resistance gene prevalence was described in the gut commensal bifidobacteria community, which is widely used in food processing and as probiotics (96). Notably, the acquisition of antibiotic resistance genes is age dependent, with a substantial increase during the first year of life.

Fittipaldi and others (97) reviewed more than 70 virulent factors in the zoonotic agent *Streptococcus suis*. Interestingly, the enzyme microbial transglutaminase, which is extensively used in the processed food industries (98–100), has been recently described as a virulence factor in this bacterium (101). Finally, the resistome present in probiotics is most probably underestimated and the double-edged sword effects of probiotics such as health benefits vs antibiotic resistance gene dissemination to the gut microbiome have to be carefully contemplated (102). The safety issues associated with the antibiotic resistance gene pool in probiotics have also been noted earlier (103, 104). Most recently, we explored the bidirectional effects of the human gut symbiont *R. hominis* in tissue models and in mono-associated mice (68). This interaction results in the concordant gene expression patterns at both sides. A set of genes considered to be important for bacterial–host colonization was induced in the bacterium. Even more interestingly, the host environment strongly induced genes involved in HGT, chemotaxis, and motility. The host responds by the enhanced expression of genes involved in innate immunity, gut barrier functions, and by increased Treg population expansion (68). Probiotics can potentially respond in a similar way, with the increased expression of genes involved in HGT, upon exposure to the gut environment.

#### TRANSGENIC FOOD PRODUCTS

The topics of potential risks associated with the intake of genetically modified organisms or the so called "transgenic organism," by humans and animals have been investigated in several of feeding trials. Interestingly, there is a shortage of support that DNA of transgenic plants, can be taken up by enteric microbiome or enter the organs other than the gastrointestinal lumen. "Neither tiny segments of transgenic DNA nor immunogenic fragments of transgenic protein were found in loin muscle samples from pigs fed a diet containing Roundup Ready soybean meal" (105). "Evaluation of the survival of transgenic plant DNA in the human gut deducted that gene transfer did not occur during the feeding experiment involving genetically modified soya" (106). No traces of endogenous soybean DNA could be traced in muscle samples of rats fed soybean meal from roundup ready or conventional soybeans (107). Likewise, no signs of transgenic DNA were traced in the milk of cows fed corn silage from an herbicide-tolerant genetically engineered products variety (108). Plasmid and genomic DNA from genetically modified plants were used in *in vitro* and *in vivo* transformation studies, but no detectable transfer of DNA was found (109). Attempts to detect DNA exchange from transgenic plants to bacteria in the gut of the tobacco hornworm (110) or bees also failed (111).

#### INDUSTRIALLY PROCESSED FOOD

There are a substantial number of microorganisms that are used in food production. It is estimated that about a quarter of all food production such as sausages, ham, cheese, and dairy products involves bacterial fermentation processes using lactic acid bacteria and other microbial and fungal strains. Here, the possibility of HGT among the bacteria used is more plausible. For example, when *Leuconostoc* and *Weissella* species, which are used as a mixed culture for aroma generation in traditional Italian and Spanish fermented cheese, were analyzed for antibiotic susceptibility, evidence for HGT was detected, demonstrating interspecies lateral gene transfer (48). In conjugation experiments performed both *in vitro* and in cheese, the transfer of erythromycin resistance between *Leuconostoc mesenteroides* and *Enterococcus faecalis* was detected.

While antibiotic resistant pathogens portray a straight threat to human and animal health due to difficulties of their uprooting (2), resistance among commensal bacteria constitutes an indirect hazard as being a reservoir of antibiotic resistance genes that can be shifted to pathogens *via* HGT. These reservoirs, therefore, represent a potential source for the dissemination of transmittable genes in bacterial ecosystems, including foodstuff. Food commensal microbes are a potential important avenue for HGT (112). For example, *Lactococcus lactis* that is associated with the Spanish traditional raw milk product was reported to transfer genetic material to lactococci and enterococci (113). Thus, food-borne commensal bacteria could be a potential source for the transfer of antimicrobial resistance genes, which is an important threat for public health (48, 114).

A recent study, which investigated samples from patients and healthy humans, farm animals and food, revealed a reduced dissemination of genes encoding extended-spectrum β-lactamase (ESBL)/pAmpC and plasmids bearing these genes from foods and farm animals to healthy humans and patients (115). Poultry and chicken meat represent a potential reservoir and a route for dissemination of these genes to humans. Although no evidence for the clonal spread of ESBL/pAmpC-producing *E. coli* from farm animals or foods to humans was found, ESBL/pAmpCproducing *E. coli* with same resistance genes and plasmids were present in farm animals, foods, and humans, suggesting HGT as a prevalent mechanism for the dissemination of antibiotic resistance genes.

Diet-driven convergence in gut microbiome functions has been investigated with 33 mammalian species and 18 humans (116). Microbiota adaptation to diet was reproducible across different mammalian lineages. Functional microbiota genes could be predicted from bacterial species assemblages, thus providing insights into the mechanisms driving the evolution of the enteric microbiome at the supra-organismal level. In the context of this review, the question is what was the contribution of horizontal gene exchange in the gut to the diet-derived convergence reported above? Redundancy in the repertoire of functional genes, for example, for complex carbohydrate utilization among various taxonomic groups suggests extensive exchange of these genes in the gut microcosm.

An entire different aspect of the industrial food production is food additives, which have been recently implicated as potential drivers of autoimmunity as affecting the integrity of intestinal epithelium tight junctions (98). Microbial transglutaminases, which are important for the survival of bacteria in nature, are extensively used in industrial food production as cross-linking agents of proteins, thus improving the texture and appearance of food products (73). They belong to the family of transglutaminases, which are functionally close to the endogenous tissue transglutaminases implicated as autoantigens in celiac disease. It has been suggested that microbial transglutaminases may be involved in the development of celiac disease (99). They have been demonstrated recently as immunogenic in celiac disease patients, representing a new marker that reflects the intestinal damage (100). In pathogens such as *S. suis*, the enzyme is a virulence factor, conferring antiphagocytic properties (101). It is not clear whether microbial transglutaminases in other bacteria may play a role in virulence. Given the extensive HGT events in the gut ecosystem and potential to enter human cells, it can participate in celiac disease progression (99) or in some other autoimmune diseases.

#### LIVESTOCK AND COMPANION ANIMALS

The issue of possible zoonotic spread of antimicrobial-resistant bacteria and the corresponding genes is complex. Ewers and others (117) reviewed data available for *E. coli* isolates from livestock and companion animals. Most of these studies analyzed the chromosomal setting, with multilocus sequence typing, and the plasmid (episomal) ESBL/AmpC genes. In contrast to the diversity of episomal ESBL/AmpC types, isolates from human and animals mainly shared identical sequence types, suggesting gene transmission pathways of zoonotic bacteria, including multiresistant ESBL-producing *E. coli* or parallel microevolution. Another work from this group revealed that urban rats might be significant in regard to the human health because of high carriage rates of *E. coli* strains that have genotypes resembling those that circulate in human patients and thus can to be regarded as zoonotic (118). There is also direct evidence that the high carriage rate of "ESBL/AmpC-producing *E. coli* by poultry at broiler farms results in a high prevalence of ESBL/AmpC-producing *E. coli* in farmers" handling them (119).

Extended-spectrum cephalosporin-resistant Enterobacteriaceae (ESCRE) are found in humans and animals and in various environments. Circulation of these bacteria among the ecological compartments, therefore, is complex, due to multiple reservoirs and different transmission routes. Moreover, Enterobacteriaceae, including ESCRE, can be a part of the normal gut microbiota of healthy humans and animals, including dogs. Most recently, evidence for the household transfer of ESBL/AmpC-producing Enterobacteriaceae between humans and dogs has been reported (120). Of note, ESCRE are also found in food-producing animals worldwide and in meat products, potentially spreading from animals to humans through the food chain. ESCRE are also found in healthy and diseased companion animals such as dogs, and the potential zoonotic risk of this is emphasized (121).

But the risk is not limited only to dogs. Various antibiotic resistance genes and MGEs in the microbiota of many livestock and companion and wild animals have been described. One of the important factors contributing to their broad dissemination could be the selective pressure of antibiotics widely used in agriculture, especially in food-producing animals (36). A recent systematic review has tried to answer the question whether human extraintestinal *E. coli* infections resistant to expandedspectrum cephalosporins arise from food-producing animals (122). Overall, there is the evidence that a proportion of human extraintestinal ESCR *E. coli* infections could exist, with poultry as a prime suspect. In contradiction to this, no evidence of clonal spread of ESBL/pAmpC-producing *E. coli* from farm animals or foods to humans has been found (115). However, ESBL/ pAmpC-producing *E. coli* with identical genetic sequences and plasmids are present in farm animals, foods and humans thus suggesting a key role of HGT in circulation of ESBL/pAmpC among the microbiota of different ecological compartments. A recent large-scale analysis has revealed that the mobile resistome transfer network is shared between the human and animal gut microbiomes as well as by various human pathogens (49). Thus, the MGE-mediated transfer of antibiotic resistance genes is a more prevalent mechanism than the clonal dissemination of antibiotic resistance.

#### SYNTHETIC BIOLOGY PRODUCTS

Synthetic biology tends to understand, reorganize, and control biological constituents to make functional units. The iterative process of designing and testing gene circuits has the potential to produce extensive valuable information into the mechanisms of the underlying functions of cells. It appears that, synthetic biology converges with disciplines such as systems biology and even classical cell biology, representing predictability to gene expression, cell biology and cellular signaling pathways. It can be simply summarized as "creating to understand." This novel strategy uses the accumulating knowledge in the prokaryotic genetics to impact the eukaryote biological behavior (123).

Well known is the fact that the majority of microbes dwelling inside the human body are non-pathogenic and some of them can be turned, after appropriate engineering, into "smart" live therapies with defined capacities for the treatment of various morbid conditions. The use of engineered prokaryotic organisms to treat diverse human pathologies is constantly expanding. Some of diseases targeted are inflammatory bowel disease, autoimmune disorders, cancer, metabolic syndrome and obesity, neuropsychiatric disease, bacterial and viral infections as well as the development of vaccines against infectious diseases (124–126). In fact, synthetic biology strategies use microbial or viral reprogramming as human therapeutics, including novel means for strict bio containment. Another potential aspect of HGT is based on data demonstrating that DNA may be transferred between somatic cells *via* the incorporation of apoptotic bodies (127). The procedure allows the transfer of viral genes that have been merged with the genome in a receptor-independent fashion. Transferred DNA is multiplied and spreading inside daughter cells, in cell that have an inactivated DNA response which may change aberrant cell progression.

Based on the above, one can foresee several new therapeutic strategies, based on synthetic biology (126). "Living pills" are the engineered bacteria having a potential to deliver, enhance, or act as therapeutic modalities to treat various diseases (128). Such a pill, harboring the engineered microbe(s), with high enzymatic activity, high output of metabolic or proteomic product and easily deliverable, might be beneficial (73). Ongoing efforts are focused on adaptation of commensal bacteria for remodeling the gut ecosystem toward disease-treating condition (129). Engineered viruses can be adopted to selectively destroy pathogenic bacteria (130, 131). The synthetic biology can also represent a new avenue for HGT between bacteria, viruses, and somatic cells. We are not at this stage as yet, but the potential risks have to be carefully evaluated beforehand by regulatory bodies overseeing food safety such as the FDA, EFSA, and other national agencies at respective countries.

#### HGT IN EUKARYOTES

The increasing body of evidence is accumulating on the HGT events in more complex eukaryotic organisms as well, which allowed them to acquire novel beneficial traits. In arthropods and nematodes, for example, the laterally acquired genes allowed to surpass plant defense mechanisms or broaden the nutritional base by acquiring the ability to degrade plant material as well as metabolizing a host-derived substrate (132–134). Traces of such lateral genetic transfer events can be found in genomes of many eukaryotes such as porifera, cnidaria, rotifers, nematodes, insects, arachnids, crustaceans, urochordates, and vertebrates (25, 135, 136). Transfer events among divergent species were mediated by various MGEs such as retroviruses, transposons and terminalrepeat retro transposons and preserved in the respective genomes.

Crisp and others (25) have recently described between ten and hundreds of potentially foreign genes that are expressed in primates, flies, nematodes, and humans. The majority of these genes are involved in metabolic pathways, suggesting that these laterally transferred genes contributed to the biochemical diversity during the early stage of invertebrate and vertebrate evolution. The authors suggested that a total of 145 human genes could be of bacterial origin, transferred to humans *via* HGT. Alignment of human genome against 53 vertebrate genomes has identified 1,467 human genome regions (2.6 M bases in total) as being more conserved with non-mammals compared with the majority of mammalian genomes (137). The authors concluded that HGT in the past might had a substantial impact on the human genome composition.

Several authors suggested that the interdomain gene transfer from bacteria and archaea into eukaryotes is more common than could be expected in the eukaryote and human genomes (137, 138). Acquisition of foreign genes *via* the mitochondrial or plastid or other organelle progenitors is not sufficient to explain this frequency and the weak-link model was proposed suggesting that the genetic transfer into multicellular eukaryotes happens during the unicellular or early developmental stages (139).

Other researchers have reservations regarding the magnitude of HGT to the human or other eukaryotic genomes (140–144). The arguments are as follows: the "foreign" DNA can represent a contamination; it can be a known mitochondrial or retro transposed gene; they represent gene loss or evolutionary rate variation; they can represent genes that evolved more rapidly in multiple lineages; and the allowance of a substantial error rate in the datasets with little statistical power. In fact, genetic contamination was reported as false signals of HGT as early as 2002 (145). A more recent controversy regarding the influx of prokaryotic genes to a eukaryotic genome is the tardigrade case (146). A substantial bacterial contamination could have contributed to the unusually high proportion of presumably HGT-acquired genes in this organism reported in the original publication. Due to the abovementioned disagreements and disputes, such studies should be subjected to close scrutiny and critical assessment, before concluding that the lateral genetic transfer represent a real biological event (141).

Besides the traces recorded in genomes, there are examples of viral and bacterial DNA incorporations into the human somatic genome that can be witnessed *in situ* (147). Integration of viral DNA into the host cell genome *via* HGT is a well-documented event in the human papilloma virus cancer development. Analysis of sequences from the Cancer Genome Atlas supports bacterial DNA integration into the nuclear chromosomes of adenocarcinoma stomach cells. Researchers from the same group have shown that cells from acute myeloid leukemia contain bacterial sequences that are present in the microbiome (148).

Based on the above information on HGT in eukaryotes (149) and among the other domains, the following is the hypothesis on the potential involvement of HGT in human chronic disease, autoimmune diseases being an example.

#### CAN HGT PLAY A ROLE IN AUTOIMMUNE DISEASE?

Autoimmune diseases have both genetic and environmental components. While the genetic component can be investigated in a more or less straightforward manner by studying genetic variations associated with these diseases, the environmental component is much more difficult to decipher because of the involvement many variables and many combinations thereof. Besides, it is clear that the genetic component could not been changed substantially during the relatively short period of human evolution to explain the recent and dramatic rise of autoimmune diseases in human populations (150). Thus, the role of environmental factors in the onset of autoimmune diseases becomes increasingly recognized (151).

One of the most essential factors that can be considered as environmental, although it resides inside us, is gut microbiota. Its composition is largely driven by the lifestyle and diet, which have been dramatically transformed during the human history. In the history of human nutrition, there have been two major changes in the diet. The first included the carbohydrate-rich Neolithic diet, with the introduction of farming (~10,000 years ago), and the more recent second that involved the introduction of industrially processed food such as flour, sugar, and other refined products (beginning from ~1850) (152–154). The view is emerging that these changes resulted in a substantial loss of gut bacterial diversity (155, 156). Presumably, the functional diversity is affected as well, thus, changing the interface of host–microbe interaction, with less appropriate functional and signaling properties of the microbiota. The resulting dysbiotic configuration potentially could make a significant contribution to the landscape of contemporary diseases, which have neither a substantial genetic component nor infectious nature (157–159). The factor contributing to the rise of these diseases could be the contemporary lifestyle with limited microbial exposure. Excessively hygienic conditions may compromise the establishment of normal microbiota and negates the immune benefits associated with it (160, 161).

As mentioned before, contemporary intensive agricultural systems with the widespread use of herbicides, insecticides, fungicides, fumigants, desiccants, harvest aids, antimicrobials, growth regulators, metals, and many other substances, are evident. The long-term health effects of the residual concentrations of these chemicals in our food are largely unknown. It has been hypothesized, for example, that the obesity epidemic in the United States may be partly due to the mass exposure of citizens to food containing low-residue antimicrobial agents (162). The resulting dysbiosis may predispose to obesity and, potentially, other diseases.

In this review, we discussed how the HGT mechanisms and processes that have been selected during the previous long-term human–microbiota coevolution may interact with the new reality potentially leading to dysbiotic conditions and even disease (98, 99, 163, 164). HGT in the human gut is intense, and the material for genetic exchange can come from various sources such as microorganisms, viruses, ingested probiotics, fermented and processed food products or industrial additives, plants, livestock, proximity to companion animals, and synthetic biology products. In the human microbiome, the microbial enzymatic machinery, especially involved in posttranslational activity, may modify naïve proteins to immunogenic ones (72, 73, 99, 165, 166). Bacterial or viral infections (167–170) could also impact the fine tuning of the intestinal homeostasis. One of the targets affected by these microbial activities is the intestinal barrier mechanism represented by the very evolutionary

conserved inter-enterocyte tight junction. The leaky gut syndrome is central to the development of various autoimmune diseases, where the luminal content can enter the circulation and initiate systemic inflammatory responses (171). The inflammation that is not properly resolved may lead to chronic inflammatory conditions including autoimmune diseases.

We propose a hypothesis here that, due to extensive inflow of genetic material to the intestine and high intensity of HGT in the gut, a substantial proportion of these genes may be expressed. This may result in the production of neo-protein or neo-peptide formation that imitates self-ones. This molecular mimicry or epitope spreading are well known pathways in autoimmunity inductions. The leaky gut condition results in translocation of the luminal content thus enhancing the inflammatory cascade. If the inflammation is not resolved properly, it is transformed into the self-sustainable chronic inflammation and tissue damage, with a risk of exposure of own antigens to the immune system. The resultant autoantibodies may exacerbate the situation further, with tissue damage and amplification of inflammatory cascades. The current hypothesis is shared, partially, with the hypothesis suggested by Robinson and others (8), on the role of the bacteria– animal HGT in carcinogenesis. It should be emphasized here that this is a hypothesis and, as such, has to be verified in future studies. It is hoped that the hypothesis will encourage the scientific community to explore the relevance of the gut HGT to the onset of chronic human diseases, including, autoimmune diseases.

#### CONCLUSION

As a product of long-term cospeciation with the host (172), commensal gut microbiota had evolved to perform many functions such as contributing to host nutrition *via* degrading dietary components inaccessible by host and synthetizing vitamins, processing and detoxifying xenobiotics, regulating host development and metabolism, conferring colonization resistance toward pathogens, and shaping and maintaining mucosal and systemic immunity. These functions are governed by the specific sets of genes, for which the gut microbiome is particularly enriched compared with other microbial ecosystems. The putative gutspecific genes include those involved in adhesion to the host proteins, in harvesting sugars of the glob series glycolipids, and in degradation of dietary or host-derived complex sugars and glycans (173). The biggest proportion of genes in the minimal gut metagenome (about 5%), however, codes for (pro) phagerelated proteins, implying the important role played by these MGEs in the human gut ecosystem.

Other processes associated with changes in lifestyle potentially contributing to the increase in HGT rates in the gut may include the use of antibiotics, the probiotics ingested, food products processed *via* fermentation, and proximity to livestock and companion animals (7, 38, 126, 174). The bestknown laterally transferred genes that had already caused serious concerns are those conferring resistance to antibiotics. Their rapid emergence and broad dissemination among taxonomically diverse species within a relatively short period of time demonstrated the extreme adaptability of bacteria (18), which is based on their naturally possessed genetic engineering tools such as MGEs (5).

Besides the responses against the selective pressure of antibiotics, there are other examples of adaptive traits acquired by gut bacteria *via* HGT, for example, in response to dietary habits. Japanese diet is traditionally reach in seaweeds and the gut metagenome of the Japanese contain porphyranases and agarases that are absent in the metagenomes of other populations (175). Presumably, in these populations, the enzymes were acquired by gut microbiota (in particular, *Bacteroides plebeius*) from marine bacteria through HGT, and they became completely functional in the new ecosystem to aid the host in digesting these dietary compounds (176).

The example above demonstrates how the microbiome may acquire genes from transient bacteria and successfully integrate them. In the current lifestyle, the microbiome becomes even more exposed to a gene inflow originating from the contemporary production of livestock, plants, industrially processed food, probiotics, and food supplements (**Figure 1**). Despite the fact that direct causal relationship between all those consumed constituents and chronic human diseases' induction is lacking, the present review highlights the potential risks associated with the modification of the evolutionary established stability of our microbiomes. Given the extensive HGT processes in the gut, assimilation of these incoming genes may affect the previously established and fine-tuned host–microbe interaction. Thus, these two factors of the contemporary lifestyle and dietary habits, i.e., a lessened exposure to natural microbiota and the increasing exposure to the previously un-encountered ones, may affect the host–microbe crosstalk and compromise the host health. Indeed, we currently witness a tremendous increase in human diseases that have a substantial gut microbiome component.

The gut microbiota is very dynamic and, on the evolutionary scale, responds almost immediately, *via* HGT mechanisms, to various selective pressures imposed, which is well exemplified by the recent history of antibiotic use, which resulted in the widespread antibiotic resistance (177). One of the current functions of the gut mobilome under continuous antibiotic selective pressure, as serving as the antibiotic resistance gene reservoir, is well established (178). Concomitantly with selection for antibiotic resistance *per se*, this pressure may result in the change of the virulome profile as well. Although the relationship between antibiotic resistance and virulence is complex (22), there is a number of examples how these two traits evolved simultaneously *via* the common mechanisms (179). It

#### REFERENCES


is not fully understood, however, what is the impact of subtler selective pressures that is applied on gut microbiota through the use of probiotics, fermented food, food with the presence xenobiotics and antibiotics, and various food supplements. The effect of genetic exchange with the microbiota of livestock and pet and companion animals is also not well understood. The mobilome-driven genome diversification in the gut is very intense (180) and, undoubtedly, the selective pressure imposed by us has already resulted in the selection of certain genes and genotypes in the microbiome. In the light of the increasing number of diseases and conditions having inflammatory and autoimmune components linked to gut microbiota (171), we have to admit that the best microbiome configuration has not been selected by our contemporary lifestyle and dietary habits. Understanding and then correcting this imbalance may greatly contribute to our health and well-being. Notably, at least in animal model, autoimmunity can be tackled by gene transfer to promote immune tolerance (181, 182). We hope that the hypothesis proposed here will encourage the medical and scientific communities to evaluate the risks associated with HGT events in the human gut.

#### ETHICS STATEMENT

The study does not involve any human or animal subjects.

#### AUTHOR CONTRIBUTIONS

AL and RA screened literature, analyzed the data, and wrote the manuscript. TM designed, edited, and revised critically the manuscript.

#### ACKNOWLEDGMENTS

The authors thank Neu Alf for drawing the figure and to Dr. Ramesh Ajay for editing the manuscript.


immunogenic in celiac disease patients. *Autoimmun Rev* (2016) 15:1111–9. doi:10.1016/j.autrev.2016.09.011


in urban rats a risk for public health? *Antimicrob Agents Chemother* (2013) 57:2424–5. doi:10.1128/AAC.02321-12


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Lerner, Matthias and Aminov. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Case of Meningitis in a Neonate Caused by an Extended-Spectrum-Beta-Lactamase-Producing Strain of Hypervirulent Klebsiella pneumoniae

Khalit S. Khaertynov<sup>1</sup> \*, Vladimir A. Anokhin<sup>1</sup> , Yuri N. Davidyuk<sup>2</sup> , Irina V. Nicolaeva<sup>1</sup> , Svetlana V. Khalioullina<sup>1</sup> , Dina R. Semyenova<sup>1</sup> , Evgeny Y. Alatyrev<sup>3</sup> , Natalia N. Skvortsova<sup>3</sup> and Levon G. Abrahamyan<sup>4</sup> \*

<sup>1</sup> Department of Children Infectious Diseases, Kazan State Medical University, Kazan, Russia, <sup>2</sup> Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, <sup>3</sup> Intensive Care Unit No 2, Republican Clinical Infectious Diseases Hospital, Kazan, Russia, <sup>4</sup> Laboratory for Animal Molecular Virology, Research Group on Infectious Diseases in Production Animals and Swine and Poultry Infectious Diseases Research Center, Faculty of Veterinary Medicine, Université de Montréal, Saint-Hyacinthe, QC, Canada

#### Edited by:

Gayane Manukyan, National Academy of Sciences of the Republic of Armenia (NAS RA), Armenia

#### Reviewed by:

Moffat Mulemena Malisheni, National University of Singapore, Singapore Kirsty Le Doare, Imperial College London, United Kingdom

#### \*Correspondence:

Khalit S. Khaertynov khalit65@rambler.ru Levon G. Abrahamyan levon.abrahamyan@umontreal.ca

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 05 July 2017 Accepted: 03 August 2017 Published: 15 August 2017

#### Citation:

Khaertynov KS, Anokhin VA, Davidyuk YN, Nicolaeva IV, Khalioullina SV, Semyenova DR, Alatyrev EY, Skvortsova NN and Abrahamyan LG (2017) Case of Meningitis in a Neonate Caused by an Extended-Spectrum-Beta-Lactamase-Producing Strain of Hypervirulent Klebsiella pneumoniae. Front. Microbiol. 8:1576. doi: 10.3389/fmicb.2017.01576 Klebsiella pneumoniae is one of the most important infectious agents among neonates. This pathogen has a potential to develop an increased antimicrobial resistance and virulence. The classic non-virulent strain of K. pneumoniae, producing an extendedspectrum beta-lactamases (ESBL), is associated with nosocomial infection mainly in preterm neonates. Hypervirulent K. pneumoniae strains are associated with invasive infection among previously healthy ambulatory patients, and most of them exhibit antimicrobial susceptibility. During the last few years, several cases of diseases caused by hypervirulent K. pneumoniae producing ESBL have been registered in different geographical regions of the world. However, reports of such cases in neonates are rare. Here, we reported that this pathogen can cause pyogenic meningitis in full-term neonate with poor prognosis. A previously healthy, full-term, 12-day-old neonate was admitted to the infectious diseases hospital with suspected meningitis. The clinical symptoms included loss of appetite, irritability, fever, seizures, and a bulging anterior fontanelle. The analysis of the cerebrospinal fluid confirmed the diagnosis of meningitis. Blood and cerebrospinal fluid cultures were positive for K. pneumoniae, producing ESBL. K. pneumoniae isolates were resistant to aminopenicillins, 3rd generation cephalosporins but were sensitive to imipenem and meropenem. The "string test" was positive. The study of the virulence factors of K. pneumoniae by PCR revealed the presence of the rmpA gene. A combination of K. pneumoniae virulence and drug resistance complicated by cerebral oedema led to the death of the neonate. We concluded that both the risk of developing severe forms of infection and the outcome of the disease due to K. pneumonia are associated with the phenotypic features of the pathogen such as its antibiotic susceptibility and virulence factors. Emergence of the ESBL-producing strain of hypervirulent K. pneumoniae could represent a new serious threat to public health, suggesting an urgent need to enhance clinical awareness and epidemiological surveillance.

Keywords: Klebsiella pneumoniae, extended-spectrum β-lactamases, hypervirulent, meningitis, neonate

## INTRODUCTION

fmicb-08-01576 August 12, 2017 Time: 15:43 # 2

#### Subject

Here, we report a case of neonatal meningitis caused by an ESBL-producing hypervirulent Klebsiella pneumoniae strain in a 12-day-old, male neonate. A 40-week-gestation male, with a birth weight of 3400 grams was born by cesarean delivery to a 24-yearold woman. The infant's Apgar scores were 8 and 9 at delivery, and he was discharged on the 5th day after delivery. The mother gave no history of infections before delivery and no complications during pregnancy. Blood and cerebrospinal fluid cultures (CSF) of the infant patient were positive for K. pneumoniae.

The aim of this study was to determine the antibiotic susceptibility and virulence factors of K. pneumoniae isolated from a neonate with purulent meningitis. We investigated the antibiotic susceptibility of K. pneumoniae, the ability of the microorganism for production of extended-spectrum β-lactamases (ESBL), and its virulent factors: the rmpA gene and the hypermucoviscosity phenotype identified by a positive string test.

The Institutional Review Board of the Republican Clinical Infectious Diseases Hospital approved this study and written informed consent was obtained from a subject's parents, according to the guidelines approved under this protocol (Federal Law "Protection of Health Right of Citizens of Russian Federation" N323- FL, 11.21.2011).

#### Bacterial Isolates

One milliliter of blood sample was collected by a sterile syringe and mixed with 20 ml of Brain-Heart Infusion broth (Conda Pronadisa, Spain). This mixture was incubated at 37◦C for 7 days, streaked onto the surface of blood agar and MacConkey agar (Oxoid, United Kingdom), and incubated at 37◦C for 24 h (Collee et al., 1996). In order to isolate K. pneumoniae from CSF, a chocolate agar and blood medium were used (Oxoid, United Kingdom). Bacterial isolates were identified according to morphological and biochemical tests (Gram stain, capsule stain, motility test, indole production test, urease production test, Methyl Red test, Voges–Proskauer test) (MacFaddin, 2000) and confirmed by matrix-assisted laser desorption-ionization time-of-flight mass spectrometry (Microflex, Bruker Daltonics, Bremen, Germany).

#### Gram Stain

Bacterial isolates were resuspended in normal saline for Gram stain. Smear was prepared on a glass slide, air dried and fixed by gentle heating. The slide was flooded with crystal violet for 1 min. The stain was washed off with excess of tap water. Gram's iodine was poured over the slide for 1 min. The slide was washed and destained with ethyl alcohol. Finally, counter staining was done with safranin for 30 s. Slides were washed again, dried and examined under the microscope.

#### For Capsule Staining

India Ink method was used. We placed a single drop of India ink on a clean microscope slide. Then, we removed some colonies from culture plate with a flamed loop and mixed them in the drop of India ink. Then, we placed the end of another clean microscope slide at an angle on the original slide containing the microorganisms. Next, we spread drop out so that it formed a thin film. After 5 min, we saturated the slide with crystal violet for 1 min and carefully rinsed it with water. The slide was air-dried for 5 min. Then, the slide was examined under the microscope using an oil immersion objective lens and looked for purple cells surrounded by a clear halo on a dark background.

#### Motility Test

Colonies of isolated microorganisms from an 18–24 h culture were inoculated into the medium by stabbing the center of the medium (HiMedia, India) to a depth of 1,5 cm. The inoculated medium was incubated at 37◦C for 18 h. A positive motility test was indicated by a diffuse zone of growth flaring from the line of inoculation.

#### For the Indole Production Test

Conventional tube method was used. Colonies of microorganism were inoculated in tryptophan broth and were incubated at 37◦C for 24 h in ambient air. Then, 0,5 ml of Kovac's reagent was added to the broth culture. The test was considered positive if a pink colored ring appeared after the addition of reagent. Negative test is indicated if no color change occurred after the addition of reagent.

#### Urease Production Test

Colonies of isolated microorganism from an overnight Brain-Heart Infusion broth were streaked onto the surface of a urea agar slant. The tube of medium was incubated at 37◦C in ambient air for 7 days. The test was positive if a pink color appeared.

#### Methyl Red and Voges–Proskauer Tests

Colonies of isolated microorganism were inoculated into Methyl Red/Voges–Proskauer broth tube. The tube of medium was incubated at 37◦C for 24 h. After incubation, we removed two – 1 mL aliquots and placed them into two small tubes: one tube was for the methyl red test and the other for the Voges–Proskauer test. For the Methyl Rose test, we added five drops of methyl red to one tube. A red color at the surface was considered a positive result. For the Voges–Proskauer test, we added 0.6mL of 5% alpha naphthol, followed by 0.2 mL of 40% potassium hydroxide and shook the tube gently. A positive test was represented by the development of a red color 15 min after the addition of the reagents.

#### Antibiotic Susceptibility Testing

The antibiotic susceptibility of K. pneumoniae isolates was determined by the Kirby-Bauer disk diffusion method according to Clinical Laboratory Standards Institute guidelines (CLSI) (Patel et al., 2015). Suspension of K. pneumoniae isolate was spread by sterile glass rods on the surface of Mueller Hinton agar (Oxoid, United Kingdom). Then antibiotic disks (Bio-Rad, France) were placed onto the surface of the inoculated Mueller Hinton agar plate. The plate was then incubated at 37◦C for 18 h. Antimicrobial susceptibility was determined by measuring the diameter of the inhibition zone according to CLSI (2015). All antibiotics used for this test are listed in **Table 1**.

#### TABLE 1 | The susceptibility of K. pneumoniae to antibiotics.

fmicb-08-01576 August 12, 2017 Time: 15:43 # 3


S, susceptible; I, intermediate; R, resistant.

#### Test for Production of Extended-Spectrum β-Lactamases (ESBL)

The K. pneumoniae isolates were tested for ESBL by using the double-disk method according CLSI (Patel et al., 2015).

Amoxicillin/clavulanate disks were placed in the center of Mueller Hinton agar plate (Oxoid, United Kingdom). The disk of ceftazidime and cefotaxime were placed at the distance of 20 mm from the amoxicillin/clavulanic acid disk. The plates were incubated aerobically at 37◦C for 18 h before the zone size recorded. A positive result was indicated when the inhibition zones around any of the cephalosporin disks was augmented in the direction of the disk containing clavulanic acid.

#### Hypermucoviscosity Testing

Single colonies, cultured on Brain Heart infusion agar plates (Conda Pronadisa, Spain), were obtained and tested for their ability to form viscous strings. The hypermucoviscosity was defined by the formation of viscous strings > 5 mm length (Yu et al., 2007; Shon et al., 2013).

#### DNA Extraction

Some colonies from the surface of MacConkey agar were suspended in 50 µl of sterile water. Total DNA was extracted from suspended cells using an extraction kit ("Litech," Russia) according to the manufacturer's recommendations.

#### PCR Detection of Virulence-Associated Genes

DNA samples were analyzed using polymerase chain reaction (PCR) with primer pair for the rmpA (50 -ACGACTTTCAAGAGAAATGA-3<sup>0</sup> forward and 5<sup>0</sup> - CATAGATGTCATAATCACAC-3<sup>0</sup> reverse). Amplification was performed using C1000 Thermo Cycler ("Bio-Rad Laboratories," United States) applying the following program: (1) DNA denaturation at 94◦C, 3 min; (2) 35 cycles at 94◦C, 30 s; 45◦C, 30 s; 72◦C, 35 s; (3) final extension at 72◦C, 5 min; (4) reaction termination at 4◦C. The amplicons were separated in 1% agarose gel and purified by using GeneJET Gel Extraction Kit (Thermo Scientific, United States) according to the manufacturer's recommendations. The PCR-products were sequenced using the 3730 DNA Analyzer (Life Technologies, United States) to confirm the presence of the rmpA gene.

#### Case

A full-term, 12-day-old, male neonate was admitted to the infectious diseases hospital with suspected meningitis on the 3rd day of illness. During the first 2 days, irritability and a loss of appetite were observed. On admission day, the infant had a temperature of 39◦C and seizures. He looked noticeably ill and sleepy. On physical examination, his anterior fontanelle was bulging. The skin was pale, without rash. Chest, abdomen, and heart examinations did not show any abnormalities. The heart rate was 156 beats per minute, respiratory rate – 40 per minute. The liver and spleen were not enlarged. Chest X-ray was normal. The initial blood analysis revealed increased C-reactive protein (100 mg/L), presepsin (2932 pg/mL) and procalcitonin (more than 10 ng/mL). The initial complete blood cell count (CBC) analysis did not reveal any changes. CBC showed a erythrocytes count of 5,4 × 1012/L, a white blood cell (WBC) count of 6,4 × 10<sup>9</sup> /L, with 22% segmented neutrophils, 62% lymphocytes, 15% monocytes, 1% eosinophils, and 235 × 10<sup>9</sup> /L platelets. A lumbar puncture was performed: CSF was turbid, physico-chemical examination showed 21 000 WBC/mm<sup>3</sup> with 90% neutrophils, 10% lymphocytes. CSF protein was 290 mg/dL, glucose – 0,3 mmol/L. Serum glucose was 8 mmol/L. The latex antigen test was negative for Haemophilus influenzae B, Neisseria meningitides, Escherichia coli, and Streptococcus group B. The cranial ultrasonography was performed and demonstrated thickening of the ventricular walls. The course of disease was complicated by the development of cerebral edema. The Pediatric Glasgow Coma Scale Score was 5. Changes in WBCs were found only on the 3rd day after hospitalization (**Table 2**). WBC examination showed 28.3 × 10<sup>9</sup> /L leukocytes with 77% neutrophils. The duration of leukocytosis and neutrophilia were 15 and 23 days, respectively. From the 2nd day of the hospitalization, the platelet count dropped to 25 × 10<sup>9</sup> /L. The duration of thrombocytopenia was 8 days. Blood and CSFs were positive for K. pneumoniae producing ESBL. Colonies of


Hb, hemoglobin; WBC, white blood cells.

FIGURE 1 | Hypermucoviscosity phenotype of Klebsiella pneumoniae. When the colonies were touched with a loop and the loop lifted vertically from the surface of the blood agar plate, the mucoid isolates adhered to the loop.

bacteria isolated on media were gray, mucoid, with diameters up to 2–4 mm, gram-negative, contained a thick capsule, and were non-motile (motility test was negative). Colonies of bacteria were positive for urease and Voges–Proskauer tests, and were itive negative for indole and Methyl Red tests. CSF culture for other bacteria was negative. The "string test" was positive (**Figure 1**). The study of the virulence factors by PCR revealed the presence of the rmpA (regulator of the mucoid phenotype) gene. The treatment of the patient included comprehensive antibiotics (ampicillin, amikacin, meropenem, cefoperazone), dexamethasone, IgM-enriched intravenous immunoglobulin, infusion therapy and mechanical ventilation. On admission day, a neonate was started on ampicillin (200 mg/kg/day) and amikacin (10 mg/kg/day). Both antibiotics a neonate received within 3 days, but did not show any clinical improvement. Following the isolation of K. pneumoniae (after 3 days), meropenem (120mg/kg/day) was administered, which a neonate received for 15 days. From the 19th day after hospitalization until the death, the patient received cefoperazone (100mg/kg/day). Despite the therapy, the patient died on the 35th day of the disease. The postmortem examination revealed purulent meningoencephalitis, ventriculitis with the outcome of total cerebral leukomalacia, scattered pulmonary atelectasis, bilateral pneumonia, and the depletion of the thymus and spleen.

#### BACKGROUND

Klebsiella pneumoniae is one of the leading causes of hospitalacquired infection and neonatal sepsis (Janda and Abbott, 2006; Jones, 2010; Camacho-Gonzalez et al., 2013). The risk of severe bacterial infections such as sepsis and meningitis in neonates is associated mainly with neonatal factors: prematurity, low birth weight, and immaturity of innate and adaptive immunity (Cuenca et al., 2013; Cortese et al., 2016). Additionally, pathogenic features of the K. pneumoniae such as virulence and antibiotic resistance can define the course and outcome of the infection (Al-Hasan et al., 2011; Jacob et al., 2013). K. pneumoniae is an opportunistic pathogen often resistant to multiple antibiotics. During the last few decades, ESBL positive K. pneumoniae isolates have been recovered worldwide, especially in intensive care units (ICU) (Canton et al., 2008; Gelband et al., 2015). ESBLs can inactivate all penicillins and cephalosporins, including 3rd generation cephalosporins (Gelband et al., 2015). In neonates, K. pneumoniae median resistance to ampicillin was 94% and cephalosporins 84% in Asia; 100% and 50% in Africa (Le Doare et al., 2015). The prevalence of ESBL-producing strains of K. pneumoniae in the United States is 23%, in some countries of Europe up to 85–100% (Gelband et al., 2015). These microorganisms can cause outbreaks of neonatal sepsis in hospitals and neonatal ICU (Haller et al., 2015; Khaertynov et al., 2016). Virulence factors of K. pneumoniae play an important role in the development of infection. The factors that are implicated in the virulence of K. pneumoniae strains include the capsular polysaccharide, lipopolysaccharide, fimbrial adhesins, and siderophores (Broberg et al., 2014; Li et al., 2014). In the mid-1980s and 1990s, reports from Taiwan described cases of disease caused by hypervirulent strains of K. pneumoniae (hv-KP) (Liu et al., 1986; Cheng et al., 1991; Wang et al., 1998). A combination of clinical and bacterial phenotypic features of the hv-KP distinguishes it from the "classic" K. pneumoniae strains (Shon et al., 2013). One of these features is its ability to cause severe invasive infection (liver abscesses, meningitis, endophthalmitis) in previously healthy ambulatory patients. The second distinctive characteristic is the hypermucoviscous phenotype, which results in mucoid colonies on agar plates. This phenotype is defined by a positive "string test." A positive string test is indicated by a microbiological inoculation loop able to generate a viscous string > 5 mm the length by stretching bacterial colonies on an agar plate (Yu et al., 2007; Shon et al., 2013). The hypermucoviscous phenotype of K. pneumoniae is associated with the presence of rmpA and rmpA2 genes (Decré et al., 2011; Bialek-Davenet et al., 2014). The hypervirulent strains of K. pneumoniae currently spread throughout the world (Decré et al., 2011; Bialek-Davenet et al., 2014). Until recently, it was believed that the classic, non-virulent strains of K. pneumoniae producing ESBL and the hv-KP strains evolved separately, and have been considered as independent of each other. However, in 2014 an hv-KP able to produce ESBL (Zhang et al., 2015) was found in China, and in 2015 a clinical case caused by hv-KP producing ESBL was reported in France (Surgers et al., 2016). It is likely the frequency of these cases will grow worldwide. These cases were mainly observed in adult patients. The reports of such cases in neonates are extremely rare (Shankar et al., 2016).

#### DISCUSSION

Neonatal infection can occur in term infants because of birth risk factors and exposure to pathogens in the community

Khaertynov et al. Case of Meningitis in a Neonate

as well. Neonatal sepsis and meningitis are more common in preterm newborns with low gestational age and low birthweight (Kavuncuoglu et al., 2013 ˘ ; Samuelsson et al., 2014). K. pneumoniae infection is a typical nosocomial infection in neonates with the combination of these factors. However, in the case reported here, the disease occurred in a fullterm baby. The mother of the baby did not have clinical signs of infectious disease. The neonatal meningitis ensues when pathogenic virulence factors overcome host defense mechanisms. For instance, the presence of a capsule at cell surface protects K. pneumoniae from opsonization and phagocytosis by macrophages and neutrophils (Li et al., 2014). On the other hand, capsular polysaccharide of K. pneumoniae suppress the early inflammatory response by inhibition of IL-8 expression through the inhibition of TLR4 signaling (Li et al., 2014). Thus, the pathogenic potential of the microorganism plays an important role in the disease outcome. In the reported case, the K. pneumoniae isolates produced ESBL and were absolutely resistant to aminopenicillins, 3rd generation cephalosporins and gentamicin, with diameters of inhibition zones of 0 mm for each of them. K. pneumoniae isolates were sensitive only to imipenem and meropenem with diameters of inhibition zones of 25 and 24 mm, respectively. Intermediate sensitivity was observed for amikacin and co-trimoxazole with diameters of inhibition zones of 16 mm and 13 mm, respectively. Additionally, these isolates were hypervirulent (i.e., positive "string test" and presence of rmpA). In this case, neonatal meningitis was caused by ESBL-producing strain of hv-KP. An unfavorable outcome has been associated with the inefficiency and late onset of antibacterial therapy, as well as the formation of a severe inflammatory reaction from the membranes and brain matter. Bacterial meningitis in neonates is an independent risk factor for mortality: mortality rates of meningitis caused by K. pneumoniae reaching 17.1% (Watson et al., 2003). We noted the absence of changes in WBC on admission day despite the massive pyogenic process in the cerebrospinal fluid. Given the fact of bacteremia, this unusual situation with the redistribution of neutrophils in newborn fluids cannot be explained only by the biological characteristics of the microorganism. Likely, the nature of the immune response to a severe invasive infection in a newborn patient takes place. The primary reaction of the newborn child in the tissues of the central nervous system is associated with the redistribution of neutrophil chemoattractants: with an increase in their concentration in the cerebrospinal fluid, their level in the blood should be reduced. We regret that we were unable to confirm this assumption. For example, simultaneous assessment of interleukin-8 in the blood and cerebrospinal fluid could have provided confirmation. However, we do not rule out that such changes are associated

#### REFERENCES

Al-Hasan, M. N., Huskins, W. C., Lahr, B. D., Eckel-Passow, J. E., and Baddour, L. M. (2011). Epidemiology and outcome of Gramnegative bloodstream infection in children: a population-based study. Epidemiol. Infect. 139, 791–796. doi: 10.1017/S0950268810001640 1640

with specific biological characteristics of the microorganism. Meningitis caused by hypervirulent K. pneumoniae is associated with potential mortality. Emergence of hypervirulent strains of K. pneumoniae producing EXBL can represent a major challenge for patient treatment. In our study, the most sensitive antibiotics to hypervirulent ESBL-producing K. pneumoniae strains were meropenem and imipenem. In cases of neonatal sepsis or neonatal meningitis caused by hypervirulent strains of K. pneumoniae producing EXBL, these two carbapenems should be considered as first-line therapy and should be administered as soon as possible.

## CONCLUDING REMARKS

The risk of developing severe forms of infection due to K. pneumonia and the outcome of the disease are associated not only with neonatal factors (low gestational age, very low birth weight), but also with the features of the microorganism: its antibiotic susceptibility and virulence. ESBL-producing strain of hypervirulent K. pneumoniae causes invasive infection (pyogenic meningitis) in full-term neonate with poor prognosis.

## AUTHOR CONTRIBUTIONS

KK design of the study and wrote the paper. Discussed the results and implications, wrote the manuscript. VA design of the study and wrote the paper. YD DNA extraction of K. pneumoniae strains with subsequent genotyping by PCR method to determine virulence factors. IN design of the study. SK wrote the paper. DS collection of clinical data. EA collection and interpretation of clinical data. NS isolation of K. pneumoniae colonies, determination of their antibiotic resistance and ability to produce of extended spectrum β-lactamase. LA wrote the paper. Discussed the results and implications, wrote the manuscript.

## ACKNOWLEDGMENTS

This study was performed according to the Russian Government Program of Competitive Growth of Kazan Federal University and subsidy allocated to Kazan Federal University for the state assignment in the sphere of scientific activities. Some of the experiments were conducted using the equipment of Interdisciplinary center for collective use and Pharmaceutical Research and Education Center, Kazan (Volga Region) Federal University, Kazan, Russia. Authors would like to thank Mackenzie Waltke for proofreading.

Bialek-Davenet, S., Criscuolo, A., Ailloud, F., Passet, V., Jones, L., Delannoy-Vieillard, A. S., et al. (2014). Genomic definition of hypervirulent and multidrug-resistant Klebsiella pneumoniae clonal groups. Emerg. Infect. Dis. 20, 1812–1820. doi: 10.3201/eid2011.14020622

Broberg, C. A., Palacios, M., and Miller, V. L. (2014). Klebsiella: a long way to go towards understanding this enigmatic jet-setter. F1000Prime Rep. 6:64. doi: 10.12703/P6-64


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Khaertynov, Anokhin, Davidyuk, Nicolaeva, Khalioullina, Semyenova, Alatyrev, Skvortsova and Abrahamyan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Viral impact in autoimmune diseases: expanding the "X Chromosome–nucleolus nexus" Hypothesis

#### *Wesley H. Brooks\**

*Department of Chemistry, University of South Florida, Tampa, FL, United States*

Viruses are suspected of significant roles in autoimmune diseases but the mechanisms are unclear. We get some insight by considering demands a virus places on host cells. Viruses not only require production of their own proteins, RNA and/or DNA, but also production of additional cellular machinery, such as ribosomes, to handle the increased demands. Since the nucleolus is a major site of RNA processing and ribonucleoprotein assembly, nucleoli are targeted by viruses, directly when viral RNA and proteins enter the nucleolus and indirectly when viruses induce increased expression of cellular polyamine genes. Polyamines are at high levels in nucleoli to assist in RNA folding. The size and activity of nucleoli increase directly with increases in polyamines. Nucleolar expansion due to abnormal increases in polyamines could disrupt nearby chromatin, such as the inactive X chromosome, leading to expression of previously sequestered DNA. Sudden expression of a large concentration of Alu elements from the disrupted inactive X can compete with RNA transcripts containing intronic Alu sequences that normally maintain nucleolar structural integrity. Such disruption of nucleolar activity can lead to misfolded RNAs, misassembled ribonucleoprotein complexes, and fragmentation of the nucleolus. Many autoantigens in lupus are, at least transiently, components of the nucleolus. Considering these effects of viruses, the "X chromosome–nucleolus nexus" hypothesis, which proposed disruption of the inactive X by the nucleolus during stress, is now expanded here to propose subsequent disruption of the nucleolus by previously sequestered Alu elements, which can fragment the nucleolus, leading to generation of autoantigens.

Keywords: autoimmune disease, polyamines, nucleolus, virus, X chromosome

#### INTRODUCTION

Previously, we presented the "X chromosome–nucleolus nexus" hypothesis (1–3). In the hypothesis it was proposed that enlargement of the nucleolus in response to cellular stress could disrupt neighboring chromatin, such as the inactive X chromosome. As a result, sequestered alleles (e.g., polyamine genes on the inactive X), elements (e.g., Alu elements), and viruses could be opened for transcription. This could lead to eventual creation of autoantigens due to overexpression of genes and elements from both the previously active chromatin and the, now, reactivated chromatin. Here is presented new details to the hypothesis, explaining how the disrupted chromatin can

#### *Edited by:*

*Juarez Antonio Simões Quaresma, Universidade Federal do Pará, Brazil*

#### *Reviewed by:*

*Jean-Marc Gallo, King's College London, United Kingdom Lydia E. Matesic, University of South Carolina, United States*

> *\*Correspondence: Wesley H. Brooks wesleybrooks@usf.edu*

#### *Specialty section:*

*This article was submitted to Microbial Immunology, a section of the journal Frontiers in Immunology*

*Received: 29 July 2017 Accepted: 13 November 2017 Published: 28 November 2017*

#### *Citation:*

*Brooks WH (2017) Viral Impact in Autoimmune Diseases: Expanding the "X Chromosome–Nucleolus Nexus" Hypothesis. Front. Immunol. 8:1657. doi: 10.3389/fimmu.2017.01657*

lead to subsequent disruption of the nucleolus, even nucleolar fragmentation, which results in ineffective nucleolar functioning, misfolded RNAs, misassembled or incompletely assembled ribonucleoprotein (RNP) complexes, and stabilization of nucleolar components in autoantigenic conformations. Many of the major autoantigens in autoimmune diseases like systemic lupus erythematosus (SLE) are, at least transiently, components of the nucleolus (e.g., splicosome subunits). Among the factors that could cause extraordinary cellular stress, viruses are highly suspected of causing such disruption in autoimmune diseases.

#### VIRAL INVOLVEMENT IN AUTOIMMUNE DISEASES

Exposomics is the study of all environmental factors which a person may encounter during their lifetime, even including prenatal exposure. These environmental factors in the exposome can include components of the diet, gut microbiota, chemicals, air pollutants, heavy metals, and infectious agents. These factors can cause cellular stress and can have a cumulative effect in cells through accumulation of genetic damage and/or disruption of epigenetic control, especially in genetically predisposed individuals, that establishes the conditions for eventual manifestation and progression of an autoimmune disease. Within the exposome is the infectome which is the collection of pathogens that may contribute to an individual's onset and progression of an autoimmune disease (4). This can be complicated by the latency of some pathogens and the synergistic effects of multiple pathogens. However, unless one is looking for pathogen antigens, the specific pathogen and its effect on the immune system may be masked by a larger response to more abundant autoantigens some of which the pathogen's antigens may mimic. In addition, it has been difficult to prove these associations since, in the case of viruses, many viruses can establish latent infections but the autoimmune disease may not manifest itself until several years after the initial infection, thus clouding the true extent of their association. For example, a mononucleosis infection, a.k.a. the "kissing disease," which involves the Epstein–Barr virus (EBV), increases the risk for subsequent appearance of multiple sclerosis (MS) but manifestation of the MS might not occur until as long as 30 years after the mononucleosis episode (5). Add to this the fact that in the interim the individual has had other infections caused by other pathogens that complicate the situation, potentially triggering the actual autoimmune disease for which an initial EBV infection set the stage. A subsequent infection with another virus could allow activation of latent viruses giving a combined stressful impact on the cell. For example, a primary infection by cytomegalovirus can lead to reactivation of latent EBV which provokes an immune response (6). The induction of one latent virus by another virus shows the potential complexity underlying autoimmune diseases. The general population has had exposure to many of the viruses associated with autoimmune diseases but for most individuals there is no autoimmune disease development, suggesting that genetic susceptibility is also important as well as possible epigenetic and environmental factors. As an example, most adults have had exposure to EBV but few develop an autoimmune disease, suggesting other factors are involved rather than just EBV. A genetic possibility for these differing responses may be based on different HLA types, for example, entry of EBV into a host cell *via* binding of the EBV's gp42 glycoprotein to human CD21 and lymphocytic antigen type HLA-DR. Other HLA sub-types may have different expression levels or have different affinity for the gp42 and not be as compliant for EBV entry. In addition, the extracellular portion of the EBV BZLF2 protein can suppress antigen presentation by binding HLA-DR delaying detection of the EBV (7).

**Table 1** lists many of the viruses that have shown associations with autoimmune diseases. We should note that there is variety in the route of transmission and entry among these viruses: (1) respiratory and oral secretions (saliva, sputum, nasal mucus) (e.g., EBV, parvovirus); (2) gut (e.g., enteroviruses); (3) insect vector transmission (e.g., mosquitos for Zika, West Nile); (4) sexual interactions (e.g., HPV, HIV); and (5) transfusions (e.g., HIV). The tissue types in which viral sequestration occurs may vary, such as EBV behind the blood–brain barrier associated with MS or possible EBV in the synovium associated with RA. We should also note that there are both RNA and DNA viruses listed in **Table 1** and most of these viruses can persist in a latent state in the host. Appearance of viral antigens does not necessarily mean that the virus is the cause of the autoimmune disease episode since it may only be the result of stress from an autoimmune disease episode that leads to subsequent activation of a hidden virus.

For most of these associations (**Table 1**), it remains to be determined if the virus is the causative agent, one of several combined contributing agents, or simply appearing as a result of impaired host cell suppression of the virus. For example, the measles virus is suspected of involvement in MS due to the appearance of antibodies to measles virus antigens in cerebrospinal fluid of MS patients (36). Whether the MS is a direct result of the measles virus or the appearance of measles antigens is a consequence of the MS or is simply coincidental is not known. There may, in fact, be another virus or another environmental agent that has disrupted the suppression of the latent measles virus. As it is, the situation is even more perplexing since the introduction of measles vaccination for the general population has not caused a significant change in the occurrence rate of MS (37). Other viruses with an infrequent association with an autoimmune response can have reemergence in other forms, such as the varicella zoster virus which causes chicken pox and which can reemerge as shingles and is associated with MS (38). And we should bear in mind that human endogenous retroviruses are suspected of involvement in serious diseases, including autoimmune diseases (25, 39).

Among the viruses listed in **Table 1**, EBV has received the most attention as a virus with links to autoimmune diseases (15, 40). An association with EBV infection has been observed in SLE (41, 42), MS (43), RA, and Sjögren's syndrome (SjS) (44) and prior EBV infection, as indicated by sero-positivity for EBV antigens, is observed in 94.2% of controls and 99.5% of MS patients (45). In addition, EBV (human gammaherpesvirus 4, HHV-4) is representative of several herpes viruses that have shown associations with autoimmune diseases. Therefore, based


#### Table 1 | Virus and autoimmune disease associations.

*Association may be causative (virus induces autoimmune disease) or result (autoimmune disease facilitates viral expression). Some associations may be due to vaccines (e.g., HPV). ACTD, autoimmune connective tissue diseases; AGS, Aicardi-Goutiéres syndrome; AIH, autoimmune hepatitis; AITD, autoimmune thyroid diseases including Hashimoto's and Graves' diseases; ALZ, Alzheimer's disease; APS, anti-phospholipid syndrome; GBS, Guilian Barré syndrome; MG, myasthenia gravis; MS, multiple sclerosis; PV, pemphigus vulgaris; RA, rheumatoid arthritis; SjS, Sjögren's syndrome; SLE, systemic lupus erythematosus; SS, systemic sclerosis; T1D, type 1 diabetes; dsDNA, double-stranded DNA; ssDNA, single-stranded DNA; ssRNA, single-stranded RNA.*

*a Others: pharyngitis, lymphadenopathy, and mononucleosis syndrome.*

*bOthers: giant cell arthritis, Wegner's granulomatosis, and polyarteritis nodosa.*

*c Others: keratitis, herpes esophagitis, and encephalitis.*

*dHuman endogenous retroviruses (HERVs) are classified via differing methods. See Ref. (35). HERV-E (Human endogenous retrovirus, group E);* 

*HRES-1 (non-HERV-E human T cell leukemia-related endogenous retrovirus).*

on current knowledge, EBV is most useful for describing possible viral involvement in autoimmune diseases in general.

One way in which a viral infection could provoke an autoimmune reaction is by disrupting the host cell's epigenetic control during a particularly strong cellular stress response to the viral activity leading to expression of previously sequestered gene alleles. Expression from those newly opened sites could lead to imbalance in the protein and RNA products. The female predominance of many autoimmune diseases suggests that the X chromosome and possibly disruption of the inactive X chromosome, a major epigenetic structure in the cell, could be of significance in such a scenario of viral disruption of epigenetic control (1, 3). One point of concern is fragile sites, which are particularly susceptible to viral insertion and DNA breaks. Fragile sites can be hundreds of thousands, even millions of base pairs in length. The X chromosome has four major fragile sites (FRAXA at Xq28; FRAXB at Xp22; FRAXC at Xq22; and FRAXD at Xq27) (1, 46). Reactivation of part or all of the inactive X chromosome could open these fragile sites for expression of hidden viruses in the fragile sites, adding to the cellular stress.

Once a virus becomes active in a cell, one of its prime targets in taking over the cell is the nucleolus. The virus is dependent on the host cell's machinery, including ribosomes and transfer RNAs (tRNA), in order for viral proteins to be synthesized. And viral RNAs need to be properly folded and assembled into ribonucleoprotein complexes (RNPs). RNA and RNP processing and assembly are major functions of the nucleolus. The virus puts additional demands on the nucleolar functions beyond the host cell's needs and, since the virus does not code for ribosomes and such, the virus needs to induce increased nucleolar capacity and activity. Localization of viral RNA and proteins to the nucleolus takes advantage of nuclear and nucleolar localization signals (NoLS) and chaperones. In the case of viral RNA transcripts, RNA pol III transcribed RNAs bind SSB/La and SSA/Ro which assist in nucleolar entry and processing along with any required refolding. For viral proteins, nuclear localization signals (NLS), which are sequences of basic amino acids, and NoLS are used but, since these signals are frequently closely positioned, it has been difficult to decipher the NoLS from the more recognizable NLS (47). Some progress has been made in determining NoLS, such as for the adeno-associated virus serotype 2 assembly activating protein (48), and for nucleolar retention signals, such as found in coronavirus nucleocapsid proteins (49).

## THE NUCLEOLUS—STRUCTURE, FUNCTION, AND DYNAMICS

The nucleolus [reviewed in Ref. (50–54)] is a prominent feature in the nucleus, appearing as a vacant area when imaging nuclear DNA content. There is, in fact, some DNA in the nucleolus. As far as active DNA in the nucleolus, nucleoli associate primarily with repetitive copies of ribosomal DNA (rDNA) genes expressing ribosomal RNAs from nucleolar organizing regions (NORs) which are located on the acrocentric autosomes 13, 14, 15, 21, and 22. Other DNA sequences may be involved at the periphery of the nucleolus transiently, such as in repair of breaks in DNA near the rDNA genes. A role in DNA repair in general is emerging for the nucleolus since many proteins involved in DNA repair have associations with the nucleolus (55). Further, centromeric DNA is associated with nucleoli as part of the nucleolar regulation of the cell cycle (56). The nucleolus does not have a membrane defining its structure but nucleoli are typically surrounded by a shell of heterochromatin established in chromosomes containing nucleolar-associated chromatin domains (NADs) (52). This then serves to define the boundaries of the nucleolus. The NADs contain satellite DNA, mostly from centromeric and pericentromeric regions of chromosomes. The NADs also contain gene poor and silent chromatin. In addition, the inactive X chromosome (a.k.a. the Barr body), a heterochromatic body found in most human female cells, is found in close proximity to nucleoli in one-third of cells throughout the cell cycle and 90% of cells in S phase suggesting a putative role for the nucleolus in maintaining X chromosome inactivation (57). The heterochromatin–nucleolar associations are facilitated, in general, by insulator proteins, CTCF (CCCTC-binding factor) and nucleophosmin and additionally, in the case of the inactive X chromosome, by X inactivation specific transcript RNA. EBV latency can be controlled by CTCF bound in the promoter region of the Epstein–Barr virus nuclear antigen 2 (EBNA-2) gene (58). Disruption of chromatin–nucleolar interactions could lead to changes in EBV latency when CTCF interactions with inserted EBV genes are disrupted. In addition, another very important point to keep in mind is that nucleolin bound to RNA polymerase II (RNA pol II) transcripts that contain intronic Alu elements appear to have a critical role in maintaining the integrity of the nucleolus (59). When Caudron-Herger and colleagues added RNA pol III transcribed Alu element sequences, even as short as 20 nucleotides, there was fragmentation of nucleoli into small nucleolar-like units that were very inefficient in carrying out nucleolar functions of RNA and RNP processing and assembly (**Figure 1**). The authors proposed that the nucleolar fragmentation was attributable to Dicer-facilitated degradation of hybridized RNA pol III Alu sequences with RNA pol II intronic Alu sequences. The work of Caudron-Herger and colleagues demonstrates a close connection between nucleolar integrity and the complexes of nucleolin with RNA pol II transcripts containing intronic Alu sequences. Another possibility for nucleolar disruption by RNA pol III Alu transcripts that Caudron-Herger and colleagues did not mention is possible competition for nucleolin between the RNA pol II intronic Alu sequences and a sudden abundance of RNA pol III Alu transcripts. We believe this could have a major role in generation of autoantigens as we will explain below.

Normally nucleoli contain three discernable sub-regions: the fibrillary centers (FCs), the dense fibrillary centers (DFCs), and the granular components (GCs). The FCs are the sites of rDNA transcription by RNA polymerase I (RNA pol I) to generate the initial ribosomal RNA transcript, the pre-rRNA. Only 50% of the ~400 rDNA repeats in the human diploid genome are transcriptionally active (60). Processing of the pre-rRNA occurs primarily in the DFCs assisted by small nucleolar RNAs (snoRNAs). Assembly of the final ribosomal subunits occurs in the GC, which has a high concentration of proteins needed to complete the RNPs (61). Other RNAs and RNPs processed and assembled in the nucleolus include: the signal recognition particle (SRP) which controls translation and localization of extracellular proteins by transporting them to the endoplasmic reticulum (ER) for eventual extracellular release; tRNAs which require extensive folding; small nuclear ribonucleoprotein complexes involved in splicing of messenger RNAs (mRNAs); and centromere components. Therefore, the nucleolus is involved directly or indirectly in many cellular functions, such as regulation of mitosis; cell-cycle progression; cell proliferation; mRNA processing *via* splicing; translation; protein localization; and various forms of stress response.

The nucleolar proteome contains over 4,500 proteins according to the nucleolar proteome database, NOPdb3.0 (62). About 30% of these proteins are involved in ribosome biogenesis. Since the demands on nucleolar output can change rapidly, the nucleolar proteome is very dynamic. In addition, the size of the nucleolus can change dramatically depending on the needs. Increased nucleolar size correlates directly with increases in polyamine synthesis (63). The polyamines, spermidine and spermine, are involved in many cellular functions but their highest concentrations are found in the nucleoli where the polyamines assist in folding of RNA transcripts and assembly of RNPs. The polyamines have a unique combination of length (spermidine ~11 Å spermine ~14 Å), flexibility (all single bonds) and high cationic charge at physiological pH (spermidine +3; spermine +4) which makes them ideal counter ions to assist in folding the negatively charged RNA transcripts in the nucleolus.

Nucleoli are very dynamic structures in the cell and cell cycle. There can be more than one nucleolus in the nucleus and,

combined, they can occupy up to 25% of the nucleus. They can expand rapidly, facilitated by increased polyamines, in response to cellular stress since the cell may need to have more ribosomes and tRNAs to synthesize new proteins to recover from the stress. However, we should remember that the nucleolus surrounds itself with heterochromatin so there is the possibility of displacement or disruption of neighboring heterochromatin due to the nucleolar dynamics (1). With regards to the cell cycle, nucleoli disappear in mitosis and reappear in telophase and early G1 forming around NORs with the rDNA genes and pre-existing rRNA and ribosomal complexes (64). In addition, CDK1 cyclin kinases have a key role in controlling nucleoli during cell cycling, and centromere complexes are generated in the nucleolus giving further importance to nucleoli in cell cycling.

## VIRAL IMPACT ON THE NUCLEOLUS

Once in the host cell, the virus can be sequestered into the host chromatin or it can initiate viral replication. In the case of EBV, the multi-functional Epstein–Barr nuclear antigen 1 (EBNA-1) protein from the EBV genome can assist in the spread and attachment of viral DNA to metaphase chromosomes (65). EBNA-1 can also disrupt the host cell's USP7-assisted stabilization of p53/ TP53, thereby preventing the host cell from entering apoptosis, setting the stage for continual viral replication (65). However, viruses do need host cell machinery produced in the nucleolus, such as ribosomes, to facilitate viral replication. Therefore, the virus will attempt to increase nucleolar activity and turn on viral gene expression. The EBV genome has a snoRNA, called v-snoRNA1, which is found in the nucleolus in infected cells (66). The v-snoRNA1 appears to be involved in activation of the viral DNA polymerase. Another EBV early gene is Epstein–Barr nuclear antigen 2 (EBNA-2) which is a transactivator of viral and host genes. EBNA-2 can associate with RNA polymerase II promoters to induce increased transcription and this includes the *MYC* gene (67, 68). MYC induces increased RNA polymerase III (RNA pol III) activity which creates viral RNA transcripts and many of the RNA transcripts for nucleolar assembled complexes (69). And MYC induces increased transcription by RNA pol I to create rRNA transcripts (70). The MYC interactome consists of approximately 15% of genes throughout the host genome (71, 72). Included among these are genes involved in polyamine synthesis: ornithine decarboxylase (*ODC*); spermidine synthase (*SDS*); and spermine synthase (*SMS*) (73–75). And so the virus induces polyamine synthesis which is directly associated with an increase in the size and activity of the nucleolus (63, 76). The cationic polyamines are ubiquitous and have many important interactions throughout the cell and local extracellular environment (e.g., spermine in neural synapses). The highest levels of polyamines are found in the nucleolus where the polyamines play a critical role in RNA folding by neutralizing anionic charges in the RNA sufficiently for intra-strand RNA–RNA interactions to form. Polyamine availability directly correlates with increased RNA expression and processing (77). And the polyamines can assist in RNP complex assembly. Mostly these are transient interactions of the polyamines but, in some cases, the polyamines remain as part of the final RNA complex, such as in tRNAs (1). Since viral genes need to be expressed and viral RNAs need to be folded and some viral proteins localize to the nucleolus for folding and assembly into the final virion, it is understandable why a virus would want to increase the host cell's polyamine content to increase nucleolar capacity and activity (78). However, the relation between viral activation and subsequent RNA synthesis is more complex and can vary among viral types. For example, poliovirus inhibits RNA pol I activity by inducing SL1 cleavage and UBF posttranslational modification (79), whereas hepatitis C virus stimulates RNA pol I activity which is involved in transcription of the rDNA genes (60).

Viruses can influence the nucleolar proteome leading to abnormal redistribution of nucleolar components to the nucleus, cytoplasm, and even cell surface (80). For example, viruses induce cell surface exposure of SSB/La which normally facilitates termination of RNA pol III transcription in the nucleus and then, along with SSA/Ro, acts as a chaperone for the RNA transcript as it is processed in the nucleolus (81). Viruses can also disrupt the cell-cycle related kinases in the nucleolus, suppressing normal cell cycling and, thereby, hijacking the nucleolus to focus on viral RNA and protein synthesis and assembly of viral RNP complexes (78). The usual effect of cellular stress, such as viral activity, on the nucleolus is to cause enlargement of the nucleolus but on some occasions the nucleolus can actually decrease in size. Inhibition of RNA pol I by poliovirus, as mentioned above, could be such a situation since a drop in rRNA transcripts would inhibit the major nucleolar function of ribosome synthesis.

#### THE "X CHROMOSOME–NUCLEOLUS NEXUS" HYPOTHESIS: DISRUPTION OF THE INACTIVE X

The stress that viral activity can put on the nucleolus can lead to extensive enlargement of the nucleolus. This could potentially disrupt the epigenetic silencing in heterochromatin neighboring the nucleolus. The inactive X chromosome, a.k.a. the Barr body, would be especially vulnerable, as we proposed in the original version of the "X chromosome–nucleolus nexus" hypothesis (1), since the Barr body is frequently found in close proximity to a nucleolus (57) and against the nuclear membrane (82), as depicted in **Figure 2A**. Sandwiched between the nucleolus and the nuclear membrane, the Barr body would not be able to avoid exposure and disruption due to the nucleolin and nucleophosmin that are involved in chromatin remodeling from an expanding nucleolus (83, 84). In addition, exposure of the chromatin to the high content of polyamines in the nucleolus could add to the disruption of chromatin since the cationic polyamines can compete with histones for DNA binding. Moreover, the polyamines have the potential to stabilize alternate DNA conformations, such as Z-DNA, which is targeted by autoantibodies in some cases of SLE and RA. Negative supercoiling stress is stored in nucleosomes as the double-stranded right-hand coiled B-DNA makes a left-handed supercoil over the surface of the nucleosome's histones. Displacement of the histones during chromatin

in 90% of cells in S phase and one-third of cells throughout the cell cycle except during mitosis when nucleoli disappear (57). This places one of the most inactive structures in the cell, the inactive X, next to one of the most active, multi-functional, and dynamic structures, the nucleolus. (B) The nucleolus can rapidly expand during cellular stress, such as viral activation. Trapped between the nucleolus and the nuclear membrane, the Xi could be disrupted by the expanding nucleolus.

remodeling could release the negative supercoiling stress, allowing it to flux through the DNA and potentially flipping into left-hand coiled Z-DNA which is also a form of negative supercoiling storage. Z-DNA appears only transiently in chromatin since most DNA is wrapped up as B-DNA in nucleosomes. Z-DNA is not flexible enough to bend around histones so it is excluded from the 145 bp bound to the surface of the histones. Since nucleosomes in human chromatin occur every 200 bp on average, there is normally little opportunity for Z-DNA to form and persist. Exposure to high levels of polyamines from the nucleolus concomitant with disruption of nucleosomes could increase the likelihood of Z-DNA persistence when there is a shift of negative supercoiling storage from nucleosomes to Z-DNA (85). In a similar manner, DNA cruciforms are formed from negative supercoiling stress but their occurrence is also suppressed by positioned nucleosomes. The Alu elements, of which there are more than one million throughout the human genome, contain sequences capable of cruciform formation (1). Since there are approximately 15 million nucleosomes associated with the human genomic DNA, there is ample stored negative supercoiling stress. This potential for rapid dynamic changes in chromatin, including disruption of higher order "stacked" nucleosomes, alternate DNA conformations, displacement of bound proteins, and DNA strand separation, is perhaps under-appreciated aspects of epigenetics and could come into play when the nucleolus encroaches on surrounding chromatin.

Normally males have only one X chromosome whereas females have two X chromosomes. Most genes on the X are not sex-related so females only need one active X chromosome. Therefore, early in embryogenesis, each female cell inactivates one X chromosome, either the maternally derived or the paternally derived X, and each daughter cell will inherit that inactivation pattern. The process of X chromosome inactivation [reviewed in Ref. (86)], which results in the heterochromatic Barr body, begins from the X inactivation center at Xq13 of the X chromosome's long arm (Xq) (**Figure 3**). Approximately 95% of genes on the Xq and 65% of genes on the short arm (Xp) are inactive and form the heterochromatic core of the Barr body (87). The surface of the Barr body would be more characteristic of euchromatin with genes that escape inactivation and some inactive genes that are surrounded by active genes or genes potentiated for activity. Especially interesting are genes on the Xp from Xp22 to the Xp telomere, including the pseudo-autosomal region 1 (PAR1). These would be at the surface of the Barr body and more readily disrupted by an expanding nucleolus under

Figure 3 | Establishment of the inactive X chromosome (Xi). Early in embryogenesis one of the two X chromosomes in female cells is inactivated by persistent expression of the X inactivation specific transcript RNA (XIST) from the X inactivation center (XIC). XIST RNA does not code for protein but remains in the nucleus and binds contiguous chromatin (i.e., the Xi, a.k.a. the Barr body), recruiting epigenetic silencing effectors (e.g., DNA methyltransferases). Approximately 95% of genes from the long arm (Xq) and 65% of genes from the short arm (Xp) are silenced. Silenced genes shown as dark blue, while genes that escape inactivation are shown as light blue [based on Ref. (89)]. The result is the Barr body which appears as a dense heterochromatic structure near the nuclear membrane. The bulk of the heterochromatic core contains Xq genes with some Xp genes, and the euchromatic-like surface layer has primarily Xp genes that are: actively expressed; potentiated for expression; or silenced but adjacent to expressed genes. Particularly interesting in the Xp is the pseudo-autosomal region 1 (PAR1) which has an abundance of Alu elements (46). In addition, Xp22 contains a "hot" LINE-1 sequence that can code for a fully functional reverse transcriptase. Xp22 also contains a fragile site (FRAXB). Fragile sites are preferential locations for viral insertions. And Xp22 on the Xi contains epigenetically silenced genes for spermine synthase (SMS) and spermidine/ spermine N1 acetyltransferase (SAT1). Overexpression of SMS and/or SAT1 that could occur with disruption of epigenetic silencing on the Xi can impact cellular methylation and polyamine types and levels. This could also impact polyamine activity in the nucleoli.

stress. SMS and spermidine/spermine N1 acetyltransferase (SAT1) are involved in polyamine synthesis and recycling, respectively, and normally SMS and SAT1, located at Xp22.1, are inactive on the Barr body (87). However, disruption of the Barr body by enlargement of the nucleolus, as shown in **Figure 2B**, could lead to reactivation of SMS and SAT1. This would result in a rapid increase in polyamine synthesis and recycling beyond what was already induced by the host cell and the viral activity. There would be an increase in acetylated polyamines by SAT1 that could interfere with RNA folding and, through oxidation, generate putrescine which, in turn, could allosterically increase S-adenosylmethionine (SAM) decarboxylase activation reducing SAM needed for DNA and protein methylation (72). Excess free polyamines can be conjugated to proteins by transglutaminases and acetylated polyamines, and the conjugated polyamines and putrescine can be oxidized to toxic acrolein. There is a close relationship between the intensity of SjS and the appearance of acrolein-conjugated proteins (88). The net effect of expression of SMS and SAT1 from the disrupted Barr body is dysregulation of polyamine levels. Add to this the fact that SAT1 can undergo super induction, meaning there could be a several 100-fold increase in polyamine acetylation. In the nucleolus, there could be an increase in polyamines and now acetylated polyamines that interfere with normal RNA folding and RNP assembly. Once SAT1 becomes active from both X chromosomes from nucleolar disruption of the Barr body, going forward there could be a drop in polyamines during subsequent stress events as super induction of SAT1 acetylates polyamines. It may follow that the nucleolus can no longer expand sufficiently to adapt to new stresses and the nucleolus can no longer work efficiently in proper folding and assembly of RNAs and RNPs, leading to creation of autoantigens. In other words, an initial stress-induced polyamine-driven expansion of the nucleolus could disrupt the Barr body leading to RNA pol III Alu transcript-driven fragmentation of the nucleolus. Subsequently, with reactivation of SAT1 from the Barr body, it may reduce polyamines and reduce the ability of the nucleolus to function normally in folding and assembly of RNAs and RNPs and fail to react effectively to future stressful events since polyamines will be rapidly acetylated as they are synthesized. This scenario could result in ongoing generation of abnormal, potentially autoantigenic RNPs due to the compromised nucleolus that lacks sufficient polyamines.

Another problem that could arise is reverse transcription. Most LINE-1 elements have mutated sufficiently so that they no longer code for functional reverse transcriptases. A few, including one in Xp22, can still produce functional reverse transcriptases but are suppressed by positioned nucleosomes (1). Reverse transcription of Alu elements could be particularly consequential. Alu elements are rich in G–C base pairs; therefore, reverse transcribed Alu DNA would require significant methylation since hypomethylated DNA would be interpreted as exogenous. LINE-1 reverse transcriptases preferentially reverse transcribe LINE-1 RNA at a rate of 1,000× and Alu RNA at a rate of 300× in comparison to other RNAs (90). The cell could quickly become inundated with hypomethylated Alu DNA. Li and Steinman reported a high content of Alu DNA (up to 55%) in the free DNA in sera of lupus patients whereas Alu elements only account for 10.8% of the human genome (91). Those authors suggested that reverse transcription could be a possible cause. We proposed that fully functional LINE-1 elements activated from disruption of the X chromosome could be involved in such reverse transcription.

## EXPANDING THE "X CHROMOSOME– NUCLEOLUS NEXUS" HYPOTHESIS: DISRUPTION OF THE NUCLEOLUS

The earlier version of the "X chromosome–nucleolus nexus" hypothesis suggested that there could be consequences from expression of previously sequestered Alu elements, particularly from the PAR1 of the X chromosome short arm where there is an exceptionally high content of Alu elements (1). We can now add detail to the hypothesis regarding what consequences could arise from RNA pol III expression of these Alu elements.

There are over 1,000,000 Alu elements spread throughout the human genome but most are suppressed by a positioned nucleosome. Displacement of the nucleosome could open the Alu element's internal RNA pol III transcription start site. RNA pol III can be quite prolific since it does not require energy (ATP), does not require extensive assembly of transcription factors, can initiate from the intragenic promoter in Alu elements, and can rapidly reinitiate to generate multiple transcripts. In addition, since Alu elements average only 300 bp and the intragenic promoter requires only about 70 bp for transcription factor binding, displacement of only one nucleosome would be all the opening needed. The abundance of RNA pol III typically found near the nucleolus, particularly the perinucleolar compartment, could rapidly generate thousands of Alu RNAs if there were a disruptive event, such as encroachment of the nucleolus into the Barr body. Especially vulnerable is the dense cluster of Alu elements in the PAR1 region near the surface of the Barr body. Whereas Alu elements comprise 10.8% of the human genome, they are at only 8% in the X chromosome. However, Alu elements account for 28.8% of the PAR1 region and 19% of the adjacent S5 region (46). Since PAR1 has approximately 2.5 million base pairs, there are estimated to be more than 2,500 Alu elements in PAR1 that could potentially flood the nucleus and nucleolus with Alu RNA transcripts (**Figure 4A**). Contrast this with the approximately 200 active ribosomal RNA genes in the nucleolus. The Alu RNAs could interfere with assembly of the SRP which contains an Alu domain that binds SRP 9/14 heterodimers (72). Free Alu RNA could compete in binding the SRP 9/14 leaving incomplete SRPs that cannot halt ribosomal activity in the cytoplasm when needed during synthesis of extracellular targeted proteins. This could lead to improper modifications (e.g., transglutamination) and localization of proteins. And, opening of the Alu elements, which have extensive intra-strand matching sequences, could also facilitate formation of cruciforms in the DNA which could be stabilized by polyamines. These cruciforms could be interpreted as autoantigens by the immune system.

Perhaps, the greatest danger from expression of Alu elements from the disrupted Barr body is deduced from the work of Caudron-Herger and colleagues mentioned previously (59). An abundance of RNA pol III Alu transcripts from the disrupted Barr body could compete with or lead to degradation of the RNA pol II intronic Alu RNA that, along with nucleolin and nucleophosmin, provides structural integrity for the nucleolus. This would lead to fragmentation of the nucleolus into inefficient subunits (**Figure 4B**). These nucleolar-derived subunits could have abnormal levels of polyamines and acetylated polyamines that cannot properly fold RNA and assemble RNPs. In fact, the needed components for assembly of an RNP like the ribosome may be unequally distributed among the nucleolar fragments preventing complete assembly. And there could be viral proteins and RNAs competing to join RNP assemblies. The RNPs and partial assemblies could be stabilized in abnormal conformations and associations by the polyamines and become autoantigenic when released from the cell. Such extracellular exposure could occur by blebbing and microparticle release as the cell enters apoptosis, NETosis or other forms of termination (92). With a loss of integrity of the nucleolus and expression of viral components, some nucleolar material could be displayed on the cell surface, such as the La protein (81). In addition, nucleolar fragmentation could lead to loss of nucleolar control of cell cycling, such as assembly of centromeres. Another problem that could arise is involvement of cyclic GMP-AMP synthase (cGAS) which detects cytosolic DNA. This includes detection of micronuclei that contain DNA from a disrupted nucleus or from DNA damage (93). Fragmentation of the nucleolus, as described above, could potentially generate such micronuclei, especially when centromere assembly and functioning are disrupted or when there are lagging chromosomes during mitotic segregation of chromosomes. The appearance of hypomethylated reverse transcribed Alu DNA in the cytosol, possibly originating from X-linked LINE-1 reverse transcription of PAR1 Alu element RNA, could also trigger the cGAS-STING pathway. Formation of cyclic GMP-AMP (cGAMP) can trigger activation of the Stimulator of Interferon Genes (STING) protein which induces transcription of interferon β (IFNβ) as part of the innate immune response in antiviral, antibacterial, and anticancer activity and is suspected of involvement in autoinflammatory and autoimmune diseases (94).

Therefore, the original "X chromosome–nucleolus nexus" hypothesis, which explained how the nucleolus could disrupt the inactive X chromosome, can now be expanded to include disruption of the nucleolus by X-linked Alu RNA transcripts that lead to nucleolar fragmentation and generation of autoantigenic material.

## THE "X CHROMOSOME–NUCLEOLUS NEXUS" HYPOTHESIS IN RELATION TO OTHER DISEASES

The "X chromosome–nucleolus nexus" hypothesis was developed primarily with SLE in mind but it could be involved in many autoimmune diseases. The mechanism could have differing effects due to the cell types and locations involved. For example, RA and SLE can have some of the same autoantigens targeted,

Figure 4 | Autoantigens generated by disruption of the nucleolus. (A) The original version of the "inactive X chromosome and nucleolus nexus" hypothesis proposed that there is disruption of the inactive X by the nucleolus due to an extraordinary expansion of the nucleolus under stress (Figure 2B). This disruption could open previously sequestered DNA, especially Alu sequences and genes in the short arm of the Xi that are located in the euchromatic-like surface layer of the Xi (1). Now, based on the work by Caudron-Herger and colleagues (59), we can propose additions to the hypothesis, that X-linked Alu transcripts generated by the abundant RNA pol III near the nucleolus can disrupt the nucleolin-RNA pol II intronic Alu complexes, either by Dicer degradation or by direct competition between the RNA pol III Alu transcripts and the intronic Alu sequences. (B) The subsequent fragmentation of the nucleolus could result in nucleolar fragments that contain conformationally abnormal autoantigenic structures due to improperly folded RNAs or improperly assembled ribonucleoprotein complexes (RNPs). For example, in some nucleolar fragments there may be insufficient quantities of ribosomal components (either RNAs or proteins) and, therefore, complete functional ribosomes cannot be formed. There may also be incorporation of viral RNA and/or viral proteins into the RNPs. Also, overexpression of X-linked spermine synthase and/or spermidine/spermine N1 acetyltransferase could result in abnormal types and levels of polyamines in the nucleolus and nucleolar fragments. For example, there may be putrescine, acetylated polyamines, and/or nuclear aggregates of polyamines in the nucleolus that interfere with RNA folding. Normally one would expect only spermine and spermidine to be present in large quantities. Extracellular release (by apoptosis, necrosis, NETosis) of these abnormal nucleolar products could provoke an autoimmune reaction that later targets the more abundant normal products due to epitope spreading.

such as Z-DNA, but RA is primarily behind the synovial membrane reducing full exposure of antigenic and autoantigenic material to the immune system. However, continual attraction of neutrophils to the same confined inflammation site in RA where the neutrophils undergo NETosis in an ineffective attempt at clearing abnormal material would produce chronic local exposure of cells and extracellular material to the neutrophil's active peptidyl arginine deiminases (PADs) producing high levels of citrullinated proteins (e.g., collagen) that eventually provokes the adaptive immune system into producing autoantibodies targeting the modified collagen as a major autoantigen in RA (85). In a similar manner, MS is confined behind the bloodbrain barrier reducing access of autoantigenic material to the immune system but, again, neutrophils continually attracted to an MS lesion could be releasing PADs that citrullinate myelin, eventually triggering autoantibodies to citrullinated myelin basic protein which is a major autoantigen in MS. SLE is a systemic disease suggesting that the immune system can more readily react to the broad array of abnormal material seen in SLE and generate autoantibodies. Compared to RA or MS, the autoantigens, autoantibodies and complexes of the two in SLE have easier access to the circulatory system allowing the reaction to spread to and deposit in different organs rather than being confined behind a membrane barrier. SjS can be a primary disease or it can be secondary to SLE, RA or MS. This suggests that there could be similar mechanisms occurring in all four of these disorders. Involvement of polyamines, possibly due to loss of epigenetic control of X-linked polyamine genes, is suspected in SjS since the appearance of acrolein conjugated proteins is related to the intensity of SjS and acrolein is an oxidation product of polyamines (88). The hypothesis may even play a role in Alzheimer's disease (ALZ) since there is a female bias in the disease and there are autoantibodies involved in ALZ (95). In addition, polyamine levels are altered in ALZ (96, 97) along with increased acrolein (98); SAM levels are greatly decreased (99); polyamines are involved in plaque formation (100) and nucleolar poly (ADP-ribose) polymerase 1 (PARP1) is decreased in ALZ (101). Cellular stress that leads to disruption of the inactive X chromosome and/or the nucleolus could play a role in the altered polyamine activity, decreases in SAM, decreases in PARP1, and appearance of acrolein.

The hypothesis as a whole or in parts (part 1: inactive X disruption and/or part 2: nucleolar disruption) could have a role in some cancers. The inactive X chromosome is often missing in tumor cells from breast and ovarian cancers (102). The inactive X may have reactivated from a decrease in methylation (possibly due to over activity of polyamine synthesis and recycling) or the inactive X was lost due to improper segregation of chromosomes to daughter cells (possibly due to centromere assembly in the nucleolus). Viruses could be involved in the loss of X inactivation since viruses increase polyamine levels in order to increase nucleolar activity for their benefit, as exemplified by EBV induction of ODC, SDS, and SMS via increased MYC activity. The subsequent reduction in SAM due to polyamine synthesis would make it difficult for the cell to maintain chromatin methylation required for epigenetic silencing in the X chromosome and other chromosomes leading to disruption of control of oncogenes and tumor suppressor genes and there is the possibility of opening previously sequestered viruses that then try to take control of the nucleoli. The inactive X has the greatest demands for methylation but it is the last chromosome to be replicated and repackaged in late S phase or even early G2 when SAM levels would have already been impacted by methylation of other chromosomes. Reactivation or loss of the inactive X chromosome could explain some of the cases of triple-negative breast cancer in which there is no overexpression of HER2, estrogen, or progesterone receptors (103). So this epigenetic scenario of Barr body disruption could explain some of the enigmatic cases of cancers. Also, keep in mind that viral disruption of nucleoli could interfere with nucleolar involvement in DNA repair, nucleolar assembly of centromeres, and alter nucleolar control of cell-cycle kinases (78). Fragmenting of nucleoli as centromeres are being formed could lead to abnormal distribution of chromosomes resulting in daughter cells of differing karyotypes, such as a parent (46,XX) cell generating daughter cells of (45,X0) and (47,XXX). Disruption of the nucleolus in tumor cells could also lead to appearance of autoantigens. Autoantigens can arise in cancers but they differ from those normally seen in autoimmune diseases such as SLE. The differences could arise from: the cell type involved (proliferating versus mature, differentiated); the nucleolar activity and content at the time of disruption; and the rapidity of the disruption (acute versus gradual accumulation). We can consider that nucleoli in proliferating cells would be heavily involved in cell cycling, such as generating centromere components, and so many of the autoantigens that arise would be expected to be related to cell cycling and suppression of apoptosis (104). Autoimmune diseases, such as SLE, give rise to autoantigens that are components more routinely found in abundance in nucleoli, such as ribosomal or splicosomal components.

There is a slightly higher risk of cancers among autoimmune patients but the risk varies with regards to the type of cancer (105). Therapeutics taken by the patient targeting the autoimmune disease could contribute to cancer development. Hematological, thyroid, lung, and vulva cancers show an increased risk with non-Hodgkin's lymphoma showing a 3× to 4× greater risk in lupus patients, while breast, endometrial, and ovarian cancers show a lower risk. For now there is no direct connection between viruses and the "X chromosome–nucleolus nexus" hypothesis to the increased risk of cancers among autoimmune disease patients but we can consider the induction by viruses of increased polyamine levels and the possible reactivation of X-linked polyamine genes as means by which competition for the cellular methyl donor, SAM, could reduce DNA methylation and open oncogenes for overexpression in proliferation competent cells. Increased nucleolar size due to increased polyamines could add to the disruption of neighboring epigenetically silenced chromatin to expose alleles for expression.

## CONCLUSION

The focus of this discussion has been on viruses and the "X chromosome–nucleolus nexus" hypothesis since we now understand the effects a virus can have on the nucleolus and how it is to the benefit of the virus to influence the nucleolar activity. And EBV has been used as the primary example of viral involvement in autoimmune diseases since it is one of the viruses most suspected of having such a role and we can connect EBV actions (e.g., increased MYC activity) to increases in polyamines that could directly impact the nucleolus and trigger the hypothesized mechanism. Other factors besides viruses, such as bacteria or chemicals, can contribute to autoimmune diseases but the means is less clear and may not closely follow the mechanism proposed. Bacteria, for example, produce putrescine and spermidine without the extensive controls on polyamine synthesis seen in eukaryotes but the bacteria can produce their own machinery and are not as dependent as viruses on the cell's nucleolus. In addition, disruption of heterochromatin, such as the inactive X chromosome, can open latent viruses. The X chromosome has four major fragile sites and fragile sites are frequently the locations chosen for viral insertion. The particulars of the fragile site and its vulnerability to viral insertion may add to the genetic susceptibility of an individual.

The previous version of the "X chromosome–nucleolus nexus" hypothesis, or simply the "nucleolus" hypothesis, explained how an overly stressed nucleolus could disrupt neighboring heterochromatin, especially the Barr body. As a result, there could be detrimental increases in synthesis and recycling of polyamines that could impact cellular methylation and potentially stabilize autoantigenic complexes of nucleolar and chromatin components. The point was made that many autoantigens in SLE are, at least transiently, components of the nucleolus but the means by which such nucleolar components could become autoantigenic was not presented. Now, with this work, the hypothesis has been extended to include disruption of the nucleolus as an additional step. A disrupted Barr body could generate an abundance of polyamines and Alu RNA from X-linked genes and elements that further stress and damage the nucleolus, making it very inefficient in its functions, even fragmenting it and possibly leading to cell death. And there may be overexpression of SAT1 that hampers the nucleolus in subsequent stress events since polyamines may be converted to a predominance of acetylated polyamines that are less effective at or even detrimental to proper nucleolar folding and assembly of RNAs and RNPs. Meanwhile during nucleolar disruption autoantigens may be created, stabilized and released extracellularly.

Further work is needed to understand how the various autoantigens provoke the autoimmune response. Is it, for example, a conformational alteration of the ribosomal subunits stabilized by polyamines, or incomplete assembly of an RNP? Or could it be "guilt by association" such as SSA/Ro bound to misfolded RNAs stabilized with polyamines that prevent proper refolding? There may be incorporation of viral components in the RNPs that make it autoantigenic. Or could it be abnormal localization and/or modification of proteins that are misdirected due to Alu RNA interference with SRP assembly (72). Epitope spreading from the autoantigenic complex to the normal endogenous protein would seem to have a role in the autoimmune response with the greater abundance of the endogenous protein then providing more of the provocation than the original autoantigenic complex. And the fragmentation of nucleoli, as described here, could lead to extracellular signaling and extracellular exposure of autoantigens. Testing of the hypothesis can use powerful approaches, such as computational molecular dynamics and single cell analysis, that have now reached sophistication that allow us to explore the interactions of nucleolar components as they are normally processed and the possibilities of how abnormalities could occur. For example, what is the effect of acetylated polyamines if they were to compete with spermidine and spermine in the nucleolar folding and assembly of RNPs? What are the interactions and resulting structures of intronic Alu RNA with nucleolin? And what is the distribution and composition of RNAs and proteins in nucleolar fragments compared to intact nucleoli? And, perhaps most important, what does this new hypothesis present as far as therapeutic targets? Certainly suppressing viral activity, MYC activity, polyamine synthesis, and

#### REFERENCES


polyamine recycling are important targets but also newer areas, such as the cGAS-STING pathway are promising targets too. The importance of Alu elements implied by this hypothesis calls into question the use of mouse models of autoimmune diseases since mice do not have the extensive amount of Alu elements seen in humans (certainly not a cluster of 28.8% seen in the PAR1 of the human X). In addition, the mouse X chromosome is telocentric (just one long arm) whereas the human X is submetacentric (a long arm and a short arm). The mouse X inactivation would be relatively consistent since it does not negotiate a centromere. X inactivation researchers complain that it is difficult to study partial X reactivation in mice due to the consistency of inactivation along the mouse X. This makes the murine X more robust under stress, at least with regards to this hypothesis (1).

We have previously discussed the short arm of the X chromosome (Xp) and especially the portion from Xp21.2 to the terminus as having a major role in SLE, particularly with regards to possible reactivation of the inactive X chromosome (1–3, 72, 82, 106, 107). This section includes: a "hot" LINE-1 (codes for a fully functional reverse transcriptase); the polyamine genes spermidine/spermine N1 acetyltransferase and SMS; and the PAR1 region with an abundance of Alu elements. Other groups are just now coming to the conclusion that the Xp arm has a major role in autoimmune diseases (108) although they have not mentioned the Alu elements, LINE-1 and polyamine genes we have mentioned and they have not made the connection to the nucleolus. It is hoped that autoimmune disease researchers will consider the "X chromosome–nucleolus nexus" hypothesis since it is the most comprehensive explanation yet for autoimmune diseases. It does involve areas with which those researchers may not be familiar, such as polyamines, X inactivation, epigenetics, and the nucleolus making it a rather complex scenario but autoimmune diseases are very complex phenomena.

#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

## FUNDING

The author has no funding other than as a Research Assistant Professor in the Chemistry Department at the University of South Florida, Tampa, FL USA.


autoimmunity that portends type 1 diabetes. *Diabetes* (2014) 63(2): 446–55. doi:10.2337/db13-0619


Alzheimer disease. *Neurosci Lett* (1996) 206:193–5. doi:10.1016/S0304- 3940(96)12451-4


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2017 Brooks. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Human Endogenous Retrovirus-K and TDP-43 Expression Bridges ALS and HIV Neuropathology

Renée N. Douville1,2 \* and Avindra Nath<sup>3</sup>

<sup>1</sup> Department of Biology, University of Winnipeg, Winnipeg, MB, Canada, <sup>2</sup> Department of Immunology, University of Manitoba, Winnipeg, MB, Canada, <sup>3</sup> Section of Infections of the Nervous System, National Institute of Neurological Diseases and Stroke, National Institutes of Health, Bethesda, MD, United States

Despite the repetitive association of endogenous retroviruses in human disease, the mechanisms behind their pathological contributions remain to be resolved. Here we discuss how neuronal human endogenous retrovirus-K (HERV-K) expression in human immunodeficiency virus (HIV)-infected individuals is a distinct pathological aspect of HIV-associated neurological conditions, such as HIV encephalitis and HIV-associated neurocognitive disorders. Enhanced neuronal HERV-K levels were observed in the majority of HIV-infected individuals, and to a higher degree in brain tissue marked by HIV replication. Moreover, we highlight an important neuropathological overlap between amyotrophic lateral sclerosis and HIV encephalitis, that being the formation of neurotoxic TDP-43 deposits in neurons. Herein, we argue for enhanced transdisciplinary research in the field of ERV biology, using an example of how HERV-K expression has novel mechanistic and therapeutic implications for HIV neuropathology.

#### Edited by:

Wesley H. Brooks, University of South Florida, United States

#### Reviewed by:

Yoshinao Kubo, Nagasaki University, Japan Amr Aswad, University of Oxford, United Kingdom Antoinette Van Der Kuyl, University of Amsterdam, Netherlands

> \*Correspondence: Renée N. Douville r.douville@uwinnipeg.ca

#### Specialty section:

This article was submitted to Microbial Immunology, a section of the journal Frontiers in Microbiology

Received: 31 July 2017 Accepted: 27 September 2017 Published: 11 October 2017

#### Citation:

Douville RN and Nath A (2017) Human Endogenous Retrovirus-K and TDP-43 Expression Bridges ALS and HIV Neuropathology. Front. Microbiol. 8:1986. doi: 10.3389/fmicb.2017.01986 Keywords: human immunodeficiency virus (HIV), human endogenous retrovirus-K (HERV-K), TDP-43, NeuroAIDS, amyotrophic lateral sclerosis (ALS)

### INTRODUCTION

Originating from ancient retroviruses that overcame host defense mechanisms and permanently integrated into the DNA of our hominid ancestors, endogenous retroviruses (ERVs) occupy over 8% of the human genome. The process of retroviral endogenation has resulted in at least 31 independently acquired ERV genera in the human genome (Belshaw et al., 2005; Blikstad et al., 2008; Subramanian et al., 2011). Despite roles in homeostasis (Brattas et al., 2017; Meyer et al., 2017), ERVs are increasingly being recognized as integral players in the pathogenesis of many human diseases [reviewed in Ruprecht et al. (2008), Balada et al. (2009), Manghera et al. (2014), Christensen (2016)]. Furthermore, concrete examples in other species indicate how ERVs may drive pathogenic mechanisms such as cellular transformation and immune dysregulation [reviewed in Kassiotis (2014), Mager and Stoye (2015)]. This is illustrated by mice deficient in select TLRs (−3, −7, −9), which exhibit endogenously derived MuLV viremia and development of acute T cell lymphoblastic leukemia (Yu et al., 2012). It also highlights the fact that both innate and adaptive immune responses are essential for control of ERV expression (Stetson et al., 2008; Hurst and Magiorkinis, 2015); conversely, ERVs also regulate immunity genes (Hurst and Magiorkinis, 2015; Chuong et al., 2016). Neurological effects caused by ERVs are also documented. Neurovirulent strains of MuLV such as Cas-Br-E cause infection of microglial cells with neurodegeneration in the brain and spinal cord which resembles amyotrophic lateral sclerosis

(ALS) and is mediated by the envelope protein of the virus (Lynch and Sharpe, 2000; Jolicoeur et al., 2003).

Ongoing clinical trials seek to determine the value of ERVs as biomarkers and therapeutic targets (clinicaltrials.gov NCT02437110 and NCT02782858) (Alfahad and Nath, 2013). Breakthroughs in one specialty are likely to translate into other disease contexts, due to the overlap of ERV expression patterns in multiple conditions (Brutting et al., 2017). Herein, we argue for enhanced transdisciplinary research in the field of ERV biology, using an example of how endogenous retrovirus-K (HERV-K) expression has novel implications for human immunodeficiency virus (HIV) neuropathology.

#### HIV INFECTION: MORE THAN ONE RETROVIRUS AT PLAY?

Even with antiretroviral therapy, the clinical treatment of HIV infection is often complicated by neurocognitive disorders (Robertson et al., 2007). The occurrence of mild neurocognitive deficits, such as those impacting daily living activities, have notably increased among HIV<sup>+</sup> populations – despite a decline in the more severe forms of HIV-associated neurocognitive disorders (HAND) with the advent of antiretroviral therapy (McArthur et al., 2010). Even worse, there is currently no targeted treatment to prevent the onset or progression of HAND. Clinical trials with a wide variety of neuroprotective drugs, antiinflammatory agents, and antioxidants have failed or shown only minor effects. Risk factors for the progression to HAND include genetic, viral, and co-morbid factors, such as drug use and increasing age (Saylor et al., 2016).

Human immunodeficiency virus enters the central nervous system (CNS) soon after initial infection. Whether HIV drives neurodegeneration by slow and progressive pathological changes or sudden alterations caused by systemic immunosuppression is unclear. In the CNS, productive HIV infection is supported by cell types such as astrocytes, microglia, and macrophages which subsequently elicit chronic pro-inflammatory responses [reviewed in Kraft-Terry et al. (2010)]. Both HIV infection and inflammatory signals mediated by cytokines such as IL-6, IL-1β, TNFα, and IFNγ, can trigger the expression of ERVs (Serra et al., 2003; Contreras-Galindo et al., 2007b), and specifically HERV-K (Manghera et al., 2016b).

Human immunodeficiency virus infection can promote ERV expression in proliferating peripheral blood mononuclear cells (PBMC) (Contreras-Galindo et al., 2007b). Longitudinal analysis of PBMC from HIV-infected patients shows that increased HERV-K expression precedes spikes of HIV replication in select individuals (Contreras-Galindo et al., 2007a). Enhanced HERV-K levels were observed in patients treated sub-optimal therapeutic doses of antiretrovirals, or those who outright failed to respond to HAART therapy (Contreras-Galindo et al., 2006, 2007a). Further evidence for a dynamically intertwined relationship between HERV-K and HIV replication stems form the observation that HIV elite controllers have robust cellular and antibody responses against the HERV-K (HML-2) capsid protein (de Mulder et al., 2017).

Until recently, it was unclear if HERV-K re-activation stays limited to peripheral tissues or is a feature of HIV neuroinvasion. Independent groups have now shown that HERV-K proteins accumulate in cortical neurons of patients with HIV infection (Bhat et al., 2014). Specifically, here we sought to determine if HIV infection in the brain is associated with the up-regulation of HERV-K expression, and additional markers of neuronal dysfunction.

#### AN EXPLORATION OF HERV-K EXPRESSION IN HIV<sup>+</sup> BRAIN TISSUE

We have previously demonstrated enhanced HERV-K expression in cortical and spinal neurons of patients with ALS (Douville et al., 2011; Li et al., 2015). Therefore, we examined the extent of cortical HERV-K expression in HIV patients with encephalitis and/or HIV-associated neurocognitive disorder (HIV-E/HAND, n = 6,), HIV patients without encephalitis (HIV, n = 9), and individuals with chronic systemic illness (controls, n = 7), through detection and localization of the viral reverse transcriptase (RT) protein within autopsy tissues (**Table 1** and **Figure 1**).

In both HIV-infected groups, increased HERV-K RT protein was detectable (using AbNova #H00002087-A01 antibody), but surprisingly expressed at similar levels (**Figure 1A**). Based on the findings of Contreras-Galindo et al. (2007a), we expected HERV-K induction to occur immediately preceding or during HIV replication in the brain. HIV-E is a multifocal disease; therefore, varying levels of HIV replication occurs within the many regions of the brain. This leads to an intrinsic sampling bias when examining excised brain specimens, as select tissue samples may not be representative of the global pattern of HIV-associated pathology. Future studies would benefit from larger sample sizes and from sampling multiple brain regions per individual to improve the chances of identifying focal lesions. Considering these caveats, we evaluated the degree of HIV replication in each tissue specimen based on the presence or absence of HIV p24 positive cells as determined by immunohistochemistry, and stratified the tissues from HIV-infected individuals accordingly. HERV-K RT expression was significantly elevated in brain tissue with p24 reactivity (**Figure 1B**). This finding is consistent with the observation that productive HIV infection in PBMC triggers HERV-K expression (Contreras-Galindo et al., 2007b; Laderoute et al., 2007). Since HIV<sup>+</sup> patients who fail to respond to HAART therapy exhibit elevated HERV-K levels in PBMC (Contreras-Galindo et al., 2007a), our results suggest that HERV-K expression in the brain may reflect inadequate drug penetration into the CNS or an unsuccessful response to antiretroviral treatment. Moreover, plasma HIV loads were associated with HERV-K RT measurements in cortical tissue (Spearman's correlation p = 0.052), supportive of the idea that systemic HIV replication favors HERV-K re-activation in both the CNS and peripheral compartment. Productive HIV infection in neocortical tissues is characteristic of HIV-E/HAND (McArthur et al., 2010), and could also play a significant role in milder neurocognitive disorders. Moreover, robust HERV-K expression


#### TABLE 1 | Cortical brain tissue specimens.

fmicb-08-01986 October 9, 2017 Time: 15:34 # 3

Tissue specimens were obtained from the California NeuroAIDS Tissue Consortium (CNTC), the Texas Repository for AIDS Neuropathogenesis Research (TRANR), the National NeuroAIDS Tissue Consortium (NNTC), the Human Brain and Spinal Fluid Resource Center (HBSFRC), the Rocky Mountain MS Center (RMMSC), and the Johns Hopkins School of Medicine Brain Bank (JHSMBB). Among confirmed cases with HIV infection (as indicated by CD4 count, plasma and cerebral spinal fluid viral loads), brain tissue samples were further classified as having HIV-encephalitis (HIV-E) based of neuropathological examination or HIV-associated dementia (HAD) based on clinical history. Post-mortem interval (PMI) is indicated in hours. No data (ND).

was seen in several HIV<sup>+</sup> patients not clinically diagnosed with HIV-E/HAND. This points to pathological but modest HIV replication and/or HIV-associated neuroinflammation driving HERV-K activity within the brain, which precedes overt clinical signs and symptoms of neurocognitive impairment.

A notable finding is that the enhanced HERV-K expression was limited to neuronal cells in HIV<sup>+</sup> specimens, yet HIV is not trophic for neurons. Local neuroinflammation is likely a key driver of HERV-K expression in the brain, and is supported in a study of ALS neuropathology (Manghera et al., 2016b). Alternatively, the enhanced neuronal expression of HERV-K may be directly mediated by viral proteins released from HIV-infected cells, such as HIV Tat (Gonzalez-Hernandez et al., 2012, 2014). HIV Tat protein can be released extracellularly from infected macrophages within the brain and then taken up by neurons [reviewed in Li et al. (2009)]. In depth analysis of the HERV-K promoter suggests that several transcription factors related to inflammation, hormone regulation, and tissue-specific signaling may also modulate the transcription of this ERV (Manghera and Douville, 2013).

However, a consequence of neuronal HERV-K upregulation may be a blockade on the progression of HIV infection in the brain. Experimental evidence suggests that HERV-K expression may be a limiting factor for HIV replication in neurons. In vitro modeling shows that HERV-K Gag and Env proteins can play an essential role as retroviral restriction factors, thus limiting HIV replication (Monde et al., 2017; Terry et al., 2017). Furin processing of HERV-K Env impacts HIV-1 production, as well as several key amino acid residues in the surface domain altering the degree of HIV production in HERV-K108 versus HERV-Kcon Env overexpressing 293T cells (Terry et al., 2017). Furthermore, it has been demonstrated that HERV-K Env (type 2) expression in neural cells is a protective mechanism against HIV replication in these cells, and that this may confer a degree of protection against a variety of insults (Bhat et al., 2014). Despite HERV-K activity impeding HIV replication in neurons, the brain remains negatively impacted by elevated HERV-K expression in neurons. Recent data show that the expression of HERV-K in mature neurons is toxic, with data also showing a neuropathological impact of HERV-K Env on mature neurons, as transgenic mice expressing HERV-K env are afflicted with severe neuronal damage and progressive motor dysfunction (Li et al., 2015). Another key point is that few studies have considered the potential roles of other HERV-K-encoded proteins on the viability and function of neurons. Clearly, we need a broader view of the impact of HERV-K in neurological disorders.

#### THE SEARCH FOR AN HERV-K-ASSOCIATED BIOMARKER

Identification of HERV-K-associated biomarkers is a key step for both future research and clinical trials in a variety of HERV-K-associated diseases. Our previous study of patients with ALS

FIGURE 1 | Endogenous retrovirus-K reverse transcriptase is induced in cortical tissue during human immunodeficiency virus (HIV) infection. (A) HIV-infected individuals, with HAND/HIV-encephalitis (HIV-E) or without HIV-E (HIV) expressed greater levels of endogenous retrovirus-K (HERV-K) reverse transcriptase protein in their cortical tissue, as compared to patients deceased with chronic systemic disease (control). Antibodies against the HERV-K reverse transcriptase protein (AbNova #H00002087-A01) and human TDP-43 (Protein Tech #10782-2-AP) were used for immunohistochemistry as previously described (Douville et al., 2011). (B) HIV replication in cortical tissue, as measured by HIV p24 protein immunostaining (mouse anti-HIV p24 Gag monoclonal, #24-4 NARRRP), is associated with significantly higher HERV-K reverse transcriptase expression. Mann–Whitney derived t-test, <sup>∗</sup>p < 0.05. (C) Significant correlation of neuronal HERV-K reverse transcriptase and TDP-43 protein levels in HIV<sup>+</sup> patients. Representative immunohistochemistry images of TDP-43 protein, endogenous retrovirus-K reverse transcriptase (HERV-K RT), nucleic as measured by DAPI staining, and neurons as measured by Nissl staining in the cortical brain tissue of HIV-infected patients (HIV+) versus patients with systemic disease (control). Arrows indicate weak TDP-43 positivity in control tissue neurons. ALS-derived neuron with a TDP-43 deposit outside the nuclear boundary is indicated with an asterisk. (D,E) A strong correlation between HERV-K RT and TDP-43 expression in brain tissues. ImageJ analysis was used to quantify the density of HERV-K RT and TDP-43 staining within individual tissue samples (D) and within individual neurons (E) of HIV-infected and controls cortical brain specimens. Spearman correlation used to assess association between HERV-K RT and TDP-43 expression patterns.

found that the extent of HERV-K RT expression was strongly correlated with TAR DNA binding protein-43 (TDP-43) in vivo (Douville et al., 2011). Further, the HERV-K LTR has four binding sites for TDP-43 which have been shown to regulate its activation (Li et al., 2015). A common event in ALS is the aberrant deposit of ubiquitinated and hyper-phosphorylated TDP-43 in the cytoplasm and nucleus of neurons (Mackenzie et al., 2007; Buratti and Baralle, 2008). Further, formation of TDP-43 aggregates has been shown to alter HERV-K RT and polyprotein levels and cellular localization of these viral proteins (Manghera et al., 2016a). Measurement of TDP-43 protein expression (Protein Tech #10782-2-AP antibody) by immunohistochemistry (**Figure 1C**) or western blot analysis (data not shown) consistently show an increase in TDP-43 levels in HIV<sup>+</sup> specimens, as compared to controls. Nuclear TDP-43 expression was enhanced on average six-fold in the cortical neurons of HIV patients, and was accompanied by enhanced TDP-43 phosphorylation (data not shown). Furthermore, co-expression of HERV-K RT and TDP-43 proteins occurred in the majority of neurons (**Figure 1C**). To examine this expression pattern quantitatively, the staining density measurements for HERV-K RT and TDP-43 were performed in tissues and co-labeled neurons. **Figures 1D,E** demonstrate that there is a significant positive correlation between TDP-43 expression and HERV-K RT expression in tissue (Spearman's correlation 0.72, p < 0.0001, n = 22) and within individual neurons (Spearman's correlation 0.72, p < 0.0001, n = 40). This data supports the idea that specific post-translational modifications of TDP-43 may alter its protein turnover rate resulting in modified nuclear expression, and the inception of cytoplasmic TDP-43 aggregates (**Figure 1C**, asterisk) (Mackenzie et al., 2007; Buratti and Baralle, 2008; Brady et al., 2011). This disruption in turn can impact HERV-K expression patterns (Manghera et al., 2016a).

As to why enhanced TDP-43 expression occurs in association with HERV-K, this phenomenon may reflect a similar transcriptional responsiveness to inflammatory signals. In silico analysis of the TDP-43 promoter reveals that like other interferon-stimulated genes, it harbors both IRF and κB binding sites and therefore may be transcriptionally up-regulated by pro-inflammatory responses. (Douville et al., 2011). HERV-K transcription and viral protein production are also strongly associated with the activity of the transcription factor complex of IRF1, and NF-κB p50 and p65 (Manghera et al., 2016b). Moreover, TDP-43 expression has been shown to be inducible during viral infection (Brasier et al., 2004). As TDP-43 was originally described as an inhibitor of HIV transcription (Ou et al., 1995), we postulated that it may act as a retroviral restriction factor, with a potential role in repressing both HIV and HERV-K provirus expression. Subsequent studies are not in agreement as to the potential of TDP-43 as a retroviral restriction factor, with some in support (Li et al., 2015; Krug et al., 2017), and others against (Nehls et al., 2014; Manghera et al., 2016a; Prudencio et al., 2017), perhaps reflecting cell-type or model-specific effects.

**Figure 1C** also highlights an important neuropathological overlap between ALS and HIV-E: the formation of nuclear and cytoplasmic TDP-43 deposits in neurons. The cleavage and subsequent aggregation of TDP-43 results in neurotoxicity (Yang et al., 2010; Che et al., 2011; Ratti and Buratti, 2016). There is a preponderance of TDP-43 pathology in over 90% of ALS cases (Irwin et al., 2015; Scotter et al., 2015), despite <1% of sporadic ALS cases having a clear genetic cause (Ling et al., 2013). Similarly, we observed that all HIV<sup>+</sup> brain tissues exhibited heightened TDP-43 expression, with select neurons containing cytoplasmic TDP-43 accumulation (**Figure 1C**, asterisk). Truncated forms of TDP-43 called TDP-25 and TDP-35 are known to seed native TDP-43 aggregation (Huang et al., 2014; Xiao et al., 2015), thus altering its capacity to perform cellular functions (Che et al., 2011; Monahan et al., 2016). With sequestration of TDP-43, the cell's ability to regulate RNA splicing is compromised (Scotter et al., 2015). Thus, HIV neuroinfection and ALS may share aberrant and altered protein deposition patterns, with pathogenic consequences.

The role of retroviruses (if any) in the cell-to-cell transmission of pathogenic TDP-43 moieties has yet to be elucidated (reviewed in Hanspal et al. (2017)]. This represents an attractive hypothesis considering that cortical neurons are known to secrete exosomes (Faure et al., 2006), and that exosomes have been shown to harbor retroviral cargo (Wurdinger et al., 2012; Vargas et al., 2014). Monitoring exosome composition could be an attractive source of clinical biomarkers for future studies on ERV-associated diseases.

## FUTURE DIRECTIONS

Despite repetitive association of ERVs in neurological diseases, such as ALS, multiple sclerosis, and schizophrenia, the mechanisms behind their pathological contributions remain to be resolved. Accumulating evidence suggests that HERV-K expression is a distinct pathological aspect of HIV-associated neurological disorders. Enhanced HERV-K RT expression in adult HIV<sup>+</sup> individuals was restricted to neurons, and most elevated within brain tissue exhibiting HIV replication. This is a similar pattern of neuropathology to that seen in ALS, where cortical neurons – including motor neurons – express HERV-K viral proteins (RT and envelope proteins) (Douville et al., 2011; Bhat et al., 2014; Li et al., 2015; Manghera et al., 2016b). Considering this notable association with motor neuron disease, it is important to point out the prevalence of motor disturbances in HIV infection. Perinatally-acquired HIV infection is associated with neurodevelopmental disturbances, including neurocognitive, gross motor, and psychomotor deficits (Blanchette et al., 2001; McGrath et al., 2006). Elevated rates of motor dysfunction and developmental delay is widely reported in HIV+ infants and children across socio-economic strata, with reported rates reaching 66.7% (Ferguson and Jelsma, 2009; Govender et al., 2011; Le Doare et al., 2012). Even seemingly asymptomatic HIV<sup>+</sup> children exhibit measurable motor deficits in standardized testing (Boivin et al., 1995). In adult HIV populations, motor symptoms were often found in conjunction with HAND, but in the post-HAART era it is less common to clinically observe deficits in motor skills and psychomotor speed (Heaton et al., 2011; Saylor et al., 2016). In both children and

adults, early initiation of combination antiretroviral therapy (cART) has been shown to improve motor function (despite remaining subnormal), but later initiation of cART failed to yield as notable clinical improvement (Le Doare et al., 2012; Kore et al., 2015; Saylor et al., 2016). The potential role of HERV-K in mediating neurodevelopmental abnormalities in the HIVinfected children needs to be explored. Importantly, HERV-K plays a critical role in early embryogenesis (Grow et al., 2015), and hence its dysregulation by HIV could potentially alter neurodevelopment.

In rare cases, an ALS-like syndrome can occasionally be caused by retroviruses such as HIV and human T cell leukemia virus type-1 (HTLV-1) (Matsuzaki et al., 2000; Verma and Berger, 2006). Antiretroviral therapy can reverse the symptoms of this ALS-like syndrome in HIV-infected individuals (Moulignier et al., 2001; von Giesen et al., 2002), suggesting that HIV replication in the CNS can drive a pathology which symptomatically resembles ALS. Indeed, recent attempts treat HIV-associated motor neuron disease with antiretroviral therapy showed promise, with reversal of recent onset symptoms or a protracted course of the illness following treatment (Bowen et al., 2016). In these patients, the clinical improvement paralleled a decrease in HERV-K viral load in plasma (Bowen et al., 2016). Future research into HIV-associated motor deficits in pediatric and adult populations should consider the potential of HERV-K in driving motor impairment.

Our results support the concept that a failure to control HIV infection, either from a lack of response to HAART therapy, receiving a sub-optimal regimen or the inability of antiretroviral drugs to adequately penetrate the CNS, may be associated with enhanced HERV-K in the brain. The use of antiretroviral drugs to suppress HIV replication in the CNS may have an indirect (or potentially direct) neuroprotective effect by limiting HERV-K-mediated pathology. There is currently a dearth of knowledge pertaining to the identification and use of antiretrovirals customized for use against HERV-K (Tyagi et al., 2017). It remains to be seen how optimization of drug regimens more tailored to inhibition of HERV-K proteins may improve upon these clinical results.

Cognitive and psychomotor symptoms in HIV have been associated with structural changes in the brain following HIV infection, enhanced inflammation and immune activation, as well as metabolic disturbances (Saylor et al., 2016). By merging transdisciplinary expertise, our findings point to an overlapping pathological contribution of TDP-43 deregulation and HERV-K re-activation in both HAND and ALS. Exploration into the pathogenic effects of TDP-43 proteinopathy in ALS has vastly

#### REFERENCES


expanded potential therapeutic options for this neurological disease, including small molecule activators of autophagy, the ubiquitin-proteasome system, and chaperone proteins (Scotter et al., 2015). Indeed, therapeutics overcoming TDP-43-mediated pathology may also be clinically useful in retrovirus-associated neurological disease. An improved understanding of the role of TDP-43 in the pathogenesis of HERV-K and HIV infection may hold benefit for either ALS or HAND, in addition to other TDP-43-associated disorders such as Alzheimer's disease (Davis et al., 2017) and prefrontal dementia (Lomen-Hoerth et al., 2003; Merrilees et al., 2010). Considering the seemingly disparate human conditions associated with ERV expression, more efforts into bridging disciplines is warranted, with the goal of elucidating shared molecular pathways and pathologies. Elevating the field of ERVs into the realm of mainstream biomedical research is worthwhile given its considerable clinical relevance. New outlooks on how ERV biology fits into our understanding of human disease may bring us closer to treating some of the most clinically difficult conditions, such as HIVassociated neurological disorders and ALS.

## ETHICS STATEMENT

De-identified autopsy brain samples were obtained from the National Neuro-AIDS Tissue Consortium (www.nntc.org). The use these tissues was determined to be IRB exempt.

#### AUTHOR CONTRIBUTIONS

RD and AN designed the study and wrote the manuscript. RD performed the experiments and analyzed the data. All authors read and approved the final manuscript.

### FUNDING

This work has been supported by grants from the National Institutes of Health (NIH).

## ACKNOWLEDGMENTS

We thank Dr. Wenxue Li for his advice on brain sample preparation for western blotting. We thank Marie-Josée Nadeau for careful review of the manuscript.

implications for present-day activity. J. Virol. 79, 12507–12514. doi: 10.1128/ JVI.79.19.12507-12514.2005




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Douville and Nath. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.