# THE GUT MICROBIOME IN HEALTH AND DISEASE

EDITED BY : Nathan W. Schmidt, Venkatakrishna Rao Jala, Michele Marie Kosiewicz and Pascale Alard PUBLISHED IN : Frontiers in Cellular and Infection Microbiology

#### 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 978-2-88963-003-5 DOI 10.3389/978-2-88963-003-5

#### 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

# THE GUT MICROBIOME IN HEALTH AND DISEASE

Topic Editors:

Nathan W. Schmidt, University of Louisville, United States Venkatakrishna Rao Jala, University of Louisville, United States Michele Marie Kosiewicz, University of Louisville, United States Pascale Alard, University of Louisville, United States

Citation: Schmidt, N. W., Jala, V. R., Kosiewicz, M. M., Alard, P., eds. (2019). The Gut Microbiome in Health and Disease. Lausanne: Frontiers Media. doi: 10.3389/978-2-88963-003-5

# Table of Contents


Mireia Lopez-Siles, Núria Enrich-Capó, Xavier Aldeguer, Miriam Sabat-Mir, Sylvia H. Duncan, L. Jesús Garcia-Gil and Margarita Martinez-Medina


Blessing O. Anonye


Qiurong Li, Chenyang Wang, Chun Tang, Xiaofan Zhao, Qin He and Jieshou Li


Hong-Li Li, Lan Lu, Xiao-Shuang Wang, Li-Yue Qin, Ping Wang, Shui-Ping Qiu, Hui Wu, Fei Huang, Bei-Bei Zhang, Hai-Lian Shi and Xiao-Jun Wu

*115 The Influence of Proton Pump Inhibitors on the Fecal Microbiome of Infants With Gastroesophageal Reflux—A Prospective Longitudinal Interventional Study*

Christoph Castellani, Georg Singer, Karl Kashofer, Andrea Huber-Zeyringer, Christina Flucher, Margarita Kaiser and Holger Till

*122* Helicobacter pylori *CagA Protein Negatively Regulates Autophagy and Promotes Inflammatory Response via c-Met-PI3K/Akt-mTOR Signaling Pathway*

Na Li, Bin Tang, Yin-ping Jia, Pan Zhu, Yuan Zhuang, Yao Fang, Qian Li, Kun Wang, Wei-jun Zhang, Gang Guo, Tong-jian Wang, You-jun Feng, Bin Qiao, Xu-hu Mao and Quan-ming Zou


Jean-Félix Sicard, Guillaume Le Bihan, Philippe Vogeleer, Mario Jacques and Josée Harel

*160 Alterations of the Gut Microbiome in Hypertension*

Qiulong Yan, Yifang Gu, Xiangchun Li, Wei Yang, Liqiu Jia, Changming Chen, Xiuyan Han, Yukun Huang, Lizhe Zhao, Peng Li, Zhiwei Fang, Junpeng Zhou, Xiuru Guan, Yanchun Ding, Shaopeng Wang, Muhammad Khan, Yi Xin, Shenghui Li and Yufang Ma

*169 Secretory Products of the Human GI Tract Microbiome and Their Potential Impact on Alzheimer's Disease (AD): Detection of Lipopolysaccharide (LPS) in AD Hippocampus*

Yuhai Zhao, Vivian Jaber and Walter J. Lukiw

*178 Molecular Characterization of the Human Stomach Microbiota in Gastric Cancer Patients*

Guoqin Yu, Javier Torres, Nan Hu, Rafael Medrano-Guzman, Roberto Herrera-Goepfert, Michael S. Humphrys, Lemin Wang, Chaoyu Wang, Ti Ding, Jacques Ravel, Philip R. Taylor, Christian C. Abnet and Alisa M. Goldstein

*189 Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis*

Gregory R. Young, Darren L. Smith, Nicholas D. Embleton, Janet E. Berrington, Edward C. Schwalbe, Stephen P. Cummings, Christopher J. van der Gast and Clare Lanyon

*199 The NAG Sensor NagC Regulates LEE Gene Expression and Contributes to Gut Colonization by* Escherichia coli *O157:H7*

Guillaume Le Bihan, Jean-Félix Sicard, Philippe Garneau, Annick Bernalier-Donadille, Alain P. Gobert, Annie Garrivier, Christine Martin, Anthony G. Hay, Francis Beaudry, Josée Harel and Grégory Jubelin

*209 Sampling Strategies for Three-Dimensional Spatial Community Structures in IBD Microbiota Research*

Shaocun Zhang, Xiaocang Cao and He Huang

#### *226 Enterotype May Drive the Dietary-Associated Cardiometabolic Risk Factors*

Ana C. F. de Moraes, Gabriel R. Fernandes, Isis T. da Silva, Bianca Almeida-Pititto, Everton P. Gomes, Alexandre da Costa Pereira and Sandra R. G. Ferreira

*235 Human Enterovirus 68 Interferes With the Host Cell Cycle to Facilitate Viral Production*

Zeng-yan Wang, Ting Zhong, Yue Wang, Feng-mei Song, Xiao-feng Yu, Li-ping Xing, Wen-yan Zhang, Jing-hua Yu, Shu-cheng Hua and Xiao-fang Yu


# Response: Commentary: Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis

Gregory R. Young<sup>1</sup> \*, Darren L. Smith<sup>1</sup> , Nicholas D. Embleton<sup>2</sup> , Janet Elizabeth Berrington<sup>2</sup> , Edward C. Schwalbe<sup>1</sup> , Stephen Paul Cummings <sup>3</sup> , Christopher J. van der Gast <sup>4</sup> and Clare Lanyon<sup>1</sup> \*

Keywords: PMA, viability analyses, microbiota, methods—analytic, bias, clinical samples

*<sup>1</sup> Faculty of Health and Life Sciences, University of Northumbria, Newcastle upon Tyne, United Kingdom, <sup>2</sup> Newcastle Neonatal Service, Newcastle upon Tyne Hospitals NHS Foundation Trust, Newcastle upon Tyne, United Kingdom, <sup>3</sup> School of Science and Engineering, Teesside University, Middlesbrough, United Kingdom, <sup>4</sup> School of Healthcare Science, Manchester Metropolitan University, Manchester, United Kingdom*

#### Edited by:

**A Commentary on**

*Pascale Alard, University of Louisville, United States*

#### Reviewed by:

*Bjoern O. Schroeder, University of Gothenburg, Sweden Natalia Shulzhenko, Oregon State University, United States*

#### \*Correspondence:

*Gregory R. Young gregory.young@northumbria.ac.uk Clare Lanyon clare.lanyon@northumbria.ac.uk*

#### Specialty section:

*This article was submitted to Microbiome in Health and Disease, a section of the journal Frontiers in Cellular and Infection Microbiology*

> Received: *28 June 2018* Accepted: *05 October 2018* Published: *24 October 2018*

#### Citation:

*Young GR, Smith DL, Embleton ND, Berrington JE, Schwalbe EC, Cummings SP, van der Gast CJ and Lanyon C (2018) Response: Commentary: Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis. Front. Cell. Infect. Microbiol. 8:374. doi: 10.3389/fcimb.2018.00374*

#### **Commentary: Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis**

by Agustí, G., and Codony, F. (2018). Front. Cell. Infect. Microbiol. 8:212. doi: 10.3389/fcimb.2018.00212

We would like to thank Agustí and Codony (2018), for their interest in our recent manuscript (Young et al., 2017), and valuable comments. We agree that non-viable cell exclusion is an important consideration, worth making when conducting analyses of microbial communities via targeted DNA sequencing and amplification approaches. The technique is especially pertinent in environments where large volumes of non-viable bacteria are expected, such as in preterm infant stool, where multiple clinical interventions manifest deliberate bacterial killing.

Whilst we agree with the general principles with regards to sample collection and handling described by Agustí and Codony (2018) we maintain that these are not always possible in real-life, clinical patient samples. For example, as patient care is the primary concern in this cohort, sample collection is convenience based rather than experimentally dictated. This requires samples to be spontaneously collected and stored prior to processing. We concede this increases the likelihood of loss of anaerobic bacterial viability. We would, however point out that the study by Brusa et al. (1989), highlighted in the commentary defines viability purely as culturability. Studies (Contreras et al., 2011; Nocker et al., 2011) have reported loss of culturability occurs at lower stress levels than loss of membrane integrity. Thus, suggesting greater cellular stress may be required to deplete DNA signals from non-viable cells, as determined during PMA-based viability assays. Moreover, we highlight in our manuscript that PMA is a conservative parameter for loss of viability. We also propose that, in the specific cohort investigated, conservative non-viable determination is preferable to over-exaggerated determination or a complete absence of it. This is especially true when non-treated samples can be analyzed in parallel.

DNA retention on microtube surfaces and subsequent inclusion in the viable communities may well be an issue in PMA-based viability determination. In our study however, samples were not heat-killed at any point. This is in contrast to the study by Agustí et al. (2017), in which DNA binding to microtube walls may have occurred prior to PMA-treatment. Furthermore, the methods described by ourselves outline transfer between sterile glass pots during initial sample collection and several microtubes and well plates from PMA-addition to incubation and photo-activation and, finally DNA isolation. We propose that these several steps are sufficient to reduce the impact of DNA retained on microtube surfaces on the assigned live DNA fraction.

In addition to the valuable comments made by Agustí and Codony (2018), where the target sequence is <300 bp we recommend combination of a nested-PCR approach, comprising initial long fragment pre-amplification, followed by subsequent target amplicon sequencing/quantification, first described by Luo et al. (2010). This increases the likelihood of encountering intercalated PMA during the DNA amplification process.

The use of PMA as a viability-dye is extremely useful in microbial community analysis. As a greater volume of literature becomes available regarding it's use, the technique will no doubt refine. From our experience with the technique, we propose several important considerations to be made when planning any study in which PMA-based non-viable cell exclusion is employed:


We would further direct anyone interested to a comprehensive review of the technique by Fittipaldi et al. (2012).

### AUTHOR CONTRIBUTIONS

GY wrote the response. All authors proofread and approved this commentary response.

### 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 © 2018 Young, Smith, Embleton, Berrington, Schwalbe, Cummings, van der Gast and Lanyon. 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.

# Alterations in the Abundance and Co-occurrence of *Akkermansia muciniphila* and *Faecalibacterium prausnitzii* in the Colonic Mucosa of Inflammatory Bowel Disease Subjects

Mireia Lopez-Siles <sup>1</sup> , Núria Enrich-Capó<sup>1</sup> , Xavier Aldeguer <sup>2</sup> , Miriam Sabat-Mir <sup>3</sup> , Sylvia H. Duncan<sup>4</sup> , L. Jesús Garcia-Gil <sup>1</sup> \* and Margarita Martinez-Medina<sup>1</sup>

#### *Edited by:*

Venkatakrishna Rao Jala, University of Louisville, United States

#### *Reviewed by:*

Susan M. Bueno, Pontificia Universidad Católica de Chile, Chile Valerio Iebba, Sapienza Università di Roma, Italy

> *\*Correspondence:* L. Jesús Garcia-Gil jesus.garcia@udg.edu

> > *Specialty section:*

This article was submitted to Microbiome in Health and Disease, a section of the journal Frontiers in Cellular and Infection Microbiology

*Received:* 15 December 2017 *Accepted:* 25 July 2018 *Published:* 07 September 2018

#### *Citation:*

Lopez-Siles M, Enrich-Capó N, Aldeguer X, Sabat-Mir M, Duncan SH, Garcia-Gil LJ and Martinez-Medina M (2018) Alterations in the Abundance and Co-occurrence of Akkermansia muciniphila and Faecalibacterium prausnitzii in the Colonic Mucosa of Inflammatory Bowel Disease Subjects. Front. Cell. Infect. Microbiol. 8:281. doi: 10.3389/fcimb.2018.00281 <sup>1</sup> Laboratory of Molecular Microbiology, Biology Department, Universitat de Girona, Girona, Spain, <sup>2</sup> Department of Gastroenterology, Hospital Dr. Josep Trueta, Girona, Spain, <sup>3</sup> Department of Gastroenterology, Hospital Santa Caterina, Girona, Spain, <sup>4</sup> Microbiology Group, Rowett Institute of Nutrition and Health, Aberdeen, United Kingdom

Akkermansia muciniphila and Faecalibacterium prausnitzii, cohabitants in the intestinal mucosa, are considered members of a healthy microbiota and reduction of both species occurs in several intestinal disorders, including inflammatory bowel disease. Little is known however about a possible link between the reduction in quantity of these species, and in which circumstances this may occur. This study aims to determine the abundances and co-occurrence of the two species in order to elucidate conditions that may compromise their presence in the gut. Loads of A. muciniphila, total F. prausnitzii and its two phylogroup (16S rRNA gene copies) were determined by quantitative polymerase chain reaction in colonic biopsies from 17 healthy controls (H), 23 patients with ulcerative colitis (UC), 31 patients with Crohn's disease (CD), 3 with irritable bowel syndrome (IBS) and 3 with colorectal cancer (CRC). Data were normalized to total bacterial 16S rRNA gene copies in the same sample. Prevalence, relative abundances and correlation analyses were performed according to type of disease and considering relevant clinical characteristics of patients such as IBD location, age of disease onset, CD behavior, current medication and activity status. Co-occurrence of both species was found in 29% of H, 65% of UC and 29% of CD. Lower levels of total F. prausnitzii and phylogroups were found in subjects with CD, compared with H subjects (P ≤ 0.044). In contrast, no differences were found with the regard to A. muciniphila abundance across different disease states, but CD patients with disease onset below 16 years of age featured a marked depletion of this species. In CD patients, correlation between A. muciniphila and total F. prausnitzii (ρ = 0.362, P = 0.045) was observed, and particularly in those with non-stricturing, non-penetrating disease behavior and under moderate immunosuppressants therapy. Altogether, this study revealed that co-occurrence of both

**8**

species differs between disease status. In addition, IBD patients featured a reduction of F. prausnitzii but similar loads of A. muciniphila when compared to H subjects, with the exception of those with early onset CD. Depletion of A. muciniphila in this subgroup of subjects suggests that it could be a potential biomarker to assist in pediatric CD diagnosis.

Keywords: *Akkermansia muciniphila*, *Faecalibacterium prausnitzii*, Crohn's disease, ulcerative colitis, inflammatory bowel diseases

#### INTRODUCTION

Crohn's disease (CD) and ulcerative colitis (UC) are the two major types of idiopathic inflammatory bowel diseases (IBD) (Mendoza Hernández et al., 2007). Both are chronic inflammatory disorders of the gut. UC typically begins in the rectum and inflammation may extend continuously to involve the entire colon. In 20% of CD patients the disease affects the colon exclusively (Silverberg et al., 2005), but the most commonly involved areas are terminal ileum and the beginning of the colon. In CD any part of the gastrointestinal tract (from the oropharynx to the anus) may be affected in a patchy pattern (Mendoza Hernández et al., 2007). Other than location, differences in the mucosal lesions exist between these conditions. Inflammation in CD can be transmural reaching the serosa, whereas inflammation in UC patients is generally restricted to the mucosa.

The inner layer of the bowel wall is a niche of particular importance, because of the spatial proximity between epithelial cells and gut bacteria, and thus the study of human intestinal mucosa biopsies provides meaningful insights of host-bacterial interactions. Numerous studies have been prompted over the last decade aiming at deciphering the exact role of gut microbiota in IBD. Nowadays there is a wide variety of clinical and experimental studies revealing microbial implication in IBD (Sartor, 2006, 2008; Seksik et al., 2006; Manichanh et al., 2012). The most replicated finding by far has been disturbances in the intestinal microbiota composition balance, situation known as dysbiosis (Gophna et al., 2006; Manichanh et al., 2006; Martinez-Medina et al., 2006; Andoh et al., 2009; Willing et al., 2009; Sokol and Seksik, 2010; Joossens et al., 2011; Mondot et al., 2011; Machiels et al., 2013). In this state, dominating species, to whom a beneficial role to preserve gut homeostasis has been attributed, become underrepresented.

Akkermansia muciniphila inhabits mainly in the mucosa, and represents between 1 and 3% of the gut microbiota (Derrien et al., 2004, 2008). A decrease of this species has been demonstrated in feces and/or biopsies of several disorders including autism, obesity, type 2 diabetes, appendicitis, and IBD (Belzer and de Vos, 2012; Everard et al., 2013). Studies in mice models have shown that gut colonization by this species affects expression of genes involved in immune response-regulatory processes (Derrien et al., 2011) as well as in host's lipid metabolism (Lukovac et al., 2014), especially in the colon. It is of note that extracellular vesicles derived from this species have a protective function that ameliorates severity of induced colitis in mice, suggesting that it has an important role in the maintenance of intestinal homeostasis (Kang et al., 2013). A. muciniphila is essential for a healthy mucus layer in the human gut in terms of mucus production and thickness (Belzer and de Vos, 2012). This species is not only important for the host, but also for gut microbial community. Its specific capability to degrade mucus results in the release of oligosaccharides and the production of propionate and acetate (Derrien et al., 2004) as well as amino acids, important cofactors and vitamins (van Passel et al., 2011) that become available for other gut symbionts. However, significant co-occurrence of this species with other bacterial taxa present in the gut has not been revealed in feces (Lozupone et al., 2012).

In turn, Faecalibacterium prausnitzii is also an abundant intestinal microorganism with a feco-mucosal distribution, and whose relative abundance can represent between 2 and 15% of intestinal bacterial communities (Swidsinski et al., 2005; Baumgart et al., 2007; Flint et al., 2012). Several studies, of fecal and/or mucosal samples, have shown that F. prausnitzii prevalence and abundance are reduced under certain disorders such as celiac disease (Swidsinski et al., 2008; De Palma et al., 2009), obesity and type 2 diabetes (Furet et al., 2010; Graessler et al., 2012), appendicitis (Swidsinski et al., 2011), chronic diarrhea (Dörffel et al., 2012), irritable bowel syndrome (IBS) of alternating type (Rajilic-Stojanovi ´ c et al., ´ 2011), colorectal cancer (CRC) (Balamurugan et al., 2008; Lopez-Siles et al., 2016), and particularly in IBD (Sokol et al., 2008, 2009; Swidsinski et al., 2008; Willing et al., 2009; Machiels et al., 2013; Lopez-Siles et al., 2014). Low abundance of this species has been linked with active IBD (Sokol et al., 2009), and some complications such as a higher risk of post-operative recurrence (Sokol et al., 2008) or pouchitis (McLaughlin et al., 2010). Other than butyrate production (which can reduce intestinal mucosa inflammation and is the main energy source for the colonocytes), additional anti-inflammatory properties have been attributed to F. prausnitzii (Sokol et al., 2008; Miquel et al., 2013; Martín et al., 2014). Both, cell and supernatant fractions of this species, have been proven to reduce severity of acute (Sokol et al., 2008; Rossi et al., 2015), chronic (Martín et al., 2014) and low grade (Martín et al., 2015) inflammation in murine models. This has been attributed to an enhancement of intestinal barrier function related with the expression of certain tight junction proteins other than claudin (Carlsson et al., 2013). F. prausnitzii also influences gut physiology through the production of mucus O-glycans, and may help to maintain suitable proportions of different cell types of secretory linage in the intestinal epithelium, as evidenced in rodent studies (Wrzosek et al., 2013). To date, it remains unclear which conditions are likely to compromise this species in the gut. Alterations in gut pH or bile salt concentration have been suggested (Lopez-Siles et al., 2012), but a break in the ecologic relations with other gut symbionts that support its presence in the gut may also contribute, but have been little studied. Co-occurrence network analysis of gut bacteria found in feces, showed that F. prausnitzii co-occurs with several members of the C. coccoides group and Bacteroidetes (Lozupone et al., 2012). As F. prausnitzii growth is stimulated by acetate (Duncan et al., 2002), its presence in the gut may also be favored by acetate producers like A. muciniphila. However, little is known about interaction between these two species.

This work is aimed at determining the variation of mucosaassociated A. muciniphila and F. prausnitzii between healthy control subjects (H) and patients suffering from IBD, in order to elucidate in which conditions imbalances of these species take place, and if both species are affected equally. Some IBS and CRC patients have been included for comparative purposes. Prevalence and abundance of mucosa-associated A. muciniphila and F. prausnitzii have been determined in colonic samples by quantitative polymerase chain reaction (qPCR). Data have been analyzed taking into account patients' most relevant clinical characteristics. Medication at sampling was also considered in order to determine if any of the current therapies are effective in restoring these species levels to those found in H. In addition, correlation analysis of their load has also been conducted to provide supporting evidence on the effect of one population over the other, or about whether or not they are influenced by similar gut factors.

#### MATERIALS AND METHODS

#### Patient Recruitment and Characteristics

The study population was a cohort consisting of Spanish volunteers including 54 IBD (31 CD and 23 UC), three IBS, three CRC patients and 17 H (**Table 1**). Subjects were gender matched for all the groups. Concerning age, CD patients were younger than those in the H group (P = 0.002), whereas CRC patients were significantly older than those with IBD (P ≤ 0.028). Besides, at disease onset, CD patients were younger than UC (mean age ± SD; UC = 37.2 ± 13.3 years, CD = 28.0 ± 12.4 years; P = 0.012).

Subjects were recruited by the Gastroenterology Services of the Hospital Universitari Dr. Josep Trueta (Girona, Spain) and the Hospital Santa Caterina (Salt, Spain) between 2006 and 2010. To avoid bias between centers, patients with IBD were diagnosed according to standard clinical, pathological, and endoscopic criteria and categorized according to the Montreal classification (Silverberg et al., 2005). IBS patients were diagnosed according to Rome III criteria (available at http://www.romecriteria.org/criteria/). The diagnosis of CRC was established by colonoscopy and biopsy, and data correlated with high risk of developing this disease was recorded. Controls consisted of subjects who underwent colonoscopy for different reasons as rectorrhagia (n = 8), colorectal cancer familial history (n = 3), and abdominal pain (n = 6), and all featured normal colonoscopy. Clinically relevant data of all the patients was collected (**Table 1**). Percentage of active patients was higher in CD patients than in UC (P = 0.036). Individuals included in this study were >18 years old, did not have any other intestinal disease and were not pregnant. Antibiotic treatment within 2 months before colonoscopy was an exclusion criterion.

### Ethics Statement

This work was approved by the Ethics Committee of Clinical Research of the Hospital Universitari Dr. Josep Trueta (Girona, Spain) and the Institut d'Assistència Sanitària of Girona (Salt, Spain) on 26th May 2006 (protocol code BACTECCU, Ref CEIC: 08/06) and 21st April 2009 (protocol code BACTODIAG, Ref CEIC: 10/08), respectively. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

#### Sample Collection and DNA Extraction

Cleansing of the gastrointestinal tract using Casenglicol <sup>R</sup> was performed prior to colonoscopy, following manufacturer's guidelines. During routine endoscopy and following standard procedures, a biopsy (<25 mg) from non-affected tissue of the colon was taken for each subject. All biopsies were immediately placed in sterile tubes without any buffer. Following completion of the whole endoscopic procedure, all samples were stored at −80◦C upon analysis.

To discard transient and loosely attached bacteria, biopsies were subjected to two mild ultrasound wash cycles, as reported previously (Martinez-Medina et al., 2006). Afterwards, DNA was extracted using the NucleoSpin <sup>R</sup> Tissue Kit (Macherey-Nagel GmbH & Co., Duren, Germany). The support protocol for Gram positive bacteria (consisting of pre-incubation during 1 h at 37◦C with buffer T1 (20 mm Tris/HCl, 2 mM EDTA, 1% Triton X-100, pH 8) supplemented with 20 mg/ml lysozyme), and the RNAse treatment step were carried out. Genomic DNA was eluted with 10 mM Tris-HCl (pH 7.4) and stored at −80◦C until use. DNA concentration and purity of the extracts were determined with a NanoDrop ND-100 spectrophotometer (NanoDrop Technologies, USA). The average purity and concentration of the DNA extracts was (mean ± SD) 1.794 ± 0.579 Abs260/280 ratio and 174.864 ± 131.829 ng/µl, respectively.

#### Quantification Standards for qPCR

Standard DNA templates from F. prausnitzii strain S3L/3 (phylogroup I), F. prausnitzii DSM 17677 (phylogroup II) and A. muciniphila (ATCCBAA-835) were prepared as genetic constructs after PCR amplification of the whole 16S rRNA as previously reported (Lane, 1991; Weisburg et al., 1991), and subsequent insertion of this gene into a pCR <sup>R</sup> 4-TOPO <sup>R</sup> cloning plasmid (Invitrogen, CA, USA) following manufacturer's guidelines.

After purification with the NucleoSpin <sup>R</sup> Plasmid kit (Macherey-Nagel GmbH & Co., Duren, Germany), plasmids were linearized with SpeI and quantified using QubitTM Quantitation Platform (Invitrogen, Carlsbad, USA). Initial target concentration was inferred taking into consideration the theoretical molecular weight (3.58 × 10<sup>6</sup> Da) and the size of recombinant plasmid (5421 pb).

#### TABLE 1 | Sample size and clinical characteristics of subjects.


IBD, Inflammatory bowel disease; TNF, tumor necrosis factor; na, not applicable.

\*Medical treatment at the time of sampling was available in 21/23 UC, and 29/31 CD patients; Age of disease onset was available for 20/23 UC patients, and 29/31 CD patients; Disease behavior at last follow-up before the time of sampling was available in 27/31 CD patients, and none had penetrating CD (B3); Maximal disease extent at the time of sampling was available in 22/23 UC and 30/31 CD patients.

§Groups were compared by appropriate statistical tests, and P-value ≤ 0.05 was considered significant, † χ 2 test, \*\*ANOVA. Analyses statistically significant are highlighted in boldface.

Standard curves were obtained from 10-fold serial dilutions of the titrated suspension of linearized plasmids. Strains used to construct each standard curve are indicated in **Table 2**. To prepare the standard curve, only dilutions within the linear dynamic range span of each reaction were used, as detailed in **Table 2**. Total bacteria 16S rRNA gene quantification was used to intercalibrate all the standard curves, in order to make sure that results obtained were comparable. Accordingly, plasmid preparations for A. muciniphila standard and phylogroups standards were run as unknown samples in a total bacterial qPCR. Specific intercalibration of total F. prausnitzii qPCR was not required as it uses the same standard curve template employed for total bacterial qPCR (**Table 2**). Quantification values obtained were compared to initial target concentration inferred from DNA concentration, and <10% variation was obtained.

#### qPCR Assays

Previously reported 16S rRNA gene-targeted primers and probes were used for total F. prausnitzii (Lopez-Siles et al., 2014), phylogroups (Lopez-Siles et al., 2016), A. muciniphila (Collado et al., 2007) and total bacterial (Furet et al., 2009) quantification through qPCR.

Amplification reactions were performed as described elsewhere (Collado et al., 2007; Furet et al., 2009; Lopez-Siles et al., 2014, 2016) with slight modifications detailed in **Table 2**. Briefly, quantifications were carried out in a total volume of 20 µl reactions containing: 1× TaqMan <sup>R</sup> Universal PCR Master Mix 2× or SYBR <sup>R</sup> Green PCR Master Mix 2× (Applied Biosystems, Foster City, CA, USA) as required, 300–900 nM of each primer and 250–300 nM of probe if necessary. Up to 50 ng of genomic DNA template was added in each reaction. All primers and probes used in this study as well as PCR conditions are detailed in **Table 2**. Total F. prausnitzii, and total bacteria primers and


Frontiers in Cellular and Infection Microbiology | www.frontiersin.org

dMelting curve consisted on 95

◦C 15 s, 60

◦C 1 min, 95

◦C 15 s, and 60

◦C 15 s (average temperature

 slope 0.58

◦C/s).

hydrolysis probes were purchased from Applied Biosystems (Foster City, CA, USA), whereas primers and hydrolysis probes for F. prausnitzii phylogroups and A. muciniphila were acquired from Biomers (Ulm, Germany). DNA of the internal amplification control (IAC) was synthesized by Bonsai technologies group (Alcobendas, Spain). All oligonucleotides were purified by HPLC. Plates and optical caps were provided by Applied Biosystems (Ref. 4323032 and Ref. 4306737, respectively).

Samples were run at least in duplicate in the same plate (**Table S1**), which was set up manually. For data analysis, the mean of the quantifications was used. Duplicates were considered valid if the standard deviation between quantification cycles (Cq) was <0.34 (i.e., a difference of <10% of the quantity was tolerated), and if not quantification was repeated. Quantification controls consisting of at least five reactions with a known number of target genes were performed to assess inter-run reproducibility. Inhibition was controlled on total F. prausnitzii quantification by adding 10<sup>3</sup> copies of an internal amplification control (IAC) template to each reaction. It was considered that there was no inhibition if the obtained C<sup>q</sup> was <0.34 different from those obtained when quantifying the IAC alone for any of the replicates. In each run, a non-amplification control (NTC) which did not contain any DNA template (either bacterial or IAC) was also included. In all cases with hydrolysis probes, NTC resulted in undetectable C<sup>q</sup> values whereas for SYBR Green assays NTC had C<sup>q</sup> >35, and melting curve analysis confirmed no specific amplification.

A 7500 Real Time PCR system (Applied Biosystems, USA) was used to perform all qPCR. The thermal profile used for each assay is detailed in **Table 2**. In summary, it consisted of a first step at 50◦C during 2 min for amperase treatment followed by a 95◦C hold for 10 min to denature DNA and activate Ampli-Taq Gold polymerase; and a further 40–50 cycles consisting of a denaturation step at 95◦C for 15 s, followed by an annealing and extension step at 60◦C (or at 64◦C for phylogroups quantification) for 1 min. When required, melting curve analysis was performed to assess whether or not fluorescence was due to specific amplification products. Data were collected and analyzed using the 7500 SDS system software version 1.4 (Applied Biosystems). Assays were performed under average PCR efficiencies of (mean ± SD) 89.9 ± 4.6% (**Figure S2**).

#### Data Normalization and Statistical Analysis

Regarding the qualitative analyses, absence of F. prausnitzii, its phylogroups or A. muciniphila was considered if no detection was obtained during the qPCR analysis, corresponding to samples that carried these bacteria below the detection limit (i.e., 106.6, 1.10, 2.39, and 374.09 16S rRNA genes per reaction for total F. prausnitzii, phylogroup I, phylogroup II, and A. muciniphila, respectively). Pearson's χ 2 test was used to compare the prevalence between groups of patients, by IBD disease location, age of disease onset and other clinically relevant data as activity, treatment and whether or not patients have had intestinal resection.

Referring to quantitative analyses, total F. prausnitzii, phylogroups and A. muciniphila 16S rRNA gene copy detected in each sample were normalized to the total bacterial 16S rRNA gene copies in the same sample. Data are given as the log<sup>10</sup> of the ratio between 16S rRNA gene copies of the target microorganism and million of total bacterial 16S rRNA genes detected in the same sample. No further correction to adjust for differences in 16S rRNA gene operons in each species was performed, as no consensus has been achieved to date for F. prausnitzii according to rrnDB (Stoddard et al., 2015). For those samples with no detection of F. prausnitzii, phylogroup or A. muciniphila, the number of copies corresponding to their respective detection limit was used for calculations of relative abundances. Kruskal–Wallis non-parametric test was applied to assess differences in variables with more than two categories such as diagnostics, CD and UC disease location, age of disease onset, and current medication. Mann–Whitney U test was used to perform pairwise comparisons of subcategories of these variables, and FDR after multiple comparisons was assessed (**Table S2**). Mann–Whitney U test was also used to compare, variables with two categories such as activity (active CD and UC patients when CDAI > 150 (Best et al., 1976) and a Mayo score >3 (Pineton de Chambrun et al., 2010), respectively), and intestinal resection. In addition, ratios between 16S rRNA gene copies of either total F. prausnitzii, phylogroup I or phylogroup II and A. muciniphila were calculated and analyzed as detailed above. Spearman correlation coefficient and significance between total F. prausnitzii, or phylogroup quantities and A. muciniphila load was calculated. The same statistical method was used to analyze the correlation between the quantity of each bacterial group, and continuous clinical data such as time (in years) since disease onset. All the statistical analyses were performed using the SPSS 15.0 statistical package (LEAD Technologies, Inc.). Significance levels were established for P ≤ 0.05.

## RESULTS

#### Prevalence of Mucosa-Associated *A. muciniphila* and *F. prausnitzii*

To assess co-occurrence of both species, prevalence of F. prausnitzii (total or separating by phylogroups), and A. muciniphila as calculated from positive determinations over total samples, was analyzed by condition, by IBD location and also taking into account relevant clinical data (**Figure 1**). Four categories of patients were established based on: detecting only F. prausnitzii, detecting only A. muciniphila, detecting both species, or none of them. Most of the subjects carried both species or F. prausnitzii alone, whilst finding A. muciniphila alone was rare and in some cases, none of the two species was found suggesting that if present, they are below the detection limit of our assays.

When analysing the cohort by diagnostics (**Figure 1A**), only statistically significant differences in the proportions of each category were observed for the co-occurrence of F. prausnitzii phylogroup I and A. muciniphila (P < 0.001). As a particularity of CD, none of the two species was detected in 51.61% of subjects of this group, which was a higher proportion than in the other groups of subjects (ranging from 4.35% to 33.3%). In 64.71% of H subjects only one species (F. prausnitzii phylopgroup I) was

FIGURE 1 | Prevalence of total F. prausnitzii, phylogroups and A. muciniphila in (A) each group of patients, (B) in Inflammatory bowel disease (IBD) subjects according to whether or not ileum is affected, and (C) in IBD by age of disease onset. H, control subjects; IBS, irritable bowel syndrome; UC, ulcerative colitis; CD, Crohn's disease; CRC, colorectal cancer.

detected, while this category was present in between 16% and 34% in the other groups of subjects. F. prausnitzii phylogroup I and A. muciniphila were found in approximately 30% of H, 33.33% of IBS and 20% of CD subjects, whereas 47.82% of UC and 66.67% of CRC patients fall into this category. Similar trends were observed for F. prausnitzii phylogroup II and A. muciniphila co-occurence analysis (P = 0.105), whereas percentages of each category were more similar between groups of subjects when analyzed in conjunction prevalence of total F. prausnitzii and A. muciniphila (P = 0.387). Of note, all subjects with IBS or CRC had at least one of the two species.

When analysing the cohort by IBD disease location, no significant differences within UC and within CD subtypes were achieved (**Figure S1**). However, a trend toward different proportions was observed. Half of subjects with distal UC (E2) were characterized by presenting only A. muciniphila. All patients with UC and also those with C-CD carried at least one species, whereas lower prevalences were found in CD patients with ileal involvement. Particularly, none of the two species were detected in 10–31% of CD patients with ileal disease location, whereas in 23.1% of I-CD and 40% of IC-CD, both species co-occurred. As the frequencies observed in C-CD patients resembled in some cases those in UC, we analyzed IBD subjects by grouping those with (either I-CD or IC-CD) or without (i.e., C-CD or UC) ileum involved (**Figure 1B**). Interestingly, when analyzing cooccurrence of F. prausnitzii phylogroup I and A. muciniphila the percentage of subjects with none or only one species detected was over 80% in those with ileal disease whereas in approximately 40% of subjects with colonic disease both species were found (P = 0.028).

When analyzing prevalence taking into account clinical data of the patients, no significant differences were observed by activity, medication or intestinal resection neither in CD nor in UC. No differences were found within CD patients according to disease behavior. Interestingly, A. muciniphila was not detected in any of the CD patients diagnosed with disease onset below 16 years of age (**Figure 1C**). It remains to be stablished if this is a common issue with UC patients, as none with early disease onset was included in our cohort.

## Abundances of Mucosa-Associated *A. muciniphila* and *F. prausnitzii*

A. muciniphila, total F. prausnitzii and its phylogroups load was compared amongst patients with different intestinal conditions (**Table 3**). Total F. prausnitzii was less abundant in CRC and IBD patients as compared to H subjects. However, statistically significant differences were achieved only for CRC (P = 0.028) and CD patients [P = 0.021, but not sustained after FDR assessment (**Table S2**)], probably because the reduction is lower for UC patients and the high variability between subjects. In CD patients, those with ileal involvement presented the lowest levels of this bacterium (P = 0.050), whereas IC-CD patients and C-CD were similar to UC (**Table 3**). Slight differences in average load were also found within UC patients although these differences were not statistically supported. Patients with ulcerative proctitis (E1) and extensive UC (E3) presented F. prausnitzii loads similar to H subjects, whereas those with E2 had abundances between CD patients and H subjects.

F. prausnitzii phylogroup I load was reduced in all groups of patients in comparison to H subjects. This reduction was particularly noticeable in CD and CRC patients (P < 0.008), while in UC and IBS patients, it was observed as well, but less apparent. When analyzing data by disease location, I-CD patients showed the most marked reduction of phylogroup I counts in comparison to other CD locations (P = 0.025). Values did not differ significantly in UC patients when analyzed by location. However, loads in E2 and E3 subjects resembled that of CD patients, while for E1 subjects, their profiles were closer to that observed in H subjects.

With respect to F. prausnitzii phylogroup II, its abundance was significantly reduced in CD patients when compared to UC subjects (P = 0.015) while similar loads were observed between all the other groups of subjects (**Table 3**). Although no differences by IBD location were found, loads tend to be lower in those with ileal involvement (either I-CD or IC-CD, P = 0.069).

Interestingly, A. muciniphila load was similar between all the groups of subjects and also no differences were observed between IBD locations (P > 0.540; **Table 3**). In all groups of subjects, total F. prausnitzii counts outnumbered A. muciniphila, but there was a high variability between subjects, even within each condition. No difference in the ratio of total F. prausnitzii: A. muciniphila was found by group of subjects, neither when analyzing by IBD subtypes according to disease location. In contrast, when calculating these ratios by F. prausnitzii phylogroups, significant differences were found between conditions (**Figure 2**). CD patients, featured lower phylogroup I:A. muciniphila ratios than H (P = 0.031), and also lower phylogroup II: A. muciniphila ratios compared to UC (P = 0.017). When analyzing IBD groups by disease location, no significant differences were observed, probably due to the high dispersion of data and to the fact that when separating by location the number of patients included within each category is reduced. Nonetheless, subjects with a larger disease extension, or with ileum involvement, tended to feature lower values of both ratios. These differences in ratios are due to differences in F. prausnitzii (or phylogroup) load, as A. muciniphila load was similar across all subjects.

#### *A. muciniphila* and *F. prausnitzii* Abundances in Relation to Patients Clinical and Treatment Data

No differences in A. muciniphila, counts were observed in either UC or CD patients according to activity status. Nevertheless, those patients with active UC featured the lowest load. F. prausnitzii and the abundance of the phylogroups did not differ between active and inactive UC patients (**Table 4**). CD patients with active disease feature lower levels of total F. prausnitzii and phylogroups in comparison to patients in remission, but differences did not achieve statistical significance either.

Resection in CD patients was not a determining factor for A. muciniphila loads, either (**Table 5**). Instead, F. prausnitzii abundance was lower in those CD patients that underwent intestinal resection, with significant statistical differences for TABLE 3 | Abundances of mucosa-associated F. prausnitzii, its phylogroups and A. muciniphila in controls (H), irritable bowel syndrome (IBS), colorectal cancer (CRC), Ulcerative Colitis (UC), and Crohn's disease (CD) patients.


§Mean log1<sup>0</sup> (16S rRNA gene copies/million bacterial 16S rRNA gene copies) ± standard deviations.

\*Statistics was calculated separately for each variable (column). Only for those analyses statistically significant (P-value in bold), pairwise comparisons were conducted, and groups of patients with similar abundances are indicated with the same superscript (a, b). Disease locations of UC and CD patients have been analyzed as independent groups. Similarly, patients' subtypes with similar abundances are indicated with the same superscript (•). In both cases, groups not sharing superscript are those with statistically different median abundance values (P-value < 0.05).

FIGURE 2 | Box and whiskers plot of the ratio total F. prausnitzii: A. muciniphila, F. prausnitzii phylogroup I: A. muciniphila, and F. prausnitzii phylogroup II: A. muciniphila: (from left to right) by group of subjects (up) and by Inflammatory Bowel Disease subtype (down). Data are represented as log10 of each ratio. The median is represented by the horizontal line in each box. Boxes cover the 25 and 75% quantiles, and bars show the 10 and 90% quantiles. Individual data are also shown. Homogeneous subgroups in each plot are indicated with the same superscript. H, control subjects; IBS, irritable bowel syndrome; UC, ulcerative colitis; CD, Croh's disease; CRC, colorectal cancer; E1, ulcerative proctitis; E2, ulcerative left-sided colitis; E3, ulcerative pancolitis; IC-CD, ileocolonic CD, I-CD, ileal CD; C-CD, colonic CD. A. muciniphila, 16S rRNA gene Akkermansia muciniphila; F. prausnitzii, 16S rRNA gene total Faecalibacterium prausnitzii; Phylogroup I, 16S rRNA gene F. prausnitzii phylogroup I; Phylogroup II, 16S rRNA gene F. prausnitzii phylogroup II.


Active CD and UC were defined when CDAI >150 (Best et al., 1976) and a Mayo score >3 (Pineton de Chambrun et al., 2010), respectively. \*Median (log<sup>10</sup> 16S rRNA gene copies/million bacterial 16S rRNA gene copies) ± standard deviations.

§UC, ulcerative colitis; CD, Crohn's disease.

TABLE 5 | F. prausnitzii, its phylogroups and A. muciniphila abundance in inflammatory bowel disease patients depending on whether or not they have had intestinal resection during the course of the disease.


\*Median (log<sup>10</sup> 16S rRNA gene copies/million bacterial 16S rRNA gene copies) ± standard deviations; na, not applicable. Analyses statistically significant are highlighted in boldface. §UC, ulcerative colitis; CD, Crohn's disease.

phylogroup II. Precisely, resected subjects had 10 times less phylogroup II than those without intestinal surgery (P = 0.018) whereas the phylogroup I load was only slightly lower in resected than non-resected patients.

The A. muciniphila load was lower in CD patients who presented with the disease below 16 years of age (**Table 6**). This group of patients also featured very low quantities of F. prausnitzii phylogroup I although statistical significance was not achieved. No differences in these bacterial loads were observed between groups of UC patients with different age of disease onset. We also analyzed disease duration, but no statistically significant correlation was found between any of the bacterial loads and time of disease duration (data not shown).

Finally, data were analyzed by taking into account the medication of the patients at the time of sampling (**Table 7**). No differences in A. muciniphila, F. prausnitzii or in phylogroups abundances were observed between medications for any IBD. However, those UC patients that received anti-tumor necrosis factor had the lowest levels of A. muciniphila. In contrast, those CD patients receiving moderate immunosupressants had lower F. prausnitzii loads than patients without treatment or receiving therapies such as mesalazine or anti-tumor necrosis factor.

#### Correlation Between *A. muciniphila* and *F. prausnitzii* Abundances

Correlation between A. muciniphila and F. prausnitzii numbers was analyzed to provide supporting evidence for a direct/indirect effect of one population over the other or about a putative common factor affecting both species populations in a given condition (**Figure 3**).

No correlation between these two species was found in H or UC patients (**Figure S3**). Therefore, in these conditions, factors of the gut environment may be differentially impacting on each species. In contrast, positive correlation of both species load was observed in CD subjects (**Figure 3**). Analysis by phylogroups indicated that A. muciniphila quantity tended to positively correlate with phylogroup I in CD patients (P = 0.060), whereas no significant correlation was observed with phylogroup II (**Figure S3**).

Moreover, no significant correlation between the two species was observed, when splitting subjects by activity status or whether or not they have had intestinal resection. Of note, a positive correlation between A. muciniphila and F. prausnitzii (particularly phylogroup I) was observed in CD patients with non-stricturing, non-penetrating disease (B1) and in those under moderate immunosuppressants (ρ ≥ 0.539, P ≤ 0.017). In contrast, in UC patients with disease onset between 17 and 40 years of age, a negative correlation between A. muciniphila and F. prausnitzii phylogroup II was observed (ρ = −0.673, P = 0.023).

#### DISCUSSION

A. muciniphila and F. prausnitzii are two symbiotic and numerically abundant members of the gut microbiota, and both have been associated with dysbiosis in several disease conditions, including IBD. Their niche is close to the intestinal mucosa,



\*Median (log1<sup>0</sup> 16S rRNA gene copies/million bacterial 16S rRNA gene copies) ± standard deviations.

§UC, ulcerative colitis; CD, Crohn's disease.

#Statistics was calculated separately for each variable (column). Groups of patients with similar abundances of A. muciniphila are indicated with the same superscript (a, b) whereas groups not sharing superscript are those with statistically different median abundance values (P < 0.05). Analyses statistically significant are highlighted in boldface.

TABLE 7 | F. prausnitzii, its phylogroups and A. muciniphila abundances in inflammatory bowel disease by medication at sampling.


\*Median (log1<sup>0</sup> 16S rRNA gene copies/million bacterial 16S rRNA gene copies) ± standard deviations.

§UC, ulcerative colitis; CD, Crohn's disease; Mod. Immsup, moderate immunosuppresants; Anti-TNF, Anti-tumor necrosis factor.

and therefore it can be hypothesized that they may play a key role in cross-talk with the host. Both species are considered to have a part in a well-functioning gut and thus are considered as promising next generation probiotics (Neef and Sanz, 2013; Martín et al., 2017). In the present study we have analyzed the prevalence and abundance of mucosa associated A. muciniphila, total F. prausnitzii and phylogroup in H and IBD subjects, taking into account the diversity of disease locations and the clinical features of patients. Some IBS and CRC have been included as well, but only for illustrative purposes given the low number of patients engaged. The abundance of both species has been previously reported to be reduced in several intestinal disorders (Belzer and de Vos, 2012; Miquel et al., 2013), but here for the first time we correlate the load of both species. Through analysis of clinical data, we consider which particular conditions this underrepresentation is favored, and whether or not the imbalance of one species is linked to changes in the abundance of the other.

Our data show that the A. muciniphila load in the mucosa of H subjects is slightly higher (2.0- to 4.5-fold, respectively) than in IBD and CRC patients, but is not statistically significant. An increase in A. muciniphila abundance in CRC patients compared to controls has been previously found in stools (Weir et al., 2013) but not in mucosal biopsies (Mira-Pascual et al., 2015), and the analysis of our limited cohort is in line with this finding. Previous studies have reported a significant decrease of this species in IBD subjects (Png et al., 2010). Methodological differences may explain the inconsistency with our findings, as we exclusively focused on colonic samples. In biopsy samples, Png and collaborators observed a reduction of this species in IBD patients that ranged between 2.9- and 3.9-fold when compared to controls, which is similar to the reduction observed in our subjects. In that study (Png et al., 2010), differences were observed depending on whether or not the tissue was affected, with the depletion being more conspicuous in inflamed tissue, and without reaching significant differences between noninflamed tissue of CD and controls, which is in line with our results as we used non-affected tissue. In addition, we have explored differences taking into account disease location, activity or intestinal resection, but no association between A. muciniphila load and these variables has been revealed. Intriguingly, CD patients who presented with disease below 16 years of age had a striking reduction of this species compared to those with disease onset at a later age. A. muciniphila has been reported to colonize the gut in early infancy, and loads in infants 1 year old are similar to that found in adults (Collado et al., 2007). Therefore, it seems likely that this depletion is not a general phenomenon that occurs in IBD or age-driven, but due to particular features of pediatric IBD that are sustained throughout the disease. In line with this, discrepancies between dysbiosis signatures in adult and

infant IBD patients have been previously reported (Hansen et al., 2012) and it remains to be explored through prospective studies if early disease onset results in long term microbial signatures. Another future application of this finding could be to explore the usefulness of A. muciniphila depletion as a biomarker to assist in pediatric IBD diagnosis.

Regarding mucosa-associated F. prausnitzii loads we have observed a marked reduction in CRC and CD patients, especially in those with ileal involvement, affecting both phylogroups of this species. Although less prominent, UC patients also featured lower F. prausnitzii abundance than H subjects. Our study is in agreement with previous reports which found F. prausnitzii to be reduced in CRC and IBD adults (Swidsinski et al., 2005, 2008; Martinez-Medina et al., 2006; Frank et al., 2007; Sokol et al., 2008, 2009; Willing et al., 2009; McLaughlin et al., 2010; Vermeiren et al., 2012; Kabeerdoss et al., 2013; Machiels et al., 2013; Miquel et al., 2013; Lopez-Siles et al., 2014, 2016). Besides, lower abundance of both F. prausnitzii phylogroups has been previously reported concerning CD patients (Jia et al., 2010; Lopez-Siles et al., 2016), which is in line with our findings. Moreover, because we have observed differences between IBD subtypes, our results support the hypothesis that patients with ileal disease location constitute a differentiated pathological entity (Willing et al., 2009). We have corroborated that the reduction of F. prausnitzii numbers compared to H subjects takes place in both active and inactive IBD patients (Willing et al., 2009), with active CD patients featuring the lowest levels of phylogroup I. Also in agreement with previous studies (Sokol et al., 2008) lower numbers of F. prausnitzii were detected in resected CD patients, but in our study, statistically significant differences were only achieved for phylogroup II, probably because the depletion was more striking. It remains unknown why there are shifts in particular subgroups of this species. To date, several articles convey the point that the genus Faecalibacterium hosts a complex diversity (Lopez-Siles et al., 2015; Benevides et al., 2017; Martín et al., 2017). This diversity has been shown mainly through phylogenetic methods, but phenotypical diversity also exists. Supporting this point, studies characterizing several strains of this species isolated from different origins have failed to find phenotypic traits that consistently distinguish members from one or other subtype (Lopez-Siles et al., 2012; Foditsch et al., 2014; Martín et al., 2017). However, the effect of host factors differentially influencing F. prausnitzii subpopulations has been poorly explored which may explain our results. Another hypothesis could be that subtypes of F. prausnitzii interact in a different manner with other members of the microbiome, which has also been scarcely studied to date.

We have explored co-occurrence and correlation between A. muciniphila and F. prausnitzii in H and IBD patients. We considered that both species may have a syntrophic relationship, thus we hypothesize that the depletion or enrichment of one would imply the same effect on the other. In particular, A. muciniphila mucolytic activity could release oligosaccharides, co-factors, vitamins, and short chain fatty acids, including acetate that juxtaposed species could use for growth. Indeed, F. prausnitzii has been proven to be able to use some oligosaccharides derived from mucus and its growth is stimulated by acetate and requires presence of vitamins in the medium (Duncan et al., 2002; Lopez-Siles et al., 2012). These compounds, can be provided by A. muciniphila, although not exclusively, and therefore establish cross-feeding interactions. A recent study based on co-culture experiments demonstrated this trophic interaction (Belzer et al., 2017). However, in most of the cases studied here, we did not find a correlation between F. prausnitzii and A. muciniphila abundances, and the two species co-occurred only in 41.5% of subjects engaged in the study. This may be because F. prausnitzii does not depend exclusively on byproducts synthesized by A. muciniphila. In agreement with that, other studies have reported that F. prausnitzii can benefit from the presence of a variety of acetate-producing species (Wrzosek et al., 2013; Rios-Covian et al., 2015). It would be interesting to determine whether this species (or other mucus-inhabiting species) increase in patients in which A. muciniphila diminishes, and thus may be partially replacing its role concerning acetate production.

Nonetheless, we observed that positive correlation between the two species happens in CD patients, and particularly for those with B1 behavior or under immunosuppressant therapy. The fact that there is a positive correlation of the two species indicates that their abundance varies in a similar way in this particular condition. The most likely scenario is that in CD the two bacteria are similarly affected by host and gut environmental factors. To support this hypothesis, both species share the characteristic that their growth is severely compromised at pH < 5.5 (Derrien et al., 2004; Lopez-Siles et al., 2012) and in turn, acidic stools have been reported for IBD patients (Nugent et al., 2001; Barkas et al., 2013). Notably, in those cases of correlation between both species, only members of phylogroup I were involved. It would be interesting to perform co-culture studies with different F. prausnitzii strains, and monitor their growth under different conditions in order to determine more accurately their relationship.

In our cohort, A. muciniphila was not detected in 57.1% of all subjects but this seems to be related to the fact that this species has a lower relative abundance in the gut, rather than to a higher sensitivity to gut disease. In contrast, A. muciniphila was more frequently found in UC patients. This could be partially explained by the fact that a higher proportion of loose mucus has been found in UC patients (Antoni et al., 2014), which is the likely niche for A. muciniphila. Another hypothesis that can not be ruled out is that some factor of UC patients favors the presence of A. muciniphila. In our limited cohort, we have also observed higher prevalence of this species in CRC group compared to controls, which is in line with previous findings (Mira-Pascual et al., 2015). However, our data points out that this higher presence does not imply an increase in the abundance at the mucosal level. In contrast, almost 90% of subjects were F. prausnitzii carriers and thus the fact that it is a second-liner in the mucosa and the fact that this species can rely on other members of the gut microbiota for cross-feeding may explain its higher ubiquity and abundance compared to A. muciniphila.

Finally, our study revealed that CD patients are characterized by a low F. prausnitzii: A. muciniphila, ratio affecting both phylogroups. This indicates that compared to H and UC, these patients have an altered proportion of beneficial microorganisms in the mucosa. Although our study does not allow us to decipher if this imbalance is a cause or a consequence of the disease, it can be an aggravator because the two species have been linked to be key for mucus integrity (Derrien et al., 2004; Wrzosek et al., 2013) and gut homeostasis. A significant depletion of both species has also been reported in children with atopic disease (Candela et al., 2012), and therapeutic strategies to restore these species needs to be explored, particularly for disorders that have in common to feature chronic inflammation. In addition, two recent studies have linked the two species with response to immunotherapy treatment (Gopalakrishnan et al., 2017; Routy et al., 2017), thus pointing out another situation in which it is relevant to have these bacteria. Further studies to assess implications in IBD treatment response would be interesting, as immunomodulators are among the usual therapies prescribed to IBD patients. Finally, further confirmation of our results in a larger cohort would be required given that we have engaged a limited number of subjects, and thus it would provide robustness to those findings not sustained after FRD assessment.

## CONCLUSIONS

IBD patients are characterized by a reduction of F. prausnitzii and a slight underrepresentation of A. muciniphila in the colonic mucosa, regardless of disease activity status. While differences in F. prausnitzii load have been observed for I-CD patients, early onset CD is characterized by a lack of A. muciniphila, but further prospective studies are required to assess if this feature is sustained long term. Positive correlation between the two species was found in CD patients, and further studies are required to elucidate which common factors alter both populations in particular gut disorders.

## AUTHOR CONTRIBUTIONS

XA, SD, LG-G, ML-S, and MM-M study concept and design. XA, NE-C, ML-S, and MS-M acquisition of data. ML-S and MM-M interpretation of data and statistical analysis. ML-S drafting the manuscript. XA, SD, NE-C, LG-G, MM-M, and MS-M critical revision of the manuscript for important intellectual content. LG-G and MM-M obtained funding. All authors have approved the final version of the manuscript and agree to be accountable for all aspects of the work, ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

## FUNDING

This work was funded by the Universitat de Girona projects MPCUdG2016-009 and GdRCompetUdG2017, and the Spanish

## REFERENCES


Ministry of Education and Science through projects SAF2006- 00414, SAF2010-15896 and SAF2013-43284-P, being the last cofunded by the European Regional Development. Dr. Sylvia H. Duncan acknowledges support from the Scottish Government Research and Environment Science and Analytical Services Division (RESAS).

## ACKNOWLEDGMENTS

We appreciate the generosity of the patients who freely gave their time and samples to make this study possible, and the theater staff of all centers for their dedication and careful sample collection. Thanks are due to MD. David Busquets and Ms. Romà Surís for assistance with samples collection and analyses, respectively. We are thankful to Ms. Natàlia Adell for statistical help.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcimb. 2018.00281/full#supplementary-material

barrier function in mice DSS colitis. Scand. J. Gastroenterol. 48, 1136–1144. doi: 10.3109/00365521.2013.828773


successional and opportunistic lifestyles of human gut symbionts. Genome Res. 22, 1974–1984. doi: 10.1101/gr.138198.112


in fecal samples from patients with irritable bowel syndrome. Gastroenterology 141, 1792–1801. doi: 10.1053/j.gastro.2011.07.043


in situ hybridization study in mice. World. J. Gastroenterol. 11, 1131–1140. doi: 10.3748/wjg.v11.i8.1131


**Conflict of Interest Statement:** XA is a consultant for AbbVie, Janssen and Takeda, and has received honoraria for lectures, including services on speakers bureaus from AbbVie, MS-D, Janssen, Takeda, Shire, Zambon and Ferring. XA, LG-G, ML-S, and MM-M, have filed a European patent for a "Method for the detection, follow up and/or classification of intestinal diseases" (application number EP15382427).

The remaining 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 Lopez-Siles, Enrich-Capó, Aldeguer, Sabat-Mir, Duncan, Garcia-Gil and Martinez-Medina. 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.

# *Akkermansia muciniphila* as a Model Case for the Development of an Improved Quantitative RPA Microbiome Assay

Heather J. Goux <sup>1</sup> , Dimple Chavan<sup>1</sup> , Mary Crum<sup>2</sup> , Katerina Kourentzi <sup>2</sup> and Richard C. Willson1,2,3 \*

<sup>1</sup> Department of Biology and Biochemistry, University of Houston, Houston, TX, United States, <sup>2</sup> Department of Chemical and Biomolecular Engineering, University of Houston, Houston, TX, United States, <sup>3</sup> Tecnológico de Monterrey-ITESM Campus Monterrey, Monterrey, Mexico

#### *Edited by:*

Pascale Alard, University of Louisville, United States

#### *Reviewed by:*

Elisabeth Margaretha Bik, uBiome, United States James E. Graham, University of Louisville, United States

> *\*Correspondence:* Richard C. Willson willson@uh.edu

#### *Specialty section:*

This article was submitted to Microbiome in Health and Disease, a section of the journal Frontiers in Cellular and Infection Microbiology

> *Received:* 11 December 2017 *Accepted:* 20 June 2018 *Published:* 12 July 2018

#### *Citation:*

Goux HJ, Chavan D, Crum M, Kourentzi K and Willson RC (2018) Akkermansia muciniphila as a Model Case for the Development of an Improved Quantitative RPA Microbiome Assay. Front. Cell. Infect. Microbiol. 8:237. doi: 10.3389/fcimb.2018.00237 Changes in the population levels of specific bacterial species within the gut microbiome have been linked to a variety of illnesses. Most assays that determine the relative abundance of specific taxa are based on amplification and sequencing of stable phylogenetic gene regions. Such lab-based analysis requires pre-analytical sample preservation and storage that have been shown to introduce biases in the characterization of microbial profiles. Recombinase polymerase amplification (RPA) is an isothermal nucleic acid amplification method that employs commercially available, easy-to-use freeze-dried enzyme pellets that can be used to analyze specimens rapidly in the field or clinic, using a portable fluorometer. Immediate analysis of diverse bacterial communities can lead to a more accurate quantification of relative bacterial abundance. In this study, we discovered that universal bacterial 16S ribosomal DNA primers give false-positive signals in RPA analysis because manufacturing host Escherichia coli DNA is present in the RPA reagents. The manufacturer of RPA reagents advises against developing an RPA assay that detects the presence of E. coli due to the presence of contaminating E. coli DNA in the reaction buffer (www.twistdx.co.uk/). We, therefore, explored four strategies to deplete or fragment extraneous DNA in RPA reagents while preserving enzyme activity: metal-chelate affinity chromatography, sonication, DNA cleavage using methylation-dependent restriction endonucleases, and DNA depletion using anti-DNA antibodies. Removing DNA with anti-DNA antibodies enabled the development of a quantitative RPA microbiome assay capable of determining the relative abundance of the physiologically-important bacterium Akkermansia muciniphila in human feces.

Keywords: RPA, gut microbiome, *Akkermansia muciniphila*, bacterial quantification, point-of-need

## INTRODUCTION

The human intestinal microbiome contains ≥10<sup>14</sup> bacteria representing over 400 species (Ott et al., 2004). Recent publications have suggested that the composition of the gut microbiota is significantly associated with health and disease (Shreiner et al., 2015; Lloyd-Price et al., 2016; Lynch and Pedersen, 2016; Duvallet et al., 2017). Relatively small changes in bacterial levels from key taxonomic groups have been linked to a wide range of illnesses, including inflammatory bowel disease, Crohn's disease, colon cancer, and hyperglycemia (Watterlot et al., 2008; Kang et al., 2013; Scher et al., 2013; Scheperjans et al., 2015; Schneeberger et al., 2015; Dao et al., 2016; Rosa et al., 2017; Wong et al., 2017). The use of proor prebiotics or fecal microbiota transplantation have been shown to alter the microbial profile of the gut, in some cases improving health (Muegge, 2011; Petrof et al., 2013; Colman and Rubin, 2014; Cui et al., 2015; Plovier et al., 2016; Routy et al., 2018). Development of a method to monitor the abundance of beneficial microorganisms within the complex milieu of the gut microbiome is therefore of considerable interest.

A variety of techniques are available for analyzing microbiota composition, including small-subunit ribosomal RNA (16S rRNA) gene sequencing, whole-metagenome shotgun sequencing, quantitative polymerase chain reaction assays, and microbial culture (Morgan and Huttenhower, 2012). All prokaryotes harbor a 16S rRNA gene, which includes both conserved sequences and species-specific hypervariable regions. Well-developed databases (e.g., GreenGenes, Ribosomal Database Project, and Silva) are available to classify 16S rRNA sequence data at high taxonomic resolution for use in microbial population profiling. Sequencing short 16S rRNA gene segments often is more cost-effective than sequencing the entire metagenome and thus enables cohort studies large enough to identify statistically significant correlations with disease states (Morgan and Huttenhower, 2012; Hermann-Bank et al., 2013; Robinson et al., 2016). Although 16S rRNA sequencing is currently the favored tool for the detection of bacterial biomarkers, qPCR is faster, cheaper, and easier to interpret, making it the preferred method for biomarker validation (Watterlot et al., 2008; Kostic et al., 2012; Kang et al., 2013; Scher et al., 2013; Zhang et al., 2016). In addition, qPCR often enables more accurate quantification of specific species than 16S rRNA sequencing (Hermann-Bank et al., 2013). However, PCR and 16S detection methods can be subject to biases associated with pre-analytical sample preservation, storage, and DNA extraction (Robinson et al., 2016), leading to inaccuracies in fecal bacterial quantification. Reducing the amount of sample handling and eliminating sample storage can lead to a more accurate estimation of bacterial abundance within the gut.

Recombinase polymerase amplification (RPA) is an isothermal amplification nucleic acid detection method suitable for analysis of samples at the point of need (Euler et al., 2012a; Abd El Wahed et al., 2013, 2015; Rosser et al., 2015; Bonney et al., 2017; Kim and Lee, 2017). Rather than heat denaturation, RPA uses recombinases (E. coli RecA) to form a complex with signal stranded oligonucleotides (30–35 nt primers) and single strand binding proteins (SSBs) assist site-specific D-loop strand invasion (Piepenburg et al., 2006) of dsDNA. At a constant temperature of 37–42◦C, Sau (Staphylococcus aureus) DNA polymerase performs primer-extension to generate a new strand of DNA. Much like PCR, the newly generated product goes on to become the template for future rounds of amplification. Incorporation of a cleavable self-quenched exo-probe or SYBR Green dye allows for real-time fluorescence monitoring of RPA assays, thus enabling quantitative analysis typically within 10–15 min (Crannell et al., 2014, 2015; Kim and Lee, 2016, 2017; Moore and Jaykus, 2017). As RPA utilizes freeze-dried, reaction-ready enzyme pellets (manufactured by TwistDx, Inc.) and a portable fluorometer, the technique can be readily adapted to field applications (Abd El Wahed et al., 2015). Immediate field-based analysis of diverse bacterial communities can lead to a more accurate quantification of relative bacterial abundance.

Although RPA can provide valuable data for quantifying bacterial taxa within the gut microbiome, transitioning from an exploratory 16S rRNA–based sequencing study to a PCR/RPAbased confirmatory study can be complicated by differences in the way taxonomic abundance is defined (Ott et al., 2004; Morgan and Huttenhower, 2012; Gloor et al., 2017). In real-time RPA, bacteria are quantified according to a standard curve with abundance defined in terms of gene copies per unit volume (Euler et al., 2012b; Crannell et al., 2014, 2015; Kim and Lee, 2017). By contrast, in 16S rRNA–based sequencing, bacterial abundance is estimated based on the fraction of total observed 16S rRNA sequences assignable to a particular taxonomic group (with the complication that the number of 16S rRNA genes per genome can vary; Robinson et al., 2016). Thus, correlations based on 16S rRNA–based sequencing are relative rather than absolute (Ott et al., 2004); as such, subsequent quantitative PCR/RPA data should ideally be reported in terms of relative abundance. One way this has been achieved in gut bacterial PCR studies is by calculating the ratio of group-specific to total 16S rRNA abundance (Hermann-Bank et al., 2013; Brukner et al., 2015; Zhang et al., 2016). This method can be easily adapted to RPA to quantify target organisms.

The importance of accurate quantification of specific organisms within the gut microbiome can be illustrated by the case of diabetes, which affected 9.4% of Americans (30.3 million individuals) in 2015 (Centers for Disease Control and Prevention, 2017). Akkermansia muciniphila is a gramnegative, anaerobic, mucin-degrading bacterium commonly found in high abundance in the human gut. Low abundance of A. muciniphila has been linked to hyperglycemia, glucose intolerance, obesity, and type 2 diabetes (Everard et al., 2011, 2013; Louis et al., 2016; Yassour et al., 2016). A. muciniphila is also a key producer of short-chain fatty acids in the gut, which have been shown to inhibit inflammation and aid in metabolic dysregulation (Dao et al., 2016). In several mouse studies, administering A. muciniphila to diabetic subjects improved their metabolic functions and aided in weight loss (Greer et al., 2016; Plovier et al., 2016; Hänninen et al., 2017). The medical community may soon need an inexpensive screening tool for identifying individuals with lower fecal A. muciniphila abundance that could benefit from therapeutic intervention. We, therefore, chose A. muciniphila as a model organism in developing a quantitative RPA microbiome assay.

During assay development, we found that non-specific amplification occurred using 16S rRNA universal primer pairs in the absence of added DNA template, reducing the accuracy of total bacterial load quantification. Amplification of residual Escherichia coli production-host DNA contained in the recombinant RPA reagents was identified as the confounding factor and was confirmed by Sanger sequencing. Relative A. muciniphila abundance is calculated from the ratio of A. muciniphila to total 16S rRNA abundance. Therefore, an inability to quantify the total bacterial load precludes accurate estimation of the relative abundance of A. muciniphila. To resolve this issue, we explored four strategies to deplete or destroy extraneous E. coli DNA while preserving RPA reagent functionality: (1) metal-chelate affinity capture; (2) sonication; (3) methylation-dependent restriction endonuclease digestion; and (4) DNA capture/removal using anti-DNA antibodies. We show that removal of interfering DNA using anti-DNA antibodies was the most effective strategy. Using RPA reagents treated with anti-DNA antibodies, we then constructed a quantitative total bacterial standard curve and determined the relative abundance of A. muciniphila in a human fecal sample to demonstrate the application of the quantitative RPA microbiome assay.

## MATERIALS AND METHODS

#### Design of Bacteria-Specific Primers

It is suggested that RPA primers be 30–35 nt long for optimal amplification of the template. Only 9% of previously published RPA primers are below 30 nt in length (Daher et al., 2016). RPA was attempted with widely-used 16S universal primers (primer set 2; **Table 1**) which resulted in a significant delay in amplification. We, therefore, undertook to design our own bacteria-specific primers 28–35 nt in length and with GC content of 30–70%. Highly-conserved regions within the 16S rRNA gene were chosen as the targets for bacteria-specific RPA primers. When designing a 16S gene specific universal primer, it is likely that large increases in primer length would lead to an increase in the number of mismatches. However, guidelines for RPA primer design are not as stringent as those for the design of PCR primers. Longer RPA primers tolerate some additional mismatches while still achieving adequate amplification (Daher et al., 2015). The National Center for Biotechnology Information BLAST tool was used to test each primer's potential match to all publicly-available bacteria genome sequences. A match was defined for this purpose as less than six total mismatches and less than four mismatches in the last six 3′ nucleotides of either the primer or the target sequence. Primer sequences complementary to human DNA were identified using BLAST and excluded in order to reduce the probability of amplifying human DNA. The free-energy calculation function of the OligoAnalyzer tool (http://www.idtdna.com/analyzer/ Applications/OligoAnalyzer) was used to assess the potential formation of secondary structures (primer dimers and hairpins) and avoid inter- or intra-molecular interactions with a 1G of <−6 kcal.mol−<sup>1</sup> .

During A. muciniphila specific primer design, two different A. muciniphila 16S rRNA gene reference sequences (American Type Culture Collection [ATCC] strain BAA-835; accession number NR\_074436.1 and accession number NR\_042817.1) were downloaded from GenBank (http://www.ncbi.nlm.nih.gov/ genbank) and aligned using SeaView (Gascuel et al., 2010) to identify species-specific conserved sequences within the 16S rRNA gene variable region. The newly-developed A. muciniphila specific primers (primer set 3, **Table 1**) were tested for speciesspecificity using an NCBI BLAST search tool to confirm zero mismatches to all 39 A. muciniphila genomes in the GenBank reference genome database. The NCBI BLAST tool was also used to test each primer's off-target specificity to all publiclyavailable bacteria genome sequences outside of A. muciniphila. The reverse primer was found to be 100% matched to two off-target organisms, Haloferula rosa and Luteolibacter algae (Verrucomicrobia phylum). Luteolibacter algae and H. rosa are two species commonly found in marine environments (Yoon et al., 2008; He et al., 2017) and are not known to inhabit the human gut in significant abundance. Using data from the U.S. NIH Human Microbiome Project and the search engine EZ Bio Cloud (https://www.ezbiocloud.net/ resources/human\_microbiome) we found no presence of L. algae and H. rosa in stool. The Akkermansia genus dominates the Verrucomicrobia population found in the human gut (Dubourg et al., 2013). When found in the gut, the abundance of these species is not enough to significantly affect the calculation of A. muciniphila abundance (0–5% of total reads; Zhang et al., 2015).

#### Genomic DNA Standards

E. coli strain 1532 (ATCC 35218) was streaked onto a 5% sheep's blood agar plate (Becton, Dickinson and Company; Franklin Lakes, NJ) and incubated for 12 h at 37◦C. Genomic DNA (gDNA) was isolated from E. coli cultures using an UltraClean Microbial DNA Isolation kit (Mo Bio Laboratories; Carlsbad, CA). The absorbance of isolated E. coli and commercially obtained A. muciniphila gDNA (strain ATCC BAA-835) was measured at 230, 260, and 280 nm using a Nanodrop 1000 (NanoDrop Instruments, Wilmington, DE). The 260/280 and 260/230 absorbance ratios were ≥2.0, confirming gDNA purity. The gDNA concentration, expressed as genome copies per µL, was calculated using the absorbance at 260 nm, the extinction coefficient of double-stranded DNA (0.020 µg <sup>−</sup><sup>1</sup> mL cm−<sup>1</sup> ), the average molar weight of a DNA base pair (650 g mol−<sup>1</sup> ), the size of each reference strain's genome (E. coli ATCC 35218, 4.64 Mbp; A. muciniphila ATCC BAA-835, 2.66 Mbp), and Avogadro's number. gDNA aliquots (10 µL) were stored at −20◦C and used as RPA standards. Ten-fold serial dilutions of both E. coli and A. muciniphila gDNA were made using nuclease-free deionized water. Finally, 2 or 5 µL of template (100–10<sup>7</sup> copies of E. coli or A. muciniphila gDNA) were added to quantitative RPA and PCR reactions.



#### Quantitative Real-Time RPA

Real-time RPA reactions were performed using a TwistAmp Basic kit (TwistDx, Cambridge, UK, TABAS03KIT) with primers purchased from Integrated DNA Technologies (Coralville, IA). Master mix (45.5µL) containing 420 nM primers (primer pair set 1 or 2, **Table 1**), SYBR green dye I (ThermoFisher Scientific, Product #S7567; 45,500-fold dilution of stock concentration), and TwistAmp rehydration buffer was prepared and distributed into TwistAmp Basic reaction tubes. Next, 2 or 5 µL of template (100–10<sup>6</sup> copies of E. coli or A. muciniphila gDNA standard) and 2.5µL of 280 mM magnesium acetate (MgAc) were added to the reaction mix to initiate the amplification reaction. Tubes were then placed into an Agilent MxPro 3005 real-time PCR machine (Agilent Technologies, Santa Clara, CA). Fluorescence (excitation, 497 nm; detection, 520 nm) was measured every 15 s for 60 min at 37◦C.

#### Strategies for Removing Extraneous DNA From RPA Reagents

#### Strategy 1: Removing DNA Using Metal-Chelate Affinity Capture

Chelating Sepharose fast-flow beads (25µL; catalog no. 17057502, GE Healthcare; cross-linked 6% agarose functionalized with iminodiacetic acid groups) were charged according to the manufacturer's protocol with Ni2<sup>+</sup> ions to enable interaction with aromatic DNA base nitrogen atoms (Murphy et al., 2003; Cano et al., 2005) and then resuspended in 70 µL of 60 mM Tris buffer (pH 7) containing 1 M NaCl. Four TwistDx TwistAmp RPA reaction pellets were rehydrated in 70 µL of TwistDx Rehydration Buffer. The bead and RPA reagent suspensions were combined and mixed on a rotator at end-over-end at 3 rpm for 2 h at 4◦C. The mixture was then centrifuged at 5,500 xg for 2 min to remove the beads, and the supernatant was aliquoted into four PCR tubes (35 µL per tube). Master mix (47.5µL) containing 505µM primers and SYBR green dye (47,500-fold dilution of stock solution) was added to each tube along with 2µL of E. coli gDNA standard (100 copies per µL). Next, 2.5µL of 280 mM MgAc was added to the PCR tubes, which were then placed into an Agilent MxPro 3005 Real-Time PCR machine and amplified at 37◦C.

#### Strategy 2: DNA Shearing by Sonication

Individual TwistAmp RPA Reaction pellets were reconstituted with 29.5µL of TwistDx Rehydration Buffer, placed on ice, and sonicated at a amplitude of 40 for 10 cycles (3 s on, 7 s off) using an ultrasonic homogenizer (Model 150V/T, Biologics Inc.). Next, 29.5µL of the sonicated suspension and 16 µL of master mix (containing 1.5µM primer set 1 and 16,000-fold dilution of SYBR green dye) were pipetted into new PCR tubes. Template (2 µL; 100 E. coli gDNA copies per µL) and 2.5 µL of 280 mM MgAc were added to initiate RPA, and the tubes were placed into an Agilent MxPro 3005 real-time PCR machine for amplification.

#### Strategy 3: Methylation-Dependent Endonuclease Digestion

RPA pellets were suspended in 29.5 µL of Rehydration Buffer with 0–140 units of DpnI restriction endonuclease (New England Biolabs, product #R0176S) and incubated for 15 or 60 min at 37◦C, with or without 14 mM MgAc, to cleave methylated E. coli gDNA at Gm6ATC restriction sites. After digestion, tubes with treated pellet solution were mixed with 47.5 µL of master mix containing 505µM primers and SYBR green dye (47,500-fold dilution of stock solution). Finally, 2 µL (100 copies) of E. coli gDNA standard and 2.5 µL of 280 mM MgAc were added to the tube caps and spun down in a microcentrifuge to simultaneously initiate the RPA reactions.

#### Strategy 4: DNA Depletion Using Anti-DNA Antibodies

Anti-dsDNA antibodies coupled to amine-modified magnetic particles were prepared as follows. Nine hundred microliters of 55.6µg mL−<sup>1</sup> anti-dsDNA antibody (#ab27156, Abcam) in 100 mM sodium acetate buffer (pH 5.4) was added to 45µL of 0.1 M NaIO4. After 30 min of incubation at room temperature (RT), oxidized antibodies were concentrated using 100-kDa Amicon Ultra centrifugal filters (Millipore, Billerica, MA, USA) and diluted to 500 µL at 100 µg mL−<sup>1</sup> in 200 mM sodium carbonate buffer (pH 9.6). Next, 3.1-µm Promag amine microspheres (200 µL; 1 × 10<sup>8</sup> particles; Bangs Laboratories Inc.) were washed and resuspended in 500µL of 200 mM sodium carbonate buffer (pH 9.6) and then added to the oxidized antibodies. After incubation at RT for 2 h, 15µL of 5 M NaCNBH3, 1 M NaOH was added to the reaction and incubated for an additional 30 min at RT. Next, 75 µL of 1 M hydroxylamine was added, and the mixture was incubated for 30 min. The antibody-functionalized magnetic particles were washed 3 times and stored at 4◦C in phosphate-buffered saline (PBS) (final concentration, 1 × 10<sup>5</sup> particles per µL).

To remove extraneous DNA, TwistAmp Basic kit freezedried pellets (TwistDx) were reconstituted in 29.5 µL of TwistDx Rehydration Buffer and transferred to a low-binding microcentrifuge tube containing 2 × 10<sup>5</sup> antibody-functionalized magnetic beads in 2 µL of buffer. The microcentrifuge tube was placed on a rotator and incubated for 30 min at 4◦C, after which the beads were removed using a magnet, and the solution was transferred to new tubes in 29.5-µL aliquots. Master mix (47.5 µL) containing 505µM primers and SYBR green dye (47,500 fold dilution of the stock solution) was added to each tube along with E. coli gDNA (100 genome copies per µL) in 2 µL. Finally, 2.5 µL of 280 mM MgAc was added to the tube caps and spun in a microcentrifuge to simultaneously initiate RPA.

#### qPCR Assays

Real-time PCR standard curves were prepared with 10-fold serial dilutions of gDNA standards (E. coli ATCC 35218 or A. muciniphila ATCC BAA-835) using previously-validated primer pairs (16S rRNA bacteria-specific primer set 2, or A. muciniphila specific primer set 4, **Table 1**). qPCR was performed in 20-µL reaction volumes containing 550 nM primers (1.1 µL of 10µM), a 1× concentration of Brilliant III Ultra-Fast SYBR green qPCR master mix (10 µL of 2× commercial stock conc.; Agilent Technologies), and 2 or 5 µL of template. Amplification curves, baselines, and threshold cycles were calculated using Agilent MxPro 3005P software as described below.

#### Threshold Time and Cycle Parameters for RPA and qPCR Reactions

MxPro software (Agilent Technologies) was used to normalize baseline fluorescence and calculate RPA threshold times from raw fluorescence data. Using a linear least mean squares algorithm, the baseline function was calculated by fitting the raw fluorescence in the first 3 min of the reaction to a firstorder function. Baseline-corrected fluorescence was obtained by subtracting the baseline from the raw fluorescence to plot an amplification curve. For each assay, threshold fluorescence was defined as the point at which the fluorescence exceeded the average baseline-corrected fluorescence by three standard deviations, located in the exponential region of the amplification curve. The threshold time or cycle was calculated as the time or cycle at which the reaction met the threshold fluorescence. For each assay, average threshold cycle (n = 3) was plotted against copies of gDNA per reaction on semi-logarithmic axes to generate a regression line.

#### Determining Total Bacterial and *A. muciniphila* Abundance

An anonymized human fecal sample from a deceased Caucasian female with a medical history of hypoglycemia and hypothyroidism was obtained from Analytical Biological Services Inc. Three micrograms of DNA (15 ng/µL) was isolated from 220 mg of the fecal sample using QIAamp DNA Stool mini kit (catalog no. 51504, Qiagen). Nucleic acid purity was confirmed by a 260 nm/280 nm absorbance ratio of 1.83 using NanodropTM. Either 3 or 7.5 ng of isolated DNA (2- or 5-µL of a 10-fold dilution of the stock concentration) was run as the template in quantitative RPA and PCR assays. Average threshold time (RPA) or cycle (PCR) and relative standard curves were used to determine the number of E. coli or A. muciniphila gDNA copies per 15 ng of the isolated DNA. Finally, the relative A. muciniphila abundance was calculated as the ratio of A. muciniphila gDNA copies per µL to the number of bacterial gDNA copies per µL.

For both A. muciniphila (ATCC BAA-835) and E. coli (ATCC 35218) gDNA standards, the 16S rRNA gene sequences from the GenBank (accession nos. NR\_074436.1 and EF436579, respectively) were aligned to the complete genome sequences of A. muciniphila (accession no. NC\_010655.1) and E. coli (accession NZ\_KK583188.1) using the NCBI BLAST. This resulted in three and seven matches (100% in identity and composition) for A. muciniphila and E. coli, respectively. Thus, to calculate the 16S rRNA gene copies for each standard, the genomic DNA copies were multiplied by the number of 16rRNA gene copies per genome (3 for A. muciniphila and 7 for E. coli).

#### Sequencing-Based Detection

The composition of the fecal microbiome was independently characterized using 16S rRNA sequencing. A QIAamp DNA Stool mini kit was used to isolate DNA from the fecal sample. SeqWright Genomic Services (Houston, TX) generated and validated a library of 300-bp amplicons (MiSeq Reagent kit v3, MS-102-3001) using the extracted DNA and a universal primer pair (primer set 5; **Table 1**) spanning the 16S rRNA V4 region (Caporaso et al., 2011). Sequencing was performed on an Illumina MiSeq instrument with 250-bp paired-end reads. Reads were then uploaded to the Sequencing Read Archive (BioProjectID: PRJNA472995) and Illumina Basespace Sequencing Hub for sequencing analysis. Kraken Metagenomics software was used to assign taxonomic labels to each read using a k-mer–based algorithm. The relative abundance of A. muciniphila was then calculated by comparing the number of reads classified as A. muciniphila to the total number of bacterial reads.

#### RESULTS

### Identification of *Escherichia coli* DNA Contamination in RPA Reagents

With unmodified RPA reagents, amplification was observed with primer set 1 in reactions containing 100–1,000,000 copies of E. coli gDNA in 10.3 min essentially independent of initial template concentration (standard deviation: 60 s, range: 10–11 min); no-template controls also showed amplification in 9.5 min (n = 1) (Figure S1). To confirm that the newly designed primer pair (primer set 1) was bacteria-specific, the amplification curves generated using primer set 1 were compared to amplification curves generated using the previously published and validated universal primer set 2 (**Table 1**, Figure S1). Amplification was observed with primer set 2 in reactions containing 100–1,000,000 copies of E. coli gDNA in 24.2 min (standard deviation: 150 s, range: 23.2–25.8 min), essentially independent of template concentration; the no-template control also showed amplification in 24.8 min (n = 1). A difference in primer length is the cause of the large jump in threshold time that is observed when no-template control reactions are tested with the two universal primer pairs. The manufacturer of the RPA reagent (TwistDx) recommends that RPA primers be 30–35 nt long. When RPA reactions are tested with PCR-size primers (15–22 nt in length), reactions will take longer to achieve threshold fluorescence than when reactions are tested with primers of the recommended length.

Amplicons generated from the no-template reaction with primer set 1 were purified using a QIAquick PCR Purification kit. To determine if primer dimers were contributing to unintended amplification, the purified RPA product was analyzed by gel electrophoresis to show the expected ∼330-bp single band. To determine the phylogenic origin of the amplicon, the product was Sanger sequenced (Genewiz, Houston, TX) and the sequencing traces analyzed with Sequence Scanner 2 (Applied Biosystems, Foster City, CA) to generate a consensus sequence. The consensus sequence was aligned against NCBI's 16S ribosomal RNA database using nBLAST to show the highest similarity (99– 100% identity) when aligned with a portion of the 16S rRNA gene in E. coli. Information provided by the manufacturer indicated that essential enzymes in RPA reagents are produced in an E. coli expression host. We, therefore, hypothesized that DNA from the E. coli vector is present in the RPA reaction pellets and amplified when 16S rRNA universal primers are used.

## Strategies for Removing Extraneous DNA From RPA Reagents

#### Removing DNA Using Metal-Chelate Affinity Capture

As the purines of single-stranded nucleic acids contain an imidazole ring similar to that of the histidines recognized in "His6-tagging," they exhibit a strong affinity for chelated transition metal ions (Murphy et al., 2003; Cano et al., 2005). Thus, we hypothesized that Ni-IDA immobilized metal affinity chromatography (IMAC) agarose beads would be an effective means of selectively binding and removing extraneous DNA from RPA reagents.

RPA reaction pellets were first reconstituted in rehydration buffer and then incubated with Ni+<sup>2</sup> -loaded IDA sepharose particles, as described in section Materials and Methods. The particles were removed by centrifugation, and the resulting supernatant was tested using primer set 1 in real-time RPA reactions spiked with 100 copies of E. coli gDNA (data not shown). Amplification curves generated using Ni-treated, and untreated reagents were indistinguishable, indicating that: (1) there was no significant decline in reaction efficiency after incubation with the IMAC resin; (2) RPA proteins exhibited low nonspecific (or His-tag) binding to Ni-IDA sepharose beads; and (3) little, if any, DNA was removed during the treatment of RPA pellets. Given our previous finding that Ni-IDA affinity is specific to single-stranded nucleic acids with exposed purine bases (Murphy et al., 2003; Cano et al., 2005), this result suggests that the majority of the residual DNA in the RPA reagents is double-stranded.

#### DNA Shearing by Sonication

Sonication is widely used in next-generation sequencing protocols to shear genomic DNA into fragments of as little as 150–200 bp. Gel electrophoresis analysis of the product generated from primer set 2, and no-template control reactions showed a single band, ∼300 bp in length. Thus, we hypothesized that sonicating the reaction pellet would render the contaminating DNA un-amplifiable in downstream RPA reactions. Notemplate reactions with sonicated reagents reached the threshold fluorescence, on average, 60 s earlier than reactions with untreated reagents (Figure S2). Thus, we concluded that: (1) RPA reagents are tolerant of the degree of sonication applied, and (2) sonicating RPA pellets was not sufficient to shear extraneous DNA to such a degree as to impair amplification.

This result is consistent with previous studies which showed increases in amplification efficiency when samples are sonicated before PCR (Golenberg et al., 1996; Veal et al., 2012). Fragmenting genomic DNA into ∼1 kb segments suggests facilitation of DNA dehybridization in GC-rich regions for quicker binding of polymerase at recognition sites and enhanced strand amplification.

#### Methylation-Dependent Endonuclease Digestion

DpnI is a methylation-dependent restriction endonuclease that cleaves DNA prepared from E. coli dam<sup>+</sup> strains but not PCRamplified DNA (Glickman, 1980; Barnes et al., 2014). The DpnI cleavage site (5′ -Gm6ATC) was found to occur twice within the region of predicted amplification in the E. coli 16S rRNA gene (Figure S3). Thus, pre-treating RPA reagents with DpnI restriction enzymes should digest and render the extraneous DNA un-amplifiable.

Laborious removal of DpnI after incubation and prior to initiation of the RPA reaction likely would be unnecessary. As RPA reactions are isothermal, they commence immediately upon mixing of the reagents. This continuous amplification leads to prompt generation of synthetic products (Piepenburg et al., 2006). If sample (reaction template) and primers are simultaneously added to the DpnI-treated pellet, a portion of the sample DNA is likely to evade DpnI cleavage, allowing it to be amplified during the first round of RPA. The resulting unmethylated synthetic products are impervious to DpnI enzymatic cleavage and are available to act as a template in subsequent rounds of amplification.

No-template RPA reactions that contained pellets pre-treated for 15 min with 20 units of DpnI showed only a 90-s delay in threshold time when compared with reactions using untreated

pellets, suggesting that some but not all of the extraneous DNA was digested during DpnI treatment (**Figure 1**). Negative control reactions with pellets incubated with DpnI in the absence of magnesium (DpnI's co-factor) showed the same threshold time as reactions using untreated RPA reagent.

Increasing the DpnI digestion time from 15 to 60 min resulted in no increase in the threshold time difference between notemplate reactions with treated and untreated RPA reagent. However, increasing the DpnI concentration 5- or 7-fold resulted in a 2.75 and 5.5-min increases in the threshold time, respectively (data not shown). Results indicate that treating RPA reagents with DpnI can digest and render a significant proportion of contaminating DNA un-amplifiable in subsequent applications.

#### DNA Depletion Using Anti-DNA Antibodies

Magnetic particles conjugated with anti-DNA antibodies with affinity for both dsDNA and ssDNA were used to treat RPA reagents, as described in section Materials and Methods. Notemplate RPA reactions that were pretreated with anti-dsDNA antibody magnetic particles showed a 3-min later threshold time than reactions involving untreated reagents (**Figure 2**). These data demonstrate that treating RPA reagents with anti-dsDNA antibody magnetic particles removes a significant proportion of extraneous DNA.

Reactions with treated reagents showed a 1.75-min earlier threshold time when spiked with 100 copies of E. coli gDNA compared with reactions with treated reagents in the absence of spiked E. coli gDNA (**Figure 2**). These data indicate that treatment with anti-dsDNA antibody magnetic particles does not significantly remove proteins necessary for RPA amplification.

Both DpnI treatment and anti-dsDNA antibody magnetic particles effectively removed extraneous DNA, with similar efficiencies, from RPA reagents. Treating RPA reagents with 100 units of DpnI showed a 2.75 min delay in threshold time when compared to reactions with untreated reagents. RPA reagents treated with anti-dsDNA antibody showed a comparable delay of 3 min in threshold time when compared to reactions with untreated reagents.

The cost of RPA reagents is ∼\$5 per reaction; these reagents can be treated to remove extraneous DNA with 100 units of DpnI at a cost of \$3.33, or using anti-DNA antibody and magnetic particles for \$2.25 per reaction. Thus later work employed the slightly less-expensive anti-DNA magnetic particles.

#### Quantitative Bacterial Detection Using RPA

To determine the detection range of the assay based on DNAdepleted reagents, 10-fold serial dilutions of E. coli gDNA standards were spiked into RPA reactions containing RPA reagents pre-treated with anti-ds DNA antibody magnetic particles (**Figure 3A**). Two replicates for concentrations ranging from 10<sup>6</sup> to 10<sup>3</sup> E. coli gDNA copies per reaction and two no-template control reactions were performed. Threshold time versus log copy number of E. coli gDNA was fit to a semi-logarithmic regression line to generate a standard curve (**Figure 3B**).

To demonstrate that the total bacterial load in a clinically relevant sample can be quantified using RPA, DNA was isolated from a human fecal sample (Analytical Biological Services, Inc.) and amplified in RPA reactions. Based on the E. coli standard curve, the total bacterial load of the fecal sample was 1.01 × 10<sup>7</sup> bacterial gDNA copies per 15 ng of isolated gDNA.

The same fecal sample was also re-analyzed using real-time PCR. A total of 2 µL of each E. coli gDNA standard dilution was spiked into PCR reactions containing previously published pan bacteria–specific primers (**Table 1**, primer set 2; n = 3). Threshold cycle was plotted against log copies of E. coli gDNA per reaction to generate a standard curve (Figure S4), which was then used to calculate the total bacterial load of the fecal sample as 5.58 × 10<sup>6</sup> bacterial gDNA copies per 15 ng of isolated DNA. The total bacterial load of the stool sample quantified by qPCR differed

by less than two-fold from the total bacterial load estimated by quantitative RPA (1.01 × 10<sup>7</sup> bacterial gDNA copies per 15 ng of isolated gDNA).

muciniphila load of the fecal sample as determined using RPA (2.99 × 10<sup>5</sup> bacterial gDNA copies per 15 ng of isolated gDNA).

#### Development of the *A. muciniphila* Assay Absolute A. muciniphila Abundance

Akkermansia muciniphila gDNA (0–1,000,000 copies; ATCC BAA-835) was amplified using real-time SYBR Green-RPA with A. muciniphila–specific primers designed in this work (**Table 1**, primer set 3). The amplification curves were used to determine threshold time values and generate a standard curve (log copy number gDNA vs. threshold cycle, **Figure 4**). Reactions run with no added template (NTC) exhibited an average threshold time of 10.8 ± 0.42 min (n = 3). Reactions spiked with 1,000 copies E. coli gDNA (ATCC 35218) exhibited an average threshold time insignificantly different from that of the NTC (10.6 min; n = 3).

Five microliters of a 10-fold dilution of gDNA isolated from the fecal sample (15 ng gDNA per µl) was run as a template in RPA reactions to yield an average threshold time of 6.05 ± 0.1 min (n = 3). The standard curve (**Figure 4**) was then used to quantify the absolute load of A. muciniphila as 2.99 × 10<sup>5</sup> gDNA copies per 15 ng of DNA isolated from the fecal sample.

The absolute A. muciniphila concentration in the fecal sample was separately determined by qPCR using primer set 4 (**Table 1**), which reportedly enables determination of the absolute abundance of A. muciniphila via qPCR (Collado et al., 2007; Schneeberger et al., 2015; Guo et al., 2016). The absolute A. muciniphila load of the fecal sample was estimated based on the qPCR semi-logarithmic regression line as 8.91 × 10<sup>4</sup> gDNA copies per reaction, or 1.78 × 10<sup>5</sup> gDNA copies per 15 ng of isolated gDNA (Figure S5), quite similar to the absolute A.

### Relative *A. muciniphila* Abundance

Relative A. muciniphila abundance was calculated as the ratio of A. muciniphila 16S copies to total bacterial 16S copies using both qPCR (using primer sets 2 and 4) and RPA (using primer sets 1 and 3) to show a relative abundance of 1.36 and 1.29%, respectively. The relative A. muciniphila abundance of the fecal sample was determined using RPA from 3 × 2.99 × 10<sup>5</sup> A. muciniphila 16S copies per 15 ng of gDNA divided by 7 × 1.01 × 10<sup>7</sup> bacterial 16S copies per 15 ng of gDNA) and from PCR 5.34 × 10<sup>5</sup> A. muciniphila 16S copies/3.91 × 10<sup>7</sup> bacterial 16S copies) as 3 × 1.78 × 10<sup>5</sup> A. muciniphila 16S copies per 15 ng of gDNA divided by 7 × 5.58 × 10<sup>6</sup> bacterial 16S copies per 15 ng of gDNA.

In 16S rRNA gene sequencing, relative A. muciniphila abundance was defined as the ratio of A. muciniphila reads to the total number of bacterial reads. In 16S rRNA sequencing, the fecal sample produced 14,358,714 reads, of which 14,285,134 (99.5%) were taxonomically classified as bacteria. A total of 297,040 reads (2.07% of all reads) were classified as A. muciniphila ATCC BAA-835–specific, in good agreement with the RPA result.

When compared to sequencing, RPA gave a slightly lower relative A. muciniphila abundance in the fecal sample. This result could have been due to off-target amplification of the bacteria-specific primers. Note that the accuracy of the relative A. muciniphila abundance RPA assay, as with all nucleic acidbased assays, is highly dependent upon the quality of the primers.

RPA assay sensitivity could perhaps be improved by increasing the primer specificity.

#### DISCUSSION

Many microbial taxonomic groups have been identified as beneficial, detrimental, or simply indicative of a wide range of health conditions. As the number of organisms of interest expands, so does the need for inexpensive, accurate, and quantitative measurement of bacterial abundance within the gut. Unlike qPCR or 16S sequencing, RPA enables field-based testing. Recombinase polymerase amplification (RPA) is an isothermal nucleic acid amplification method that employs commercially available, easy-to-use freeze-dried, reaction-ready enzyme pellets that can be used to analyze specimens rapidly in the field, using a portable fluorometer. Analyzing complex microbial communities immediately after sampling can lead to a more accurate quantification of relative bacterial abundance. RPA is a method that is quickly growing in usage and range of applications (Piepenburg et al., 2008; Kim and Easley, 2011; Loo et al., 2013; Shin et al., 2013; Xu et al., 2014; Tortajada-Genaro et al., 2015; Daher et al., 2016; Yamanaka et al., 2017).

While many groups have developed RPA assays that demonstrate species-specific detection (Euler et al., 2012b; Ahmed et al., 2014; Krõlov et al., 2014; Clancy et al., 2015; Liljander et al., 2015; Cabada et al., 2017), there has been no development of an RPA assay that performs bacterial-specific detection. This gap is due to a high level of E. coli DNA in commercial RPA reagents. This paper is the first to offer a solution to this contamination. After successfully removing the contaminating DNA, we proceeded to demonstrate successful RPA quantification of bacteria in a stool sample.

In this proof-of-concept study we enabled the estimation of the relative abundance of A. muciniphila in human fecal DNA and demonstrated the promise of RPA as an inexpensive and accurate tool for measuring gut microbiome marker organisms. A. muciniphila appears to play a pivotal role in insulin resistance and inflammation. Furthermore, low abundance of A. muciniphila in the human gut is found to correlate with obesity. More recent studies indicate that A. muciniphila can influence the effectiveness of certain cancer immunotherapy drugs (Gopalakrishnan et al., 2018; Routy et al., 2018). Altering patients' gut microbiota to be rich in A. muciniphila may increase the fraction of individuals who respond to cancer immunotherapies. Given the short time A. muciniphila has been studied, the multitude of studies that support its role as a beneficial bacterium suggests its important role in human health. The medical community may soon need an inexpensive screening tool for identifying individuals with low fecal A. muciniphila abundance. To further assess the analytical accuracy and sensitivity of Akkermansia muciniphila detection with RPA more fecal samples should be tested.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

HG helped conceive the project, designed and executed experiments, and wrote the manuscript. DC helped to design and perform experiments, and edited the manuscript. MC helped to interpret and conceive the project, and edited the manuscript. KK interpreted experiments, and helped to write and edit the manuscript. RW conceived the project, interpreted experiments, and helped to write and edit the manuscript.

#### ACKNOWLEDGMENTS

This work was supported by the FEMSA Foundation and the National Science Foundation, www.nsf.gov, (Grant Number 1759440).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcimb. 2018.00237/full#supplementary-material

pathogenic Leptospira. Int. J. Environ. Res. Public Health 11, 4953–4964. doi: 10.3390/ijerph110504953


load with associated clinical implications. Diagn. Microbiol. Infect. Dis. 83, 1–6. doi: 10.1016/j.diagmicrobio.2015.04.005


**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.

The reviewer JG and handling Editor declared their shared affiliation.

Copyright © 2018 Goux, Chavan, Crum, Kourentzi and Willson. 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) 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.

# Intestinal Inflammation in Chilean Infants Fed With Bovine Formula vs. Breast Milk and Its Association With Their Gut Microbiota

Juan C. Ossa\*, Dominique Yáñez, Romina Valenzuela, Pablo Gallardo, Yalda Lucero and Mauricio J. Farfán\*

Departamento de Pediatría y Cirugía Infantil, Facultad de Medicina, Hospital Dr. Luis Calvo Mackenna, Universidad de Chile, Santiago, Chile

Introduction: Compared to bovine formula (BF), breast milk (BM) has unique properties. In the newborn intestine, there is a homeostatic balance between the counterparts of the immune system, which allows a physiological inflammation, modulated by the gut microbiota. Many studies have attempted to understand the effect of BF vs. BM, and the changes in the gut microbiota, but few also focus on intestinal inflammation.

Methods: We conducted a cohort study of newborn infants during their first 3 months. In stool samples taken at 1 and 3 months (timepoints T1 and T3), we quantified calprotectin, IL-8 and α1-antitrypsin by ELISA and we evaluated the expression of IL8 and IL1β genes by RT-qPCR. To determine the microbiota composition, the 16S rRNA gene was amplified and sequenced using 454 pyrosequencing. Sequences were clustered into operational taxonomic units (OTUs).

#### Results: In total 15 BM and 10 BF infants were enrolled. In the BM group, we found calprotectin and α1-antitrypsin levels were significantly elevated at T3 compared to T1; no differences were found between T1 and T3 in the BF group. A comparison between the BM and BF groups showed that calprotectin levels at T1 were lower in the BM than the BF group; this difference was not observed at T3. For IL-8 levels, we found no differences between groups. A gene expression analysis of the IL8 and IL1β genes showed that infants from the BF group at T1 have a significantly increased expression of these markers compared to the BM group. Gut microbiota analyses revealed that the phylum Bacteroidetes was higher in BM than BF, whereas Firmicutes were higher in BF. A redundancy analysis and ANOVA showed BM has a community structure statistically different to BF at T1 but not at T3. Compared to BF, BM at T1 showed a higher representation of Enterococcus, Streptococcus, Enterobacter, Lactococcus, and Propionibacterium.

Conclusions: We found a basal state of inflammation in the infants' intestine based on inflammation markers. One month after birth, infants receiving BF exhibited higher levels of inflammation compared to BM.

Keywords: intestinal inflammation, bovine formula, breast milk, gut microbiota, infant cohort

#### Edited by:

Pascale Alard, University of Louisville, United States

#### Reviewed by:

Young Min Kwon, University of Arkansas, United States Erdong Cheng, University of Pittsburgh Cancer Institute, United States

#### \*Correspondence:

Juan C. Ossa jcossa@gmail.com Mauricio J. Farfán mfarfan@med.uchile.cl

Received: 12 December 2017 Accepted: 17 May 2018 Published: 21 June 2018

#### Citation:

Ossa JC, Yáñez D, Valenzuela R, Gallardo P, Lucero Y and Farfán MJ (2018) Intestinal Inflammation in Chilean Infants Fed With Bovine Formula vs. Breast Milk and Its Association With Their Gut Microbiota. Front. Cell. Infect. Microbiol. 8:190. doi: 10.3389/fcimb.2018.00190

## INTRODUCTION

Breast milk (BM) has been and will continue to be the ideal type of nutrition for every term or pre-term newborn. The WHO recommends exclusive breastfeeding for the first 6 months of life, with supplemental breastfeeding until 2 years old and beyond (Hoddinott et al., 2008). Compared to bovine formula milk (BF), BM contains nutrients, hormones, growth factors, immunoglobulins, cytokines and bacteria which confer protection against many diseases, such as necrotizing enterocolitis, respiratory and gastrointestinal infections, allergy, celiac disease, obesity, diabetes type I and II (Horta et al., 2007; Le Huërou-Luron et al., 2010). In the healthy newborn intestine, the counterparts of the immune system allow the mucosa to display a physiological inflammation, a result of the immune response to diet and bacteria in the intestinal lumen (Fiocchi, 1997). Gut microbiota starts to develop in utero, inherited from the mother, and is later influenced by the mode of delivery and the newborn feeding pattern (Bäckhed et al., 2012). Nowadays, gut microbiota plays a key role in maturation and maintenance of the immune system, food metabolism, intestinal epithelial cell homeostasis, protection against pathogens and neural development of the gut-brain axis (Hill and Artis, 2010; Lathrop et al., 2011). The shift in the composition of a healthy microbiota to an unhealthy one is called dysbiosis. Currently, many enteric and non-enteric diseases have been associated with dysbiosis of the gut microbiota (Arrieta et al., 2014).

Most studies have attempted to understand the effect of BF or BM on the gut microbiota composition showing that BF feeding is associated with microbiota with lower abundance of Bacteroides, and higher Clostridia compared to BM-fed infants. Regarding Bifidobacteria, there is a controversy as to whether they are lower in number and frequency in BF than BM. Also, BF-fed infants exhibit higher counts of Enterobacteriaceae than BM-fed infants (Guaraldi and Salvatori, 2012; Fan et al., 2014). Other studies have shown the effect of diet on the intestinal cell homeostasis in healthy neonates by either analyzing gene expression from exfoliated epithelial cells or protein levels in stools. Although several inflammatory markers as well as inflammatory gene expression have been evaluated in stool or serum in infant, few studies have addressed diet in the newborn looking at intestinal inflammation and its relation to changes in gut microbiota composition (Chapkin et al., 2010; Savino et al., 2010). Considering the above, using a non-invasive technique based on stool analysis, we conducted a 3-month cohort study of newborn infants who are either in the exclusive BM or BF to determine inflammatory markers in stool and gut microbiota composition.

## METHODS

## Study Design

We conducted a 3-month cohort study following 2 groups of newborn infants who are fed exclusively either with BM or BF. All infants were recruited from the maternity ward of Hospital Luis Tisné in Santiago, Chile. In order to associate the effect of diet with inflammation marker levels and the microbiota composition over the 3-month period, a collection of stool samples were taken at 1 (timepoint T1) and 3 (timepoint T3) months of life, within a range of ±5 days. The timepoints were chosen due to the limited knowledge about the intestinal inflammation and its association with gut microbiota under 6 months.

#### Patients

Inclusion Criteria. We enrolled infants born at term (38–42 weeks of gestation), vaginally delivered, and described healthy at the time of discharge from the hospital. For enrollment, all the infants had to be receiving BM or BF exclusively. The BF group also included infants who were in BM and receiving formula supplements ≥20% of the volume ingested that day. Exclusion Criteria. We excluded from the study infants or mothers who during the study period received antibiotics, probiotics, steroidal or non-steroidal anti-inflammatory drugs 1 month prior to enrollment. Also, we excluded from the study mothers hospitalized other than for delivery, for surgical intervention, serious infection, or with any sign or symptom of infection or gastrointestinal disease (diarrhea, vomiting, fever).

### Clinical Assessment

During recruitment, a complete clinical interview was done to take the history regarding pregnancy, delivery mode, birth weight, frequency and quantity of feeding, stool frequency, family composition, number of siblings, history of allergy, gastrointestinal and other systemic disorders in the family, family income and household environment (number of rooms, water supply and pets in the home).

#### Sample Collection

A stool sample was obtained from the infants during the endpoints described above. Samples were collected during a clinical visit to our center, the Hospital Luis Calvo Mackenna, Santiago, Chile. In case that the infant has no stool during the visit, a home kit for the parents was given to take the sample and store it in a sterile container to be transported to our center within the following 6 h. Stool samples were divided into at least 4 aliquots and stored at −80◦C.

#### Ethics

This study was done in accordance with the recommendations of the Declaration of Helsinki. The study protocol was approved by the ethical committees of the Servicio de Salud Metropolitano Oriente and Hospital Luis Tisné. Written informed consent was obtained from all parents on behalf of their infants.

### Inflammatory Protein Markers Determination

ELISA commercial kits for stool samples were used for the analysis of calprotectin (IDK <sup>R</sup> calprotectin ELISA, Immunodiagnostik, Germany) and α1-antitrypsin (IDK <sup>R</sup>

α1-antitrypsin ELISA, Immunodiagnostik, Germany). Samples were processed as directed by the manufacturer. For IL-8, we determined the concentration of these markers by ELISA as previously described (Harrington et al., 2005).

#### Gene Expression Assay

RNA was isolated from the stool sample using the Stool Total RNA Purification Kit (Norgen Biotek) according to the manufacturer's instructions. Then, cDNA was synthesized using the First Strand cDNA Synthesis Kit (ThermoFisher Scientific) and RT-qPCR was carried out with the TaqMan Gene Expression Assay (ThermoFisher Scientific) and specific TaqMan probes for IL8, IL1β, and GADPH genes as previously described (Bennett et al., 2009; Chapkin et al., 2010). The GADPH gene was used as a housekeeping gene to normalize the expression of IL8 and IL1β. Changes in cycle threshold (1CT) values for each gene were obtained at T1 and T3. The mean of the 1CT of the BM group was used as a reference for the fold expression changes in the BF group.

## Pyrosequencing and Operational Taxonomic Unit (OTU) Assignment

Total DNA was extracted from stool samples using the QIAamp Fast DNA Stool Mini Kit (Qiagen) and stored at −20◦C until PCR amplification. The 16S rRNA gene was amplified in a twostep process. First, the 16S rRNA gene was amplified using the primers GM3 and 1492R, and then a nested PCR was performed using the GM3-PS forward primer and a different 907-PS reverse primer for each sample in a 7-cycle reaction as described (Gallardo et al., 2017). Amplicons were purified and the concentration of the purified product was determined. Equimolar mixtures of the amplicons (10–12 samples each) were shipped to Macrogen Inc. (Seoul, Korea) for pyrosequencing. Pyrosequencing of each mix was done through 454 GS-FLX using a 1/8 plate. Sequence trimming and OTU assignment were performed by Macrogen using CD-HIT-DUP and QIIME (Caporaso et al., 2010) according to their standard protocol [cutoff of 97% of sequence identity at species level for OTU assignment and using the 11th version of RDP-16s rDNA database as reference (http://rdp.cme.msu.edu/index.jsp)]. The sequence data reported in this study have been deposited in the European Nucleotide Archive (ENA) database, under accession number PRJEB25846.

#### Statistics

Gene expression and protein results are expressed as means ± standard error of the mean (SEM). Comparison of results between multiple groups was performed using one-way analysis of variance (ANOVA) and the Tukey-Kramer multiple comparisons test for results between the different experimental groups. For the OTU comparison analysis we used a non-parametric bootstrapping method, the Mann-Whitney u test. Differences with a P < 0.05 were considered statistically significant. Analyses were performed using Prism6 (GraphPad, San Diego, California, USA). Redundancy analysis (RDA) of OTU composition was done using the "vegan" package of version 3.4.2 of the R software, as described by Gallardo et al. (2017). The abundance of each taxa was normalized by the total diversity per sample prior to any group comparison. For the taxonomic analysis, the abundance of each taxa was normalized by the total diversity per sample prior to any group comparison.

## RESULTS

#### Patient Demographic

During the study, 25 patients were recruited, 15 in the BM and 10 in the BF group. There were no differences between gender or birth weight and lengths. The number of stools/day was lower in the BF group (2.5 vs. 4.4), but this difference was not statistically significant. All the patients had the same average number of relatives in the home. Regarding pets and allergies, all the families had pets and 50% of the patients in both groups had a history of first-grade atopy (atopic dermatitis or asthma or allergic rhinitis or food allergy). Finally, family income was similar in both groups. **Table 1** summarizes these findings.

## BF-Fed Infants Showed Higher Intestinal Inflammation Than BM-Fed Infants

We quantified the levels of inflammatory markers calprotectin, IL-8 and α1-antitrypsin in stool samples by ELISA. In the BM group, we found calprotectin and α1-antitrypsin levels were significantly higher at T3 than at T1. No differences were found between T1 and T3 in the BF group (**Figures 1A,B**). A comparison of the BM and BF groups showed that calprotectin levels at T1 were lower in the BM group; this difference was not observed at T3 (**Figure 1A**). For IL-8 levels, we found no difference between groups (**Figure 1C**).

Gene expression analysis of IL8 and IL1β genes showed that infants from the BF group have a significantly increased expression of these markers (2.3 ± 0.45 and 2.3 ± 0.4-folds, respectively) at T1 compared to the BM group. At T3, no difference was found in the expression of these genes (**Figure 2**).

#### BF-Fed Infants Harbor a Different Intestinal Microbiota Than BM-Fed Infants

The taxonomic analysis showed a total number of 140 OTUs, 28 exclusively present in the BM group and 23 in the BF group. Shared OTUs among groups was only 2 at T1, but at T3 the number of shared OTUs increased to 48. When comparing the



ns, not significant.

microbiota at T1, BM-fed newborns had 22 exclusive OTUs compared to 6 in the BF group (**Figure 3**). At phylum level, we found the BM group had a lower Firmicutes proportion (20 and 23%) than the BF group (38 and 42%) at T1 and T3. By contrast, a higher Bacteroides presence at T1 and T3 was found in the BF group (38 and 47%) than in the BM group (16 and 25%), respectively. The Firmicutes/Bacteroides ratio was lower in the BM group than the BF group at T1 (0.5 vs. 2.4) and at T3 (0.5 vs. 1.7). The proportions of Proteobacteria and Actinobacteria were similar among all groups. At genus level, the most abundant genera in all groups were Escherichia/Shigella and Bacteroides, followed by Parabacteroides, Lechnospiracea, and Veillonela (**Figure 4**).

A comparison of OTUs at genus level showed that the BF group had a higher representation of Enterococcus (p = 0.001), Streptococcus (p = 0.001), Enterobacter (p = 0.01), Lactococcus (p = 0.03) and Propionibacterium (p = 0.04**)** than the BM group at T1. At T3, the BM group had a higher representation of Sutterella (p = 0.04) and Parabacteroides (p = 0.04), whereas BF-fed infants had higher number of Streptococus (p = 0.01).

A RDA analysis of OTU composition showed a significant difference between BM and BF groups at T1 (**Figure 5A**), but not at T3 (**Figure 5B**).

#### DISCUSSION

Breastfeeding confers important benefits on the infant and protection from many diseases, most of them associated with changes in the intestinal tract environment. Few studies have endeavored to address the effect of diet and intestinal inflammation on the newborn. Here, we have shown in a cohort study that BF-fed infants have a higher intestinal inflammation defined by an increased concentration of calprotectin and α1-antitrypsin in stool samples taken 1 month after birth (T1) compared to an infant fed exclusively with BM. Even though these differences were not observed 2 months later (T3), our data support the role of BM in the low-grade inflammation compared to BF-fed infants 1 month after birth. Calprotectin and α1-antitrypsin are markers that specifically express protein loss and inflammation as seen in several gastrointestinal disorders, such as allergies and inflammatory bowel diseases (Poullis et al., 2002; Saarinen et al., 2002). Previous reports have compared the calprotectin levels in stools in healthy infants, BM vs., BF (median age 51 days old). Interestingly, the stool calprotectin level was higher in the BM group than in the BF group, suggesting a possible degree of local inflammation in the intestine in the BM infants (Savino et al., 2010). In another study, no differences were found in the stool calprotectin level between BM and BF newborns at 3 months old (Rosti et al., 2011). Similar to our findings but with a different approach, Kainonen et al., using a cohort of infants fed with BM and BF at high risk for the development of allergies, compared the levels of INF-γ, TNF-α, and IL-2 (proinflammatory), IL-5 and IL-4 (allergy) and IL-10 and TGF-β2 (anti-inflammatory). The BM group showed significantly lower proinflammatory markers in serum compared to the BF group, and the TGF-β2 levels in the BF group were significantly lower than in the BM group. These findings lasted up to 1 year, despite supplementation with solid food in both groups. Finally, they suggested BM had an immunomodulatory role "protecting" against inflammation (Kainonen et al., 2012). Possible mechanisms in breast milk dampening inflammation might involve its components, including immunoglobulins, cytokines such as IL-10, defensing, macrophage colony stimulating factors secreted

by mammary epithelial cells and TGFβ produced by leukocytes present in the milk (Hennet and Borsig, 2016). In light of our results, the contribution of calprotectin and α1-antitrypsin in the gut homeostasis in infants merits further investigation.

We also quantified IL-8 in stools, but no significant differences were found between the groups. Although IL-8 is a pivotal molecule that orchestrates tissue inflammation, its role in intestinal inflammation in healthy children is not well characterized. In order to clarify the involvement of this cytokine, we decided to evaluate the expression of the IL8 gene and another pro-inflammatory gene, IL-1β, in stool samples. We found a significantly increased expression of both genes in the BF group compared to the BM group at T1 but not at T3, suggesting that both genes might have a role in BM in ameliorating inflammation in the first month.

Microbiota has emerged as an important environmental factor in the inflammation of several gut diseases in children (Lu and Ni, 2015). In healthy infants, microbiota might play an important role in gut homeostasis. Our data suggest that the gut microbiota of the BM group clearly differs from the BF infants at T1 (**Figure 5**). BF harbors more Firmicutes and fewer Bacteroidetes, exhibiting a higher Firmicutes/Bacteroidetes (F/B) ratio than the BM group at T1 and T3, which is in line with previous data in healthy newborns (Mariat et al., 2009). A similar increase in Firmicutes has been seen in babies initially born by cesarean section (Hill et al., 2017). Interestingly, obesity is associated with an abundance of Firmicutes and a depletion of Bacteroidetes, where the interrelation of short-chain fatty acids fermented by bacteria plus the lipopolysaccharide from gram negative bacteria will induce inflammation and obesity (Chakraborti, 2015; Koliada et al., 2017). At genus level, we found that the BF group had a significantly higher amount of Enterococcus, Enterobacter and also Streptococcus than the BM group at T1, and all of these had been shown to be responsible for sepsis in early neonates as well in animal models receiving BF (Nakayama et al., 2003; Simonsen et al., 2014). These observations might be associated with a difference in calprotectin levels found in the BM group at this timepoint. Interestingly, in the BM group the calprotectin and α1-antitrypsin levels were higher at T3 than at T1, and these genera were found to be more abundant at T3 than at T1, suggesting that species belonging to these genera might be linked to gut inflammation. Our data support the idea that diet induces microbial changes that can induce inflammation and that BM reduces the inflammation burden, modulating the microbiota and thus maintaining intestinal homeostasis.

Our study has limitations. The number of patients included in the study was small, a situation that might be explained by the number of mothers who currently breastfeed their child exclusively in the first months of life. In the gut microbiota analysis, we could not identify any Bifidobacterium known to be present in newborn samples, despite other studies having found differences in their abundance between BF and BM (Penders et al., 2006; Hascoët et al., 2011). These results may possibly be attributed to the sequencing platform used (454 pyrosequencing). New sequencing platforms will make it possible to overcome this issue in future projects. Another limitation was the number of cytokines and genes evaluated. Although calprotectin, IL-8 and α1-antitrypsin are key markers of intestinal inflammation, there are several markers that could be explored. The development of new multiplex analyte platforms validated for stool samples will provide a broader picture of the molecules involved in intestinal homeostasis in infants.

In conclusion, using non-invasive methods in stools we found a basal state of inflammation baseline in the infant's intestine based on inflammation markers. At 1 month after birth, infants receiving BF exhibited higher levels of inflammation than BM in terms of changes in the microbiota composition. These results

FIGURE 4 | Community profile at different taxonomic levels. Relative abundance of taxa at Phylum (A) and Genus (B) level of the 10 most abundant taxa. Each color represents a different taxonomic unit. Less representative taxa were grouped as "other".

FIGURE 5 | Redundancy analysis at timepoints T1 and T3. A redundancy analysis (RDA) was conducted using sample classification as the explanatory matrix and relative OTU diversity as the response matrix at (A) T1, 1 month and (B) T3, 3 months. Data was normalized with a double square root transformation. Sample grouping and axis significance were analyzed by ANOVA (RDA T1 p = 0.003, RDA T3 p = 0.238).

might signify that BM has a protective role in ameliorating inflammation, modulating the intestinal microbiota during the first months of life.

#### AUTHOR CONTRIBUTIONS

JO participated in the study design, data acquisition, data interpretation, manuscript writing and final approval of the manuscript. DY participated in the sample collection and analysis. RV participated in the cohort enrollment and sample

### REFERENCES


collection. PG participated in the microbiota data analysis and manuscript writing. YL participated in the study design and data interpretation. MF participated in the study design, data acquisition, data interpretation, manuscript writing and final approval of the manuscript.

#### ACKNOWLEDGMENTS

This work was supported by FONDECYT grants 11130374 to JO and 1160426 to MF.


**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 Ossa, Yáñez, Valenzuela, Gallardo, Lucero and Farfán. 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: Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis

Gemma Agustí <sup>1</sup> and Francesc Codony <sup>2</sup> \*

<sup>1</sup> Departament d'Òptica i Optometria, Universitat Politècnica de Catalunya-Barcelona Tech, Terrassa, Spain, <sup>2</sup> Laboratori Municipal - Aigües de Mataró, Mataró, Spain

Keywords: viability PCR, live-dead distinction, analysis bias, microbiome, PMA

#### **A commentary on**

#### **Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis**

by Young, G. R., Smith, D. L., Embleton, N. D., Berrington, J. E., Schwalbe, E. C., Cummings, S. P., et al. (2017). Front. Cell. Infect. Microbiol. 7:237. doi: 10.3389/fcimb.2017.00237

The recent work by Young et al. (2017) demonstrates the importance of obtaining accurate data when analyzing the microbial community in real-life complex samples. These authors showed that an analysis of gut microbiota samples from preterm infants who at risk of necrotizing enterocolitis (NEC) and sepsis could be biased if the analysis includes DNA from dead cells. The study used a viability PCR (vPCR) approach to overcome this obstacle. Notably, vPCR uses specific photoreactive dye that cannot cross intact cell membranes. When a sample is incubated with this dye and exposed to light, the dye irreversibly binds to DNA in dead cells that have damaged cell membranes. This step effectively neutralizes the DNA and prevents it from being detected by PCR.

Regarding the importance of removing DNA from dead cells during microbial analysis, similar conclusions have been drawn by studies of, for example, environmental samples (Carini et al., 2016) and clinical specimens (Rogers et al., 2013). We agree with this premise and encourage others to use vPCR in microbial analysis, but caution that proper sample preparation is required for this technique.

vPCR methodology is theoretically simple to use since it requires only a reagent and a light source; however, as with many techniques, a lack of practical experience with the technique led the authors to overlook some important considerations. vPCR is still relatively new, with the first paper by Nogva et al. (2003). The scientific community has developed strategies and general rules for optimizing the vPCR workflow; some of these strategies were reviewed by Fittipaldi et al. (2012). More recent work has demonstrated that during sample manipulation, plastic can contribute to false positive results. Indeed, the walls of microtubes can retain free DNA, thereby preventing the nucleic acid from interacting with the reagent. Accordingly, the DNA fraction on the walls of the microtubes is not neutralized, and when it is detected, it is incorrectly assigned to the live DNA fraction (Agustí et al., 2017). This is only one example of how technical issues during early sample treatment steps can affect the conclusions of the analysis. For this reason, it is our opinion that despite the valuable conclusions, some microbiome analyses based on vPCR have hidden weaknesses.

In the paper by Young et al. (2017), the authors analyzed stool samples that were suspended in PBS, but the pH and the oxygen levels of standard PBS solutions were not optimal for

#### Edited by:

Pascale Alard, University of Louisville, United States

#### Reviewed by:

Morgan Langille, Dalhousie University, Canada

#### \*Correspondence:

Francesc Codony francesc.codony@aiguesmataro.cat

> Received: 22 June 2017 Accepted: 05 June 2018 Published: 20 June 2018

#### Citation:

Agustí G and Codony F (2018) Commentary: Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis. Front. Cell. Infect. Microbiol. 8:212. doi: 10.3389/fcimb.2018.00212 ensuring the viability of the cells in the stool samples. Therefore, the sample handling steps themselves could have affected the viability of some cells in the specimen. For example, only half of the anaerobic microorganisms from the mammalian large bowel survive exposure to O<sup>2</sup> for 5 min, and this percentage decreases to 3–5% after 20 min of exposure (Brusa et al., 1989). This effect is quite predictable and easy to understand; however, other important aspects, such as centrifugation (Peterson et al., 2012), the time between sample collection and storage **(**Cuthbertson et al., 2014), and freeze/thawing (Cuthbertson, 2015) can also lead to bias in culture-independent analysis.

The authors evaluated the effect of freezing the samples and concluded that freeze/thawing did not change the results for the stool samples. However, factors other than storage were not

#### REFERENCES


considered, so it remains unclear whether their conclusions were influenced by additional biases.

The work of Young et al. is valuable, and the conclusions very clearly show the importance of overcoming bias due to DNA from dead cells in a microbial diversity analysis. We suggest that an additional warning about the importance of appropriate sample handling would be helpful to readers who may wish to conduct similar analyses.

#### AUTHOR CONTRIBUTIONS

Both authors reviewed the original work and wrote this commentary.


**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 Agustí and Codony. 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: Bacteriophage transfer during faecal microbiota transplantation in Clostridium difficile infection is associated with treatment outcome

#### Blessing O. Anonye1,2 \*

<sup>1</sup> Microbiology and Infection Unit, Division of Biomedical Sciences, Warwick Medical School, Coventry, United Kingdom, <sup>2</sup> Warwick Integrative Synthetic Biology Centre, School of Life Sciences, University of Warwick, Coventry, United Kingdom

Keywords: bacteriophages, gut microbiota, fecal microbiota transplantation, C. difficile infection, Caudovirales

#### **A commentary on**

#### **Bacteriophage transfer during faecal microbiota transplantation in Clostridium difficile infection is associated with treatment outcome**

by Zuo T., Wong S. H., Lam K., Lui R., Cheung K., Tang W., et al. (2017). Gut 67, 634–643. doi: 10.1136/gutjnl-2017-313952

#### Edited by:

Till Strowig, Helmholtz-Zentrum für Infektionsforschung, Germany

#### Reviewed by:

Joseph Sorg, Texas A&M University, United States Xingmin Sun, University of South Florida, United States V. K. Viswanathan, University of Arizona, United States

#### \*Correspondence:

Blessing O. Anonye b.anonye@warwick.ac.uk

Received: 25 October 2017 Accepted: 19 March 2018 Published: 04 April 2018

#### Citation:

Anonye BO (2018) Commentary: Bacteriophage transfer during faecal microbiota transplantation in Clostridium difficile infection is associated with treatment outcome. Front. Cell. Infect. Microbiol. 8:104. doi: 10.3389/fcimb.2018.00104 Fecal microbiota transplantation (FMT) has been used as a treatmentof last resort for recurrent C. difficile infections (CDI) with a cure rate of 85–90% after the first FMT (van Nood et al., 2013; Jiang et al., 2017). Several studies have examined the changes that occur in the bacterial community that leads to reestablishment of the intestinal microbiota (Fuentes et al., 2014; Seekatz et al., 2014; Staley et al., 2016). However, studies that have investigated the role of viruses in FMT are limited (Broecker et al., 2016a,b, 2017; Ott et al., 2017).

Recently, Zuo and colleagues performed metagenomics sequencing of virus like particles on fecal samples from patients with CDI and healthy controls to determine if bacteriophages were associated with restoration of the intestinal microbiota after FMT (Zuo et al., 2017). Prior to FMT, the patients had increased abundance of Caudovirales with decreased diversity, richness and evenness when compared to healthy controls. Longitudinal studies of patients who received FMT (n = 9) and standard therapy, vancomycin (n = 5) demonstrated that FMT led to the transfer of viruses from the donor to the recipients (Zuo et al., 2017).

The patients that were administered FMT were divided into two groups of "responders" and "non-responders" based on whether they were cured of CDI or recurred after FMT. In particular, after FMT, they noticed a significant decrease in the abundance of Caudovirales and increase in richness of donor-derived Caudovirales in the enteric virome of the patients (Zuo et al., 2017). There was a correlation between donor viral richness and the patients responding to FMT. Of the two-thirds that were cured after FMT, four of the donors had higher Caudovirales richness when compared to the non-responders group where the donor Caudovirales richness was lower (Zuo et al., 2017). When compared to the non-responders, the remaining two donors had a similar (n = 1 for non-responder donor) or slightly higher Caudovirale richness.

Furthermore, lower abundance of the family, Microviridae was observed in the patients before FMT when compared to the controls but this increased after FMT. Fifteen viral species were found to be enriched between FMT responders and non-responders. Viral species belonging to the Microviridae family such as Eel River Basin pequenovirus was the most abundant in the responders (Zuo et al., 2017).

Moreover, Zuo et al. performed 16S rRNA gene sequencing to determine changes in the bacterial community and noted increase in Lachnospiraceae and Ruminococcaceae families (Zuo et al., 2017). However, there was no significant difference between donor transferred bacteria between FMT responders and non-responders. Interestingly, vancomycin treatment had no significant effect on the viral community (Caudovirales) of those who responded to the antibiotic therapy. However, the bacterial community was significantly affected (Zuo et al., 2017).

Broecker et al. investigated the long term bacterial and virome changes in a patient after FMT for recurrent CDI, and found the virome at several months post-FMT related to the donor virome (Broecker et al., 2016a,b). Similarly, Ott and colleagues recently demonstrated that sterile fecal filtrates from donor feces was effective in treating recurrent CDI in five patients (Ott et al., 2017). They showed that the "phagebiota" of a recipient at 6 weeks post-FMT was similar to the fecal filtrate from the donor (Ott et al., 2017). These findings indicate that apart from live bacteria, other components of the microbiota such as bacteriophages, antimicrobial compounds or metabolites contribute to reestablishment of the intestinal microbiota in FMT.

There is no doubt that bacteriophages play a role in the intestinal microbiota with a potential to alter the composition and function of the host microbiota. The question is what constitutes a healthy gut phageome and how do they influence the human gut microbiota? Most of the phages in healthy human gut microbiota belong to the Caudovirales order and from the family Microviridae which have double and single stranded DNA respectively (Kim et al., 2011; Manrique et al., 2016). As seen above in recurrent CDI, increased diversity, richness and evenness of the Caudovirales was implicated in the efficacy of FMT. However, in other intestinal diseases, it is not clear cut as to the role of Caudovirales likely due to other risk factors involved (Norman et al., 2015). For example, Caudovirales richness was observed in patients with inflammatory bowel disease (Norman et al., 2015).

#### REFERENCES


The next question one may be tempted to ask is, how stable is the "phagebiota" in individuals and what are the effects of antibiotic perturbations on the phageome? Research by Ly et al. showed that transmission of viruses was common between members of the same household over a 6-month period (Ly et al., 2016). Treatment of healthy individuals living in a particular household with the antibiotics, amoxicillin or azithromycin for 7 days did not affect the composition of viruses in the intestinal microbiota (Ly et al., 2016).

These studies have highlighted the underappreciated role viruses play in addition to bacterial colonization in the intestinal microbiota. However, as individual phages are specific in action toward their bacterial host, it would be advantageous to isolate phages that target pathogenic bacteria such as C. difficile, though this is not trivial. Indeed, previous work revealed that using a single phage, 8CD27 (Meader et al., 2010, 2013) or a combination of phages led to the inhibition of C. difficile growth in vitro and in vivo (Nale et al., 2016a,b). Recently, a combination of four phages was found to totally inhibit C. difficile growth in a batch fermentation model spiked with feces from four healthy volunteers (Nale et al., 2018).

Much work remains to be done on phage therapy for C. difficile infection. The ability to develop a synthetic mixture of phages as treatment for infectious diseases, will go a long way in this era of antibiotic resistance. Not only will this be beneficial in severe CDI, but could also be useful as a therapy for other diseases related to the intestinal microbiota.

#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

#### FUNDING

BA received a small Warwick Integrative Synthetic Biology (WISB) grant. WISB is a BBSRC/EPSRC Synthetic Biology Research Centre (grant ref: BB/M017982/1) funded under the UK Research Councils' Synthetic Biology for Growth programme.

transplant in recurrent Clostridium difficile infection. ISME J. 8, 1621–1633. doi: 10.1038/ismej.2014.13


**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 © 2018 Anonye. 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.

# Mycotoxin: Its Impact on Gut Health and Microbiota

#### Winnie-Pui-Pui Liew and Sabran Mohd-Redzwan\*

Department of Nutrition and Dietetics, Faculty of Medicine and Health Sciences, Universiti Putra Malaysia, Serdang, Malaysia

The secondary metabolites produced by fungi known as mycotoxins, are capable of causing mycotoxicosis (diseases and death) in human and animals. Contamination of feedstuffs as well as food commodities by fungi occurs frequently in a natural manner and is accompanied by the presence of mycotoxins. The occurrence of mycotoxins' contamination is further stimulated by the on-going global warming as reflected in some findings. This review comprehensively discussed the role of mycotoxins (trichothecenes, zearalenone, fumonisins, ochratoxins, and aflatoxins) toward gut health and gut microbiota. Certainly, mycotoxins cause perturbation in the gut, particularly in the intestinal epithelial. Recent insights have generated an entirely new perspective where there is a bi-directional relationship exists between mycotoxins and gut microbiota, thus suggesting that our gut microbiota might be involved in the development of mycotoxicosis. The bacteria–xenobiotic interplay for the host is highlighted in this review article. It is now well established that a healthy gut microbiota is largely responsible for the overall health of the host. Findings revealed that the gut microbiota is capable of eliminating mycotoxin from the host naturally, provided that the host is healthy with a balance gut microbiota. Moreover, mycotoxins have been demonstrated for modulation of gut microbiota composition, and such alteration in gut microbiota can be observed up to species level in some of the studies. Most, if not all, of the reported effects of mycotoxins, are negative in terms of intestinal health, where beneficial bacteria are eliminated accompanied by an increase of the gut pathogen. The interactions between gut microbiota and mycotoxins have a significant role in the development of mycotoxicosis, particularly hepatocellular carcinoma. Such knowledge potentially drives the development of novel and innovative strategies for the prevention and therapy of mycotoxin contamination and mycotoxicosis.

Keywords: mycotoxicosis, intestine, hepatocellular carcinoma, trichothecene, zearalenone, fumonisin, ochratoxin, aflatoxin

## BACKGROUND

The momentum of scientific paper publication toward mycotoxin is an increasing trend where 16,821 papers were recorded in Scopus since the first mycotoxin, aflatoxin (AF) was identified in the year 1965. Data clearly showed the significance of mycotoxin research which will be further discussed later in this review paper. Nevertheless, the global health issue arose from mycotoxin is still frequently ignored in many low-income countries, where mycotoxins affect staple foods (Wild and Gong, 2010). The exposure is long-term and often at high

#### Edited by:

Venkatakrishna Rao Jala, University of Louisville, United States

#### Reviewed by:

Gabriela Del Valle Perdigon, Consejo Nacional de Ciencia y Tecnología - CONICET, Argentina Alinne Castro, Universidade Católica Dom Bosco, Brazil

#### \*Correspondence:

Sabran Mohd-Redzwan mohdredzwan@upm.edu.my; mohd.redzwan.sabran@gmail.com

> Received: 10 October 2017 Accepted: 12 February 2018 Published: 26 February 2018

#### Citation:

Liew W-P-P and Mohd-Redzwan S (2018) Mycotoxin: Its Impact on Gut Health and Microbiota. Front. Cell. Infect. Microbiol. 8:60. doi: 10.3389/fcimb.2018.00060 doses, regretfully these particular regions are the least regulated in terms of agricultural practices and human exposure. The attention only has been paid in the richer nations of the world, to meet stringent import regulations on mycotoxin contamination (Battilani et al., 2016). To date, the world still desires for a more accurate evidence-based on mycotoxins and human health, as well as a better biomarker of exposure and data from studies of disease distribution. Current data are valid to justify and respond to reduce exposure in vulnerable populations (Freire and da Rocha, 2017). The implementation of more practical and affordable mycotoxin removal techniques at the household level to effectively reduce exposure are becoming increasingly important. When mycotoxins are introduced into the organism from food, they first come to interact with the gastrointestinal (GI) tract (Assunção et al., 2016). The GI tract is where the gut microbiota resides: it is known for its role in modulating the immune system and digestive processes. Gut microbiota work in concert with the GI tract protects the host from the toxicity of mycotoxins. Accordingly, integration of microbialbased approaches through maintaining a healthy gut microbiota is highly demanded.

## MYCOTOXINS

Mycotoxins, the low molecular mass (MW ∼700 Da) secondary metabolites mainly produced by Aspergillus, Penicillium, and Fusarium are highly noxious substances on animals and humans. However, not all mycotoxin are classified as such, for example, Penicillin, is widely used an antibiotic (Speight, 2012). The structural form of mycotoxins varies from simple four C compounds, e.g., moniliformin, to complex substances such as the phomopsins (Zain, 2011). Fungal proliferation and production of mycotoxins rise naturally due to environmental factors, especially during tropical conditions (Mohd-Redzwan et al., 2013). Besides, the downstream processing such as poor harvesting practices, improper storage and less than optimal conditions during transportation, processing, and marketing can also contribute to the growth of fungi and increase the risk of the major food spoilage agent caused by mycotoxin production (Khazaeli et al., 2014). Due to their ubiquitous nature of fungi, mycotoxins have been increasingly attracting the concern of health organizations where their occurrence in foods cannot be ignored and already poses risk to consumers (Jahanian, 2016).

#### IMPORTANCE OF RESEARCH ON MYCOTOXIN

#### Occurrence of Mycotoxicosis

Notably, it has been estimated that 25% of the world's crops such as nuts, cereals, and rice are contaminated by mold and fungal growth, as reviewed by the United Nations Food and Agriculture Organization and the World Health Organization (Pandya and Arade, 2016). The toxic effect of mycotoxins on animal and human health is referred to as mycotoxicosis. Exposure to mycotoxins is mostly by ingestion but also occurs by the dermal and inhalation routes. The extent of adverse effects of mycotoxins on human or animals health mainly depends on the extent of exposure (dosage and period), type of mycotoxins, physiological and nutritional status as well as possible synergistic effects of other chemicals to which the animals or humans are exposed (Gajecka et al., 2013). In 1960, interest on mycotoxins was initiated by the occurrence of Turkey X disease caused by AF, which killed more than 100,000 turkeys. Subsequently, it was found that AFs are carcinogenic and cause hepatocellular carcinoma (HCC) in animals and humans, and this has stimulated research on mycotoxins (Peraica et al., 1999). Since then, around 400 mycotoxins are known, but AFs, ochratoxins, zearalenone (ZEA), fumonisins (FBs) and trichothecenes are mostly focused on public health issues (Ates et al., 2013). Mycotoxin exposure is not only limited to pure mycotoxins but also masked mycotoxin which formed when plants protect themselves by conjugating mycotoxins to biopolymers. In addition, some people are more susceptible to getting mycotoxicosis than others, and this is due to the pharmacogenetic variability where specific gene mutations such as cytochrome p450 (CYP 450) genes could either increase or decrease the metabolic activity (cytotoxicity) of the challenging mycotoxins (Sun et al., 2016). For instance, in both in vivo (Muhammad et al., 2017) and in vitro (Lewis et al., 2000) studies, CYPs' 1A2 and 3A4 appear as the most important enzymes that increased metabolism of AFB1 to its active form, AFB1-8,9 epoxide and subsequently to AFB1-DNA adduct formation, in which the biomarker has been linked to the development of liver cancer (Ceccaroli et al., 2015).

Chronic mycotoxicosis causes a greater impact on human health. Mycotoxin can induce diverse and powerful toxic effects in test systems: some are carcinogenic, mutagenic, teratogenic, estrogenic, hemorrhagic, immunotoxic, nephrotoxic, hepatotoxic, dermatoxic and neurotoxic (Milicevi ´ c et al., ´ 2010). Frequently, mycotoxicosis remains unrecognized by medical professionals. Mycotoxicosis can be weighed when a disease appears in several persons, with no obvious connection to a known etiological agent, such as microorganisms (Viegas et al., 2015).

#### Future Prospect: Impact of Growing Population and Ongoing Climate Change on Mycotoxin

By the year 2030, the world's population is estimated to reach 8.2 billion people, and with 842 million people estimated as having been undernourished in the period of the year 2011– 2013, food supply will definitely present a growing challenge in the next decades (FAO, 2014). This scenario will, in turn, have a tremendous negative impact on food supply (FAO, 2014). It is worth to note that the presence of hazardous substances (e.g., mycotoxins) also limits or reduces the marketability of food products in international markets (Anater et al., 2016).

There is now widespread consensus that the earth is warming at an unprecedented rate (Medina et al., 2015). The geographic distribution and production of the crop, as well as the phyllosphere microflora of crops, are expected to be strongly affected by climate change. For instance, mycotoxigenic Aspergillus flavus are able to grow under high temperatures and drought conditions. The resilient growth of A. flavus under extreme heat and dry condition is an expected and emerging dilemma mainly in the Mediterranean and other temperate regions (Logrieco et al., 2003). For example, the impacts of climate change have been observed in Serbia, where no contamination occurred previously, but prolonged hot and dry weather in the year 2012 resulted in 69% of maizes contaminated with AFs (Medina et al., 2015). A similar case also found in Hungary, where the increase in AFs contamination may be due to climate change conditions (Dobolyi et al., 2013).

The world's largest agri-food exporters include countries such as Brazil and Argentina and parts of Asia including China and India are identified as hot spots for impacts of climate change (Ray et al., 2012). Thus, from a food security perspective, a more accurate prediction of impacts of climate change on mycotoxins need to be addressed to prevent compromised food sustainability which possibly resulting in negative social consequences.

## GASTROINTESTINAL TRACT

The GI tract is an organ within humans and other animals which responsible for food ingestion, digestion, energy and nutrients absorption, immune response, as well as elimination of waste products (feces) (Celi et al., 2017). The architecture of the GI tract is intended to facilitate these functions. The basic feature of GI tract is a muscular tube lined by a mucous membrane and comprised four layers forming a continuous passage. All segments of the GI tract are divided into four layers: mucosa, submucosa, muscularis propria, and serosa (Jaladanki and Wang, 2016). The mucosa is made up of three layers (epithelium, lamina propria, and muscular mucosae). The entire mucosa rests on the submucosa, beneath which is the muscularis propria. The outermost layer is named as the serosa. The complex infolding at mucosa layer forms an immense surface area for the most efficient nutrient absorption. The submucosa contains arteries, veins, inflammatory cells, lymphatics, and autonomic nerves. The muscularis mucosa is a thin layer of smooth muscle that forms the basis of peristalsis. While, the serosa is made of connective tissue that contains blood vessels, nerves, and fat (Jaladanki and Wang, 2016).

The epithelium layer at the innermost of mucosa is of vital importance for intestinal barrier function. The intestinal epithelium is one layer of thin cells lining the gut lumen. The epithelial contains enterocytes, enteroendocrine, and goblet cells at villi, whereas the Paneth cells, located under the crypts (Fink and Koo, 2016). It acts as a barrier to block the entry of harmful agents such as pathogens, toxins, and foreign antigens. Besides, it is also an important site for nutrient absorption including electrolytes, dietary nutrients, and water via its selective permeable membrane (Constantinescu and Chou, 2016). Each intestinal epithelial cell is connected by desmosomes, tight junctions (TJs), and adherens junctions (AJs). The AJs and desmosomes are responsible for the mechanical linkage of adjacent cells. Whereas, the TJs control the intercellular space and regulate selective paracellular ionic solute transport (Capaldo et al., 2014). Above the epithelium lies a complex microflora which is recognized as gut microbiota and the role of gut microbiota will be discussed later in this review article. The selective permeable barrier of mucosal epithelium establishes the interplay between the intestinal immune system and the luminal contents.

## Mycotoxins and Gut Health

Upon ingestion of contaminated food or feed, the GI tract is particularly affected by mycotoxin. Generally, intestinal barrier in the GI tract functions as a filter against harmful mycotoxins. However, some mycotoxins have been found to exert their detrimental effects in the GI tract. For example, mycotoxins can alter the normal intestinal functions such as barrier function and nutrient absorption. Some mycotoxins also affect the histomorphology of intestine. The impacts of mycotoxins include trichothecenes, zearalenone, fumonisins, ochratoxins, and AFs on general and gut health will be comprehensively reviewed.

#### Trichothecenes

Fusarium graminearum is the main fungi species that produces tricothecenes. All tricothecenes contain an epoxide at the C12, C13 positions, which is responsible for their toxicological activity (Nathanail et al., 2015). T-2 toxin (Type A) and DON (Type B) are the major mycotoxins that cause toxicity to humans and animals via oral ingestion (Nathanail et al., 2015).

During World War II, a biological weapon caused an acute syndrome consists of cough, sore throat, dyspnea, bloody nasal discharge, and fever was reported by Soviet scientists (Pitt and Miller, 2016). Twenty years later, T-2 mycotoxin was discovered when civilians consumed wheat that was unintentionally contaminated with Fusarium fungi (Pitt and Miller, 2016). A human toxicosis due to ingestion of moldy rice contaminated with T-2 toxin has been reported in China. According to Wang Z. et al. (1993), 65% of patients developed food poisoning symptoms such as chills, nausea, abdominal distension, dizziness, vomiting, thoracic stuffiness, abdominal pain, and diarrhea. Similar to T-2 toxicity, victims of DON outbreak suffered from vomiting syndromes (Etzel, 2014). Several outbreaks of acute DON toxicity in human have been reported in India, China, and the USA (Etzel, 2014).

Trichothecenes toxic effects in animals (dairy cattle, swines, broilers, and rats) include decreased plasma glucose, reduced blood cell and leukocyte count, weight loss, alimentary toxic aleukia, as well as pathological changes in the liver and stomach (Adhikari et al., 2017). The mechanism involved in T-2 and DON toxicity is generally via oxidative stress-mediated deoxyribonucleic acid (DNA) damage and apoptosis (Wu et al., 2014). Furthermore, T-2 and DON are well-known inhibitors of protein synthesis resulting from the binding of peptidyltransferase, which is located in the 60s ribosomal subunit (Yang et al., 2017).

In the GI tract, a decreased absorption of glucose was observed following T-2 and DON intoxication resulted from suppressed SGLT1 (glucose transporter) mRNA expression. Apart from the glucose absorption, SGLT1 also responsible for water reabsorption, thus reduction of SGLT1 transporter induces diarrhea as well (Grenier and Applegate, 2013).

The weight loss effect of trichothecenes involved neuroendocrine factors and cytokines. DON and T-2 elevated concentrations of the indoleamines, serotonin and 5-hydroxy-3-indoleacetic acid (HIAA) in all brain regions (Wang J. et al., 1993). These neuroendocrine factors can affect the secretion of both anorexigenic and/or orexigenic hormones (Maresca, 2013). Through increasing gene expression of anorexia-inducing proinflammatory cytokines such as interleukin-1β (IL-1β), interleukin-6 (IL-6) and tumor necrosis factor-α (TNF-α), trichothecenes exacerbate the condition of anorexia (Wu et al., 2015). In addition, DON and T-2 also induced the release of the satiety hormones, peptide YY (PYY) and cholecystokinin (CCK), which are critical mediators of anorexia (Wu et al., 2015).

Using animal models, trichothecenes was found to induce necrotic lesions in the GI tract (Kolf-Clauw et al., 2013). A shortening of villi height was also observed in trichothecenestreated animals (swine, poultry, and rat model). The changes on villi were due to activation of the apoptotic pathway by trichothecenes, which in turn leads to nutrition malabsorption (Alizadeh et al., 2015). Furthermore, results obtained from in vivo and in vitro studies showed that trichothecenes increased intestinal permeability. Using porcine epithelial cell, trichothecenes increased the intestinal permeability by lowering tight junction proteins expression (Osselaere et al., 2013). In addition, previous studies revealed a significant (P < 0.05) decreased in the number of goblet cells that secrete mucin in trichothecenes-treated animals. Mucin is primarily involved in the gut barrier function (Pinton and Oswald, 2014). The disruption in the integrity of intestinal epithelium allows the entry of the pathogen into the gut lumen (Lessard et al., 2015). Besides, trichothecenes have been linked with a decreased level of IL-8 in the intestine, which is responsible for pathogen removal (Kadota et al., 2013). Overall, trichothecenes exert negative impacts on GI tracts specifically on the gut absorption, integrity, and immunity.

#### Zearalenone

Zearalenone (ZEA) is a mycotoxin that primarily produced by Fusarium graminearum and Fusarium culmorum in foods and feeds. The high rate of co-occurrence of ZEA with FBs and DON indicates that these mycotoxins might be involved in a wide range of synergistic and additive interactions. ZEA has been linked to scabby grain toxicosis occurred in Japan, China, Australia, and the USA, with symptoms including nausea, vomiting, and diarrhea (Liao et al., 2009).

It is well recognized that ZEA is a non-steroidal estrogenic mycotoxin that is implicated in the reproductive disorders of farm animals (swines, cattle, and sheep) and hyperoestrogenic syndromes in humans (Kotowicz et al., 2014). Toxicological studies of ZEA revealed its effects on the reproductive system, including enlarged uterus, altered reproductive tract, decreased fertility, as well as abnormal level of progesterone and estradiol. Besides, the ingestion of ZEA during pregnancy reduced fetal weight and survival rate of embryo (Zhang et al., 2014). This phenomenon can be explained through the structure of ZEA. ZEA has a structure which allows it to bind to the mammalian estrogen receptor, although with lower affinity compared to the natural-occurring estrogens (Hueza et al., 2014). Besides, ZEA has also been shown to be hepatotoxic, haematotoxic, immunotoxic and genotoxic (Zhou et al., 2017).

Although the reproductive organ is the main target for ZEA to induce toxicity, the adverse effects of ZEA on GI tracts have been reported. The effects of ZEA ingestion on the GI tract are not as detrimental compared to the other mycotoxins. Studies using intestinal epithelial cells showed that ZEA induced cell death without altering the cell integrity as indicated by transepithelial electrical resistance (Marin et al., 2015). In contrast, it was discovered that the metabolites of ZEA (α- and β-zearalenol) significantly (P < 0.05) decreased the cell integrity. The study showed that ZEA and its metabolites acted differently in the gut (Marin et al., 2015). Abassi et al. (2016) demonstrated that ZEA enhanced cell proliferation, increased colony formation and fastened cell migration of colon carcinoma cell line HCT116. Another study also showed that ZEA downregulated the expression of tumor-suppressor genes (PCDH11X, DKK1, and TC5313860) in intestinal cells (Taranu et al., 2015). In fact, the modulation of gene expression was responsible for the carcinogenic effects of ZEA. Nevertheless, swines ingested ZEA did not showed changes in the height of villi, the thickness of the mucosa, and number of goblet cells (Gajecka et al., 2016a,b; Lewczuk et al., 2016). In brief, ZEA plays a negative role in gut health although no apparent histological changes have been observed.

#### Fumonisins

Fusarium verticillioides is the major producer of fumonisins, where fumonisin B1 (FB1) is the most abundant in nature (Lerda, 2017). In contrast to most mycotoxins, which are hydrophobic in nature, fumonisins are hydrophilic compounds, which hinders its discovery until 1988 (Gelderblom et al., 1988). Human epidemiological studies in South Africa, Italy, and China revealed that the esophageal cancer is related to the intake of corn grains containing fumonisins (Chilaka et al., 2017). Another epidemic of neural tube defects (birth defects of the brain, spine, or spinal cord) occurred along the Texas-Mexico border, China and South Africa were also found to be associated with fumonisinscontaminated corn consumption (Ortiz et al., 2015). In animals, fumonisins have been found to cause pulmonary edema and hydrothorax in swines; leukoencephalomalacia in equine; and HCC in rats (da Rocha et al., 2014).

FB1 shares the same structure as cellular sphingolipids (Masching et al., 2016). Sphingolipids are responsible for neurological and immunological diseases, as well as cancer. Normal degradation of sphingolipids to ceramide requires sphingomyelinase and ceramidase (Boini et al., 2017). However, FB1 disturbs sphingolipids metabolism via ceramide synthase inhibition which leads to sphingosine accumulation in cells (Masching et al., 2016). FB1 elevates sphingosine levels in urine, serum, kidney, liver, and small intestine. The abnormal turnover of sphingosine induced cytotoxicity, oxidative stress, apoptosis in cells (Hahn et al., 2015).

Using intestinal cell lines (IPEC-1, Caco-2, and HT29), Minervini et al. (2014) found FB1 decreased the cell viability and proliferation in a concentration-dependent manner. A possible mechanism has been suggested through the accumulation of sphinganine by FB1. In the intestinal epithelial cells, sphinganine accumulation blocked G0/G1 phase in cell and resulted in growth inhibition and apoptosis (Angius et al., 2015). The accumulation of sphinganine also altered glycoprotein distribution in the jejunum and caused an increase in transepithelial passage of FB1 (Yamazoe et al., 2017). In addition, FB1 altered the integrity of intestinal barrier by suppressing tight junction (TJ) protein expression level (Romero et al., 2016). The increase in intestinal permeability, in turn, promotes translocation of bacteria (Kelly et al., 2015).

Besides, high levels of FB1 also induced an overgrowth in intestinal goblet cell of broiler and swine (Alassane-Kpembi and Oswald, 2015). Goblet-cell hyperplasia is associated with increased mucin secretion. However, continuous hypersecretion of mucins might deplete the number of goblet cells, resulting in devastation of mucus barrier (Johansson and Hansson, 2016). Previous studies conducted using intestinal cell lines (IPEC-1, Caco-2, and HT29) showed that FB1 was able to regulate immune responses. Upon LPS exposure to the FB1-treated cell line, a reduction in IL-8 synthesis was detected (Minervini et al., 2014). Such reduction could be responsible for a low number of polymorphonuclear leukocytes (PMNs) recruited to infection sites, thus leading to the ineffective elimination of pathogen from the gut (Brazil et al., 2013). Generally, in the gut, FB1 increased intestinal cell apoptosis, reduced intestinal barrier and caused immune dysfunction.

#### Ochratoxin

Ochratoxin is mainly produced by Aspergillus species and Penicillium species. Ochratoxin A (OTA) is the most prevalent and relevant fungal toxin of this group (Liuzzi et al., 2017). The main target site of OTA is kidney. Previous findings from animals showed OTA is a potent renal carcinogen (Russo et al., 2016). The International Agency for Research on Cancer (IARC) categorized OTA as possibly carcinogenic to humans under Group 2B carcinogen. Apart from that, OTA is an immunosuppressive, teratogenic, and nephrotoxic compound (Ladeira et al., 2017).

In human studies, OTA is associated with kidney diseases, such as Balkan endemic nephropathy (BEN). BEN is a chronic tubulointerstitial disease which slowly progressed into terminal renal failure. Indeed, a 15 years study confirmed that BEN is correlated with upper urothelial tract cancer (Rouprêt et al., 2015). Furthermore, OTA has been associated with the occurrence of upper urothelial tract cancer (Fahmy et al., 2014). However, a systemic review by Bui-Klimke and Wu (2015) revealed that there is no significant evidence for human health risks associated with OTA exposure based on the epidemiological data. The modes of toxic action of OTA are identified through the blockage of protein synthesis and energy production, the formation of DNA adduct formation, apoptosis, as well as the induction of oxidative stress (Koszegi and Poór, 2016 ˝ ). Moreover, recent studies showed OTA triggered autism via epigenetic mechanism (Mezzelani et al., 2016).

Other than its adverse effects on the kidney, previous studies also revealed the gut changes induced by OTA. OTA altered nutrition absorption in the intestine. In vitro studies demonstrated that OTA decreased glucose absorption via SGLT1 transporter (Peraica et al., 2011). In addition, OTA-treated animals experienced faster and more harmful parasite infections (provoked by Eimeria acervulina and E. adenoeides) in chicks and turkey compared to control. The results showed that animals fed with OTA had higher lesion and oocyst indexes in the intestine and more damage at mucosa (Manafi et al., 2011). This can be explained by the increased intestinal permeability in the presence of OTA. In addition, results from immunoblotting and immunofluorescence showed that the expression of TJ proteins responsible for intestinal integrity was significantly (P < 0.05) suppressed by OTA (McLaughlin et al., 2004). Besides, OTAinduced oxidative stress also can alter intestinal permeability (Anderson et al., 2016). The oxidative stress induced by OTA has been found to be associated with the apoptosis in the intestinal IPEC-J2 cells (Wang et al., 2017). A study by Solcan et al. (2015) on OTA-fed broilers revealed there is a decrease in villi height and increase in apoptosis of intestinal epithelial cells. Similar results were obtained from another study using broiler model (Qu et al., 2017). Inflammation pathway in the intestine was also affected by OTA. The expression of inflammation-related cytokines (IL-8, IL-6, IL-17A, IL-12, and IL-18) was significantly (P < 0.05) decreased in the intestine of the piglets exposed to the toxin (Marin et al., 2017). The alteration of immune system renders the gut vulnerability to infection. OTA exerted its effect on gut via the reduction of nutrient absorption, disruption of intestinal permeability, cell apoptosis, and modulation of immune system.

#### Aflatoxin

AF is a mycotoxin produced by Aspergillus flavus and Aspergillus parasiticus. The most common mycotoxin found in human food and animal feed is AFB1. In fact, AFB1 is the most potent hepatocarcinogen recognized in mammals and listed as Group I carcinogen by IARC (Muhammad et al., 2017). Liver is the main target site of AFB1. Cumulative evidences from human and animals revealed a strong linkage occurs between AFB1 and HCC. While, acute aflatoxicosis induced abdominal pain, vomiting, edema, and death (Mohd-Redzwan et al., 2013).

Aflatoxicosis outbreak has been recorded four times in Kenya from 2004 to 2014, with near to 600 individuals were affected and 211 deaths were reported from the tragic outbreak (Awuor et al., 2017). As discussed earlier, p450 enzymes in the liver metabolize AFB1 into AFB1-8,9-exo-epoxide. The highly reactive exo-epoxides form derivatives with DNA, RNA and proteins which subsequently react with the p53 tumor suppressor gene. The reaction generates AFB1-N7-Gua which is then converted to its stabilized form, AFB1-formamidopyrimidine (AFB1- FABY) adduct. AFB1-FABY causes transversion of guanine (G) to thymine (T), which leads to mutation and malignant transformation (Kew, 2013). In addition to the hepatotoxicity of AFB1 mentioned above, other adverse effects include growth retardation, immunosuppression, and genotoxicity have been reported (Kumar et al., 2017).

Like most of the mycotoxins, AFB1 compromised the health of GI tracts. Colon cell line (Caco-2) was used in in vitro experiment to determine the AFB1 toxicity in the intestine. AFB1 significantly (P < 0.05) inhibited cell growth, increased lactate dehydrogenase activity and caused genetic damage. It is found that the mechanism of AFB1 cytotoxicity are associated with reactive oxygen species (ROS) generation, which leads to the damage of cell membrane and DNA (Zhang et al., 2015). Besides, transepithelial electrical resistance assay showed a reduction in intestinal Caco-2 cells' integrity after AFB1 treatment (Romero et al., 2016).

Similar result has been observed in in vivo study where AFB1 affects intestinal barrier function in broiler model as indicated by an increased ratio of lactulose to rhamnose ratio in the plasma (Chen et al., 2016). Several studies on broiler exposed to AFB1 showed that the density (weight/length) of intestine was reduced (Hossein and Gurbuz, 2015). An increase of apoptotic events was found in the jejunum, accompanied with elevated apoptotic markers (Bax and caspase-3) mRNA expression level. Moreover, the increased apoptosis was corresponded to a lower jejunal villi height as found in the other studies (Peng et al., 2014; Zheng et al., 2017). Another study by Akinrinmade et al. (2016) demonstrated intestinal injuries induced by AFB1 in rats. In the AFB1-fed rat, leucocyte and lymphocyte infiltration were observed at lamina propria of the intestinal mucosa. In the duodenum and ileum, AFB1 exposure caused intestinal lesions such as the development of sub-epithelial space and villi degeneration. The adverse effects on the gut from AFB1 exposure include the disruption of intestinal barrier, cell proliferation, cell apoptosis, and immune system. Although AFB1 is the most life-threatening mycotoxin, yet its toxicity on the gut is comparable to the other mycotoxins.

## GUT MICROBIOTA

The gut microbiota represents an ensemble of microorganisms including bacteria, viruses, and fungi that harbor within the GI tracts of living organisms. In the past, gut microbiota studies have been focused on the association of single pathogenic organisms with human health. Non-pathogenic microbes were previously thought to be benign as compared to the pathogens (Holmes et al., 2011). Nevertheless, the gut microbiota has recently become a blooming research area (Hoffmann et al., 2013).

The rapid rise of remarkable and cost-effective nextgeneration DNA sequencing methods provide an effective approach to study the composition of the host microbiota. Metagenomic sequencing and amplicon sequencing using specific genes markers are established to replace cultureindependent methods for host microbiota analysis (Kuczynski et al., 2012). The amplified sequences resulting read abundance, which reflects the microbial diversity. The advances in molecular biology have provided innovative ways to entangle the complex microbial communities. Using the methods, it has been revealed that majority of microorganisms resides in the gut cannot be cultured outside the host. In the human, for instance, approximately 80% of the total bacterial species in the gut failed to be cultured under laboratory conditions (Guarner and Malagelada, 2003). Besides, these methods also revealed differences in gut bacterial community between anatomical sites, between individuals, and between healthy and diseased states. The findings have completely transformed the view of mammals biology (Weinstock, 2012). As such, growing interest has led to an increasing research into the communities of non-pathogenic microbes that inhabit the human body, and the need to describe the genomes of these organisms to understand the human microbiota has been recognized.

The composition of the gut microbiota varies significantly at the relative ratios of dominant phyla, genera, and species. In particular, stable and healthy gut microbiota is generally indicated by the rich diversity of gut bacteria (Mosca et al., 2016). The rapid development of gut microbiota study has revealed its significant role in maintaining human health. The involvement of gut microbiota in nutrition, metabolism, and immune function has been well established. The gut microbiota allows the host to metabolize a vast range of dietary substrates. For example, the metabolism of carbohydrate is a major catalytic function of the microbiota. The gut microbiota (specifically Bacteroides, Bifidobacterium, Enterobacterium, Fecalibacterium, and Roseburia) assist in the fermentation of complex polysaccharides that escaped proximal digestion (Jandhyala et al., 2015). The fermentation process produces monosaccharides and short chain fatty acids (SCFAs) which include acetate, butyrate, and propionate that are rich energy sources for the host (Jandhyala et al., 2015). Similarly, metabolism pathways of proteins, bile, and phytochemicals, as well as vitamins synthesis by the microbiota, has been also elucidated (Rowland et al., 2017).

Apart from that, the gut microbiota protects the host against infections via several mechanisms. The microbes reside in the gut modulate the population of pathogenic microorganisms via competitive exclusion for attachment sites and nutrient (Donaldson et al., 2016). The significance of gut microbiota in the development of immunity can be readily appreciated from the study of germ-free (GF) mice. By comparison to normal mice, GF mice which have a lack of microbiota have been shown to exhibit irregularities in cytokines profile, contain poorly formed local and systemic lymphoid structures as well as the abnormal level of immune cells (Sekirov et al., 2010). Furthermore, accumulating recent data demonstrated that the functions of gut microbiota further extend beyond the gut. Mechanisms studies suggesting that microbial metabolites have taken the role by sending signals to peripheral organs, including the liver, adipose tissue, pancreas, cardiovascular system, lung, and even to the brain (Lv et al., 2017).

It is widely recognized that the composition of gut microbiota in newborns is obtained from mothers during delivery (Dominguez-Bello et al., 2010). While the alteration in the composition of the gut microbiota is mediated by numerous factors including dietary changes (Cani and Everard, 2016), development of disorders and diseases (Hand et al., 2016), genetics as well as stressful experiences (Karl et al., 2017). Sufficient evidence has revealed the inter-relationship between dietary habit and the intestinal microbiota composition (Cani and Everard, 2016). For example, high fat diet renders the host to harbor a gut microbiota enriched in the phylum Firmicutes and depleted in Bacteriodetes. Besides, high fat diet has been also identified to promote proliferation of specific bacterial strains such as Enterobacteriaceae, which may increase intestinal lipopolysaccharide and subsequently increase gut permeability as well as triggering inflammation (Bibbò et al., 2016). On the other hand, a diet rich in fiber has been observed to modulate gut microbiota by altering fermentative metabolites and intestinal pH. Fermentation of fiber by the colonic microbiota produces SCFAs, wherein the metabolites play a significant role in regulating pH in the intestine. It has been demonstrated that a decrease in pH is able to significantly (P < 0.05) decrease the population of Bacteriodetes spp. and members of Enterobacteriaceae while promoting the growth of beneficial butyrate-producing microorganisms (Duncan et al., 2009).

On the other hand, studies revealed that the host genotype contributes remarkably to the resemblances in the gut microbial taxa. The genes linked with microbial taxa are particularly responsible for diet sensing, immunity, and metabolism (Goodrich et al., 2016b). Moreover, data showed a family belongs to the Firmicutes and Christensenellaceae has the highest heritability. The presence of Christensenellaceae in the gut is frequently associated with low serum triglyceride levels observed in lean and healthy human phenotype (van Opstal and Bordenstein, 2015). In genotype factor studies, monozygotic twins which developed from one zygote are often used as the subjects. As compared to dizygotic twins, it has been shown that there is a higher carriage of Methanobrevibacter smithii in monozygotic twins. M. smithii has also found to be related with leanness (Goodrich et al., 2016a). Besides, numerous studies have linked genetic loci with population of gut bacteria in mice and humans. For instance, LCT gene which encodes for lactase-phlorizin hydrolase. Single nucleotide polymorphisms (SNPs) in LCT are directly correlated with lactose intolerance and the abundance of lactose-metabolizing bacteria, specifically Bifidobacterium (Lerner et al., 2017).

Recently, the occurrence of diseases has often been linked consequentially to dysbiosis of gut microbiota. A wide range of gut microbiota-related diseases have been revealed include autism, asthma, colon cancer, Crohn's Disease, irritable bowel syndrome (IBS), food allergies, cardiovascular disease, obesity, diabetes, eczema and hepatic encephalopathy, mental disorders (Kamada et al., 2013; Sommer and Bäckhed, 2013; Korem et al., 2015). For instance, bacterial overgrowth in the small intestine is commonly observed in patients with IBS. An investigation using 16S rRNA-based microbiota profiling approaches on IBS subjects revealed quantitative and qualitative changes in both mucosal and fecal gut microbiota (Simrén et al., 2012; Collins, 2014). The findings suggested that the gut microbiota balance was compromised in the events of gut inflammation. The dysbiosis of gut microbiota initiates mucosal innate immune responses and increases intestinal permeability. Subsequently, translocation of pathogens occurs, and harmful metabolites can enter the intestinal epithelium. Such events in the gut further exacerbate the severity of diseases (Collins, 2014). Apart from that, evidence showed that the gut microbiota-derived products (SCFA, neurotransmitters, enzymes, and toxins) can be absorbed from the gut, and subsequently affect the metabolic phenotype of the host (Lee and Hase, 2014). In addition, metabolites from the host are transported into the gut via the enterohepatic circulation and serve as substrates for the microbiota. These processes give rise to an interspecies cross-talk between the host genome and the gut microbiota (Lee and Hase, 2014).

Regrettably, fewer studies have discussed on the importance of the minorities such as virus, commensal fungi, archaea, and protozoa. The gut virome comprised plant-derived viruses, giant viruses, and abundant (90%) bacteriophages. Pathogenicity of gut viruses includes gastroenteritis, pneumonitis, and diarrhea (Scarpellini et al., 2015). Whereas, fungal communities in the gut mainly consist of the Ascomycota, Basidiomycota, and Zygomycota. Candida species have been primarily associated in inflammatory bowel diseases (IBD), Crohn's disease (CD), ulcerative colitis (UC), obesity and gut inflammation (Sam et al., 2017). Undeniably, the gut microbiota influences almost all system resides in our body, which is of vital importance for survival, therefore, maintaining a balanced microbiota is essentially important.

#### The Bi-directional Interaction between Mycotoxin and the Gut Microbiota

Gut microbiota represents an important bridge between environmental substances and host metabolism. Findings found that gut microbiota, particularly in animals have profound interactions with ingested mycotoxins. Microbes reside in the gut aid host in the mycotoxin removal process through metabolizing or binding to the mycotoxins. Although some microbes possess the mycotoxin removal ability, it is noteworthy to mention that bacteria from the same genus, however, are unable to remove mycotoxin. Interestingly, few studies also demonstrated that mycotoxins can alter the gut microbiota. Such findings suggested that there is a bi-directional interaction occurs between mycotoxin and the gut microbiota. Evidence of disturbance on gut microbiota modulation induced by mycotoxin only had been studied on animal and the results have been summarized (**Table 1**). The changes in gut microbiota can be observed up to species level in some of the studies using advance molecular approaches. However, the compositions of gut microbiota are greatly influenced by various factors during the experiment. Confounding factors affecting microbial composition and function may include diet (Cani and Everard, 2016), the exposure of environmental chemical and antibiotics (Claus et al., 2017), genetic background (Goodrich et al., 2016b), as well as the mental health condition (stress) of the host (Karl et al., 2017). These factors can explain that the microbiota in same species may not be able to reduce the level of mycotoxins. Besides, the changes in gut microbiota due to the presence of mycotoxin may contribute by some uncontrolled variables. Contrasting data obtained from different studies in this review further highlight the significance of confounding factors toward the outcome of studies. Nonetheless, a well-controlled study designs is essential to ensure repeatable studies with consistent results.

#### Deoxynivalenol

A study has demonstrated the ability of gut microbiota to remove deoxynivalenol (DON) using an in vitro study (He et al., 1992). It was reported that the microorganisms in large intestines


TABLE

1


vivo

experiments:

Gut

microbiota

alteration

by

mycotoxins.

(Continued)


Frontiers in Cellular and Infection Microbiology | www.frontiersin.org

of broilers were able to transform DON into de-epoxy-DON. The gut microbes of broiler were further shown to transform DON to the less toxic metabolite, de-epoxy-DON via epoxide reductase. Similar findings have been observed in other studies using microbial content from the broiler intestine (Lun et al., 1988; Young et al., 2007). Some other studies have also shown that intestinal microorganisms of other animal species including rat (Worrell et al., 1989) and swine (Kollarczik et al., 1994) possess the same ability. However, no alteration was found when swines' intestines content was used in the study conducted by He et al. (1992).

Apart from these, mycotoxin degrading bacteria had been isolated from intestinal content for extensive studies. Microbiological selection strategies guided by PCR DGGE (denaturing gradient gel electrophoresis) had been employed to isolate DON-transforming bacteria and the isolates obtained belong to four different bacterial groups; Anaerofilum, Bacillus Clostridiales, and Collinsella (Yu H. et al., 2010). In addition, a microbial community, namely microbial culture C133 from catfish digesta was screened, and capable to completely transformed DON to de-epoxy-DON after 96 h incubation (Guan et al., 2009). Several intestinal bacterial strains have been identified as biological trichothecene detoxification agent via in vitro screening and microbial analyses in other recent reviews (Hathout and Aly, 2014). Interestingly, a study showed the ability to metabolize DON can be obtained via gut microbiota transfer in swine. However, the transfer of gut microbiota revealed no changes in the DNA-profiles of the gut bacterial composition (Eriksen et al., 2002).

Several studies have been done on the interaction of DON toward the gut microbiota. Consumption of feed contaminated with DON for a time duration of 4 weeks has been shown to exert minor effect on the total number of fecal aerobic mesophilic bacteria and anaerobic sulfite-reducing bacteria in swine. Although there was no effect of DON on microbial diversity, the richness index was significantly (P < 0.05) increased by DON exposure (Waché et al., 2008). In another study by Saint-Cyr et al. (2013), GF male rats were inoculated with fecal flora from healthy human, in order to investigate the human gut microbial changes induced by DON. By using real-time PCR quantification, a significant increase of Bacteroides/Prevotella group and decreased concentration levels of Escherichia coli were observed after feeding the rats with DON for 4 weeks (P < 0.05). In human, the shift in the proportion of Bacteroides is highly associated with diseases as individuals with Crohn's disease or celiac diseases often exhibit a higher abundance of Bacteroides than healthy individuals (Kamada, 2016). A recent study, however, revealed that DON has no significant changes in the diversity and relative abundance of gut microbiota based on 16S rRNA microbiota analysis (Payros et al., 2017).

#### T-2 Toxin

Gratz et al. (2017) demonstrated that masked T-2 toxin was released as a parent mycotoxin by human gut microbiota, and thereby contribute to mycotoxin exposure. Trichothecene mycotoxins are generally known as ribotoxic stress inducer which effectively blocks eukaryotic 28S rRNA. Thus, in theoretical aspect, the T-2 toxin would not interfere with bacterial protein translation and growth as suggested by Schmeits et al. (2014). Nevertheless, this is in contrast to the observation seen in a study conducted by Tenk et al. (1982). It was shown that the administration of T-2 toxin for 1 week was sufficient to induce a substantial increase in the aerobic bacteria count in the intestine of swines and rats (Tenk et al., 1982). While the bacterial populations have been shown to be greatly affected by trichothecene, the mechanism which causes the perturbation of bacterial population remains to be elucidated.

#### Zearalenone

An in vitro experiment demonstrated that zearalenone compounds were converted into unknown metabolites by human gut microbiota (Gratz et al., 2017). The first study on the effect of ZEA on gut microbiota has been carried out by Piotrowska et al. (2014). The changes in gut microbiota were evaluated using Biolog EcoPlate method which only allows the quantification of culturable bacteria. After 6 weeks of ZEA ingestion, the data showed the concentration of Clostridium perfringens, Enterobacteriaceae, and E. coli was significantly reduced (P < 0.05).

#### Fumonisins

Using capillary electrophoresis single-stranded conformation polymorphism (CE-SSCP), it is shown that fumonisins decreased the fecal microbiota SSCP-profiles similarity of the fumonisinstreated swines, compared to the untreated control group. The results indicated that there is an increase in the diversity of microbiota. The balance of the digestive microbiota was transiently but markedly affected after 63 days of chronic exposure to fumonisin, which consists of a mixture of FB1 and FB2 (Burel et al., 2013).

#### Ochratoxin A

A study investigating the effect of OTA on gut microbiota using bioreactors has been carried out by Ouethrani et al. (2013), in which each bioreactor represents different parts of the adult human gut (Ouethrani et al., 2013). Based on the study, the gut microbiota degradation of OTA and microbiota diversity alteration were observed only at the descending colon after 1-week exposure to OTA. PCR-TTGE targeting Lactobacilli populations showed that Lactobacillus reuteri present during the start-up period, was permanently disappeared at the end of the OTA treatment period accompanied by some minor changes in the bifidobacteria population. The alteration was explained by a significant reduction in acetic, butyric and total SCFA concentration (P < 0.05). The reduction of beneficial microbes, lactobacillus and bifidobacteria indicated that the OTA shifted the microbiota balance and possibly led to impaired immunity.

In vivo study in rat demonstrated that OTA treatment decreased the diversity of the gut microbiota (Guo et al., 2014). The authors also reported that the relative abundance of Lactobacillaceae was increased whereas the Bacteroidaceae was decreased. Moreover, at the genus level, the OTA decreased the population of Bacteroides, Dorea, Escherichia, Oribacterium, Ruminococcus, and Syntrophococcus, while increased the number of Lactobacillus. The results showed Lactobacillus was more resistant to OTA and Lactobacillus may play a role in OTA detoxification process. Apart from that, it was also reported that the total facultative anaerobes were increased by the OTA treatment. In fact, the increase in facultative anaerobes was often observed in individuals with health complication (Shimizu et al., 2011) as well as in the elderly (Rudi and Avershina, 2015). This may further suggest that the OTA may cause negative effects on the host health via gut microbiota modulation.

#### Aflatoxin

In contrast to the intense research on AFB1 untoward effects, little information is available in regard to the outcomes of AFB1 on the gut microbiota. The findings from Wang et al. (2015) suggested that AFB1 could alter the gut microbiota in a dosedependent manner. AFB1 decreased phylogenetical diversity but increased evenness of community composition. Although there was no changes at the phylum level, some lactic acid bacteria were significantly (P < 0.05) reduced in the presence of AFB1 (Wang et al., 2015). The reduction of LAB in animals treated with a lower dosage of AFB1 explains the severe immune malfunction induced by lower dosage of AFB1 (Qian et al., 2014).

A recent study showed that AFB1 at a dosage level of 1 ppm significantly (P < 0.05) reduced total LAB in the broiler. In contrast, the total number of Gram-negative bacteria and LAB were significantly (P < 0.05) increased in the group of broiler fed with 1.5 and 2 ppm of AFB1 (Galarza-Seeber et al., 2016). Besides, it has also been reported that 2.5 ppm of AFB1 increased the production of total volatile fatty acids in broilers, which suggested the association of higher dosage of AFB1 with a higher prevalence of LAB in the intestine (Kubena et al., 2001). Interestingly, a separate study showed greater numbers of bacterial mutants were recovered from mice exposed to AFB1 (Rowland, 1988). The data implies that the genotoxic effects of AFB1 not only affecting the host, but the gut microbiota as well.

#### Combination of Mycotoxins

On the other hand, the effect of a combination of mycotoxins on the modulation of gut microbiota has also been investigated. The exposure of gilts to ZEA and DON was found to pose an adverse impact on mesophilic aerobic bacteria. In particular, the amounts of C. perfringens, E. coli, and other bacteria in the family Enterobacteriaceae were reduced significantly after the 6th week of the experiment (P < 0.05). Nevertheless, the biodiversity of microorganisms in the gut was increased. Apart from that, an increase in the metabolism of amino acid by the gut microbiota was also observed. It was suggested that the increased metabolism of amino acid may be detrimental due to the formation of biogenic amines and procarcinogenic compounds (Piotrowska et al., 2014). Besides, a study confirmed AF and fumonisin mixture increased Shiga Toxin-producing E. coli (STEC) level in fecal (Baines et al., 2013). The composition of STEC-secreted cytotoxin was affected as reflected in the elevated concentration of intracellular Ca2<sup>+</sup> with a corresponding increase in cytotoxicity. Mycotoxins are capable of altering the microbial balance of the intestine. Furthermore, the possible pathway proposed is via oxidative stress induced by mycotoxins (Vinderola and Ritieni, 2014). Nonetheless, the mechanisms by which mycotoxins affect the gut bacterial composition however remain unclear.

## THE ROLE OF GUT MICROBIOTA IN THE DEVELOPMENT OF MYCOTOXICOSIS: HCC

Chronic mycotoxicosis, such as HCC results from a high dosage of mycotoxins' contamination. Such pathogenesis generally involves the formation of DNA adducts, regulation of DNA methylation, and alteration of gene expression (Dai et al., 2017; Zhu et al., 2017). Interestingly, gut microbiota perturbation is found to be one of the factors influencing mycotoxininduced HCC and its association is described in **Figure 1**. The development of HCC in mice induced by a combination of diethylnitrosamine (DEN) and hepatotoxin carbon tetrachloride (CCl4), a model that features several characteristics of chronically injured livers in which human HCC mostly arises, is prevented via gut sterilization. The same study also showed that mice that were grown in specific GF conditions demonstrated fewer and smaller tumors as compared with those grown under specific pathogen free (SPF) conditions (Dapito et al., 2012). In a toxic model of hepatocarcinogenesis, Yu L. X. et al. (2010) found that the depletion of host microflora suppresses tumor formation. Treatment of rats with antibiotic targeting gramnegative organisms (polymyxin B and neomycin) markedly reduced the size and number of HCC nodules after injection of DEN, which induces HCC.

Some specific bacterial species are also found to be correlated with HCC development. Studies showed that the intestinal colonization by Helicobacter hepaticus induced HCC, and the DNA of Helicobacter ssp. is only present in liver biopsies from HCC patients, not in control samples (Gargano and Hughes, 2014). Findings from both animal and human studies demonstrated that liver cirrhosis and HCC stimulate an intestinal dysbiosis as well as a significant increase population of the E. coli and Atopobium cluster, coupled with a significant (P < 0.05) reduction in the percentages of beneficial microbes such as Lactobacillus group, Bifidobacterium group, and Enterococcus group (Zhang et al., 2012). Besides, hepatocarcinogenesis is found to be related to the increased lipopolysaccharides (LPS) levels which are commonly produced by pathogens in several studies (Zhang et al., 2012).

Probiotics are known for their roles in gut health and microbiota restoration. In addition, many strains of probiotics possess the ability to reduce the level of mycotoxins, particularly via binding. Treatment with probiotics mixture, Prohep [Lactobacillus rhamnosus GG, heat-inactivated VSL#3, and viable E. coli Nissle 1917 (1:1:1)] successfully relieved the microbial imbalance and hepatic inflammation, which further decreased liver tumor growth (Li et al., 2016). A human study by El-Nezami et al. (2006) demonstrated a statistically significant decrease (up to 55% at 5th week; P < 0.05) of urinary AFB-N7-guanine level in the probiotic (L. rhamnosus LC705 and Propionibacterium freudenreichii subsp. shermanii)

mixture group compared to the placebo group. Similar finding was found by Mohd Redzwan et al. (2016) where serum AFB1-lys level were significantly lower (P < 0·05) in the Lactobacillus casei Shirota supplemented individuals. Besides, hepatic transcriptome in AFB1-induced HCC was positively altered by probiotics (Monson et al., 2015). Probiotic supplement reduces the biologically available effective toxic dose of mycotoxin coupled with its gut microbiota normalization ability, offer an effective dietary approach to decrease the risk of liver cancer. As shown in these studies, the restoration of gut microbiota equilibrium offers protection and treatment effects in HCC whereas the occurrence of HCC is linked to the higher abundance of pathogens as illustrated in **Figure 1**. The linkage of microbiota and HCC is undeniably important to understand the mechanism involved in the pathogenesis of HCC.

#### CONCLUSIONS

This concise review has attempted to draw together the keyworks to highlight the crucial interaction between mycotoxins, the gut, and the gut microbiota in human and animal health. The mycotoxins and gut microbiota studies have revealed meaningful interactions. The uptake of mycotoxin and subsequent tissue distribution are governed by GI tract absorption, and the presence of microbiota at the GI tract can affect the intestinal barrier causing different (maximal or limited) bioavailability of these fungal compounds. The gut microbiota can vary within the same species, thus different reactions toward mycotoxin can be observed as discussed in this review article. In addition, mycotoxins disrupt the gut microbiota balance, and thereby dysregulate intestinal functions and impair local immune response, which may eventually result in systemic toxicity that leads to chronic mycotoxicosis, HCC. The severity of HCC condition can be positively governed by restoration of gut microbiota balance and gut health via probiotics administration. Probiotic which generally helps restore the natural harmony of gut microbiota coupled with its mycotoxins reducing ability could increase its health-promoting value. Regardless, more studies are needed to elucidate the interaction between the gut microbiota and mycotoxin and the implication of such interaction for mycotoxicosis prevention/treatment.

## AUTHOR CONTRIBUTIONS

SM-R and W-P-PL provided the conception and the structure of the article. W-P-PL wrote the draft. SM-R and W-P-PL revised the article and approved the final version to be published.

## FUNDING

Research grant GP-IPM/2016/9480100 from Universiti Putra Malaysia.

## ACKNOWLEDGMENTS

The authors would like to acknowledge financial support through research grant GP-IPM/2016/9480100 from Universiti Putra Malaysia (UPM). W-P-PL would like to thank UPM [Graduate Research Fellowship, GRF] and Ministry of Higher Education Malaysia (MoHE) [MyBrain15 Program] for the scholarships.

## REFERENCES


status, intestinal microbiota and sensitivity to Salmonella. Toxins 5, 841–864. doi: 10.3390/toxins5040841


typhimurium colonization as affected by aflatoxins and T-2 toxin. Poult. Sci. 80, 411–417. doi: 10.1093/ps/80.4.411


alteration of sphingolipid metabolism in turkey and swine. Toxins 8:84. doi: 10.3390/toxins8030084


Sglt2 in kidneys of ochratoxin A-treated rats. Toxicol. Lett. 205:S275. doi: 10.1016/j.toxlet.2011.05.933


**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 Liew and Mohd-Redzwan. 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.

# *Clostridioides difficile* Biology: Sporulation, Germination, and Corresponding Therapies for *C. difficile* Infection

#### Duolong Zhu<sup>1</sup> , Joseph A. Sorg<sup>2</sup> and Xingmin Sun<sup>1</sup> \*

*<sup>1</sup> Department of Molecular Medicine, Morsani College of Medicine, University of South Florida, Tampa, FL, United States, <sup>2</sup> Department of Biology, Texas A&M University, College Station, TX, United States*

*Clostridioides difficile* is a Gram-positive, spore-forming, toxin-producing anaerobe, and an important nosocomial pathogen. Due to the strictly anaerobic nature of the vegetative form, spores are the main morphotype of infection and transmission of the disease. Spore formation and their subsequent germination play critical roles in *C. difficile* infection (CDI) progress. Under suitable conditions, *C. difficile* spores will germinate and outgrow to produce the pathogenic vegetative form. During CDI, *C. difficile* produces toxins (TcdA and TcdB) that are required to initiate the disease. Meanwhile, it also produces spores that are responsible for the persistence and recurrence of *C. difficile* in patients. Recent studies have shed light on the regulatory mechanisms of *C. difficile* sporulation and germination. This review is to summarize recent advances on the regulation of sporulation/germination in *C. difficile* and the corresponding therapeutic strategies that are aimed at these important processes.

#### *Edited by:*

*Nathan W. Schmidt, University of Louisville, United States*

#### *Reviewed by:*

*Paul Edward Carlson, Food and Drug Administration, United States Peter Mullany, University College London, United Kingdom*

> *\*Correspondence: Xingmin Sun*

*sun5@health.usf.edu*

*Received: 30 October 2017 Accepted: 23 January 2018 Published: 08 February 2018*

#### *Citation:*

*Zhu D, Sorg JA and Sun X (2018) Clostridioides difficile Biology: Sporulation, Germination, and Corresponding Therapies for C. difficile Infection. Front. Cell. Infect. Microbiol. 8:29. doi: 10.3389/fcimb.2018.00029* Keywords: *C. difficile*, spores, germination, CDI, sporulation

## INTRODUCTION

Clostridioides difficile (formerly Clostridium difficile; Lawson et al., 2016; Oren and Garrity, 2016) is a Gram-positive, spore-forming, toxin-producing, anaerobic bacterium which has established itself as a leading cause of nosocomial antibiotic-associated diarrhea in the developed countries (Sebaihia et al., 2006). It is found widely in the mammalian gastrointestinal (GI) tract and can cause toxin-mediated C. difficile infections (CDI) that range from mild diarrhea to pseudomembranous colitis and potential death (Lessa et al., 2012). C. difficile causes over 500,000 infections per year in the United States alone, resulting in an estimated 29,000 deaths and an estimated cost of \$1–3 billion (Dubberke and Olsen, 2012; Lessa et al., 2015). Currently, antibiotics are the standard treatments for CDI (i.e., vancomycin, metronidazole, or fidaxomicin; Evans and Safdar, 2015). Though effective, CDI recurrence after the initial treatment can still reach up to 15–35% in treated patients (Leffler and Lamont, 2015). Though recurrence is not fully understood, one of the reasons for high recurrence rate is that C. difficile spores may still be present within the patients gut and germinate to the vegetative form after completion or discontinuation of antibiotic treatment (Cornely et al., 2012). Meanwhile, poor host immune response to C. difficile and frequent disruption of the normal gut flora may also contribute to the high recurrence rate (Johnson, 2009). Due to the inherent antibiotic resistance of C. difficile cells and high prevalence of CDI in some hospitals, the Centers for Disease Control and Prevention (CDC) has listed C. difficile as "an urgent threat" regarding the antibiotic associated threats to the United States (Centres for Disease Control and Prevention (US), 2013).

Because C. difficile is an obligate anaerobic pathogen, the vegetative cells are unable to survive outside of a host in the aerobic environment. When C. difficile cells meet certain environmental stimuli (e.g., nutrient deprivation, quorum sensing, and other unidentified stress factors), they will initiate a sporulation pathway to produce sufficient dormant spores to survive in extreme situations (Setlow, 2006; Rodriguez-Palacios and LeJeune, 2011; Deakin et al., 2012; Higgins and Dworkin, 2012). C. difficile pathogenesis relies on the formation of aerotolerant dormant spores which allows C. difficile to persist within the host and to disseminate through patientto-patient contact/environmental contamination (Britton and Young, 2012). In the host GI tract, the dormant spores must germinate from dormancy to form the actively growing vegetative cells which produce the toxins that cause the primary symptoms of the disease. Under suitable conditions, when germinant receptors sense the presence of small molecules (germinants), spore germination will be induced (Sorg and Sonenshein, 2008).

Recent studies have focused on the regulatory mechanisms of C. difficile sporulation/germination to gain insight into these important processes. However, when compared to other wellstudied organisms such as Bacillus subtilis and Clostridium perfringens, our knowledge of C. difficile spore biology still lags far behind. In this review, we will discuss recent progresses in the field of C. difficile spore biology, specifically on the sporulation and germination processes and their implications for CDI treatment.

## *C. DIFFICILE* SPORULATION

#### Sporulation Program

Though the signals/molecules that trigger C. difficile sporulation have not been identified, based on studies in other organisms, it is likely that environmental stimuli such as nutrient limitation, quorum sensing, and other unidentified stress factors are involved (Higgins and Dworkin, 2012). In fact, though the mechanism is not well-defined, a recent report has suggested that quorum sensing is important for C. difficile spore formation (Darkoh et al., 2016). As described in other spore-forming bacteria (e.g., B. subtilis), the main process of C. difficile sporulation contains four morphogenetic stages (**Figure 1**; Edwards and McBride, 2014; Gil et al., 2017): (I) an asymmetric septation generates a smaller compartment (SC) and a larger mother cell (MC); (II) the MC engulfs the SC (now the forespore) in a phagocytic-like event resulting in a forespore being wholly contained within the MC's cytoplasm; (III) the spore cortex and coat layers are assembled; (IV) the MC lyses and releases the mature spore into the surrounding environment. Though the mechanisms that initiate spore formation may differ between organisms, the overall spore architecture is conserved among endospore-forming bacteria. Located in the center of the mature spore is the core. The spore core contains the genomic DNA, mRNA, ribosomes, protein, and is very rich in pyridine-2,6-dicarboxylic acid (DPA), commonly as a calcium salt (CaDPA). The spore core is surrounded by an inner membrane, a peptidoglycan-containing germ cell wall, a specialized peptidoglycan-containing cortex, an outer membrane and layers of coat protein (**Figure 1**; Edwards and McBride, 2014; Gil et al., 2017). In some C. difficile strains, an exosporium layer surrounds the coat, but not all spore-forming bacteria and not all C. difficile strains have this layer (thus this layer is not shown in **Figure 1**).

## Regulator CodY and CcpA

Environmental stimuli (e.g., nutrient deprivation or quorum sensing) could trigger C. difficile sporulation. Previous studies in Bacillus and Clostridioides species have revealed that the CodY and CcpA nutritional sensor proteins work as negative regulators of sporulation (**Figure 2**; Duncan et al., 1995; Hofmeister et al., 1995; Karow et al., 1995; Londoño-Vallejo and Stragier, 1995; Antunes et al., 2012; Nawrocki et al., 2016; Serrano et al., 2016). Among the genes CodY regulates are genes involved in spore formation including spo0A, rapA, rapC, rapE, sinI/R, sigH, and kinB. Recently, Edwards et al. demonstrated that the oligopeptide permease genes app and opp, and the putative sporulation regulator genes sinI and sinR, were regulated by CodY to suppress the initiation of C. difficile sporulation (Edwards et al., 2014). Previous studies indicated that the variability of CodYdependent regulation is an important contributor to virulence and sporulation in current epidemic isolates (Bennett et al., 2007; Majerczyk et al., 2008; van Schaik et al., 2009). But, to date, the regulatory mechanisms by which CodY affects sporulation are not fully understood because the factors that initiate sporulation in C. difficile are still being identified.

CcpA, a LacI family DNA-binding transcriptional regulator, works as a global transcriptional regulator that responds to the availability of carbohydrates (Deutscher et al., 2006). The CcpA sequence and structure are conserved in C. difficile, and has high homology to other pathogens (identity ≥62% analyzed with NCBI website), such as Staphylococcus aureus, Clostridium perfringens, and Clostridium perfringens. CcpA represses the use of alternative carbon sources and positively regulates sugar uptake, fermentation, and amino acid metabolism (Fujita, 2009). In the past few years, CcpA has been shown to regulate several virulence-associated genes. For example, it regulates the expression of the S. aureus α-hemolysin (hla), enterotoxins A, B, and C (sea, seb, and sec) genes, the C. perfringens enterotoxin (cpe) gene, and the Bacillus anthracis atxA and protective antigen (pagA) genes (Varga et al., 2004; Seidl et al., 2006; Chiang et al., 2011). Moreover, CcpA also plays critical role in the control of colonization, antibiotic resistance, and biofilm formation (Seidl et al., 2006; Varga et al., 2008). In C. difficile, CcpA directly regulates the PaLoc genes (tcdR, tcdB, tcdA, and tcdC) to mediate glucose-dependent repression of toxin production and indirectly regulates C. difficile sporulation (Antunes et al., 2011).

#### Sporulation Progress

Studies have revealed the master transcriptional regulator Spo0A plays the critical role during C. difficile sporulation (Deakin et al., 2012). In all studied endospore-forming bacteria, Spo0A must be phosphorylated (Spo0A-P) by a histidine kinase to become

activated. In Bacilli, these histidine kinases (Kin) are found on the plasma membrane and lead to the phosphorylation of Spo0A through Spo0F/Spo0B phosphotransfer system. C. difficile does not encode orthologs of these kinases or the phosphotransfer system. However, previous studies have demonstrated five putative orphan histidine kinases {CD1352 [CD630\_13520; cprK (McBride and Sonenshein, 2011)], CD1492 (CD630\_14920), CD1579 (CD630\_15790), CD1949 (CD630\_19490), and CD2492 (CD630\_24920)} in C. difficile strain 630 genome that could potentially phosphorylate Spo0A (**Figure 2**; Underwood et al., 2009). A ClosTron mutation in CD2492 (CD630\_24920) resulted in a decreased capacity of the resulting strain to generate spores compared to the WT parent. However, this mutant still generated spores (∼4%) suggesting that other histidine kinases can phosphorylate Spo0A or lead to Spo0A phosphorylation (Underwood et al., 2009). In support of this hypothesis, CD1579 (CD630\_15790) was shown to autophosphorylate and transfer a phosphate directly to Spo0A (Underwood et al., 2009). Importantly though, the authors did not complement their CD2492 ClosTron mutation, which could have polar effects on downstream genes. In contrast, Childress et al. found that a markerless deletion of CD1492 (CD630\_14920) was an inhibitor of sporulation and suppresses spore formation (Childress et al., 2016). This phenotype could be complemented by expression of the wild type allele. Currently, the function of the other putative orphan histidine kinases and their ability to phosphorylate Spo0A are unclear.

Recently, RstA was found to be a novel, positive regulator of sporulation initiation in C. difficile (**Figure 2**; Edwards et al., 2016). RstA positively affects the initiation of C. difficile sporulation through its peptide-interacting domain (TPR), and negatively regulates toxin production and mobility by affecting the flagellar-specific sigma factor (SigD) expression. But a detailed pathway on the regulation of sporulation initiation by RstA is not fully appreciated. The authors hypothesized that RstA may be a C. difficile global transcriptional regulator, similar to the broad physiological roles that the RNPP (Rap/NprR/PlcR/PrgX) proteins play in other bacteria (Edwards et al., 2016).

Spo0A functions as a critical regulator for sporulation by regulating sporulation-specific RNA polymerase sigma factors, especially for σ E , σ F , σ <sup>G</sup>, and σ <sup>K</sup> (Fimlaid and Shen, 2015). These σ factors activate compartment-specific transcriptional regulation during B. subtilis sporulation and are also conserved in Clostridium species. σ E and σ <sup>K</sup> are MC-specific, and σ F and σ <sup>G</sup> are specific to the developing forespore. The sporulation regulatory pathway of sigma factors in C. difficile is illustrated in **Figure 2**: (1) σ F is activated in the forespore soon after polar septation, and it controls early stages of development in this compartment. σ <sup>F</sup> becomes active when the anti-sigma factor SpoIIAB (ADP form) binds to the anti-anti-sigma factor SpoIIAA in its unphosphorylated form, while SpoIIE catalyzes dephosphorylation; (2) σ F activity leads to expression of SpoIIR, which interacts with the membrane-bound protease SpoIIGA (SpoIIGA is responsible for the cleavage of pro-σ E through transseptum signaling, yielding active σ E ); (3) after σ F and σ <sup>E</sup> become specifically active in the forespore and MC, respectively, the MC engulfs the forespore; (4) σ E activity leads to expression of SpoIIIA-H, which works with σ F -controlled SpoIIQ to form a channel in the inner and outer forespore membranes. SpoIIIAH and SpoIIQ localize to the asymmetric septum and the engulfing membranes and interact in the intermembrane space via their extracytoplasmic domains; (5) σ E -controlled SpoIIID activates σ <sup>K</sup> in the MC (Haraldsen and Sonenshein, 2003; Fimlaid et al., 2013; Pereira et al., 2013; Paredes-Sabja et al., 2014; Saujet et al., 2014). Though many of the factors that control spore formation are conserved in C. difficile, there are some differences in the sporulation program between C. difficile and B. subtilis. For instance (**Figure 2**), pro-σ k is not encoded by C. difficile, but the mature σ k is produced directly in C. difficile, σ E activation is dispensable for σ <sup>G</sup> activation, σ <sup>G</sup> activation is dispensable for σ <sup>K</sup> activation, and σ <sup>K</sup> is responsible for transcribing the germinant receptors while σ <sup>G</sup> is responsible in B. subtilis (Fimlaid et al., 2013; Pereira et al., 2013). Importantly, the FS line of gene expression occurs largely independently of the MC line of gene expression. Moreover, σ <sup>G</sup> can be activated before σ E and σ <sup>K</sup>, indicating that the order/sequence of sigma factor

CD1492 (CD630\_14920), CcpA, and CodY were the negative regulators in sporulation pathways. Function of SpoIIQ-SpoIIIAH complex was characterized by Pinho group, recently (Serrano et al., 2016). Spo0A and Spo0A-P were scheduled in orange boxes, four sigma factors were colored in blue. Dashed boxes indicate that the function of the proteins in regulation pathways has not been identified. Black arrows indicate the regulatory relationship between the factors has been confirmed, dashed arrows indicate the regulatory relationship between the factors has not been tested, red arrows, stops and boxes indicate the function of proteins and the correlation of factors has been confirmed recently. Question marks indicate that there is suggestive, but no conclusive experimental evidence.

activation is not as tightly controlled in C. difficile as it is in B. subtilis.

Finally, and in another departure from the model of spore formation in B. subtilis, a recent article by Ribis and colleagues used a TargeTron-based gene disruption demonstrated that the SpoVM protein is not required for spore formation/maturation (Ribis et al., 2017). SpoVM is a small protein that is expressed in the MC that recognizes the positive curvature of outer membrane of the developing forespore and embeds itself there. In B. subtilis, SpoVM recruits the SpoIVA scaffolding protein which polymerizes and surrounds the forespore. Subsequently, the coat is deposited onto the polymerized SpoIVA protein. In C. difficile, a spoVM mutation resulted in a modest defect in spore production (<5-fold), but their resistance properties are not different from a wildtype spore. This phenotype could be complemented through chromosomal complementation of the wild type allele. However, and importantly, the mutation in spoVM lead to a mislocalization of the coat proteins to one pole of the developing forespore and the coat extended into the MC cytoplasm; SpoIVA still polymerized on the surface of the forespore.

## *C. DIFFICILE* SPORE GERMINATION

## Germination Program

In most organisms, spore germination is induced when specific germinant receptors sense the presence of small molecules (germinants; Setlow, 2003). To date, germination has been moststudied in Bacillus spp. and it contains three main steps (Paredes-Sabja et al., 2011, 2014): (I) germinant (e.g., nucleosides, sugars, amino acids, and/or ions) binding with their cognate Ger-type receptors (GerAA-AB-AC) at the inner spore membrane to trigger the release of monovalent cations (H+, Na+, and K+) and the large amount of CaDPA stored within the core, in exchange for water; (II) CaDPA release and core rehydration leads to the activation of spore cortex lytic enzymes (SCLEs) SleB and CwlJ; (III) activated SleB and CwlJ degrade the peptidoglycan cortex layer, which allows for full core rehydration and resumption of metabolism in the spore core.

## Germinant Recognition/Signaling

Germination of C. difficile spores is the first step for initiating CDI. C. difficile spore germination is activated in response to certain host-derived bile salt germinants [e.g., taurocholic acid (TCA)/cholic acid derivatives] and amino acids (e.g., glycine or alanine; Sorg and Sonenshein, 2008). Chenodeoxycholic acidderivatives (a compound structurally similar to cholic acid but lacking the 12α-hydroxyl group) are competitive inhibitors of cholic acid-mediated germination (Francis et al., 2013b). Though the Ger-type germinant receptors have been widely studied in many organisms, including C. perfringens and Clostridium botulinum/sporogenes, C. difficile does not encode orthologs of the gerA germinant. Instead, C. difficile spores use the subtilisinlike, CspC pseudoprotease as the bile acid germinant receptor (**Figure 3**; Paredes-Sabja et al., 2008; Francis et al., 2013a, 2015; Wang S. W. et al., 2015; Bhattacharjee et al., 2016; Francis and Sorg, 2016). C. difficile packages three subtilisin-like serine proteases proteins, CspA, CspB, and CspC, into the spore. In C. difficile, CspB, and CspA are encoded as a cspBA gene fusion, where the CspA portion of CspBA lacks an intact catalytic triad (Adams et al., 2013; Kevorkian et al., 2016). The CspBA fusion protein undergoes interdomain cleavage during spore formation, leading to the separation of CspB and CspA, which are transported into the spore by unknown mechanisms. cspC is encoded downstream of cspBA and, similar to cspA, encodes an incomplete catalytic triad. Despite the loss of apparent catalytic activity, cspC (and cspA) important for C. difficile spore germination. Interruption of the cspC coding region through ethyl methanesulfonate (EMS)-generated SNPs and TargeTron methods abrogates spore germination, and certain SNPs in the cspC sequence also affect germinant specificity (Francis et al., 2013a). Similarly, though cspA lacks an intact catalytic triad, cspA is essential for spore germination by controlling the levels of CspC into the developing spore (Francis et al., 2013a; Kevorkian et al., 2016). Only CspB contains an intact catalytic triad and,

FIGURE 3 | Regulation pathways of *C. difficile* spore germination. This figure was drawn based on the references (Paredes-Sabja et al., 2011; Fimlaid et al., 2013). Regulation of box GerG, GerS, CspA, and SpoVAC texted in red were drawn in this figure based on the recent advances in *C. difficile* spore germination. GerS, GerG, and SpoVAC proteins were characterized by Shen group, recently (Fimlaid et al., 2015; Donnelly et al., 2016, 2017). CspC-germinant receptor and completion of germination were scheduled in the orange boxes. Black arrows indicate the regulatory relationship between the factors has been confirmed, dashed arrows indicate the regulatory relationship between the factors has not been tested. Thick black/red arrow indicates central signal pathway in germination progress. Question marks indicate that there is suggestive, but no conclusive experimental evidence.

thus, is hypothesized to be important for activating the SCLE, pro-SleC, to its active, cortex-degrading form (Kevorkian et al., 2016).

Activation of the cortex hydrolase SleC depends on the CspB protease, which cleaves the N-terminal pro sequence from the protein. Activated SleC degrades the cortex leading to CaDPA release from the spore core in response to osmotic swelling sensed at the inner spore membrane as a result of cortex degradation (Francis and Sorg, 2016). The osmotic pressure at the inner spore membrane is regulated by SpoVAC (a mechanosensing protein), which allows CaDPA release from the core (Velásquez et al., 2014; Donnelly et al., 2016; Francis and Sorg, 2016). Strikingly, inactivation of either the CspC or SleC inhibited cortex degradation and CaDPA release. These results suggest that the CspC is required for CaDPA release and that cortex degradation precedes CaDPA release, opposite to what occurs in B. subtilis (Francis and Sorg, 2016). These studies suggest that the process of C. difficile spore germination appears to occur in an outside-in manner, while in B. subtilis, the signal appears to travel from the inside-out (Francis and Sorg, 2016).

## GerG and GerS Regulators of Spore Germination

Recently, GerG and GerS were identified as important players in C. difficile spore germination (**Figure 3**; Fimlaid et al., 2015; Donnelly et al., 2017). Donnelly et al. identified the C. difficilespecific protein GerG as an important player in the C. difficile germination process (Donnelly et al., 2017). A deletion of C. difficile gerG resulted in spores with germination defects and reduced responsiveness to bile salt germinants. This phenotype was likely due to the decrease in the incorporation of the CspC, CspB, and CspA germination proteins into spores; this phenotype could be complemented in trans. Similarly, Fimlaid et al. identified another regulator of C. difficile spore germination using TargeTron-based gene disruption (Fimlaid et al., 2015). The GerS lipoprotein functions as a critical regulator in C. difficile spore germination (Fimlaid et al., 2015). In this study, the gerS mutant has a severe germination defect and fails to degrade cortex; this phenotype could be complemented in trans. Interestingly, C. difficile gerS mutant spores still cleave pro-SleC to its active form, suggesting that either cortex is not appropriately modified for SleC-recognition or that SleC is bound to other proteins that GerS regulates (Fimlaid et al., 2015). Importantly, loss of GerS attenuated the C. difficile virulence in the hamster infection model (Fimlaid et al., 2015). Because GerG and GerS are found exclusively in C. difficile, GerG and GerS proteins could be the potential targets to develop C. difficilespecific anti-infective therapies.

## Activators and Inhibitors of *C. difficile* Spore Germination

Bile-acid mediated germination is essential for C. difficile spore germination and CDI in mammalian GI tract. Bile acids are the end products of cholesterol metabolism in liver and are essential for lipoprotein, glucose, drug, and energy metabolism (Chiang, 2009; Howerton et al., 2011). In humans, cholic acid (CA) and chenodeoxycholic acid (CDCA) are two main primary bile acids (PBAs) that are conjugated with either taurine or glycine. Though most of the bile acids secreted into the gut are reabsorbed and recycled back to the liver to be used in other rounds of digestion, some escape hepatic recirculation and enter the large intestine where they become acted upon by the colonic microbiome. Here, the conjugated bile acids become deconjugated due to the action of bile salt hydrolases that are expressed on the cell surfaces of many different bacteria. Subsequently, a small subset of the colonic microbiome will take up and 7α-dehydroxylate the PBAs to form secondary bile acids (SBAs; Ridlon et al., 2006). About 50 different chemically distinct SBAs [e.g., deoxycholate (DCA), lithocholate (LCA), ursodeoxycholate (UDCA), isodeoxycholate (iDCA), and isolithocholate (iLCA)] can be found in human large intestine (Setchell et al., 1983). Recently, Thanissery et al. have analyzed the impact of gut microbial derived SBAs on C. difficile life cycle, specifically, the differences in inhibition efficiency of spore germination, growth, and toxin activity among of DCA, iDCA, LCA, iLCA, UDCA, ωMCA, and HDCA in clinically relevant C. difficile strains R20291 and CD196 (ribotype 027), M68 and CF5 (017), 630 (012), BI9 (001), and M120 (078) (Thanissery et al., 2017). Not surprisingly, the authors found these cholic acid- and chenodeoxycholic acid-derivatives all impacted the C. difficile life cycle; the sensitivity varied by strain and SBA.

Although bile acids are essential to activate C. difficile spore germination, they are not sufficient to activate germination on their own. Amino acid co-germinants are also required for spore germination (Sorg and Sonenshein, 2008; Howerton et al., 2011; Shrestha and Sorg, 2017; Shrestha et al., 2017). However, different amino acids function as co-germinants with different spore germination efficiencies. Glycine is the most effective cogerminant in C. difficile, while alanine is most-often used as co-germinant in B. subtilis and other organisms. In B. subtilis, L-alanine interacts with the GerAA-AB-AC germinant receptor to trigger CaDPA release from the spore core and subsequent cortex hydrolysis. However, D-alanine competitively-inhibits Lalanine-mediated spore germination in B. subtilis (Yasuda and Tochikubo, 1984). In C. difficile, L-alanine can also function as a co-germinant with TCA to stimulate spore germination (Shrestha et al., 2017). Though D-alanine is unable to inhibit Lalanine-mediated C. difficile spore germination, unlike what is observed in B. subtilis, D-alanine can work as a co-germinant to trigger C. difficile spore germination in defined medium (Shrestha and Sorg, 2017; Shrestha et al., 2017). In order for D-alanine to function as a good co-germinant, an alanine racemase (Alr2) should be present in the C. difficile spore. Alr2 interconverts L-alanine and D-alanine (Shrestha et al., 2017). Interestingly, C. difficile Alr2 can also interconvert L- and Dserine, and both of these amino acids can act as co-germinants for C. difficile spore germination (Shrestha et al., 2017). Building on this work, Shrestha et al. found that many different amino acids are co-germinants when tested at 37◦C (Shrestha and Sorg, 2017). In this work, two different C. difficile strains responded to a hierarchy of amino acid co-germinants. For UK1 and M68 strains, glycine was the most effective co-germinant (EC<sup>50</sup> = ∼200µM) and L-alanine, taurine, and L-glutamine were also good co-germinants (Shrestha and Sorg, 2017). Interestingly, amino acids that regulate important physiological processes were not co-germinants (L-isoleucine, L-leucine and L-valine).

Recently, Kochan et al. identified a critical role for Ca2<sup>+</sup> during C. difficile spore germination (Kochan et al., 2017). In their study, they found that C. difficile spores cannot germinate in rich medium supplemented with TCA but without Ca2+, indicating that Ca2<sup>+</sup> is indispensable for spore germination. The authors suggested that it works together with glycine to stimulate germination; however, Ca2<sup>+</sup> may play a role in the activity of the CspB serine protease, the CspC germinant receptor, the CspA pseudoprotease, or in the activity of the cortex hydrolase. Other subtilisin-like proteases require Ca2<sup>+</sup> for activity (Siezen and Leunissen, 1997) and some cortexdegrading enzymes also require Ca2+. Though no Ca2<sup>+</sup> was found in the CspB crystal structure, the structures of CspC and CspA have yet to be determined. Thus, Ca2<sup>+</sup> may not function as a co-germinant with glycine, but, rather, as an essential cofactor for C. difficile spore germination. However, and importantly, the role of Ca2<sup>+</sup> during C. difficile spore germination was also verified in the murine model. Ex vivo assays with mouse ileal contents that were depleted with chelex resin (to remove Ca2+) did not support germination of C. difficile spores (Kochan et al., 2017). This work provided a novel potential strategy for CDI control by modulating intestinal Ca2<sup>+</sup> concentration.

In summary, although several main components of spore sporulation/germination machinery of C. difficile have been identified and characterized, several questions remain regarding how C. difficile decides when to enter the sporulation pathway. Moreover, though the Csp pseudoproteases are important for germination, how they interact with and transmit the bile acid signal are still unknown. Further detailed work is necessary to characterize these important aspects of C. difficile physiology.

## TREATMENTS OF CDI BASED ON SPORULATION/GERMINATION

Currently, the standard treatment of CDI is the use of vancomycin, metronidazole, or fidaxomicin, each of which has some level of recurring disease due to the continued insult to the colonic microbiome and the presence of spores within the colon/environment (Allen et al., 2013). To meet this challenge, non-antibiotic and immune-based therapies against CDI have been developed, such as anti-toxins, vaccines, fecal microbiota transplant (FMT), and anti-germination-based compounds (Gerding et al., 2008; Howerton et al., 2013; Kociolek and Gerding, 2016). Many anti-toxins and vaccines for CDI have been developed in the past two decades (Cox et al., 2013; Monteiro et al., 2013; Mathur et al., 2014; Zhao et al., 2014; Wang Y. K. et al., 2015; Yang et al., 2015; Qiu et al., 2016). Though these treatments can effectively decrease the morbidity and mortality of CDI, most of the anti-toxins and vaccines cannot suppress C. difficile colonization and kill C. difficile spores. Therefore, with these treatments, there are still risks of potential CDI relapse in the host.

Instead of merely neutralizing C. difficile toxins in host, strategies which can directly decrease C. difficile colonization, kill the vegetative cells, and suppress sporulation/germination are desirable treatments for CDI. FMT is an effective strategy to reconstruct the gut microbiota to suppress C. difficile colonization, especially for patients who have multiple bouts of recurring disease and who have failed conventional treatment methods (Borody and Khoruts, 2012; Weingarden et al., 2014; Khoruts and Sadowsky, 2016; Kim et al., 2016). Although FMT is deemed relatively safe and low-cost, the unappealing aesthetics of the procedure is often a concern of patients (Sampath et al., 2013; Varier et al., 2015). Because the C. difficile spore form is necessary for dissemination and persistence, sporulation/germination are critical steps for CDI. Thus, it is worth developing therapeutic strategies for disrupting C. difficile disease transmission and spread according to C. difficile spore biology. Basing on the progress of C. difficile spore germination, the PBA CDCA and secondary bile acids LCA, UDCA, and iLCA are potent inhibitors of C. difficile spore germination (Sorg and Sonenshein, 2010; Zhang and Klaassen, 2010; Heeg et al., 2012). Moreover, several mouse-derived bile acids, such as α-muricholic acid, β-muricholic acid, and ω-muricholic acid inhibit C. difficile spore germination and growth (Francis et al., 2013b). Excitingly, synthesized bile acid analogs, such as CAmSA, methylchenodeoxycholic acid diacetate, and compound 21b (derived from UDCA) have been identified to inhibit C. difficile spore germination (Sorg and Sonenshein, 2009; Howerton et al., 2013; Stoltz et al., 2017). Of these compounds, CAmSA showed promise in inhibiting/delaying C. difficile disease in a mouse model of CDI. These anti-germinationbased strategies could work in a couple of different ways. (i) High risk patients who are to be treated with antibiotics could also take an anti-germinant to prevent the germination of spores within the host's gut. This patient continues to take the anti-germinant during and post-antibiotic treatment so that the normal, colonic, microbiome has a chance to repopulate and provide natural protection against CDI. (ii) Patients with CDI could take the recommended course of antibiotics plus the anti-germinants. This strategy would prevent recurring disease by allowing the microbiome to re-establish colonization resistance post-antibiotic treatment. Because both strategies block germination, and thus downstream events (vegetative growth, toxin production, and spore formation), anti-germination therapy would limit the presence of spores within the surrounding environment because C. difficile would not have a chance to expand in population and produce spores. In contrast germination-inducing strategies are a viable option for environmental cleanup; inducing in vivo germination has the potential for toxin-production and, thus, exacerbation of symptoms. Due to the inherent nature of the dormant spore, harsh chemicals (e.g., bleach) are required to clean environmental surfaces. But by germinating the spores in the environment, the germinated spores become susceptible to a wider range of sanitizing agents (Nerandzic and Donskey, 2010, 2013, 2017; Nerandzic et al., 2016). More studies should be investigated for further application of germination inhibitors.

#### REFERENCES


## CONCLUDING AND REMARKS

Although much has been learned about the sporulation/germination processes of C. difficile and the different therapeutic strategies for CDI, many key questions related to regulation pathways of sporulation/germination processes remain unanswered. Thus, much work remains to be done to further understand C. difficile spore biology and develop new efficient approaches for CDI treatment: (1) It is expected that further work will allow us to fully understand the mechanisms of the initiation of sporulation by identifying the proteins that are involved in Spo0A phosphorylation; (2) Due to the relevance of spore germination with CDI progression, it is worth defining how the bile acid germinant receptor, CspC, and the unidentified glycine germinant receptor regulate CaDPA release and cortex degradation; (3) More alternative therapeutic strategies for CDI disease need to be developed based on the knowledge of C. difficile sporulation/germination.

## AUTHOR CONTRIBUTIONS

All authors listed, have made a substantial, direct, and intellectual contribution to the work; DZ wrote and revised this manuscript; JS and XS revised this manuscript.

### ACKNOWLEDGMENTS

This work was supported in part by National Institutes of Health grants (K01-DK092352, R21-AI113470, R03-DK112004, R01-AI132711) to XS, and was also supported by awards 5R01AI116895 and 1U01AI124290 to JS from the National Institute of Allergy and Infectious Diseases.


monoclonal antitoxin antibodies actoxumab and bezlotoxumab. Infect. Immun. 83, 822–831. doi: 10.1128/IAI.02897-14


**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 Zhu, Sorg and Sun. 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.

# Identification and Characterization of Blood and Neutrophil-Associated Microbiomes in Patients with Severe Acute Pancreatitis Using Next-Generation Sequencing

#### Qiurong Li\* † , Chenyang Wang† , Chun Tang† , Xiaofan Zhao, Qin He and Jieshou Li

*Research Institute of General Surgery, Jinling Hospital, Medical School, Nanjing University, Nanjing, China*

#### Edited by:

*Michele Marie Kosiewicz, University of Louisville, United States*

#### Reviewed by:

*Renate Lux, UCLA School of Dentistry, United States Michael F. Minnick, University of Montana, United States*

> \*Correspondence: *Qiurong Li liqiurongjue@126.com*

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

Received: *05 May 2017* Accepted: *09 January 2018* Published: *23 January 2018*

#### Citation:

*Li Q, Wang C, Tang C, Zhao X, He Q and Li J (2018) Identification and Characterization of Blood and Neutrophil-Associated Microbiomes in Patients with Severe Acute Pancreatitis Using Next-Generation Sequencing. Front. Cell. Infect. Microbiol. 8:5. doi: 10.3389/fcimb.2018.00005* Infectious complications are a leading cause of death for patients with severe acute pancreatitis (SAP). Yet, our knowledge about details of the blood microbial landscape in SAP patients remains limited. Recently, some studies have reported that the peripheral circulation harbors a diverse bacterial community in healthy and septic subjects. The objective of this study was to examine the presence of the blood bacterial microbiome in SAP patients and its potential role in the development of infectious complications. Here we conducted a prospective observational study on a cohort of 50 SAP patients and 12 healthy subjects to profile the bacterial composition in the blood. The patients were subgrouped into uninfected (*n* = 17), infected (*n* = 16), and septic (*n* = 17) cases. Applying 16S rDNA-based next-generation sequencing technique, we investigated blood and neutrophil-associated microbiomes in SAP patients, and assessed their connections with immunological alterations. Based on the sequencing data, a diverse bacterial microbiota was found in peripheral blood and neutrophils from the healthy and SAP subjects. As compared to healthy controls, the blood and neutrophil-associated microbiomes in the patients were significantly altered, with an expansion in Bacteroidetes and Firmicutes as well as a decrease in Actinobacteria. Variations in the microbiome composition in patients were associated with immunological disorders, including altered lymphocyte subgroups, elevated levels of serum cytokines and altered proteomic profiles of neutrophils. However, no significant compositional difference was observed between the patient subgroups, implying that the microbiota alterations might not be linked to presence/absence of infectious complications in SAP. Together, we present an initial description of the blood and neutrophil-associated bacterial profiles in SAP patients, offering novel evidence for the existence of the blood microbiome. Identification of the blood microbiome provides novel insights into characteristics and diagnostics of bacteremia in the patients. Further study is required to assess the possible implications of the blood microbiome in health and diseases.

Keywords: blood microbiome, neutrophil-associated microbiome, severe acute pancreatitis, sepsis, nextgeneration sequencing, proteomics

## INTRODUCTION

Severe acute pancreatitis (SAP) is a terrible disease, associated with a mortality rate in the range of 20–50% (Nathens et al., 2004; Garg et al., 2005; Noor et al., 2011). Among the cases with SAP, up to 80% of deaths are attributed to infectious complications and multiple organ dysfunction syndromes (MODS) (Garg et al., 2005). Despite antibiotic prophylaxis in the management of SAP, the incidence of systemic infections is still surprisingly high (Villatoro et al., 2010). Infectious complications have become a major concern in SAP, especially cases of pancreatic necrosis (Medich et al., 1993). Accumulating evidence has suggested that systemic infections in SAP patients are mainly derived from invasion by gut organisms (Schmid et al., 1999; Noor et al., 2011). Yet, the molecular mechanisms behind the development of systemic infections in SAP are not fully known.

During the last two decades, bacterial translocation has been thought to be a major source causing systemic infection and MODS in SAP patients (MacFie et al., 1999; Cicalese et al., 2001). Enteric bacteria could cross the impaired intestinal barrier to reach peripheral circulation, leading to infected pancreatic necrosis and sepsis (Ammori et al., 1999). Just like in patients who underwent sepsis following severe trauma, burn, major operative intervention and other causes, cultures of blood specimens in SAP patients complicated by sepsis are often negative, even in the presence of infected pancreatic necrosis (Sainio et al., 1995; Ammori et al., 2003). As a result, specific interventions against infections are probably delayed in some cases, causing lethal complications. It is quite possible that enteric bacteria may translocate into systemic circulation, but escape from detection by standard culture methods. Utilization of new techniques, such as polymerase chain reaction or matrix-assisted laser desorption ionization–time of flight, to some extent, has improved the ability to detect bloodstream pathogens (Ecker et al., 2010; Buehler et al., 2016), however, our knowledge on the blood-microbial landscape in septic patients is still limited. Appling a cultureindependent method, we observed that bloodstream invasion by multiple gut organisms (commonly 5–8 bacterial species) contributed to the development of bacteremia in SAP patients (Li et al., 2013a). Owing to insensitivity of this approach to low-abundance microbial sequences (Muyzer et al., 1993), the bacterial taxonomic richness in the blood of the patients was most likely underestimated. As a result, a deeper exploration of the blood bacterial composition and diversity with emerging molecular techniques might be needed. Recent studies with 16S rDNA-based high-throughput sequencing showed that a diverse microbiota is present in the blood of septic patients (Grumaz et al., 2016; Gosiewski et al., 2017) and healthy individuals (Païssé et al., 2016). Although its biological and clinical significance remains to be further explored, discovery of a blood microbiome might represent an important step toward a better understanding of the microbial world of the human body and its relationships with health and diseases. Therefore, it is urgently needed to ascertain whether a rich microbiota is harbored in blood of SAP patients using culture-independent next-generation sequencing techniques to better understand the development of bacteremia and infectious complications.

The complex network of immune cells and specialized molecules has evolved to defend against pathogens. Various types of immune cells, including neutrophils, monocytes, and lymphocytes, could integrate microbial signals to govern the inflammatory response, together maintaining the homeostasis of systemic circulation (Akira et al., 2006). Recent studies have revealed that sepsis is probably the sequelae of a cascade of events starting with a local inflammatory response against organisms derived from the gut (Bosmann and Ward, 2013). When pathogens invade, the innate immune system recognizes microbial molecules and kicks off an inflammatory response (Akira et al., 2006). Activation of neutrophils could induce excessive production of pro-inflammatory cytokines and disrupt the balance of the pro- and anti-inflammatory response, leading to an overwhelming imbalance (Delano et al., 2011; Bosmann and Ward, 2013). As the first-line responder against pathogens, the neutrophils play a central role for elimination of bacterial infection. Some pathogens can be engulfed and gain entry to the cells (Appelberg, 2007), shaping the neutrophil-associated microbiomes (NAMs). The functional impairment of neutrophils could disrupt the dynamic balance between internalization and clearance of pathogens (Amulic et al., 2012; Hotchkiss et al., 2013; Matthew et al., 2016), likely altering the membership of the neutrophil-associated microbiome and causing infection. However, the composition of the community in neutrophils and its role in sepsis remains uncharacterized. Elucidation of changes of the NAMs and the potential connection with host immunological disorders would be helpful for better understanding the mechanism of sepsis pathogenesis in SAP patients.

Here we performed 16S rDNA-based sequencing on the blood and neutrophils of SAP patients to profile the microbial landscape in peripheral circulation and estimated its potential connection with the development of bacteremia and infectious complications. In addition, we dissected the immune cell repertoires in blood and the proteomic profiles of neutrophils through a fluorescence activated cell sorting (FACS) approach and quantitative proteomics analysis, and examined the relationships of blood microbiota alterations with immunological disorders in SAP patients.

#### MATERIALS AND METHODS

#### Ethics Statement

All participants provided written informed consent upon enrollment. Studies were approved by the Human Subjects Institutional Committee of Jinling Hospital and were conducted in accordance with all relevant guidelines and regulations.

#### Study Populations and Experimental Design

Fifty patients who underwent SAP and admitted to the Department of General Surgery, Jinling Hospital, between March 2014 and March 2016, were enrolled in this study. Acute pancreatitis was diagnosed in accordance with clinical symptoms and at least 3 times the upper limit of normal value in serum amylase or evidence on computed tomographic scan of the abdomen (Bradley, 1993). SAP is defined as the presence or absence of organ failure and/or local complications, such as pancreatic necrosis, abscess or pseudocyst (Bradley, 1993). Based on the presence or absence of infectious symptoms, the patients were distributed into three subgroups: uninfected (n = 17), infected (n = 16) and septic (n = 17) (**Table 1**). Septic patients were identified according to the presence of suspected or documented infections and an acute increase in the Sequential (Sepsis-related) Organ Failure Assessment (SOFA) score of 2 points or more (Singer et al., 2016). The infected patients were defined as having suspected or documented infections but an absence of emerging organ dysfunction, and uninfected cases showing no infectious signs. The patients' clinical characteristics, such as demographic data, clinical diagnoses, comorbidities, vital signs, hematologic and chemical data, blood gas analyses, blood cultures, Acute Physiology and Chronic Health Examination-II (APACHE-II) score and SOFA scores, were recorded (Supplementary Table 1). Twelve healthy volunteers who had no infectious signs and no elevated level of serum CRP served as control subjects. Subjects <18 or more than 70 years old were excluded in this study. Blood samples from enrolled patients and healthy subjects were collected for high-throughput sequencing, quantitative proteomics analysis of peripheral neutrophils, and measurement of blood immune cell subpopulations.

#### Sampling and Neutrophil Isolation

Peripheral blood was sterilely collected at the days in which sepsis was definitely diagnosed and immediately delivered to our laboratory. The sample was then divided into 3 aliquots of 200 µL in the biosafety cabinet and stored at −80◦C. Another portion (2 mL) of each sample was immediately used for isolation of neutrophils with commercially Ficolldextran reagents. Neutrophil-rich pellets were subjected to hypotonic lysis of the remaining erythrocytes with E-lysis. Cell pellets were resuspended in DMEM supplemented with 10% FCS (heat inactivated). Cells were incubated in polypropylene tubes (Falcon/Becton Dickinson, Cambridge, UK) to prevent adherence. The purity of neutrophils was >95%, as assessed by CD16<sup>+</sup> cell by flow cytometry. Isolated neutrophils were stored at −80◦C until DNA extraction and proteomics analysis.

#### DNA Extraction and Polymerase Chain Reaction (PCR)

DNA from whole blood and isolated neutrophils was extracted with the QIAamp DNA Mini Kit (Qiagen, Valencia, CA) following the manufacturer's instructions. For extraction of neutrophil DNA, a bead-beating step (FastPrep machine for 45 s at setting 5.5, Bio 101) after the addition of the RLT buffer was done to enhance cell lysis. The hypervariable V3 region of the 16S rRNA gene was amplified using universal primer set 357f (5′ -TACGGGAGGCAGCAG-3′ ) and 518r (5′ - ATTACCGCGGCTGCTGG-3′ ) (Li et al., 2013b). An aliquot of DNA (100 ng) recovered from the blood and neutrophils was added into a reaction mixture, and PCR reactions were carried out with a touchdown thermocycling program. The cycling was as follows: initial denaturation at 94◦C for 5 min, 30 s at 94◦C (denaturation), 30 s at 65◦C (annealing), and 30 s at 68◦C (elongation) with a 0.5◦C touchdown every second cycle during annealing for 20 cycles, followed by 15 cycles with an annealing temperature of 56◦C and a final cycle consisting of 5 min at 68◦C. The purity and correct size of the resulting PCR amplicons (approximately 190 bp) were assessed on 1% agarose gels, stained with ethidium bromide (5µg/mL) and visualized under UV light. To ascertain the specificity of the primers (no eukaryotic, mitochondrial, or Archea DNA targeted), we sequenced the PCR products (∼10 ng each sample) by Sanger technologies. The sequencing results showed that only 16S rDNA bacterial fragments were yielded from the amplification, confirming the specificity of the primers.

To ensure absence of false positive amplifications, we conducted real-time quantitative PCR with the primer set (357f/518r) to test for possible bacterial contaminants from the reagents (PCR mixtures, solutions for DNA extraction and sterile water) and consumables. The standard curve for the quantification was performed by generating a series of 10-fold dilutions from 5 × 10<sup>3</sup> to 5 × 10<sup>8</sup> of 16S rRNA gene copies per reaction using plasmid DNA containing the complete 16S rRNA gene sequence of an E. coli DH5α strain. Amplifications of samples and standard dilutions were performed in duplicates on the AB7500 real time PCR system (Life Technologies, CA). The quality of the amplifications was assessed by melting curves. The data showed that the background signal, represented by negative controls (NC), was far lower than those of blood samples (Supplementary Figure 1), indicating the absence of bacterial DNA contaminants from reagents and consumables.

#### DNA Library Construction and 16S rRNA Gene Sequencing

PCR products were purified with Agencourt AMPure beads (BeckmanCoulter). An aliquot (50 ng) of purified DNA was used for construction of barcoded libraries using the Ion Plus Fragment Library Kit (Life Technologies). In this step, a samplespecific "DNA molecular tag (barcode)," a 14-base semirandom sequence, was intended to uniquely identify original template molecules. DNA concentrations of the libraries were estimated with a Qubit dsDNA HS kit. Libraries for each run were diluted to 26 pM for template preparation. Emulsion PCR was carried out using the Ion OneTouchTM 200 Template Kit v2 (Life Technologies). Sequencing of amplicon libraries was conducted on 316 v2 chips using the Ion Torrent PGM system with the Ion Sequencing 200 kit (Life Technologies). After sequencing, the individual sequence reads were filtered by the PGM software to remove low quality and polyclonal sequences. All PGM qualityapproved, trimmed and filtered data were exported as FASTQ files.

The sequence data were deposited in the NCBI Bioproject and the Sequence Read Archive with accession codes PRJNA428535 and SRP128069, respectively.

#### Sequence Processing and Quality Control

Sequenced files were online converted to FASTA format and were filtered to remove low-quality sequences with the Galaxy

#### TABLE 1 | Demographics of study population.


*N/A, no available.*

projects (https://usegalaxy.org/). A mean quality score of ≥20 and a minimum length of 150 bp for the coupled V3 region were required. The resulting sequences were then aligned online to operational taxonomic units (OTUs) (97% identity) with the CD-HIT (http://weizhong-lab.ucsd.edu/cdhit\_454/cgi-bin/index.cgi) (Li and Godzik, 2006). Sequences that did not match the defined core region of the seed alignment were manually removed. OTUs were classified taxonomically to bacterial genera using the Ribosomal Database Project (RDP) classifier with a 50% bootstrap threshold (http://rdp.cme.msu.edu/classifier/classifier. jsp) (Wang et al., 2007).

### Protein Extraction, Digestion, and iTRAQ Labeling

Isolated neutrophils were thawed and resuspended in lysis buffer of cold acetone containing 10% trichloroacetic acid (TCA) and 10 mM dithiothreitol (DTT) and sonicated for 3 times on ice to enhance cell lysis. The protein pellets were then collected by centrifuging and resuspended in a buffer (7 M urea, 2 M thiourea, 4% CHAPS, 30 mM, Tris–HCl pH 8.0), containing 1 mM phenylmethylsulfonyl fluoride (PMSF), 2 mM ethylenediaminetetraacetic acid (EDTA) and 10 mM DTT. Samples were again sonicated and centrifuged and subsequently, the supernatant was reduced and alkylated by 10 mM DTT and 55 mM iodoacetamide (IAA). The treated proteins were precipitated with chilled acetone (1:4) at −20◦C overnight. The precipitants were resuspended in 500 mM triethylammonium bicarbonate (TEAB), then sonicated and centrifuged as above. The protein content of the supernatant was determined using the Bradford method. The resulting proteins (∼100 µg) of each sample were digested by and then labeled with 6-plex iTRAQ reagents containing 6 different stable-isotope (126–131) covalent mass tags (Applied Biosystems) according to the manufacturer's protocol.

## Peptide Fractionation and Mass Spectrometry (MS) Analysis

The labeled peptides were pooled, eluted and resolved into 10 fractions using an Ultremex SCX column containing 5 µm particles (Phenomenex, USA). The eluted fractions were desalted using a Strata X C18 column (Phenomenex, USA) and dried under vacuum. Each fraction was resuspended in a certain volume of buffer A (2% acetonitrile, 0.1% formic acid, pH3.0) and centrifuged at 20,000 × g for 10 min. The final concentration of peptide was about 0.5 µg/µl on average in each fraction. Supernatant was loaded on a Nano ACQUITY UPLC system using the autosampler. The peptides were subjected to nanoelectrospray ionization followed by tandem mass spectrometry (MS/MS) in a LTQ-Orbitrap (Thermo Fisher Science, USA) coupled online to the HPLC. Intact peptides were detected in the Orbitrap at a resolution of 60,000. Peptides were selected for MS/MS using high energy collision dissociation (HCD) operating mode with a normalized collision energy setting of 45%. LC-MS/MS was operated in positive ion mode as described. The analytical condition was set at a linear gradient from 0 to 60% of buffer B (CH3CN) in 150 min, and a flow rate of 200 nL/min. Ion fragments were detected on the LTQ. A data-dependent procedure that alternated between one MS scan followed by eight MS/MS scans was applied for the eight most abundant precursor ions above a threshold ion count of 5,000 in the MS survey scan. The electrospray voltage applied was 1,500 V. Automatic gain control (AGC) was used to prevent overfilling of the ion trap; 1 × 10<sup>4</sup> ions were accumulated in the ion trap for generation of HCD spectra. For MS scans, the m/z scan range was 350 to 2,000 Da.

### MS Data Processing, Protein Quantization, and Functional Annotation

The MS/MS spectra acquired from precursor ions were submitted to Maxquant (version 1.2.2.5) using the following search parameters: the database for the search was Uniprot proteome (version 20140418); the enzyme was trypsin (full cleavage); dimethylation labeling for quantification; the dynamic modifications were set for oxidized Met (+16); carbamidomethylation of cysteine was set as static modification; MS/MS tolerance was set at 10 ppm; the minimum peptide length was 6; the false detection rates for both peptides and proteins were all set below 0.01. All identified peptides had an ion score above the identity threshold, and a protein was considered identified if at least one such unique peptide match was apparent for the protein. Individual quantitative samples were normalized within each acquisition run according to the algorithm described in i-Tracker (Shadforth et al., 2005). The logic algorithm for set operations was applied to further screen for differentially expressed proteins identified in the present study. Gene Ontology (GO) functional annotation was carried out using Blast2GO software (Conesa et al., 2005).

#### Flow Cytometry

Blood specimens were freshly collected for assessment of lymphocyte subsets, cell apoptosis, and HLA-DR expression on monocytes. For measurement of cell apoptosis, mononuclear cells were prepared by density gradient centrifugation with Ficoll-Paque (Stem cell Technologies), and were then stained by Annexin V-PE, 7-ADD-PerCP, and fluorescein-labeled mAbs against CD4, or CD8, respectively. Assessment of T lymphocyte subsets, HLA-DR and T-helper (Th) cells was performed using commercial kits (BD Biosciences). After activation with phorbol myristate acetate and ionomycin, immunostaining was performed using fluorescein-labeled mAbs against CD4, interferon-gama (INF-γ), interleukin (IL)-4 and IL-17 (BD Pharmingen). A logical gate combining CD4<sup>+</sup> cells and their scatter properties was used for the phenotypes of Th1, Th2, and Th17 cells. The proportions of the peripheral naïve CD4 T cells (CD3+CD4+CCR7+CD45RAhighCD28+), naïve CD8 T cells (CD3+CD8+CCR7+CD45RAhighCD28+), memory CD4 T cells (CD3+CD4+CD45RA−), memory CD8 T cells (CD3+CD8+CD45RA−), effector memory (CD3+CD8+CCR7−CD45RA−), and terminal effector memory (CD3+CD8+CCR7−CD45RAhighCD28−) were also measured, respectively (Qi et al., 2014). All data were acquired on a Becton Dickinson FACSCanto II and analyzed with CellQuest software (BD Biosciences).

#### Assay of Serum Cytokines

Measurement of serum TNF-α, IL-1β, IL-6, IL-10, IL-18, and IFN-γ levels was performed in duplicate with enzyme-linked immunosorbent assay (ELISA) kits (R&D Systems, Abingdon, UK).

#### Statistical Analysis

Quantitative data are represented as means ± standard deviation (SD). Statistical analysis was performed by one-way analysis of variance (ANOVA) followed by the Holm-Sidak test using SPSS software (version 12.0). A P < 0.05 was considered significant. The species richness in the bacterial microbiota was estimated by the OTU numbers at the same sequencing depth, which was compared to reflect the difference of the microbiota diversity between groups. Correlation between two variances was estimated using linear regression analysis with a Pearson's test in R software (http://www.r-project.org/). Heatmaps were generated for non-scaled, non-normalized titer data using a Euclidean distance function with complete linkage clustering or non-clustering in R using the package pheatmap (version 3.1.1). Principal component analysis was conducted with Canoco software for Windows 4.5 (Microcomputer Power, Ithaca, NY). The output matrix containing the relative abundance of OTUs per sample was processed with the linear discriminant analysis effect size (LEfSe) algorithm (Segata et al., 2011) using an alpha cutoff of 0.05 and an effect size cutoff of 2.0.

## RESULTS

## Characterization of Bacterial Microbiomes in Peripheral Blood of SAP Patients

To characterize the bacterial communities possibly present in systemic circulation, we sequenced 16S rRNA gene amplicons of blood samples from 50 patients with SAP and 12 healthy subjects. The patients were classified into three groups: uninfected (n = 17), infected (n = 16), and septic (n = 17) (**Table 1** and Supplementary Table 1). Prior to sequencing, we conducted a quantitative polymerase chain reaction (qPCR) assay to determine the concentration of 16S rRNA gene copies in the whole blood of each subject (Supplementary Figure 1). The numbers of 16S rDNA copies in blood were significantly higher in the infected and septic patients than in healthy controls (on average 1.38 × 10<sup>8</sup> copies per milliliter) (P < 0.01) (Supplementary Figure 1D). After sequencing the 16S rDNA amplicons, we generated over 80,000 sequences for each sample, and unique sequences were clustered into 14,526 OTUs (97% ID). Taxonomic classification showed that these OTUs were assigned into 335 bacterial genera, mainly affiliated with the four major phyla: Proteobacteria, Bacteroidetes, Firmicutes and Actinobacteria (Supplementary Figure 2A). An average of 204 OTUs (range: 50–694) were obtained from the patients' samples (Supplementary Figure 2B), indicating that a diverse bacterial microbiome, rather than one or several bacterial species as thought previously, was present in peripheral circulation of the patients. Surprisingly, we also observed a highly diverse microbiota in the blood of healthy subjects (on average 360 OTUs, range: 265–520) (Supplementary Figure 2B). As compared to healthy controls, the species richness of the blood microbiota, estimated by the number of OTUs (at the same sequencing depth), was significantly reduced in SAP patients (P < 0.05) (Supplementary Figure 2B). Additionally, the composition of the microbiota was profoundly distinct between the patients and healthy controls, characterized by severe depletion of Actinobacteria (P < 0.0001) and an overgrowth of Bacteroidetes in the former (P < 0.05) (Supplementary Figure 2C). At the class level, we saw a marked increase in Bacteroidia and Clostridia, and a striking reduction in Actinobacteriae, Flavobacteriia and Bacilli in the patients vs. healthy controls (P < 0.05) (Supplementary Figure 2C). Principal component analysis (PCA) of the weighted UniFrac distances, based on the data of genus-level relative abundance, showed a clear separation of the patients' samples from healthy controls along PC1 and PC2, indicating the differences of the microbiota profiles between them (Supplementary Figure 2D). However, no significant difference was found in the microbiota structures across the patient groups, as scattered distribution of their sample dots in the PCA plot.

### Potential Source of the Blood Microbiota in SAP Patients

To track the possible source of blood microbiomes, we compared our sequences to the 16S rRNA gene dataset from the National Center for Biotechnology Information (NCBI) (The Human Microbiome Project Consortium, 2012). Consequently, an average of 87.0% (range: 73.5–98.3%) of the blood microbiome memberships across individuals were taxonomically classified as known commensal or pathogenic bacteria colonizing the human gut, far higher than those from other sites (Supplementary Figure 3A). Given the dominance of the putative gut-derived organisms within the niches, we estimated their contribution to the microbiome-wide alterations in SAP patients (**Figure 1A**). The counts of the OTUs likely affiliated with the gut organisms was markedly declined in patients relative to the healthy controls (P < 0.05) (Supplementary Figure 3B). Comparison of phylumlevel proportions indicated pronounced variations of the putative gut-derived organisms in patients, displaying similar patterns vs. the microbiome-wide changes (**Figure 1B** and Supplementary Figure 2C). The Bacteroidetes was overrepresented while Actinobacteria was markedly decreased in all patient groups (P < 0.05) (**Figure 1B**), followed by a striking increase in the ratios of the relative abundance between both phyla compared to those of healthy subjects (Supplementary Figure 4A). The ratios between Firmicutes and Actinobacteria rose (Supplementary Figure 4B), which was mainly due to significant decline in Actinobacteria in the blood of the patients. Such changes in the predominant phyla provided compelling evidence indicating that the putative gut-derived bacterial composition was significantly altered in patients (Supplementary Figure 3C). Class-level analyses showed highly consistent changes with those of the aggregate microbiota (**Figure 1B** and Supplementary Figure 2C), suggesting that shifts of the putative gut-derived organisms contributed predominantly to the blood microbiome-wide alterations in patients. At the genus level, 20 of the putative gut-derived bacterial genera, including Bacteroides (10.5 ± 9.2%), Escherichia/Shigella (9.0 ± 4.6%), Acinetobacter (8.9 ± 4.9%), Stenotrophomonas (also likely derived from soils) (8.5 ± 4.8%), Serratia (4.9 ± 2.8%), Pseudomonas (also likely as soil-derived) (4.5 ± 2.8%), Rhizobium (also likely as soil-derived) (3.0 ± 2.4%), Prevotella (1.9 ± 2.1%), Corynebacterium (also likely derived from skin or soils) (1.5 ± 1.7%) and so on, dominated the blood microbiota in patients (average relative abundance >1%) (**Figure 1C**). These bacterial genera were also abundant in healthy controls; however, their proportions were significantly distinct from those of patients (**Figure 1C**). The genera Bacteroides, Stenotrophomonas, Serratia, Rhizobium, Prevotella, Staphylococcus, and Paracoccus were markedly expanded in the blood of the patients, regardless of the illness severities (P < 0.05, vs. healthy controls) (**Figure 1C**). Interestingly, some bacterial genera, such as Rhizobium, Phascolarctobacterium, Alistipes, Parabacteroides, Faecalibacterium, Paraprevotella, and Clostridium XlVa, were present in the patients, but not detected in healthy subjects. In contrast, the genera Acinetobacter, Lactococcus, Dietzia, Flavobacterium, Pseudomonas, Corynebacterium, Sphingobium, and Brevundimonas, were consistently decreased in the patients (P < 0.05, vs. healthy controls). Of them, some bacterial genera were commonly considered to be potentially pathogenic or probiotic, and were probably of special importance for the blood microbiome alterations in patients (Supplementary Figure 5). Like the blood microbiota-wide analysis, the sample dots, representing the putative gut-derived bacterial communities, displayed highly similar distributions in the PCA plots (**Figure 1D** and Supplementary Figure 2D), indicating that altered abundance of putative gut-originated organisms was predominantly responsible for the microbiome-wide alteration in patients. To further identify taxonomic differences in the microbiomes between the patients and healthy subjects, we conducted a linear discriminant analysis (LDA) effect size

bacterial genera identified by taxonomic classification. The data represent the Log10 values of the operational taxonomic unit (OTU) counts each genus. The clustering relationships across the blood samples are shown in the upper panel. (B) The pie charts indicating the composition of the putative gut-derived organisms in the blood microbiota at the phylum and class levels. (C) Comparisons in the relative proportions of the top 30 bacterial genera among groups. (D) The plot of principal component analysis (PCA) showing the difference of the microbial community structures.

(LEfSe) algorithm. Consistent with the findings mentioned above (**Figure 1B**), the changes of the blood microbiome in patients were primarily sourced from the classes Bacteroidia, Clostridia, Actinobacteriae, Flavobacteriia and Bacilli (Supplementary Figure 6).

## Identification of Diverse Bacterial Microbiome within Neutrophils

To define the configurations of neutrophil-associated microbiomes, also termed as NAMs, we isolated peripheral neutrophils from the patients and healthy subjects, and sequenced the 16S rDNA amplicons obtained from the cells. We observed that a surprisingly diverse microbiome was present within the neutrophils both in patients and in healthy subjects, as indicated by several hundreds to thousands of unique OTUs. Similar to the blood microbiomes, the OTUs obtained from the neutrophil specimens were mainly classified into the four phyla: Proteobacteria, Bacteroidetes, Firmicutes and Actinobacteria (Supplementary Figure 7A). A notable expansion of OTUs numbers was seen in the infected or septic patients (P < 0.01) (Supplementary Figure 7B), implying that the neutrophils in these cases might capture more diverse bacteria than in healthy controls. Viewing the microbiota profile, we found that the increase of species richness in infected or septic patients was mainly due to over-presence of certain bacteria belonging to the phyla Bacteroidetes and Firmicutes (P < 0.05, vs. healthy controls) (Supplementary Figure 7C). The proportions of the classes Bacteroidia, Clostridia, and Negativicutes were significantly expanded in the patients, whilst Actinobacteriae, Flavobacteriia, Gammaproteobacteria and Betaproteobacteria were markedly declined (P < 0.05, vs. healthy controls) (Supplementary Figure 7C). The PCA plots showed that the majority of the patients' sample dots, were distant from those of the healthy controls, suggesting that the NAM shifted toward aberrant configuration in patients (Supplementary Figure 7D).

Next we determined the potential origin of the neutrophilassociated microbiome, obtaining similar results with that of the blood microbiota. The organisms that presumably derived from the gut constituted the major component of the neutrophilassociated microbiome, with high abundance proportions (on average 83.1% of the aggregate, range: 63.7 to 94.8%) across all samples (Supplementary Figure 8A). The putative gutderived organisms in neutrophils were profoundly diverse (**Figures 2A,B**), and especially in infected and septic patients, the counts of the OTUs and bacterial genera were far more than in healthy controls (P < 0.01) (**Figure 2C**). As compared to healthy controls, the most significant shifts of the bacterial communities in neutrophils of patients were the increases in the Bacteroidetes and Firmicutes, together with a profound reduction in Actinobacteria and Proteobacteria (P < 0.01) (**Figure 2D**). The ratios between Bacteroidetes or Firmicutes and Actinobacteria were significantly higher in the patients than those of healthy microbiotas (P < 0.01) (Supplementary Figure 8B). At the class level, the bacterial microbiomes in neutrophils differed markedly between the patients and healthy individuals, mainly characterized by an expansion in Bacteroidia, Clostridia and Negativicutes, as well as a reduction in Actinobacteriae, Flavobacteriia, Gammaproteobacteria, and Betaproteobacteria (P < 0.01) (**Figure 2D**). The proper proportions of such bacterial taxons appeared to be disrupted, indicating significant alterations in the neutrophil-associated microbiomes of the patients. The PCA plot, based on the relative abundance of the putative gut- derived bacterial genera, indicated that the microbiome memberships were distinguished between healthy controls and patients (**Figure 2E**). To explore the bacterial phylotypes associated with the shifts of the neutrophilassociated microbiomes in the patients, we compared the relative abundance of some specific bacterial genera that were often included in the potentially pathogenic and probiotic organisms (Supplementary Figure 9A). Of them, the potentially pathogenic organisms, including Bacteroides, Stenotrophomonas, Clostridium XIVa, Fusobacterium, Eubacterium, and Serratia were more enriched in the patients than healthy individuals (P < 0.05) (Supplementary Figure 9B). However, the genera Acinetobacter, Lactococcus, Corynebacterium, Flavobacterium, Pseudomonas, Bifidobacterium, Legionella, and Anaerococcus were significantly less abundant in the patients (P < 0.05, vs. Healthy controls) (Supplementary Figure 9B). Further, we conducted the linear discriminant analysis (LDA) effect size (LEfSe), indicating the variations of the neutrophil-associated microbiomes in patients (Supplementary Figure 10). In total, the neutrophil-associated microbiome in the patients was distinct from that of healthy subjects, which probably had important implications in the development of bacteremia and systemic infection.

#### Associations between Immune Traits and Blood Microbiome in SAP Patients

Next we characterized a wide range of circulating immune cell subtypes in the peripheral blood samples from the patients and healthy controls. As shown in the **Figure 3A**, the percentages of CD4<sup>+</sup> and CD8<sup>+</sup> T lymphocyte subsets were both reduced in the septic patients (P < 0.01, vs. healthy controls). The proportions of naïve T cells (CD4+, CD8+) were also decreased strikingly in the infected and septic cases, whilst the memory T cells were increased compared to those of healthy subjects (P < 0.01). In addition, the counts of the total lymphocytes, CD4<sup>+</sup> and CD8<sup>+</sup> T lymphocytes were significantly decreased in the patients (P < 0.01, vs. healthy controls), reaching the minimum values in septic cases (**Figure 3C**). By contrast, a significant increase in CD8<sup>+</sup> T cell apoptosis was observed in the infected and septic patients (P < 0.001) (**Figure 3B**). The apoptosis of CD4<sup>+</sup> T cells was also increased in uninfected, infected and septic cases, while a significantly statistical difference was only found between septic patients and healthy controls (P < 0.05). Of special note, the septic patients displayed a strong proinflammatory cytokine profile, characterized by increased release of IL-1β, IL-2, IL-6, TNF-α, and IFN-γ in the serum (P < 0.001) (**Figure 3B**).

The blood and neutrophil-associated microbiomes of the patients were distinctfrom those of healthy subjects, raising a question of whether shifts of the microbiomes were related to the immunological disorders in patients. To address it, we examined relationships between specific clades of the microbiomes and the immunological parameters in patients (**Figure 4**). We observed that some bacterial genera associated with the neutrophils, such as Acinetobacter, Bacteriodes, Stenotrophomonas, Serratia, Pseudomonas, Chryseobacterium, Methylobacterium, Clostridium, Enterococcus, Lactococcus, and Oscillibacter, etc., were correlated either positively or negatively with T lymphocyte subsets, especially naïve and memory T cells, in the septic patients (**Figure 4**). Similarly, some of the bacterial phylotypes showed strong correlation with the immunological traits in the uninfected and infected patients. The majority of these bacteria in the neutrophil-associated microbiomes were closely associated with the changes of serum cytokine levels in patients (Supplementary Figure 11). However, the bacterial genera that closely correlated to the lymphocyte subsets and serum cytokine concentrations were relatively fewer in healthy controls. In addition, some bacterial taxa in the peripheral blood were significantly correlative with the immunological parameters both in patients and in healthy subjects (Supplementary Figures 12, 13). Clearly, the changes of the blood and neutrophil-associated microbiomes were strongly linked to immunological disorders of patients.

FIGURE 2 | of the counts the OTUs and genera affiliated with the putative gut-derived organisms in the NAMs. \**P* < 0.05; \*\**P* < 0.01; \*\*\**P* < 0.001. (D) Changes of the predominant bacterial composition in the NAMs at the phylum and class levels. The letters "H," "U," "I," and "S" represent the healthy, uninfected, infected and septic groups, respectively. (E) Principal component analysis (PCA) of weighted UniFrac distances, based on the relative abundance of each genus, displaying the compositional differences of the NAMs.

## Changes of Neutrophil Proteomic Profiles and Its Connection with Alterations of Blood Microbiome

We further performed comparative proteomic analysis on peripheral neutrophils derived from healthy and SAP subjects in an attempt to link cell functional changes to the microbiome alterations. A total of 296 proteins were characterized as differentially expressed, were then annotated and clustered into six categories involved in biological functions of neutrophils, including innate immune defense, immune response, cytokine

release, cell apoptosis, cell structure, and metabolic activity (Supplementary Table 2). As shown in the **Figure 5A**, the expression profiles of these proteins were dramatically varied in SAP patients with sepsis and healthy controls. Of these, the proteins closely involved in bactericidal activities of neutrophil, such as lysozyme C (LYZ), eosinophil cationic protein (RNASE3), myeloperoxidase (MPO), neutrophil defensin

3 (DEFA4), properdin (CFP) and bactericidal permeabilityincreasing protein (BPI), were significantly down-regulated in septic patients, indicating that the dysfunction of innate immunity might be present in septic cases. We also compared the expression profiles of some known immune response-associated proteins, as characterized by significant down-regulations for a vast majority of these proteins in SAP patients (**Figure 5A**).

indicate the correlation coefficient between variances.

The changes in the protein expressions were likely, at least in part, involved in the immunosuppression in SAP patients with sepsis. Of special note, matrix metallopeptidase 9 (MMP9), a protein prompting leukocyte transendothelial migration, was significantly declined in septic patients, suggesting the presence of neutrophil dysfunction. The expression of the proteins involved in modulation of proinflammatory cytokines was strikingly up-regulated in septic patients, probably having implication for establishment of the uncontrollable inflammatory response. Similarly, the expression of some proteins associated with cell apoptosis was increased in septic patients, which may help to explain delayed neutrophil apoptosis in sepsis. In addition, an over-representation in metabolism-related proteins was observed in the septic patients, when compared against the data set from healthy persons. Basing on the observations, we provide indirect evidence indicating that the function of neutrophils might be impaired in SAP patients with sepsis.

Next we conducted a Pearson's correlation analysis to associate the differentially expressed proteins of neutrophils with microbiome memberships (**Figure 5B**). Our data showed that changes of the protein expression in neutrophils were closely associated with specific bacterial taxons of the microbiome. As shown in **Figure 5B,** most of the proteins in the functional clusters of innate immune defense, immune response and cytokine production correlated negatively with compositional changes of the neutrophil-associated microbiome, especially the memberships of the phyla Bacteroidetes and Firmicutes. On the contrary, a majority of the proteins involved in the apoptosis and metabolic activities of neutrophils correlated positively with the members of the microbiomes.

#### DISCUSSION

To our knowledge, this is the first prospective observational case series exploring the microbial landscape in peripheral blood and neutrophils in SAP patients, as well as their potential links with the immunological disorders of the patients. Through this study, we identified diverse bacterial microbiomes within peripheral blood and neutrophils, and discovered that the microbiome is altered in SAP patients. More importantly, the abundance changes in certain members of the microbiomes are closely linked to the immunological disorders of the patients. Our findings provide emerging evidence supporting the presence of the blood microbiome and give us novel insights into induction of bacteremia in SAP patients.

Recent studies have begun to document that human blood contains an authentic microbiome, which could contribute significantly to the development of sepsis (Grumaz et al., 2016; Gosiewski et al., 2017) and several chronic diseases (Amar et al., 2013; Rajendhran et al., 2013; Dinakaran et al., 2014). Yet, whether a diverse bacterial community is present in blood of SAP patients remains an unanswered question. Through the fingerprinting approach of 16S rRNA amplicons, we have recently demonstrated that multiple bacterial species are commonly seen in blood of patients with SAP (Li et al., 2013a), but without deep sequencing data it was not possible to define the microbial landscape in blood and its potential role in pathogenesis of sepsis. Herein we conduct 16S rRNA gene sequencing to characterize the bacterial profile present in blood of SAP patients. We show that the peripheral blood in SAP patients has a diverse bacterial microbiota, dominated by the phyla Proteobacteria, Actinobacteria, Bacteroidetes, and Firmicutes, which is consistent with previous reports (Amar et al., 2013; Rajendhran et al., 2013; Dinakaran et al., 2014). Furthermore, the blood microbiome in SAP patients appears dysbiotic, as revealed by loss of species richness and changes of predominant bacterial taxa relative to healthy controls. Together with the findings, we suggest that the perturbation of blood microbiota, which likely represents a disease-provoking state, might be involved in the progression of sepsis. Prior studies have reported that dysbiosis of the blood microbiome is an independent risk factor of cardiovascular disease, indicating its potentially pathological role (Amar et al., 2013; Rajendhran et al., 2013; Dinakaran et al., 2014). In total, our investigations based on culture-independent techniques have shown a previously unappreciated complexity of the blood bacterial microbiome in SAP patients, also forcing reconsideration of the mechanisms of sepsis pathogenesis and exacerbations.

Intestinal dysbiosis and bacterial translocation are common in critically ill patients, and there is strong evidence that the translocation of bacteria and their products across the intestinal barrier drives the progression of sepsis (Dickson, 2016; Alverdy and Krezalek, 2017). Our understanding of the concept of translocation of one or several organisms from the gut is founded on culture-based studies (MacFie et al., 1999; MacFie, 2004). Recently, Dickson and colleagues have demonstrated that the lung microbiome is enriched with gut-associated bacteria in sepsis and acute respiratory distress syndrome, providing strong evidence for gut–lung translocation of bacterial microbiota (Dickson et al., 2016). The observations prompted us to reconsider the current opinion of bacterial translocation from the gut to systemic circulation. Our data presented here reveal that the blood microbiome is mainly composed of gut-associated organisms in SAP patients, which is similar to the results in the lung microbiome (Dickson et al., 2016). It is therefore speculated that a bacterial consortium from the gut, rather than single or several organisms, might migrate into systemic circulation during sepsis. A longitudinal study of paired stool and blood specimens will be required to determine the true prevalence for gut-blood translocation of bacterial microbiota in sepsis. Nonetheless, the significant correlation between circulating bacteria and clinical manifestations offers evidence that the blood microbiome may play an important role in sepsis, even in the absence of gut–blood translocation.

Innate immune cells and microorganisms in blood are highly interactive, maintaining a delicate balance between defending against infection and eliciting an excessive inflammatory response (Nathan, 2006). Given the critical role of neutrophils in eradication of pathogens, we sought to characterize the composition of neutrophil-associated microbiomes and explore potential roles in the alteration of blood microbiome in SAP. We show that the neutrophils contain a diverse microbiome, and the microbiota composition is significantly altered in SAP patients, consistent with the findings from the blood microbiome. Proteomic analysis of the neutrophils shows that the function of neutrophils is likely disturbed, especially in septic patients, as revealed by reframing of the protein profiles. The bactericidal effector function of neutrophils, represented by decreased antibacterial peptide production (Flannagan et al., 2009), is probably restricted in SAP patients with sepsis. Another prominent feature of neutrophils is decreased expression of MMP9, possibly contributing to impairment of cell migration during systemic infections (Kolaczkowska et al., 2009). The observations indicate that the neutrophil function is probably collapsed, leading to impaired intracellular bacterial clearance and compositional shifts of NAMs in sepsis. Further investigations with neutrophil killing/phagocytosis assays will be needed to provide direct evidence linking the blood microbiome alterations seen in SAP patients with neutrophil dysfunction.

Unlike other organs, the blood was originally presumed to be sterile, and microbes were thought to be present in circulation only in sepsis cases. But, in recent years, the presence of bacterial 16S rRNA genes has been reported in the circulation of healthy individuals (Potgieter et al., 2015; Païssé et al., 2016). In this study, we have added evidence suggesting that the peripheral blood harbors a diverse bacterial microbiota in healthy individuals, consistent with the previous reports (Païssé et al., 2016). Interestingly, the blood microbiota is enriched with the putative gut-derived bacteria, implying that translocation of intestinal bacteria may play a critical role in shaping the unique microbiota. Previous studies have demonstrated that intestinal bacterial translocation could be a normal physiological event, and under healthy conditions the enteric organisms can move across the "intact" intestinal epithelium into normally sterile tissues including blood, contributing to development of systemic and/or organ immune system (Brenchley and Douek, 2012; Wiest et al., 2014). Quite contrary to the traditional concept that circulating blood is sterile in health, the blood may harbor a commensal microbiome, which is likely to be altered during disease.

In spite of compelling evidence indicating the existence of blood microbiome in SAP patients, our study has several limitations. The primary limitation is that the results are based on sequencing of 16S rRNA genes, which is not sufficient to determine if a live microbiota is present in the circulation. It is required for metatranscriptomic studies to validate the presence/absence of a live bacterial microbiome in the blood. Another limitation is that 16S rDNA-based community profiling only provides information on the microbiota composition. Future studies will be needed to move from taxonomic description to detailed, metagenomics-based functional characterizations of the microbiota. Additionally, the number of individuals sampled in our study is relatively small, and thus it

#### REFERENCES


will be important to validate these observations with increasingly larger and more sophisticated human cohort studies.

In summary, we have presented compelling evidence that the blood contains a diverse bacterial microbiome in SAP patients. We also have identified the unique compositional signature of the blood and neutrophil-associated microbiomes that could distinguish SAP patients from healthy controls. More importantly, our data reveal potential links between the alterations in the blood and neutrophil-associated microbiomes and the immunological disorders in SAP patients. Yet, the possible involvement of blood microbiome variations in the development of systemic infection remains uncertain in SAP patients. Here we have only just begun to delineate the outline of the blood microbiome, and further investigations would likely provide novel insights into its roles in health and diseases.

#### AUTHOR CONTRIBUTIONS

QL, CW, and JL conceived of the work. QL and CW designed experiments and analyzed data. CW, CT, XZ, and QH enrolled the patients, collected samples, performed the experiments, and prepared the figures and tables. CW wrote the article and revised it critically for important intellectual content. QL reviewed and revised the manuscript. All authors read and approved the final version of the manuscript for submission.

#### FUNDING

This work was supported by the grants from the National Basic Research Program (973 Program) in China (2013CB531403) and National High-tech R&D Program (863 Program) of China (2012AA021007).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcimb. 2018.00005/full#supplementary-material

DNA in the blood. Pancreas 26, 18–22. doi: 10.1097/00006676-200301000- 00004


Atlanta, GA, September 11 through 13, 1992. Arch. Surg. 128, 586–590. doi: 10.1001/archsurg.1993.01420170122019


in cynomolgus monkeys after alemtuzumab treatment. Am. J. Transplant. 13, 899–910. doi: 10.1111/ajt.12148


**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 Li, Wang, Tang, Zhao, He 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.

# Microbiota Modulate Anxiety-Like Behavior and Endocrine Abnormalities in Hypothalamic-Pituitary-Adrenal Axis

Ran Huo1, 2, 3†, Benhua Zeng4†, Li Zeng2, 5†, Ke Cheng1, 2†, Bo Li 1, 2, 3, Yuanyuan Luo1, 2 , Haiyang Wang<sup>2</sup> , Chanjuan Zhou<sup>1</sup> , Liang Fang<sup>1</sup> , Wenxia Li <sup>4</sup> , Rong Niu<sup>4</sup> , Hong Wei <sup>4</sup> \* and Peng Xie1, 2, 3, 5 \*

<sup>1</sup> Department of Neurology, Yongchuan Hospital, Chongqing Medical University, Chongqing, China, <sup>2</sup> Institute of Neuroscience and the Collaborative Innovation Center for Brain Science, Chongqing Medical University, Chongqing, China, <sup>3</sup> Key Laboratory of Clinical Laboratory Diagnostics (Ministry of Education), Department of Laboratory Medicine, Chongqing Medical University, Chongqing, China, <sup>4</sup> Department of Laboratory Animal Science, College of Basic Medical Sciences, Third Military Medical University, Chongqing, China, <sup>5</sup> Department of Neurology, First Affiliated Hospital of Chongqing Medical University, Chongqing Medical University, Chongqing, China

#### Edited by:

Michele Marie Kosiewicz, University of Louisville, United States

#### Reviewed by:

Valerio Iebba, Sapienza Università di Roma, Italy Guoqiang Zhu, Yangzhou University, China

#### \*Correspondence:

Hong Wei weihong63528@163.com Peng Xie xiepeng@cqmu.edu.cn

† These authors have contributed equally to this work.

Received: 23 July 2017 Accepted: 13 November 2017 Published: 30 November 2017

#### Citation:

Huo R, Zeng B, Zeng L, Cheng K, Li B, Luo Y, Wang H, Zhou C, Fang L, Li W, Niu R, Wei H and Xie P (2017) Microbiota Modulate Anxiety-Like Behavior and Endocrine Abnormalities in Hypothalamic-Pituitary-Adrenal Axis. Front. Cell. Infect. Microbiol. 7:489.

doi: 10.3389/fcimb.2017.00489

Intestinal microbes are an important system in the human body, with significant effects on behavior. An increasing body of research indicates that intestinal microbes affect brain function and neurogenesis, including sensitivity to stress. To investigate the effects of microbial colonization on behavior, we examined behavioral changes associated with hormones and hormone receptors in the hypothalamic-pituitary-adrenal (HPA) axis under stress. We tested germ-free (GF) mice and specific pathogen-free (SPF) mice, divided into four groups. A chronic restraint stress (CRS) protocol was utilized to induce external pressure in two stress groups by restraining mice in a conical centrifuge tube for 4 h per day for 21 days. After CRS, Initially, GF restraint-stressed mice explored more time than SPF restraint-stressed mice in the center and total distance of the OFT. Moreover, the CRH, ACTH, CORT, and ALD levels in HPA axis of GF restraint-stressed mice exhibited a significantly greater increase than those of SPF restraint-stressed mice. Finally, the Crhr1 mRNA levels of GF CRS mice were increased compared with SPF CRS mice. However, the Nr3c2 mRNA levels of GF CRS mice were decreased compared with SPF CRS mice. All results revealed that SPF mice exhibited more anxiety-like behavior than GF mice under the same external stress. Moreover, we also found that GF mice exhibited significant differences in, hormones, and hormone receptors compared with SPF mice. In conclusion, Imbalances of the HPA axis caused by intestinal microbes could affect the neuroendocrine system in the brain, resulting in an anxiety-like behavioral phenotype. This study suggested that intervention into intestinal microflora may provide a new approach for treating stress-related diseases.

Keywords: intestinal microbes, HPA axis, CRS model, microbiota-gut-brain axis, stress-related diseases

## INTRODUCTION

The intestine is the largest system in the mammalian body, containing 100 trillion organisms. Intestinal microbial flora are established in early life in mammals, and affect the host's physiological function (Grenham et al., 2011; Lozupone et al., 2012; Heitlinger et al., 2017). Recent studies also have reported that intestinal microbial steady-state imbalances can cause a range of metabolic diseases (Wen et al., 2008; Henaomejia et al., 2012; Koren et al., 2012). A number of studies have explored the mechanisms of intestinal microorganisms, and a range of microbe-related diseases have been discovered and explored in neuropsychiatric subjects. However, the precise mechanisms of action of intestinal microbial flora remain unclear. Among the known pathogenetic mechanisms, several mental illnesses have been linked to the hypothalamic-pituitary-adrenal (HPA) axis (Schatzberg et al., 2014; Fries et al., 2015).

According to the long-standing HPA axis imbalance theory, hormone imbalance is closely associated with psychiatric diseases. A range of factors, including exercise, anxiolytic drugs, and sexual experience, can interfere with the secretion of stress hormones related to the HPA axis (Romero, 2004). Meanwhile, stress-related psychiatric disorders are closely related to imbalances in the HPA axis (Jacobson, 2014; van Bodegom et al., 2017), including anxiety disorders, social anxiety disorder, and post-traumatic stress disorder (Wirtz et al., 2007). Several studies have reported that changes in HPA axis hormones vary between stimulus type and rat variety, and can be used as an index of the intensity of a stressor (Girotti et al., 2006). In addition, one study found that plasma hormone levels (adrenocorticotropic hormone, ACTH; cortisol, CORT) were increased in the HPA axis after exposure to various stressors for 30 min (Hueston et al., 2011) and decreased to baseline levels within a certain time after the termination of acute stress (Dhabhar et al., 1997). The glucocorticoid receptor (GR) and the mineralocorticoid receptor (MR) mediate regulation of CORT gene expression (Arriza et al., 1987), which illustrates that hormonal changes in the HPA axis may are associated with changes in receptor levels. Interestingly, previous studies have found microbes are closely connection between HPA axis and behavior (Moya-Pérez et al., 2017).

In recent studies, GF mice are widely used as a tool for assessing the role of intestinal microbes, which have been found to affect mouse brain function and behavior (Luczynski et al., 2016). In addition, an increasing body of research has examined the effects of intestinal microbes in the HPA axis and microbiota-gut-brain axis using GF animals and antibiotic intervention (Foster, 2015; Zeng et al., 2016). Studies in which stool is transplanted from patients into the intestine of germfree (GF) mice have revealed that gut microbiota can affect animals' behavior through the microbiota-gut-brain axis (Bercik et al., 2011; Cryan and Dinan, 2012; Zheng et al., 2016b). In the HPA axis, the hypothalamus is considered the starting point of the HPA axis, and previous studies have shown that levels of hormone concentration and hormone receptors in this brain region are altered under acute pressure (Crumeyrolle-Arias et al., 2014; Zhu et al., 2014). To create artificial chronic stress, the chronic restraint stress (CRS) model is classical and widely used to induce external pressure to detect the relationship between chronic pressure and diseases (Andrus et al., 2012). On the basis of this previous research, we hypothesized that intestinal microbial stabilization disorders would affect behavioral changes through the HPA axis using the CRS model in mice.

In the current study, to assess the effects of intestinal microbes on the HPA axis, we first examined behavior, hormone levels and receptor expression in the HPA axis using the CRS model in both GF and SPF mice. Then behavior was analyzed to assess whether differences in intestinal microbes play an important role in behavioral changes in mice.

### MATERIALS AND METHODS

#### Animals

GF Kunming (KM) and SPF KM mice (male; 6 weeks old) were provided by the Experimental Animal Center of the Third Military Medical University (Chongqing, China) and bred at the Experimental Animal Center of the Third Military Medical University (GB 14922.2-2011). GF mice were kept and subjected to the CRS protocol in sterile isolators until the beginning of the behavioral tests. Weekly fecal samples were collected from GF mice and monitored using cultures of aerobic and anaerobic microbes to ensure the reliability of sterile feeding conditions. SPF mice were kept and subjected to the CRS protocol in barrier system with 10,000 cleanliness level and noise ≤60 dB. All animals were group-housed in Macrolon cages (37 cm long, 26 cm wide, 17 cm high) and fed with autoclaved chow and water. Animal room conditions were maintained with a constant temperature of 22 ± 2 ◦C, relative humidity 55 ± 5% under a 12 h light-12 h dark cycles (lights on at 8:00 a.m.). The experimental protocols were in accord with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication No. 80-23), revised in 1996. Moreover, the Ethics Committee of Chongqing Medical University approved all the experiments.

#### Chronic Restraint Stress (CRS) Procedure

All mice were acclimatized to the standard experimental environment for 7 days before the test session (Liu et al., 2016). GF and SPF mice were subjected to an established chronic physical restraint protocol. They were placed in the 50 ml multiple breathable hollows (0.5 mm diameter, 12 holes) conical centrifuge tubes (Wong et al., 2016). This restraint vessel was adapted to the animal's body size, and no pain was involved. Mice were restrained in the pipe for 4 h (from 13:00 to 17:00), with 20 h of rest time each day for 21 days. Mice were deprived of food and water during restraint then given food and water after each restraint experiment (Zafir and Banu, 2007). Mice were released into the cage to receive water and food immediately after the experiment. This restraint procedure was approved by the Ethics Committee of Chongqing Medical University. The details of the experimental procedure are shown in **Figure 1**.

#### Behavioral Procedures

In each experiment, GF and SPF mice (n = 28–32 in each group) were removed from the bacteria isolator, and

placed in the experimental environment for at least 1 h. The whole experimental environment was insulated to maintain a temperature of 22 ± 2 ◦C, and humidity of 55 ± 5%. The trajectory of each mouse was recorded with a video tracking system linked to a computer. Trajectories were analyzed and quantified using the SMART2.5 software package (Panlab, Barcelona, Spain).

#### Open Field Test (OFT)

Mice were gently placed in the center of the apparatus and allowed to move freely. The device was constructed from opaque black paper (45 × 45 × 45 cm), and had no distinctive odor. The position placing each mouse was the fixed edge of the device. After each test, 70% ethanol was utilized to clean feces and remove odor. The test time was 6 min: 1 min to adapt, and 5 min for testing. The whole experimental process was recorded with a video tracking system. Correlative indices were measured in the last 5 min (Kim et al., 2012; Zhang et al., 2016; Zhou et al., 2016).

#### Sample Collection and Preparation

After the experimental period, mice were euthanized with 10% chloral hydrate (400 mg/kg; Chen et al., 2015). Mice were perfused with ice physiological saline (0.9% NaCl, Nongfu Spring Company Limited, Hangzhou, China). The whole brain was dissected and immediately placed in liquid nitrogen. All tissue samples were stored in a refrigerator at −80◦C (Wang et al., 2016).

#### Hormonal Measurement

To quantify changes in HPA axis hormones in the hypothalamus tissue, the concentrations of ACTH, corticotropin-releasing hormone (CRH), CORT, and aldosterone (ALD) were analyzed using an enzyme-linked immunosorbent (ELISA) kit (ACTH, least detectable dose, 0.22 pg/ml, percent coefficient of variation, 5.38%, MD Bioproducts, USA; CRH, least detectable dose, 0.19 ng/ml, percent coefficient of variation, 6.54%; CORT, least detectable dose, 0.19 ng/ml, percent coefficient of variation, 6.66%; ALD, least detectable dose, 18.75 pg/ml, percent coefficient of variation, 4.73%; Elabscience Biotechnology Co., Ltd. China). Hypothalamus tissue was weighed, then minced into small pieces, which were homogenized in 1 g: 9 ml phosphate-buffered saline (PBS; Hyclone Co., USA) with protease inhibitor (Roche, Germany). We allowed samples and reagents to equilibrate to room temperature (22–25◦C) before performing the assay. Each procedure was carried out according to the kit instructions, on ice.

Hormonal concentrations from each sample were calculated from the standard curve using CurveExpert 1.30 software (Daniel G. Hyams Co., USA) in accordance with the manufacturer's recommendations and normalized for hypothalamus tissue homogenate protein measured with the BCA method using enhanced BCA protein assay kit (Beyotime Co., China). The ELISA reaction was recorded at the corresponding wavelength using a microplate reader (Bio-Rad Co., USA).

#### Hormone-Related Receptor Measurement

Total RNA was extracted from frozen hypothalamus tissue using an RNA mini kit (Ambion, USA) on ice. All experimental procedures were in accord with the kit instructions, followed by reverse-transcribed into DNA using a PrimeScript RT Reagent Kit (Takara, Toyoto, Japan). The cycling conditions were: three cycles of reverse transcription reaction at 37◦C for 15 min and reverse transcriptase inactivation reaction at 85◦C for 5 s. The mRNA values in the hypothalamus were quantified using qRT-PCR (Roche, Germany). The cycling conditions were: 10 min preincubation at 95◦C and 40 cycles of DNA amplification at 95◦C for 10 s, 60◦C for 30 s, and 72◦C for 35 s. Primer sequences were acquired using Primerbank (Harvard, USA), and synthesized by a biotechnology company (Sangon Biotech, Shanghai, China). The primer sequences for Crhr1fwd were as follows: 5′ -gggcagcccgtgtgaattatt-3′ , rev: 5′ -atgacggcaatgtggtagtg c-3′ ; for Crhr2fwd:5′ -catccaccacgtccgagac-3′ , rev:5′ -ctcgccaggatt gacaaagaa-3′ ; for Mc2rfwd:5′ -acaccgcaagaaataactccg-3′ , rev:5′ aggaggacaatcaagttctcca-3′ ; for Nr3c1fwd:5′ -agctccccctggtagagac-3 ′ , rev:5′ -ggtgaagacgcagaaaccttg-3′ ; for Nr3c2fwd:5′ -gaagagcccc tctgtttgcag-3′ , rev:5′ -tccttgagtgatgggactgtg-3′ ; for Gapdhfwd:5′ - AGGTCGGTGTGAACGGATTTG-3′ , rev:5′ -TGTAGACCATG TAGTTGAGGTCA-3′ . The corresponding mRNA content was standardized with Gapdh mRNA, and data expression was normalized with respect to the corresponding control group. All data were quantified with LightCycler 96 SW 1.1 analysis software (Roche, Germany). Hormonal receptor levels (Crhr1, Crhr2, Mc2r, Nr3c1, Nr3c2) were analyzed using quantitative realtime polymerase chain reaction (qRT-PCR; Roche, Germany) assay.

#### Statistical Analysis

All data were calculated as single data points superimposed to boxplots. The ELISA data, PCR data, and behavioral data were analyzed using two-way analysis of variance (ANOVA) assay with SPSS 20.0 (IBM North America, New York, NY, USA). In all cases, p < 0.05 were considered statistically significant.

#### RESULTS

## Changes in Behaviors between GF and SPF Mice

In order to determine whether the microbial colonization can alter behavior in mice, the OFT was used to assess behavior. Two-way ANOVA revealed that the SPF stressed control group moved a shorter total distance in the OFT compared with the GF stressed group (p < 0.01). Other groups did not show significant differences (**Figure 2A**). GF non-stressed control animals moved a significantly greater distance than SPF non-stressed mice in the center of the OFT (p < 0.01), and GF stressed animals moved a greater distance than GF non-stressed mice (p < 0.001). Mice in the SPF CRS group moved a greater distance in the center than those in the SPF non-CRS group (p < 0.001; **Figure 2B**). GF control mice spent less time exploring than GF stressed animals (p < 0.01)in the center of the OFT. In addition, the results revealed that GF control mice spent more time exploring than SPF control animals (p < 0.001), and GF restraint-stressed mice explored more time than SPF restraint-stressed mice (p < 0.001) in the center of the OFT (**Figure 2C**).

FIGURE 4 | Validation of hormone receptors and mineralocorticoid receptor (MR) / glucocorticoid receptor (GR) expression changes in the hypothalamus. Expression of hormone receptors and MR/GR was assessed in GF mice, GF CRS mice, SPF mice, and SPF CRS mice (n = 7–8 in each group). The data were analyzed using two-way ANOVA. Crhr1, corticotropin releasing hormone receptor 1, CRFR1; Crhr2, corticotropin releasing hormone receptor 2, CRFR2; Mc2r, melanocortin 2 receptor, ACTHR; Nr3c1, nuclear receptor subfamily 3, group C, member 1, glucocorticoid receptor, GR; Nr3c2, nuclear receptor subfamily 3, group C, member 2, mineralocorticoid receptor, MR. (A) shows the hormone receptor change in HPA axis. [A, F(3, 25) = 0.315; B, F(3, 25) = 5.012;C, F(3, 25) = 5.005; D, F(3, 25) = 0.373; E, F(3, 25) = 1.813]. The MR/GR expression ratio was calculated to assess receptor disorder [F(3, 25) = 1.711, F]. The values were presented as single data points superimposed to boxplots \*p < 0.05, \*\*p < 0.01. All data were normalized to SPF control mice.

#### Hormonal Dysfunction of the HPA Axis

We examined hormone and receptor levels in hypothalamic tissue. To assess changes in HPA axis-related hormones, hormonal levels were measured using ELISA in hypothalamus homogenates. As shown in **Figure 3A**, the CRH levels of GF restraint-stressed mice exhibited a significantly greater increase than those of SPF restraint-stressed mice (p < 0.05). The concentrations of ACTH in GF restraint-stressed mice homogenates were higher than in the GF control group (p < 0.01) and SPF restraint-stressed mice (p < 0.01; **Figure 3B**). The results revealed a trend toward increased CORT concentration in the GF restraint-stressed group compared with the SPF restraint-stressed group (p < 0.05; **Figure 3C**). ALD levels in GF restraint-stressed mice were also increased in hypothalamus homogenates compared with GF control group (p < 0.05) and SPF restraint-stressed mice (p < 0.01; **Figure 3D**).

#### Changes in Hormone Receptor mRNA

To investigate the link between hormone levels and hormone receptor mRNA, receptor mRNA in the mouse hypothalamus was quantified using qRT-PCR. **Figure 4** shows the changes in receptor levels among the groups. As expected, the Crhr1 mRNA levels of GF CRS mice were increased compared with GF control (p < 0.05) and SPF CRS mice (p < 0.05; **Figure 4A**). However, Nr3c1 mRNA expression in SPF control mice was decreased compared with SPF CRS mice (p < 0.05; **Figure 4D**), and Nr3c2 mRNA expression was decreased in GF CRS mice

FIGURE 5 | Function and mutual adjustment of HPA axis. The HPA axis contains three cell types that secrete three different hormones: neurons of the PVN in the hypothalamus secrete CRH, endocrine cells in the pituitary secrete ACTH, and zona fasciculata cells in the adrenal cortex secrete cortisol. Stress, drugs, and diseases produce positive feedback regulation of neurons in the of the medial parvocellular portion of the PVN. In addition, cortisol that could result in metabolic effects produces direct negative feedback suppression of endocrine cells in the pituitary and CRH neurons of the PVN in the hypothalamus, respectively. HPA axis, hypothalamic-pituitary-adrenal axis; PVN, hypothalamic paraventricular nucleus; CRH, corticotropin-releasing hormone; ACTH, adrenocorticotropic hormone.

compared with SPF CRS mice (p < 0.01; **Figure 4E**). de Kloet (2014) demonstrated that the MR/GR balance plays an important role in mediating the function of CORT in the brain, and that dysfunction of MR/GR expression can occur in specific pathological, emotional, and cognitive conditions (Brinks et al., 2007). To detect whether MR/GR expression had changed, we calculated the ratio of MR to GR and found that MR/GR decreased in GF CRS mice compared with GF control mice (p < 0.01). A decrease was also found in SPF CRS mice (p < 0.01; **Figure 4F**).

### DISCUSSION

In our study, behavioral tests showed that GF control mice exhibited an increase in the distance traveled and time spent in the center of the OFT compared with SPF control mice, consistent with previous reports (Zeng et al., 2016; Zheng et al., 2016b). The GF mice without non-intestinal microbial colonization moved a greater total distance in the OFT and spent more time in the center, compared with SPF mice with intestinal microbes after CRS. This finding indicates that SPF mice with intestinal microbes exhibited increased anxiety-like behavior under the same pressure. Previous studies demonstrated that GF F344 rats were more likely to exhibit anxiety-like behavior than SPF rats (Crumeyrolle-Arias et al., 2014; Desbonnet et al., 2014; Wong et al., 2016; Zheng et al., 2016a,b). However, some studies found no relationship between intestinal microbes and animal behavior. The effects of intestinal microbes and physiological state on psychopathology are still debated. We then found that behavioral changes were largely consistent with changes in hormones, both in the presence of intestinal microbes and non-intestinal microbes. In addition, the results showed that hormones in GF CRS mice were significantly upregulated compared with SPF CRS mice in the HPA axis, in accord with previous reports (Sudo et al., 2004). This mechanism may be related to changes in CRH-signaling, glucocorticoids, or GR, which mediate behavior in the central nervous system (Owens and Nemeroff, 1991).

Although, some previous studies reported that anxiety- and trauma-related disorders were not consistent with simultaneous changes in the HPA axis, it is well established that these disorders are associated with an imbalance in the HPA axis (Smith et al., 1989; Baker et al., 1999; Jacobson, 2014). The current results revealed that GF CRS mice exhibited anti-anxiety behavior accompanied by HPA axis over-activity compared with SPF CRS mice. This novel finding may be related to our use of hypothalamus tissue, whereas many previous studies used plasma. The HPA axis has a complex feedback mechanism, and intestinal microbes may regulate behavior through the endocrine system, which may subsequently induce overactivity in the HPA axis.

Our examination of hormone receptors also indicated hormone dysfunction in the hypothalamus. Previous studies reported that hormone receptor gene knockout mice and mice given hormone receptor antagonists exhibited modulation of stress-coping behaviors (Boyle et al., 2005). GR is widely expressed in most cell types throughout the body (De Kloet et al., 2000). GR and MR act as ligand-activated transcription factors and affect gene transcription, playing an important role in glucocorticoid function (Reul and Kloet, 1985). In addition, researchers have reported that changes of GR or MR levels in the hippocampus are associated with HPA axis dysfunction in moodrelated illness, although findings have been inconsistent, with some studies finding that GR mRNA is decreased in depression, and other studies reporting that GR mRNA in the hippocampus was unchanged (Webster et al., 2002). At the same time, downregulated MR and GR expression, and changes in MR/GR ratio have been reported in stress-induced rats (Medina et al., 2013). In the current study, we also used the MR/GR expression ratio to assess receptor diversification in the hypothalamus after behavioral changes. The results revealed changes in MR/GR expression and the action of intestinal microbes.

Intestinal microbes constitute a large and complex ecosystem in the intestinal wall of animals, affecting physiological and neuronal function, as well as animal behavior, via the microbiotagut-brain axis and metabolites. Taken together with the behavioral and hormonal variations described above, the current results indicate that intestinal microbes play a critical role in influencing behavior and HPA-axis regulatory imbalance under external stress. Recent research suggests that intestinal microbes affect the host's physiology, metabolism and immunology, as well as nervous system development and brain function, through the microbiota-brain-gut axis (Collins and Bercik, 2009; Fu et al., 2015; Yano et al., 2015). Interestingly, Bercik et al. (2011)reported that adult mice given microbial agents via oral absorption showed changes in exploratory behavior and brain-derived neurotrophic factor (BDNF) expression in the hippocampus, while no change was observed with intraperitoneal injection of the same agent.

Studies have also reported that high intestinal permeability, bacterial translocation, and inflammatory factors are an important factor in mental disorders. Intestinal microflora mediate a series of neurotrophic factors, BDNF, and proteins (Ait-Belgnaoui et al., 2005). Intestinal microbial immune disorders are associated with aberrant neurodevelopment, and inappropriate use of antibiotics inhibits short-chain-fatty-acids (SCFAs) and the interaction between toll-like receptors and Treg cells. Moreover, the HPA axis (**Figure 5**) is affected by the peripheral nervous system (PNS), infection, and stress. The proportion of carbohydrates in food and dietary structure can also affect HPA axis activity (Keating et al., 2004; Glover et al., 2010; Ronald et al., 2010; Smith et al., 2013).

In addition, we speculated that intestinal microbes might cause intestinal metabolic changes through the intestinal microbial-gut-brain axis pathway. Metabolites may then pass through the intestinal wall, into blood circulation and through the blood-brain-barrier (BBB). The central nervous system (CNS) may then be affected by products of bacterial metabolism, causing hormone and receptor dysfunction, as well as behavioral changes.

First, intestinal microbes through enterochromaffin (EC) cells control the synthesis of 5-HT, which could be involved in brain function (Yano et al., 2015). Second, microbes may have an important relationship with the CNS through the inflammatory pathway, possibly activating local or systemic immune responses through the vagus nerve to influence the activity of the braingut axis (Borovikova et al., 2000; Wang et al., 2003). Third, SCFAs produced by intestinal bacterial fermentation have an immunomodulatory function, stimulating the link between the sympathetic nerve and nerve cells through G-protein-coupled receptor 41 (GPR41) and 43 (GPR43; Kimura et al., 2011). This might regulate the balance of microgliacytes, and mediate the release of intestinal peptide from endocrine cells to affect braingut axis activity (Wren and Bloom, 2007). In addition, this may mediate 5-HT synthesis in EC cells, which provides the CNS termination signal (Yano et al., 2015). Finally, intestinal microbes regulate tryptophan metabolism, which affects brain function and plays an important role in serotonin synthesis in the CNS (Ben-Ari, 2013). Moreover, intestinal microbes may also produce dopamine, γ-aminobutyric acid, histamine and acetylcholine, regulating the function of CNS and the stability of the HPA axis (Thomas et al., 2012; Barrett et al., 2014). In accord with this notion, the microbiota-gut-brain axis is considered to function as a bidirectional regulation mechanism of animal behavior (Wong et al., 2016).

The current study involved several limitations that should be considered. First, we did not use multiple behavioral paradigms to examine behavior more comprehensively. Moreover, this experiment did not clarify which intestinal microbial flora induced behavioral and endocrine changes in mice. Thus, more in-depth examination of the possible mechanisms involved should be conducted in follow-up research. In addition, in future studies we plan to re-colonize known microorganisms or probiotics into the intestine to regulate the connection between the intestine and the brain in mice, then utilize the corresponding intestinal microbe antibiotics, hormone or receptor antagonists to interfere with the connection, to further reveal the functional mechanisms of microorganisms in the HPA axis.

## CONCLUSIONS

Based on previous research, in the current study we predicted that intestinal microbes would be an important factor in balancing the HPA axis. Imbalances of the HPA axis caused by intestinal microbes can affect the neuroendocrine system in the brain, resulting in an anxiety-like behavioral phenotype. The current findings suggest the possibility that novel treatments could be developed for stress-related diseases, including anxiety disorders, by direct or indirect intervention in intestinal microbial flora with currently available drug treatments.

## AUTHOR CONTRIBUTIONS

RH, BZ, BL, YL, HyW, CZ, LF, WL, and RN: Performed experiments; LZ, RH, PX, and HW: Designed the study; RH and KC: Wrote the manuscript; All authors reviewed and approved the manuscript prior to its submission.

#### FUNDING

This work was supported by The National Key Research and Development Program of China (2017YFA0505700), the National Natural Science Foundation of China (grant no. 81401140 and grant no. 81601207), and China Postdoctoral Science Foundation funded project (2017M612923).

#### REFERENCES


#### ACKNOWLEDGMENTS

We thank the Third Military Medical University (Chongqing, China) for providing animals and experimental conditions.

differences in reactivity to acute stress. Neuroendocrinology 65, 360–368. doi: 10.1159/000127196


behavior. Int. J. Neuropsychopharmacol. 19:pyw020. doi: 10.1093/ijnp/ pyw020


**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 Huo, Zeng, Zeng, Cheng, Li, Luo, Wang, Zhou, Fang, Li, Niu, Wei and Xie. 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.

# Alteration of Gut Microbiota and Inflammatory Cytokine/Chemokine Profiles in 5-Fluorouracil Induced Intestinal Mucositis

Hong-Li Li † , Lan Lu† , Xiao-Shuang Wang, Li-Yue Qin, Ping Wang, Shui-Ping Qiu, Hui Wu, Fei Huang, Bei-Bei Zhang, Hai-Lian Shi\* and Xiao-Jun Wu\*

Shanghai Key Laboratory of Compound Chinese Medicines, The Ministry of Education (MOE) Key Laboratory for Standardization of Chinese Medicines, Institute of Chinese Materia Medica, Shanghai University of Traditional Chinese Medicine, Shanghai, China

#### Edited by:

Michele Marie Kosiewicz, University of Louisville, United States

#### Reviewed by:

Valerio Iebba, Sapienza Università di Roma, Italy Nour Eissa, University of Manitoba, Canada

#### \*Correspondence:

Hai-Lian Shi shihailian2003@163.com Xiao-Jun Wu xiaojunwu320@126.com

† These authors have contributed equally to this work.

Received: 20 January 2017 Accepted: 09 October 2017 Published: 26 October 2017

#### Citation:

Li H-L, Lu L, Wang X-S, Qin L-Y, Wang P, Qiu S-P, Wu H, Huang F, Zhang B-B, Shi H-L and Wu X-J (2017) Alteration of Gut Microbiota and Inflammatory Cytokine/Chemokine Profiles in 5-Fluorouracil Induced Intestinal Mucositis. Front. Cell. Infect. Microbiol. 7:455.

doi: 10.3389/fcimb.2017.00455

Disturbed homeostasis of gut microbiota has been suggested to be closely associated with 5-fluorouracil (5-Fu) induced mucositis. However, current knowledge of the overall profiles of 5-Fu-disturbed gut microbiota is limited, and so far there is no direct convincing evidence proving the causality between 5-Fu-disturbed microbiota and colonic mucositis. In mice, in agreement with previous reports, 5-Fu resulted in severe colonic mucositis indicated by weight loss, diarrhea, bloody stool, shortened colon, and infiltration of inflammatory cells. It significantly changed the profiles of inflammatory cytokines/chemokines in serum and colon. Adhesion molecules such as vascular cell adhesion molecule-1 (VCAM-1), intercellular adhesion molecule-1 (ICAM-1), and VE-Cadherin were increased. While tight junction protein occludin was reduced, however, zonula occludens-1 (ZO-1) and junctional adhesion molecule-A (JAM-A) were increased in colonic tissues of 5-Fu treated mice. Meanwhile, inflammation related signaling pathways including NF-κB and mitogen activated protein kinase (MAPKs) in the colon were activated. Further study disclosed that 5-Fu diminished bacterial community richness and diversity, leading to the relative lower abundance of Firmicutes and decreased Firmicutes/Bacteroidetes (F/B) ratio in feces and cecum contents. 5-Fu also reduced the proportion of Proteobacteria, Tenericutes, Cyanobacteria, and Candidate division TM7, but increased that of Verrucomicrobia and Actinobacteria in feces and/or cecum contents. The fecal transplant from healthy mice prevented body weight loss and colon shortening of 5-Fu treated mice. In addition, the fecal transplant from 5-Fu treated mice reduced body weight and colon length of vancomycinpretreated mice. Taken together, our study demonstrated that gut microbiota was actively involved in the pathological process of 5-Fu induced intestinal mucositis, suggesting potential attenuation of 5-Fu induced intestinal mucositis by manipulating gut microbiota homeostasis.

Keywords: gut microbiota, 5-fluorouracil, intestinal mucositis, inflammatory chemokines/cytokines, fecal transplantation

## INTRODUCTION

Gastrointestinal microbiota plays an important role in the maintenance of human health (Figlewicz, 2008; Zhao, 2013; Patel et al., 2016). Healthy gastrointestinal microbiota characterized by high rich and diverse bacteria (Vandeputte et al., 2016) interacts with mucosal epithelium and is responsible for normal substance metabolism, immune response and intestinal angiogenesis (Stringer et al., 2009a,b; Candela et al., 2014). Disturbed gut microbiota has been revealed to induce many disorders, such as metabolic diseases (obesity and diabetes) (Philippot et al., 2013), inflammatory bowel diseases (Tung et al., 2011), multiple sclerosis, and even psychiactric diseases such as depression (Wang and Kasper, 2014). More and more evidences suggest that sustained homeostasis of gut microbiota seems to benefit the recovery of many diseases.

Cancer chemotherapeutic agents have been found to interfere with the homeostasis of gut microbiota. For instance, irinotecan, a cytotoxic chemotherapy agent for colon cancer, can induce the alteration of β-glucuronidase producing bacteria of intestinal microflora (Stringer et al., 2007). Ipilimumab, a CTLA-4 blocker, even has to exert its anticancer effect through the interaction between Bacteroides fragilis (B. fragilis) and B.fragilis-specific T cells (Vétizou et al., 2015). 5-fluorouracil (5-Fu), the first-line chemotherapeutic agent for the therapy of metastatic colorectal cancer, induces gastrointestinal adverse events such as diarrhea, hemorrhage and intestinal mucositis in clinic, which not only diminish its therapeutic efficacy but also increase patient's suffering (Sonis et al., 2004; Stringer et al., 2009b). Administration of probiotics ameliorates 5-Fu induced intestinal mucositis in mice (Justino et al., 2014; Yeung et al., 2015), suggesting possible causality between the gastrointestinal microbiota and the disease. In rats, 5-Fu treatment changes the relative abundance of microbiota from several genera in gastrointestine, including Clostridium, Lactobacillus, Enterococcus, Bacteroides, Straphylococcus, Streptococcus, and Escherichia (Stringer et al., 2007). However, due to the limited techniques at that time, the profiling of the gastrointestinal microbiota was incomplete. In addition, the detailed function of gut microbiota in 5-Fu-induced gastrointestinal mucositis has not been well clarified yet.

In 5-Fu-induced mucositis rodents, chemokines/cytokines such as chemokine-1, 2, 9 (CXCL1, CXCL2, CXCL9), and interleukine-4 (IL-4) are elevated, which is accompanied with intestinal epithelium damage. Further study disclosed that CXCL9 is closely related to the intestinal damage, while IL-4 as a pro-inflammatory cytokine can increase intestinal epithelium permeability (Prisciandaro et al., 2012; Soares et al., 2013; Wang and Kasper, 2014; Lu et al., 2015; Sakai et al., 2016). NF-κB and mitogen activated protein kinase (MAPK) pathways can be activated in the small intestine of 5-Fu induced mucositis (Liu et al., 2013). However, the reciprocal association among the overall profiles of 5-Fu-induced inflammatory cytokines/chemokines, alteration of tight junction and adhesion proteins and cellular signaling pathways has not been elucidated, especially in colon tissue.

Although disturbed homeostasis of gut microbiota has been suggested to be closely associated with the adverse effect of 5-Fu, current knowledge of the overall profiles of 5-Fu-disturbed gut microbiota is limited, and so far there is no direct convincing evidence that can prove the causality between 5-Fu-disturbed microbiota and colonic mucositis. The present study was aimed to provide the overall profile of 5-Fu-disturbed gut microbiota by direct sequencing of 16S rRNA gene in cecum contents and feces of colonic mucositis mice using high throughput Miseq sequencing technologies. Meanwhile, the influence of 5-Fu on the inflammatory cytokines/chemokines, adhesion molecules, tight junction molecules as well as MAPK and NF-κB pathways in colonic tissues of mice was investigated. And the fecal transplantation experiments were conducted to elucidate the causality between gut microbiota and colonic mucositis. Our findings confirmed the important role of gut microbiota in 5- Fu induced intestinal mucositis and may provide novel therapy regimen for patients suffered from 5-Fu induced intestinal mucositis.

#### MATERIALS AND METHODS

#### Animals and Mucositis Induction

Male BALB/c mice, 4-week old, obtained from Shanghai SLAC Laboratory Animal Co. Ltd. (SYXK2014-008, Shanghai, China) were housed under a 12 h light/dark cycle at room temperature (23 ± 2 ◦C) with access to food and water ad libitum. Two weeks later, the mice were randomly divided into two groups, namely control group and 5-Fu group (n = 10/group). According to Huang et al's method (Huang et al., 2009), to induce mucositis, the 5-Fu group mice were intraperitoneally administered with 5-Fu (50 mg/kg) once daily for 3 days. Meanwhile, the control group mice were intraperitoneally administered with 0.9% saline. All animal experiments were conducted complying with the Institutional Animal Care guidelines approved by the Experimental Animal Ethical Committee of Shanghai University of Traditional Chinese Medicine.

#### Mucositis Assessment and Samples Collection

Body weight, diarrhea and bloody stool of mice were recorded daily for the assessment of mucositis. Diarrhea grade was evaluated based on the consistency of stool, using the modified parameters as described previously (Leocádio et al., 2015): 0, normal; 1, slightly wet; 2, moderate wet; 3, loose; 4, watery stool. At the last day (day 7), the grade of blood stool was assessed by a commercial testing paper (BASO diagnostics Inc. China) with the following scores: 0, normal; 1, slight bleeding; 2, moderate

**Abbreviations:** 5-Fu, 5-fluorouracil; MPO, myeloperoxidase; IFN-γ, interferonγ; IL-1/6, interleukin-1/6; TNF-α, tumor necrosis factor-α; CXCL1/5/9/13, chemokine (C-X-C motif) ligand 1/5/9/13; IL-22 R1, interleukin-22 receptor 1; IL-12 R2, interleukin-12 receptor 2; VCAM-1, vascular cell adhesion molecule-1; ICAM-1, intercellular adhesion molecule-1; G-CSF, granulocyte colony stimulating factor; GM-CSF, granulocyte-macrophage colony stimulating factor; BSA, bovine serum albumin; MAPK, mitogen activated protein kinase; ERK, extracellular signal-regulated kinase; SAPK/JNK, stress activated protein kinase/jun N-terminal kinase; p-ERK, phosphorylated extracellular signalregulated kinase; p-JNK, phosphorylated jun N-terminal kinase; iNOS, inducible NO synthase; ZO-1, zonula occludens-1; JAM-A, junctional adhesion molecule-A.

bleeding; 3, severe bleeding; 4, visible bleeding. Meanwhile, the feces were collected and stored at −80◦C. Then the mice were sacrificed under anesthesia, and the entire small intestine and colon were excised after removal of fat tissue and their length were measured. The colon tissues near the cecum were either fixed in 10% formalin (w/v) or snap frozen in liquid nitrogen for further analysis.

#### Protein Chip analysis

Colon tissues were homogenized with cell lysis buffer containing protease inhibitor cocktail on ice and centrifuged at 12,000 rpm for 15 min at 4◦C. The supernatant was collected and subjected to concentration measurement using BCA method. Afterwards, all protein samples were diluted to the same concentration. Inflammatory/anti-inflammatory cytokines in the samples were measured by RayBio <sup>R</sup> Mouse Cytokine Antibody Arrays according to the manufacturer's protocol.

#### Histopathological Assessment

Fixed colon tissue samples were embedded in paraffin, sectioned in 4 µm-thick slices, and stained with hematoxylin-eosin. The morphological alteration and inflammatory cell infiltration were observed under microscope (Olympus BX61VS).

#### Immunohistochemistry

The endogenous peroxidases in 4 µm-thick slices were deactivated by incubation with 3% H2O<sup>2</sup> for 10 min. For antigen retrieval, the sections were soaked in 10 mM citrate buffer solution (pH 6.0) and heated twice in a microwave oven. After washed thoroughly with PBS (pH7.4), the sections were blocked with 3% BSA in tris buffered saline (TBS) for 20 min, then incubated with anti-myeloperoxidase antibody (anti-MPO) (1:200, #SH0022, Skyhobio) and anti-p65 (1:400, #SH0023, Skyhobio) antibodies overnight at 4◦C followed by incubation with HRP-conjugated secondary antibody (#K5007, Dako) for 50 min. The sections were further incubated with DAB-H2O<sup>2</sup> solution (#K5007, Dako), counterstained with hematoxylin, dehydrated with ethanol and sealed in resinene for microscopic observation.

## Quantitative Polymerase Chain Reaction (qPCR)

Total RNA was extracted from colon tissues using TRIzol reagent (Life Technologies). cDNA was generated from total RNA with the RevertAid First Strand cDNA Synthesis Kit (Thermo). The primers (GeneRay) used in PCR amplification were listed in **Table 1**. Quantitative PCR was performed with SYBR Premix EX Taq under the following conditions: 95◦C, 30 s; then followed by 40 cycles (95◦C, 5 s; 60◦C, 34 s); finally 95◦C, 15 s; 60◦C, 1 min; 95◦C, 15 s. Quantity of target genes calculated by the comparative C<sup>t</sup> method was normalized to that of β-actin (internal reference) in the same sample (Araújo et al., 2015).

#### Multiplex Immunoassays

Serum was collected by centrifugation at 4,000 rpm for 10 min at 4◦C. Colon segments were homogenized in cell lysis buffer, then the supernatants were collected through centrifugation at 12,000 rpm for 15 min. Concentrations of 13 cytokines in TABLE 1 | The primers used in qPCR analysis.


the supernatants and serum were measured by ProcartaPlex <sup>R</sup> Mix&Match Mouse 13-plex [including interleukin-6 (IL-6), tumor necrosis factor-α (TNF-α), interleukin-10 (IL-10), interleukin-12p-70 (IL-12p70), interleukin-21 (IL-21), interleukin-22 (IL-22), interleukin-31 (IL-31), granulocyte colony stimulating factor (G-CSF), granulocyte-macrophage colony stimulating factor (GM-CSF), Leptin, RANTES, chemokine-5 (CXCL5), and chemokine-1 (CXCL1)] according to the manufacturer's recommendation.

## ELISA Assay

Concentration of CXCL9 in serum and supernatants of colonic tissues was quantified by CXCL9 ELISA assay kit (Abcam, UK) according to manufacturer's instruction.

#### Western Blot Analysis

Colon tissues were homogenized and lysed in RIPA buffer supplemented with protease inhibitor cocktail on ice. After centrifugation at 12,000 rpm for 15 min at 4◦C, the supernatant was collected and its protein concentration was determined by BCA method. Total protein (60 µg) from each sample was separated by SDS-PAGE and transferred onto PVDF membrane by wet transfer approach. Then PVDF membranes were blocked with 5% (w/v) bovine serum albumin (BSA) solution and incubated with different primary antibodies against p-p65 (1:1000, #3033L, Cell Signal Technology), p-IκBα (1:500, #2859S, CST), p-p38 MAPK (1:1000, #4511, CST), phosphorylated extracellular signal-regulated kinase (p-ERK1/2, 1:1000, #9154, CST), phosphorylated jun N-terminal kinase (p-JNK, 1:1000, #4668, CST), p38 MAPK (1:1000, #9212, CST), extracellular signal-regulated kinase (ERK1/2, 1:1000, #4695, CST), stress activated protein kinase/jun Nterminal kinase (SAPK/JNK, 1:1000, #9252S, CST), inducible NO synthase (iNOS, 1:1000, #ab204017, Abcam), VCAM-1(1:2000, #3540-1, Epitomics), ICAM-1 (1:1000, #3482-1, Epitomics), Occludin (1:2000, #GTX85016, GeneTex), ZO-1 (1:500, #ab59720, Abcam), JAM-A (1:500, # sc-37049, Santa cruz) and β-actin (1:2000, #12413, CST) overnight at 4◦C. After washed with 1 × PBS containing 0.1% (v/v) Tween-20, the membranes were incubated with respective secondary antibodies. The protein bands were visualized with ECL-prime kit. Quantification of target protein was performed by measuring integral optic density of respective target proteins with Tanon Gis software.

#### 16S rRNA Miseq Sequencing and Bioinformatic Analysis

Microbial genomic DNA was extracted from cecum contents and feces using a QIAamp DNA Stool Mini Kit according to the manufacturer's instructions. The resultant DNA extracts were used for the PCR amplification. Quantification of the PCR products was performed on FTC-3000TM real-time PCR instrument. The V3-V4 region of 16S rRNA gene of gut microbiota was sequenced using Illumina MiSeq 2 × 300 bp high throughput platform. The bioinformatic analysis was conducted as described previously (MacIntyre et al., 2015). The generated 16S rRNA gene sequences were analyzed using the bioinformatic software package Mothur with MiSeq SOP Pipeline. The paired reads were assembled using make.contigs. Screen.seqs command was used to remove low quality reads using the following filtering parameters, maxambig = 0, minlength = 200 and maxlength = 580, maxhomop = 8. The remained sequences were simplified using the unique.seqs command to generate a unique set of sequences, then aligned with the SILVA databases (version 119). The screen.seqs command was implemented again to keep within our defined criteria using the following parameters: start = 12,878, end = 28,464. The filter.seqs was used to remove empty columns from our alignment. Further de-noise sequences were pre-clustered using the pre.cluster command (http://www. mothur.org/wiki/Pre.cluster) allowing for up to 4 differences between sequences. Then reads were checked for chimeras using UCHIME algorithm and the chimeric sequences were removed by the chimera.uchime command with default parameters. To classify (classify.seqs) the sequences, the SILVA 119 database was used with a confidence threshold of 80%. The non-bacterial sequences were deleted. The distance matrix between the aligned sequences was generated by the dist.seqs command. Finally, these sequences were clustered to OTUs (operational taxonomic units) at 97% sequence identity (furthest neighbor method). A majority of consensus taxonomy for each OTU was obtained by the classify.otu command with default parameters.

#### Fecal Transplantation

For healthy fecal transplantation experiment, 24 mice were randomly divided into three groups: Control, 5-Fu and 5- Fu+feces (n = 8/group). Both 5-Fu group and 5-Fu+feces group mice were injected intraperitoneally with 5-Fu (50 mg/kg/day) for 3 days. For 5-Fu+feces group mice, they were additionally administered with the fecal suspension from normal mice via oral gavage from day 1 to day 7 once a day. For 5-Fu-treated fecal transplantation experiment, 40 mice were randomly divided

Li et al. 5-Fu Disturbed Gut Microbiota Homeostasis

into four groups, namely Control, 5-Fu, Con-feces and 5-Fufeces (n = 10/group). The mice in Control and 5-Fu groups were treated as aforementioned. Fecal pellets from Control group and 5-Fu group mice were collected and suspended in sterile PBS. For Con-feces group and 5-Fu-feces group mice, they were pretreated with vancomycin (100 mg/kg) for 3 days (Ubeda et al., 2013; Warn et al., 2016), then were administered with respective fecal suspension from Control group or 5-Fu group mice by oral gavage for 11 days. Body weight, diarrhea, and bloody stool of mice were recorded daily. At last, the mice were sacrificed under anesthesia and the length of entire colon after removal of fat tissue was measured.

#### Statistical Analysis

Each value was presented as mean ± S.E.M. Differences between two groups were analyzed by un-paired Student's t-test using PrismDemo 5. In all cases, the value of P < 0.05 was considered statistically significant.

## RESULTS

## 5-Fu Induced Colonic Mucositis

Consistent with previous studies (Pereira et al., 2016), body weight of 5-Fu-treated mice was dramatically decreased from day 2 to day 7 after 5-Fu treatment (**Figure 1A**, P < 0.05 or P < 0.001), compared with the control mice. Meanwhile, severe diarrhea was found in 5-Fu group mice from day 5 to day 7 (**Figure 1B**, P < 0.001). At day 7, severe bloody stool was found in 5-Fu treated mice (**Figure 1C**). Shortened intestine indicates the increased contraction ability (Dou et al., 2013), while the shortened colon is closely associated with severe diarrhea. In our experiments, the small intestine length of 5-Fu treated mice was not changed compared to that of the control (**Figure 1D**). By contrast, the colon length of 5-Fu treated mice was significantly shortened (**Figures 1E,F**, P < 0.001) and the cecum of 5-Fu treated mice seemed to be smaller (**Figure 1E**). Moreover, 5-Fu treatment injured mucosal epithelium and disrupted crypt-villus structures, which was accompanied with enhance cellular infiltration (HE staining) and neutrophil (MPO staining) infiltration (**Figures 1G,H**).

### 5-Fu Altered Inflammatory Cytokine and Chemokine Profiles

Although previous studies exposed the alteration of several inflammatory factors in 5-Fu induced intestinal mucositis (Justino et al., 2014; Lu et al., 2015), the changed profile of the other inflammatory factors involved in the process has not been explored. In present study, a mouse inflammation antibody array (40 inflammatory factors) was employed to preliminarily examine the alteration of inflammatory factor profile. As shown in **Figure 2A**, compared to the control, 5-Fu seemed to elevate the protein levels of KC (CXCL1), LIX (CXCL5), MIG (CXCL9), B-lymphocyte chemoattractant (BLC), IL-6 and sTNFR I (>1.5 fold) but decrease that of G-CSF, IL-12p40/p70, RANETS, CD30L, Fractalkine, IL-10, IL12p70, Leptin, and TIMP-2 (>1.3 fold) in colonic tissues. In terms of mRNA expression of the cytokines/chemokines, 5-Fu treatment induced the mRNA

FIGURE 1 | 5-Fu induced mucositis and colon shortening in mice. (A) 5-Fu induced body weight changes. Data were plotted as percentage of initial body weight. (B) The occurrence of diarrhea. Data represented the evaluation scores of diarrhea. (C) The bloody stool events measured by BASO testing paper. (D) The small intestine length. (E,F) The colon length. (G) HE staining of colonic sections. (H) MPO staining of colonic sections. Values were expressed as mean ± S.E.M (n = 10/group). Data were analyzed by t-test. \*P < 0.05, \*\*\*P < 0.001 vs. control group.

expression of G-CSF, CD11b, iNOS, COX-2, interferon-γ (IFNγ), IL-1β, IL-6, TNF-α, CXCL5, CXCL9, CXCL13, and CXCL1 (**Figures 2B,C**, P < 0.05, P < 0.01 or P < 0.001), but decreased that of TIMP2 and RNATES in colonic tissues. Moreover, as shown in **Figure 2D**, 5-Fu treatment modulated the mRNA expression of cytokine/chemokine receptors, as it upregulated the mRNA expression of chemokine (C-X-C motif) receptor 2, 3 (CXCR2, CXCR3), sTNFR I, sTNFR II and interleukin-22 receptor 1 (IL-22R1), however, down-regulated that of interleukin-10 receptor 2 (IL-10R2). In order to further confirm the changes of inflammatory factors, the multiplex immunoassays and ELISA assay were performed, respectively. As illustrated in **Figures 2E,F**, in serum of 5-Fu-induced mice, the protein levels of CXCL9, CXCL1 (KC), CXCL5, IL-22, IL-6, TNF-α, GM-CSF, and G-CSF were significantly increased (P < 0.05, P < 0.01, or P < 0.001), but that of RNATES, Leptin, and IL-31 were significantly decreased (P < 0.05, P < 0.01, or P < 0.001). Similarly, in colonic tissues of 5-Fu-induced mice, the protein levels of IL-22, G-CSF, IL-6, TNF-α, CXCL1 (KC), and CXCL5 were significantly elevated (**Figures 2G,H**, P < 0.01 or P < 0.001), while that of Leptin was significantly reduced (p < 0.001). By contrast, protein level of IL-12p70 did not change in both serum and colonic tissues.

#### 5-Fu Modulated the Expression of Tight Junctions (TJ) and Adhesion Proteins

Tight junction supports the integral intestinal epithelial barrier structure and barrier function, which is disrupted under inflammation (Capaldo et al., 2017; Chang et al., 2017). Adhesion molecules mediate the attachment of lymphocytes, neutrophils and inflammatory cells to the endothelial cells under inflammatory condition (Erbeldinger et al., 2017; Kim et al., 2017). As shown in **Figure 3**, 5-Fu treatment induced significant mRNA expression of adhesion molecules, VCAM-1, ICAM-1,

and VE-Cadherin (P < 0.001, P < 0.001, and P < 0.05) as well as the protein expression of VCAM-1 and ICAM-1 (P < 0.001) in colon. However, in terms of tight junction proteins, 5-Fu decreased mRNA and protein expression of occludin (P < 0.001, P < 0.001). But 5-Fu increased the protein level of JAM-A and ZO-1 (P < 0.001 and P < 0.01).

## 5-Fu Activated MAPK and NF-κB Pathway Signaling

MAPK and NF-κB pathways are closely associated with inflammation (Park et al., 2013). To determine whether MAPK and NF-κB pathways were involved in 5-Fu-induced colonic mucositis, we further assessed the effect of 5-Fu treatment on the activation of signaling molecules, including ERK1/2, JNK, p38 MAPK, IκB and NF-κB. As shown in **Figure 4**, 5-Fu enhanced the phosphorylation of ERK1/2, JNK, p38 MAPK, IκB and NFκB as well as the protein expression of iNOS in the colon (P < 0.001, or P < 0.01). Moreover, 5-Fu treatment increased the expression of activated NF-κB in the intestinal epithelial cells (**Figure 4E**). All of these results indicated that 5-Fu treatment resulted in the activation of MAPK and NF-κB signaling pathways.

#### 5-Fu Altered Bacterial Diversity and Community Composition

Gut microbiota has been indicated in inflammatory bowel disease (Terán-Ventura et al., 2014; Patel et al., 2016). Alteration of gut microbiota composition may affect the function of mucosal immune system, resulting in the intestinal inflammation (Autenrieth and Baumgart, 2017; Etienne-Mesmin et al., 2017; Holleran et al., 2017). Therefore, to clarify the change of gut microbiota of 5-Fu treated mice, the diversity and composition of gut microbiota in cecum contents and feces were analyzed by Miseq sequencing. The Chao community richness and Shannon diversity were used to estimate within-community diversity (α-diversity). Sequencing of 16S rRNA gene V3-V4 region of gut microbiota showed that 5-Fu greatly decreased the community richness of microbiota in both feces and cecum contents, compared with the controls (**Figure 5A**, P < 0.001). It significantly decreased the Shannon diversity in cecum contents but not that in feces of mice (**Figures 5B,C**, P < 0.01). Unweighted UniFrac PCoA analysis demonstrated that there was a significant difference between control and 5-Fu treated mice regarding beta-diversity at OTUs level (**Figures 5D,E**). These results indicated that 5-Fu treatment led to the richness and diversity loss in the bacterial community, especially in cecum contents.

The four major phyla in the feces and cecum contents were Bacteroidetes, Verrucomicrobia, Firmicutes, and Proteobacteria (**Figures 6A,B**, Table S1), among which Bacteroidetes and Verrucomicrobia were the relatively abundant ones. 5-Fu treatment remarkably decreased the relative abundance of Firmicutes, Proteobacteria, and Cyanobacteria at phyla level in feces (P < 0.05 or P < 0.01). However, 5-Fu increased the

abundance of Verrucomicrobia (P < 0.05), although it also reduced that of Firmicutes and Cyanobacteria (P < 0.01) in cecum contents. In addition, 5-Fu significantly decreased the ratio of Firmicutes/Bacteroidetes (F/B) in cecum contents and feces (**Figure 6C**, p < 0.001, p < 0.05). Further correlation analysis (**Figures 6D,E**) showed that F/B ratio positively correlated with body weight change (Spearman's R = 0.7761, P < 0.001 in cecum; Spearman's R = 0.6525, P < 0.05 in feces). More information about gut microbiota in cecum contents and feces could be found in Supplementary Data (Tables S1–12).

#### Disturbed Gut Microbiota was Involved in Body Weight Loss and Colon Shortening in 5-Fu Induced Colonic Mucositis

As shown in **Figure 7A**, from day 4, fecal transplantation significantly rescued the body weight loss of mice induced by 5- Fu treatment (P < 0.05). Furthermore, at day 7, fecal microbiota

feces and cecum contents. Chao is an estimator of the community richness. (B,C) 5-Fu decreased the diversity of gut microbiota in feces (B, Shannon index curves and Shannon diversity histogram) and cecum contents (C, Shannon index curves and Shannon diversity histogram). (D,E) The unweighted UniFrac PCoA results of feces (D) and cecum contents (E) for beta-diversity at OTUs level. Values were expressed as mean ± S.E.M (n = 5/group, feces; n = 10/group, cecum contents). Data were analyzed by t-test. \*\*P < 0.01, \*\*\*P < 0.001 vs. control group.

transplantation prevented the shortening of colon induced by 5- Fu treatment (**Figures 7B,C**) (P < 0.01). In another experiment, to assess the effect of fecal microbiota on 5-Fu induced colonic mucositis, the vancomycin-pretreated mice were transplanted with feces from Control and 5-Fu group mice, respectively. As shown in **Figures 7D–F**, compared to mice transplanted with normal feces, mice transplanted with feces from 5-Fu treated mice showed significant body weight loss and shortened colon. These results implicated that disturbed gut microbiota contributed to the induction of intestinal mucositis in 5-Fu treated mice.

#### DISCUSSION

Although previous studies have indicated that gut microbiota plays an important role in 5-Fu induced gastrointestinal mucositis (Stringer et al., 2009b; Chang et al., 2012; Gao et al., 2014), however, none of them described the causal relationship in a systemic way. In present study, we analyzed the alteration of gut microbiota and inflammatory cytokine/chemokine profiles with relatively systemic methods. Our findings showed that, besides small intestine mucositis, 5-Fu also induced colonic mucositis. Both gut microbiota and inflammatory cytokine/chemokine profiles were altered significantly, which was accompanied with mucosal barrier disruption and inflammatory signaling pathway activation. Further studies revealed that fecal transplant from healthy mice alleviated the severity of colonic mucositis, while that from 5-Fu treated mice seemed to induce significant symptoms of colonic mucositis. Our results indicated that the recovery of homeostasis of gut microbiota by fecal transplantation might facilitate the relief of gastrointestinal mucositis induced by 5-Fu.

Pro-inflammatory cytokines and anti-inflammatory factors play a critical role in inflammatory bowel diseases (Stringer et al., 2009b). The expression levels of IL-6, TNF-α, IL-1β, IFNγ, CXCL1 were shown to increase in small intestine of 5-Fuinduced intestinal mucositis mice (Soares et al., 2011, 2013; Chang et al., 2012; Yasuda et al., 2013; Yeung et al., 2015). In present study, IL-6, TNF-α, IL-1β were remarkably increased in serum and/or colon tissue at both mRNA and protein levels in 5-Fu induced colonic mucositis mice. Meanwhile, 5- Fu elevated the levels of IFN-γ, G-CSF, GM-CSF, and CD11b,

while decreased that of RNATES and IL-31. Interestingly, leptin, a hormone produced and secreted by adipose tissue, muscle and stomach, was also detected in colonic tissue. And, 5-Fu could significantly decrease leptin both in serum and colonic tissues. Leptin treatment has been shown to promote intestinal recovery and enhance enterocyte turnover in a rat model of methotrexateinduced mucositis (Sukhotnik et al., 2009). The decreased leptin in serum and colon might partly reflect the excerbation of colonic mucositis induced by 5-Fu. CXCL9 treatment has been disclosed to attenuate 5-Fu induced mucositis (Han et al., 2011), however, it exacerbates 5-Fu induced acute intestinal damage (Lu et al.,

2015). Therefore, further investigation is needed to clarify the effect of CXCL9 on 5-Fu induced mucositis. So far the role of CXCL5 in 5-Fu induced mucositis has not been elucidated yet. CXCL13 mediates T cell recruitment and participates in the regulation of inflammatory response (Hui et al., 2015). IL-22 produced by T cells and NK cells participates in tumorigenesis and tumor progression, and mediates chemoresistance (Wu et al., 2014), which is enhanced in colon of 5-Fu induced mice (Sakai et al., 2013). In present study, 5-Fu treatment significantly modulated the levels of CXCL5, CXCL9, CXCL13, and IL-22 in serum and/or colonic tissues. Moreover, 5-Fu increased gene

correlation between body weight changes and F/B ratio in cecum contents. (E) The correlation between body weight changes and F/B ratio in feces. Values were expressed as mean ± S.E.M (n = 5/group, feces; n = 10/group, cecum contents). Data were analyzed by t-test. \*P < 0.05, \*\*\*P < 0.001 vs. control group.

expression of CXCR2 (receptor of CXCL1, CXCL5), CXCR3 (receptor of CXCL9), sTNFRI and sTNFRII (receptors of TNFα), and IL-22R1 (receptor of IL-22), while reduced that of IL-10R2 (receptor of IL-10) in colonic tissue. All of these results implicated that 5-Fu induced colonic mucositis along with significant inflammatory responses.

10/group). Data were analyzed by t-test. \*P < 0.05, \*\*P < 0.01 vs. control group.

TJs maintain the intestinal mucosal barrier (Yang et al., 2015). Reduction of TJs expression always indicates the increased intestinal epithelial permeability (Park et al., 2015; Yang et al., 2015). Chemotherapeutic drug could increase intestinal epithelial barrier permeability via reducing protein expression of TJs (Beutheu Youmba et al., 2012). Inflammatory infiltration is a characteristic of mucositis, which is triggered by the increased adhesion molecules in intestinal endothelia that attract the circulating inflammatory cells including neutrophils, T lymphocyte cells, B lymphocyte cells to gather in the inflammatory sites (Erbeldinger et al., 2017; Kim et al., 2017). The inflammatory cells further accelerate the modification of tight junction, thereby increase intestinal permeability leading to the disruption of mucosal barrier (Leocádio et al., 2015). The elevated pro-inflammatory cytokines induced by 5-Fu have been shown to account for the loss of tight junction proteins of small intestine, such as occludin and claudin-1, and result in diarrhea (Patel et al., 2016). In colonic tissues, our results demonstrated that 5-Fu treatment disrupted tight junction as the expression of occludin was down-regulated at both mRNA and protein levels. Surprisingly, ZO-1 protein was upregulated by 5-Fu in our study. It is well-known that ZO-1 is a cytoplasmic scaffolding protein, which is breakdown or redistributed under TNF-α-induced inflammatory condition (Chen et al., 2017; Watari et al., 2017). However it did not show significant change in mucosa of 5-Fuinduced small intestine (Song et al., 2013) or irinotecan-induced gut toxicity (Wardill et al., 2014), suggesting that its specific role on mucosal barrier under these inflammatory conditions. We don't know whether there is a compensatory mechanism in ZO-1, because of the decreased expression of occludin in 5-Fu induced colonic mucositis. Meanwhile, adhesion proteins such as ICAM-1, VCAM-1, JAM-A, and VE-Cadherin were increased by 5-Fu. These results indicated that 5-Fu treatment might increase the colonic epithelial barrier permeability through decreasing TJ proteins and up-regulating adhesion proteins to recruit inflammatory cells to colonic epithelium, and then enhance the translocation of gut microbiota in the mucosa to promote the inflammation.

NF-κB and MAPK pathways can be activated by many inflammatory chemokines/cytokines, which may result in a pro-inflammatory chemokines/cytokines positive feedback (Zimmerman et al., 2008; Tung et al., 2011; Chang et al., 2012; Jiang et al., 2012; Song et al., 2013; Candela et al., 2014; Dou et al., 2014; Yeung et al., 2015). And their activation in the small intestine has been shown to be involved in 5-Fu induced mucositis (Liu et al., 2013). Also, methotrexate (MTX) treatment increased intestinal permeability partially related to the decreased TJs protein expression through MAPK and NF-κB pathways (Beutheu Youmba et al., 2012). In agreement with the report, in our study, enhanced phosphorylation of NFκB and MAPK pathway molecules were also found in colonic tissue of 5-Fu treated mice, indicating their active participation in regulating expression of tight junction proteins and colonic proinflammatory cytokines/chemokines in the pathogenesis of 5-Fu induced colonic mucositis.

Disturbed gut microbiota, in either diversity or abundance, has been found to play an important role in the pathological development of inflammatory bowel diseases (Juste et al., 2014; Rangel et al., 2016). At genus level, 5-Fu treatment has been shown to decrease Clostridium, Lactobacillus, Streptococcus and Enterococcus and increase Escherichia in rat jejunum or colon (Stringer et al., 2007, 2009b). Different from the findings in rat, in present study, 5-Fu significantly decreased Odoribacter, Candidatus Saccharimonas and Marvinbryantia, and increased Helicobacter and Thalassospira in mouse feces. Moreover, 5- Fu significantly changed the abundance of Blautia, Alistipes, Coprococcus, Roseburia, Akkermansia, Bilophila, Candidatus Saccharimonas, and Mucispirillum in mouse cecum contents. The difference between our findings and previous reports might reflect the variable microbiota profiles affected by multiple factors such as environment, diet, gender, age, and species. At phylum level, microbiota low in Firmicutes has been disclosed to enhance the intestinal sensitivity to inflammation (Natividad et al., 2015). Decreased abundance of Ruminococcaceae and Lachnospiraceae families belong to Firmicutes phylum is found to be associated with inflammatory states (Knip and Siljander, 2016). On the contrary, cocktail of Ruminococcaceae and Lachnospiraceae families can efficiently reverse experimental colitis induced by dextran sodium sulfate (DSS) (Natividad et al., 2015). The ratio of Firmicutes/Bacteroidetes seems to be important for the maintenance of physiological state as the relative abundance of Firmicutes/Bacteroidetes influences body weight of animals, especially in metabolic diseases (Turnbaugh et al., 2006; Kassinen et al., 2007; Remely et al., 2014). Proteobacteria has been shown to play a crucial and active role in overall gut metabolism and host response despite their low abundance (Pérez-Cobas et al., 2013). Cyanobacteria inhibits inflammation by production of anti-inflammatory pitinoic acids B and C (Montaser et al., 2013). On the contrary, Verrucomicrobia appears to contribute to inflammation as their abundance bloomed in mice treated with DSS (Nagalingam et al., 2011). In present study, 5- Fu reduced the richness and diversity of gut microbiota, the relative abundance of Lachnospiraceae and Ruminococcaceae families (Tables S4, S10) accompanied with a lower ratio of Firmicutes/Bacteroidetes. Moreover, 5-Fu lessened the relative abundance of Proteobacteria, Candidate division TM7 and Cyanobacteria while increased that of Verrucomicrobia and

#### REFERENCES

An, J., and Ha, E. M. (2016). Combination therapy of Lactobacillus plantarum Supernatant and 5-Fluouracil increases chemosensitivity in colorectal cancer cells. J. Microbiol. Biotechnol. 26, 1490–1503. doi: 10.4014/jmb.1605.05024

Actinobacteria. These results implicated the active involvement of the microbiota in 5-Fu induced colonic mucositis. We also found that there was significant positive correlation between body weight changes and F/B ratio in feces and cecum contents. Our further results showed that fecal transplantation from normal mice could partly reverse the body weight loss and colon length decrease of 5-Fu treated mice. Moreover, feces from 5- Fu treated mice could also result in body weight loss and colon length decrease in normal mice pretreated with vancomycin. These results demonstrated that gut microbiota dysfunction at least partly accounted for the mucositis induced by 5-Fu. And the increased colonic epithelial barrier permeability induced by 5-Fu, would promote the translocation of gut bacteria in the intestinal mucosa, then to increase the inflammatory response (Escobedo et al., 2014; Mayer et al., 2015; Severance et al., 2016; Leung and Yimlamai, 2017).

In summary, ourresults indicated that 5-Fu induced mucositis might be partly mediated by the disturbance of gut microbiota. Since modulation gut microbiota by administration of probiotics or certain gut microbiota metabolite seemed to benefit 5-Fuinduced mucositis (Justino et al., 2014; González-Sarrías et al., 2015; An and Ha, 2016; Flórez et al., 2016), our findings provided potential novel therapeutic strategy for patients suffered from 5- Fu induced intestinal mucositis by manipulation of specific gut microbiota.

#### AUTHOR CONTRIBUTIONS

HS and XWu designed all the experiments, analyzed data and wrote the paper, and the performances of HS and XWu were equal in this study. HL and LL carried out the main experiments, and the performances of HL and LL were equal in this study; XWang, LQ, PW, SQ, and HWu performed parts of experiments. FH, and BZ provided valuable suggestions for this study and helped to draft the manuscript. All authors read and approved the final manuscript.

#### ACKNOWLEDGMENTS

This work was supported by the National Natural Science Foundation of China (81603354, 81673626), Shanghai Eastern Scholar Program (2013-59), and Shanghai E-research Institute of Bioactive Constituent in TCM plan.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcimb. 2017.00455/full#supplementary-material

Araújo, C. V., Lazzarotto, C. R., Aquino, C. C., Figueiredo, I. L., Costa, T. B., Alves, L. A., et al. (2015). Alanyl-glutamine attenuates 5 fluorouracil-induced intestinal mucositis in apolipoprotein E-deficient mice. Braz J. Med. Biol. Res, 48, 493–501. doi: 10.1590/1414-431X201 44360


**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, Lu, Wang, Qin, Wang, Qiu, Wu, Huang, Zhang, Shi and Wu. 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 Influence of Proton Pump Inhibitors on the Fecal Microbiome of Infants with Gastroesophageal Reflux—A Prospective Longitudinal Interventional Study

Christoph Castellani <sup>1</sup> , Georg Singer <sup>1</sup> \*, Karl Kashofer <sup>2</sup> , Andrea Huber-Zeyringer <sup>1</sup> , Christina Flucher <sup>1</sup> , Margarita Kaiser <sup>1</sup> and Holger Till <sup>1</sup>

*<sup>1</sup> Department of Paediatric and Adolescent Surgery, Medical University of Graz, Graz, Austria, <sup>2</sup> Institute of Pathology, Medical University of Graz, Graz, Austria*

Proton pump inhibitors (PPIs) are the standard therapy for gastroesophageal reflux disease. In adults, PPI treatment is associated with *Clostridium difficile* infections (CDI). In contrast to adults the microbiome of infants develops from sterility at birth toward an adult-like profile in the first years of life. The effect of PPIs on this developing microbiome has never been studied. The aim of the present study was to determine the effect of oral PPIs on the fecal microbiome in infants with gastroesophageal reflux disease (GERD). In this prospective longitudinal study 12 infants with proven GERD received oral PPIs for a mean period of 18 weeks (range 8–44). Stool samples were collected before ("before PPI") and 4 weeks after initiation of PPI therapy ("on PPI"). A third sample was obtained 4 weeks after PPI discontinuation ("after PPI"). The fecal microbiome was determined by NGS based 16S rDNA sequencing. This trial was registered with clinicaltrials.gov (NCT02359604). In a comparison of "before PPI" and "on PPI" neither α- nor β-diversity changed significantly. On the genus level, however, the relative abundances showed a decrease of *Lactobacillus* and *Stenotrophomonas* and an increase of *Haemophilus*. After PPI therapy there was a significant increase of α- and β-diversity. Additionally, the relative abundances of the phyla Firmicutes, Bacteroidetes, and Proteobacteria were significantly changed and correlated to patients' age and the introduction of solid foods. PPI treatment has only minor effects on the fecal microbiome. After discontinuation of PPI treatment the fecal microbiome correlated to patients' age and nutrition.

#### Keywords: proton pump inhibitors, microbiome, infants, GERD, Clostridium difficile

## INTRODUCTION

Gastroesophageal reflux (GER) is a common finding in infants caused by temporary relaxations of the immature lower esophageal sphincter (LES) (Vandenplas et al., 2007). With maturation of the LES in the first year of life GER events often decrease. Some infants, however, may develop gastroesophageal reflux disease (GERD) associated with vomiting, feeding problems, pain, esophagitis, failure to thrive and/or recurrent respiratory infections (Rudolph et al., 2001; Colletti and Di Lorenzo, 2003).

#### Edited by:

*Nathan W. Schmidt, University of Louisville, United States*

#### Reviewed by:

*Gena D. Tribble, University of Texas Health Science Center at Houston, United States Xingmin Sun, University of South Florida, United States*

> \*Correspondence: *Georg Singer georg.singer@medunigraz.at*

Received: *12 July 2017* Accepted: *29 September 2017* Published: *11 October 2017*

#### Citation:

*Castellani C, Singer G, Kashofer K, Huber-Zeyringer A, Flucher C, Kaiser M and Till H (2017) The Influence of Proton Pump Inhibitors on the Fecal Microbiome of Infants with Gastroesophageal Reflux—A Prospective Longitudinal Interventional Study. Front. Cell. Infect. Microbiol. 7:444. doi: 10.3389/fcimb.2017.00444*

In these infants conservative therapy includes upright positioning, increased feeding frequencies with lower amounts and food thickeners (Hollwarth, 2012). Nevertheless, some children may require acid suppression therapy with proton pump inhibitors (PPI). In adults, possible side effects of prolonged PPI therapy include an increased risk of community acquired enteritis and Clostridium difficile infections (CDI) (Janarthanan et al., 2012; Bouwknegt et al., 2014; McDonald et al., 2015).

The influence of PPI therapy on the intestinal microbiome has only been studied in adults under PPI therapy demonstrating dramatic changes of both the gastric and esophageal microbial communities (Amir et al., 2014). Furthermore, examinations of fecal samples have shown an increased abundance of Enterococcae and Streptococcae as well as decreased Clostridiales, associated with an increased risk of CDI (Freedberg et al., 2015b). Recent reports also describe an increased risk of CDI infections in infants under acid suppression treatment (Brown et al., 2015; Freedberg et al., 2015a). The exact pathophysiological mechanism of this association, however, is poorly understood. The two most common theories are (1) PPI directly affect the microbial environment by increasing the gastric pH and/or (2) PPI directly target bacterial proton pumps containing P-type ATPase enzymes (Vesper et al., 2009).

The gut microbiome in infancy develops from sterility at birth to an adult-like profile [dominated by the phyla Firmicutes (50–70% total bacterial numbers), Bacteroidetes (10– 30%), Proteobacteria (up to 10%) and Actinobacteria (up to 10%), (Eckburg et al., 2005)] within the first years of life (Palmer et al., 2007; Yatsunenko et al., 2012). In this period a longitudinal investigation of fecal samples has revealed an increase of the total number of colonizing bacteria as well as unstable and heterogenic relative abundances of the different phyla (Palmer et al., 2007). Thus, data derived from the "stable" microbiome in adults are not representative for infants (Palmer et al., 2007; Yatsunenko et al., 2012).

PPI-associated changes of the microbiome have not been studied in infancy yet. Therefore, the aim of this prospective longitudinal interventional investigation was to assess the influence of oral PPI therapy on the fecal microbiome of infants with proven GERD.

#### MATERIALS AND METHODS

According to our institutional protocol all patients with suspected GERD undergo 24 h-pH-impedance monitoring (24 h-pH-MII). After ethical approval (Ethical Committee of the Medical University of Graz, 26-429 ex 13/14) and informed consent of parents or legal guardians patients younger than 1 year with proven GERD were enrolled in this study between November 2014 and August 2016. Patients with relevant additional diagnoses were excluded. This trial was registered with clinicaltrials.gov (NCT02359604).

A first stool sample was taken before initiation of PPI therapy, stored in a PSP <sup>R</sup> Spin Stool DNA Kit (Stratec molecular GmbH, Berlin, Germany) and frozen at −21◦C until further processing ("before PPI" sample). According to our protocol all patients received 1 mg/kg body weight oral esomeprazole daily. After 4 weeks of PPI treatment a second stool sample was collected and stored as described ("on PPI" sample). The duration of PPI therapy depended on the patients' clinical symptoms. A third sample was collected 4 weeks after discontinuation of PPI therapy ("after PPI" sample). None of the patients received antibiotics or other acid suppressants during the course of the study. Dietary habits were recorded.

#### DNA Isolation, 16s Library Preparation and Sequencing

Frozen stool samples were thawed and a peanut sized stool sample was thoroughly mixed in 500 µl PBS. 250 µl of the suspension were mixed with 250 µl bacterial lysis buffer from the MagnaPure LC DNA Isolation Kit III (Bacteria, Fungi) (Roche, Mannheim, Germany) and transferred to MagnaLyser green bead tubes (Roche, Mannheim, Germany) for mechanical lysis performed two times at 6,500 rpm for 30 s in a MagnaLyser instrument (Roche, Mannheim, Germany). After bead beating 25 µl lysozyme (100 mg/ml) were added to the samples for enzymatic lysis and incubated at 37◦C for 30 min followed by incubation with 43 µl Proteinase K (20 mg/ml) at 65◦C for 1 h. Heat inactivation of enzymes was performed at 95◦C for 10 min. Samples were centrifuged at 13,000 rpm for 5 min and 100 µl of the lysed samples were transferred to the Magna Pure instrument and DNA was purified according to manufacturer's instructions. PCR and library preparation with hypervariable regions V1-2 were performed as described before (Klymiuk et al., 2016) with 2 µl of total DNA per 25 µl PCR reaction in triplicates using primers 27f (AGAGTTTGATCCTGGCTCAG) and 357r (CTGCTGCCTYCCGTA) yielding a 330 bp long insert. Triplicates were pooled, amplification was verified by checking on a 1% agarose gel and sequencing library was normalized, indexed, and quantified according to Klymiuk et al. (2016). The pooled sample library was sequenced on a MiSeqII desktop sequencer (Illumina, Eindhoven, Netherlands) with v3 600 cycles chemistry (Illumina, Eindhoven, Netherlands) according to the manufacturer's instructions at 6 pM with 20% PhiX (Illumina, Eindhoven, Netherlands) in one run.

Sequence reads were submitted to the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra/?term=SRP119055).

#### Microbiome Analysis

Sequencing reads were processed with scripts of the QIIME platform. Briefly, reads were clustered to Operational Taxonomic Units (OTU) using the pick\_open\_reference\_otus.py script and uclust algorithm based on the greengenes database (gg\_otus-13\_8-release) and a 97% identity threshold. OTUs were visualized as OTU tables, bar charts and PCOA plots. Alpha diversity measurements (observed species and chao1) and beta-diversity measurements (unweighted unifrac) were derived using the respective QIIME tools. Group significance for all categories was determined with the Adonis test, while individual species difference was quantified by Kruskall-Wallis tests and pairwise comparisons by Mann-Whitney-U-test. The Adonis test computes R2 (effect size) and pseudo-P values of categories by first identifying the relevant centroids of the data and then calculating the squared deviations from these points. After that, significance tests are performed using F-tests based on sequential sums of squares from permutations of the raw data. Adonis tests were performed in R (2.15.1) using the vegan package. Significance of differences in alpha diversity was calculated by non-parametric two-sample t-test using Monte Carlo permutations to calculate the p-value. Lefse analysis was performed for all categories as described previously (Segata et al., 2011).

#### RESULTS

36 stool samples (n = 12 "before PPI", n = 12 "on PPI," n = 12 "after PPI") of 12 patients (8 male, 4 female) were included. The mean gestational age was 38 weeks (STD 2.0; range 35–41 weeks). Patients had a mean birth weight of 2,794 g (STD 468; range 2,100–3,688 g) and a mean birth length of 48.8 cm (STD 2.6; range 44–53 cm).

Patients were included at a mean age of 5.2 months (STD 3.2; range 0.5–10.2 months). All patients suffered from GERD. The data of their 24 h-pH-MII is shown in **Table 1**. The patients' nutrition at the time of stool sampling is displayed in **Table 2**. The mean duration of PPI treatment was 18 weeks (STD 11; range 8–44).

### PPI Therapy Had No Influence on α- and β-Diversity

In the within-individual comparison ("before PPI" vs. "on PPI"), oral PPI treatment did not influence α-diversity (Chao1 index; p = 0.729, **Figure 1**). Additionally, β-diversity did not change when comparing "before PPI" and "on PPI" (unweighted UniFrac; p = 0.913).

#### PPI Treatment Caused Only Minimal Changes in the Fecal Microbiome

Taxa summary plots at the phylum and class level at the different time points tested are depicted in **Figure 2**. On the genus level, PPI therapy caused a significant decrease of Lactobacillus and of Stenotrophomonas. Additionally, there was a significant increase of Haemophilus (**Table 3**). Although Streptococcus increased under PPI therapy none of the bacteria associated with an elevated risk of CDI in adults (Streptococcus, Enterococcus,


*AET, acid exposure time (pH* < *4); ABET, acidic bolus exposure time (pH* < *4); WABET, weakly acidic bolus exposure time (4* ≤ *pH* < *7); NABET, non-acidic bolus exposure time (pH* ≥ *7); TBET, total bolus exposure time; NRA, number of acidic refluxes; NRWA, number of weakly acidic refluxes; NRNA, number of non-acidic refluxes; NRT, total number of refluxes.*

Clostridiaceae) were significantly altered (**Figure 3**). There was no significant correlation between the results of impedance testing and the corresponding microbiome ("before PPI") (p > 0.10).

#### The Third Sample ("after PPI") Showed Increasing α- und β-Diversities and Altered Relative Abundances in Correlation with Patients' Age and Dietary Habits

The α-diversity significantly increased over the time of the experiment (p = 0.003 for the comparison "before PPI" to "after

TABLE 2 | Nutrition of the infants at the three different time points of stool sampling (*n* = 12).


*MM, mother's milk; FM, formula; SF, solid food.*

FIGURE 1 | α-diversities (Chao1 index at 12,000 reads) at the three time points tested (*n* = 12 per time point). In the within-individual comparison there was no statistically significant difference after 4 weeks of PPI treatment (*p* = 0.729). Four weeks after discontinuation of PPI therapy α-diversities were significantly increased (*p* = 0.003 for "after PPI" vs. "before PPI" and "on PPI." Lines connect individuals.

PPI" and p = 0.003 for the comparison "on PPI" and "after PPI"). Furthermore, the β-diversity significantly changed throughout the experiment (p = 0.003 comparing "before PPI" and "after PPI" and p = 0.001 comparing "on PPI" and "after PPI"). For the relative abundances the majority of changes was also seen in comparison to the third sample and occurred in the Firmicutes phylum (**Table 3**). A correlation was found between these microbial changes and the patients' age (p = 0.062) and nutrition (p = 0.001).

#### DISCUSSION

This study is the first to address fecal microbial changes under PPI treatment in infants. In contrast to most studies reported in adults GERD was not only suspected in our patients, but proven by impedance monitoring prior to enrollment and PPI therapy. Notably, we were able to find a completely different response to PPIs in infants than previously described in adults.

Up to now the majority of studies investigating the effect of PPI on gut microbiota have compared adult PPI users to nonusers. Two large series have reported significant decreases of the overall fecal microbial diversity under PPI treatment (Imhann et al., 2016; Jackson et al., 2016). However, to assess the exact effect of PPIs on individuals a longitudinal investigation of the same patients on and off PPI is required. Presently there is only one study in 12 adults which has addressed this issue reporting no significant changes of the fecal microbial diversity (Freedberg et al., 2015b). Similarly, our investigation with the same sample size in infants showed no significant changes of α- and β-diversity under PPI treatment.

Regarding relative bacterial abundances under PPI several reports with varying findings in adult patients have been published. Overall, Streptococcus, Enterococcus, and Clostridiales were most commonly affected in this population. Imhan et al., for instance, have described significant increases of Enterococcus, Streptococcus, and E. coli under PPI therapy (Imhann et al., 2016). Another group has found increased abundances of Streptococcaceae, Lactobacillaceae, Pasteurellaceae, Corynebacteriaceae, and Micrococcaceae (amongst others) and decreases of Lachnospiraceae, Ruminococcaceae, and Erysipelotrichaceae (Jackson et al., 2016). Furthermore, an observation of fecal samples of long-term PPI users has revealed an increase of Lachnospiraceae, Erysipelotrichaceae, and Streptococcaceae (Clooney et al., 2016). In contrast to adults our investigation in infants revealed only minor changes of the relative microbial abundances under PPI therapy.

In adults, an increased risk of Clostridium difficile infections (CDI) under PPI therapy has been postulated. In detail, increases of Enterococcae and Streptococcae combined with decreases of Clostridiales were reported in association with an increased risk of CDI (Freedberg et al., 2015b). Similarly, recent pediatric studies have reported an increased risk of CDI under acid suppression therapy (Brown et al., 2015). First reports about associations between PPI and CDI infections in infants (Freedberg et al., 2015a) rely on culture-based retrospective investigations only. In our series we have found a nonsignificant increase of Streptococcus and Clostridiaceae under PPI treatment (compare **Figure 3**). These results further fuel recent controversial discussions regarding the association between CDI and PPI (Leffler and Lamont, 2015; Faleck et al., 2016).

The majority of changes in our series were seen when comparing the microbiome between "before PPI" and "on PPI" to "after PPI." While we cannot rule out the possibility that the removal of PPI treatment may cause a temporary flux in diversity, other studies showing an increasing diversity with increasing infants' age (Hill et al., 2017) support the physiological development of the intestinal microbiome as the underlying reason for this finding. Additionally, the correlation between the relative abundances and patients' age/nutrition also suggests the developing microbiome of infants as the most likely cause for these alterations.

One possible limitation of the present study includes the lack of a control group without PPI treatment. However, recent investigations of the developing microbiome have shown a marked variability and heterogeneity of the fecal microbiome within the first years of life (Palmer et al., 2007; Yatsunenko et al., 2012). This makes the selection of infants for a representative

TABLE 3 | Mean relative abundances (RA) at the different levels.


\**p* < *0.05 vs. "before PPI";* #*p* < *0.05 vs. "on PPI." Significant increases under PPI therapy are colored in blue, significant decreases in red.*

and comparable control group difficult. In accordance to the literature we have found a constant increase of α-diversity in our samples (Palmer et al., 2007). This further substantiates the developing microbiome as a reason for the changes encountered in the third sample of this series. In our longitudinal intraindividual comparison, the same patient is investigated on and off PPI and thus serves as his/her own control. Another limitation is the relatively small number of included infants. However, the sample size resembles that of the only longitudinal study in adult patients (Freedberg et al., 2015b). Our setting also takes care of the heterogeneity of the microbiome in infants and the different age of our patients upon inclusion because the patients are compared by dependent tests within themselves (intra-individual).

Finally, we have only measured the fecal microbiome and could not include samples from other parts of the gastrointestinal (GI) tract. The fecal microbiome is easily accessible and the obtained results can be compared to adult studies (Freedberg et al., 2015b; Clooney et al., 2016; Imhann et al., 2016; Jackson et al., 2016). However, the fecal microbiome does not necessarily represent the whole GI-tract (Haange et al., 2012) and possible alterations of the microbial diversity of the upper GI-tract under PPI treatment are subject for further investigations.

Since we were not able to demonstrate any relevant changes under PPI therapy one might question whether a PPI dosage of 1 mg/kg/day esomeprazole was sufficient. Theoretically,

#### REFERENCES


our findings could be caused by inadequate acid suppression. Although we did not repeat impedance monitoring "on PPI" we could demonstrate that our institutional protocol and PPI dosage was effective in a previous examination (Castellani et al., 2014). Additionally, the symptoms caused by the GERD resolved in all our patients under PPI therapy suggesting adequate acid suppression.

In conclusion, oral PPI therapy did not have relevant impact on the development of the infant fecal microbiome at a sensitive time of life in our series. Microbial changes associated with an increased risk of CDI infection described in adults did not reach statistical significance in this study. The majority of alterations occurred through the course of time and is correlated to patients' age and nutrition representing the normal development of the microbiome. Future studies are required to investigate possible microbial changes of the upper GI-tract under PPI treatment.

#### AUTHOR CONTRIBUTIONS

CC and GS performed the statistics and wrote the manuscript; KK performed the microbiome analysis and performed the biostatistics; AH recruited the patients and collected the samples; CF and MK analyzed the data and assisted to draft the manuscript; HT coordinated the project and critically reviewed the manuscript.


Yatsunenko, T., Rey, F. E., Manary, M. J., Trehan, I., Dominguez-Bello, M. G., Contreras, M., et al. (2012). Human gut microbiome viewed across age and geography. Nature 486, 222–227. doi: 10.1038/nature 11053

**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 Castellani, Singer, Kashofer, Huber-Zeyringer, Flucher, Kaiser and Till. 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.

# *Helicobacter pylori* CagA Protein Negatively Regulates Autophagy and Promotes Inflammatory Response via c-Met-PI3K/Akt-mTOR Signaling Pathway

Na Li 1, 2, 3†, Bin Tang1, 2, 4†, Yin-ping Jia<sup>1</sup> , Pan Zhu<sup>1</sup> , Yuan Zhuang<sup>2</sup> , Yao Fang1, 2, Qian Li <sup>1</sup> , Kun Wang1, 2, Wei-jun Zhang<sup>2</sup> , Gang Guo<sup>2</sup> , Tong-jian Wang<sup>3</sup> , You-jun Feng<sup>5</sup> , Bin Qiao<sup>3</sup> \*, Xu-hu Mao<sup>1</sup> \* and Quan-ming Zou<sup>2</sup> \*

*<sup>1</sup> Department of Clinical Microbiology and Immunology, Southwest Hospital & College of Medical Laboratory Science, Third Military Medical University, Chongqing, China, <sup>2</sup> Department of Microbiology and Biochemical Pharmacy, National Engineering Research Center for Immunobiological Products, College of Pharmacy, Third Military Medical University, Chongqing, China, <sup>3</sup> Institute of Cardiovascular Disease, General Hospital of Jinan Military Region, Jinan, China, <sup>4</sup> Emei Sanatorium of PLA Rocket Force, Emeishan, China, <sup>5</sup> Department of Medical Microbiology and Parasitology, Zhejiang University School of Medicine, Hangzhou, China*

#### *Edited by:*

*Pascale Alard, University of Louisville, United States*

#### *Reviewed by:*

*Mario M. D'Elios, University of Florence, Italy Nagendran Tharmalingam, Brown University, United States*

#### *\*Correspondence:*

*Bin Qiao cijnmd@126.com Xu-hu Mao mxh95xy@tom.com Quan-ming Zou qmzou2007@163.com*

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

*Received: 23 July 2017 Accepted: 08 September 2017 Published: 21 September 2017*

#### *Citation:*

*Li N, Tang B, Jia Y, Zhu P, Zhuang Y, Fang Y, Li Q, Wang K, Zhang W, Guo G, Wang T, Feng Y, Qiao B, Mao X and Zou Q (2017) Helicobacter pylori CagA Protein Negatively Regulates Autophagy and Promotes Inflammatory Response via c-Met-PI3K/Akt-mTOR Signaling Pathway. Front. Cell. Infect. Microbiol. 7:417. doi: 10.3389/fcimb.2017.00417* Cytotoxin-associated-gene A (CagA) of *Helicobacter pylori* (*H. pylori*) is a virulence factor that plays critical roles in *H. pylori*-induced gastric inflammation. In the present study, gastric biopsies were used for genotyping *cagA* and *vacA* genes, determining the autophagic activity, and the severity of gastric inflammation response. It was revealed that autophagy in gastric mucosal tissues infected with *cagA*<sup>+</sup> *H. pylori* strains was lower than the levels produced by *cagA*<sup>−</sup> *H. pylori* strains, accompanied with accumulation of SQSTM1 and decreased LAMP1 expression. *In vitro,* deletion mutant of *cagA* gene resulted in increased autophagic activity, and decreased expression of SQSTM1 and cytokines, whereas over-expression of CagA down-regulated the starvation-induced autophagy, and induced more production of the cytokines. Moreover, the production of the cytokines was increased by inhibition of autophagy, but decreased by enhancement of autophagy. Deletion of CagA decreased the ability to activate Akt kinase at Ser-473 site and increased autophagy. c-Met siRNA significantly affected CagA-mediated autophagy, and decreased the level of p-Akt, p-mTOR, and p-S6. Both c-Met siRNA and MK-2206 could reverse inflammatory response. *H. pylori* CagA protein negatively regulates autophagy and promotes the inflammation in *H. pylori* infection, which is regulated by c-Met-PI3K/Akt-mTOR signaling pathway activation.

Keywords: *Helicobacter pylori*, autophagy, CagA, c-Met, SQSTM1

## INTRODUCTION

Helicobacter pylori (H. pylori) is a Gram-negative bacterium causing gastritis, peptic ulcer disease and gastric adenocarcinoma (Suerbaum and Michetti, 2002). Although H. pylori could induce strong inflammation, it is not able to clear the bacterium, resulting in persistent infection. Cytotoxin-associated gene A (CagA), one of H. pylori virulence factors, is an effector secreted by the type IV secretion system into gastric epithelial cells, and undergoes tyrosine phosphorylation, and activates a series of intracellular signal transduction reactions, resulting in severe tissue inflammation and damage (Gunn et al., 1998). Generally, H. pylori strains expressing CagA protein is more virulent, and leading to severe gastritis (Fischer et al., 2001). CagA is able to activate the transcription factor, NF-κB, and translocate it into the nucleus, where it upregulates transcription of interlukin-8 (IL-8), a chemotactic and inflammatory cytokine (Brandt et al., 2005). However, the specific mechanism that CagA-positive strains induce inflammation remains unclear.

Macroautophagy (hereafter autophagy) has an important role in controlling intracellular environment. The damaged cell organelles, proteins, or invading microorganisms are sequestered into autophagosomes, and finally delivered into autolysosomes for degradation (Shintani and Klionsky, 2004). In the case of infection of pathogenic microorganism, the final consequence of the infection was decided by the evolving struggle between the host cells and invading microbes, and autophagy plays a critical role in the struggle. A number of important pathogens could be degraded by autolysosomes, such as, Listeria monocytogenes (Py et al., 2007), group A Streptococcus (Nakagawa et al., 2004), and Francisella tularensis (Cremer et al., 2009). However, some pathogenic bacteria also develop some mechanisms to subvert autophagy to survive in cells, eventually leading to the occurrence of various diseases, such as, Shigella (Kayath et al., 2010) and Mycobacterium tuberculosis (Shin et al., 2010).

It has been demonstrated that the induction of autophagosome formation or autophagy depends on the vacuolating cytotoxin (VacA), which is another virulent factor of H. pylori (Terebiznik et al., 2009). In turn, autophagy can eliminate intracellular H. pylori and may decrease the stability of intracellular VacA and ameliorate toxin-mediated cellular vacuolation (Terebiznik et al., 2009), despite the fact that autophagy is not sufficient to block vacuole biogenesis and pathogenesis (Zavros and Rogler, 2012). Recently, Tsugawa et al. showed that intracellular CagA is degraded by autophagy and short lived in AGS cells (Tsugawa et al., 2012), but whether or not autophagy regulated by CagA in H. pylori-induced gastric inflammation have never been explored.

Therefore, the purpose of this article was to determine the effect of CagA on autophagy of gastric epithelial cells and the production of autophagy-regulated proinflammatory cytokines in H. pylori infection.

### MATERIALS AND METHODS

#### Patients and Specimens

Consecutive patients who underwent upper endoscopy due to dyspeptic symptoms at Southwest Hospital, Chongqing, China during January 2013 and December 2014, were recruited. One hundred and six (49 women and 57 men with age of 43 ± 20 years) patients were eligible for enrollment into the H. pylori positive group if they had a positive [13C] urea breath test, a positive rapid urease test, and H. pylori culture. Eleven (six women and five men with age of 35 ± 20 years) with normal gastric mucosa were eligible for enrollment into the H. pylori negative group, and the clinical characteristics are shown in Supplementary Table 2. The study was approved by the Institutional Review Board at Third Military Medical University, and all patients signed informed consent before participation. All experiments were performed in accordance with relevant guidelines and regulations.

H. pylori was successfully isolated from 106 patients, and genotyping for cagA and vacA was performed for 106 isolates. All H. pylori strains carry the vacA gene. To exclude the effect of VacA, the toxigenic vacA genotype (vacAs1m1, 42 cases), expressing a functional VacA toxic, were excluded from the study. The rest of the cases include: normal control (11 cases), cagA−/vacAs1m2 (7 cases), cagA+/vacAs1m2 (57 cases). To ensure that approximately equal numbers of each group, 23 selective patients were chosen randomly for analyzing autophagy and inflammation, dividing into normal control (8 cases), cagA−/vacAs1m2 (7 cases), cagA+/vacAs1m2 (8 cases).

#### Evaluation of Inflammation Score for Gastric Biopsy Samples

The selected gastric biopsy samples among the genotype subgroups were obtained to perform H&E staining. The intensity of inflammation was evaluated independently by two pathologists according to previously established criteria. The degree of neutrophil infiltration, mononuclear cell infiltration, atrophy, and metaplasia was assessed according to the updated Sydney classification as follows: 0, absent; 1, minimal; 2, mild; 3, moderate; 4, marked. So the biopsies were from different stages of gastritis (Dixon et al., 1996).

#### Genotyping for cagA and vacA Genes

H. pylori infection status was detected by rapid urease test, bacterial culture, <sup>13</sup>C-urea breath test, and histological examination (Vaira et al., 1999). In patients with positive culture, H. pylori isolates were subcultured for a maximum of five passages, and genomic DNA was extracted to genotype for the cagA and vacA genes, as previously described (Argent et al., 2008). The primers used for PCR amplification and nucleotide sequencing are listed in Supplementary Table 1.

#### Cell Line and *H. pylori* Strains

AGS (a human gastric cancer cell line) purchased from the cell bank of Chinese Academy of Sciences, were cultured in F12 cell culture medium (Gibco, Grand Island, NY, USA, #11765-054) supplemented with 10% FBS (Gibco, #10099-141) in a humidified incubator (5% CO2) at 37◦C. The starvation condition was established by culturing the cells with serum-free medium for 4 h.

The wide-type cagA+/vacA<sup>+</sup> H. pylori strain, NCTC11637 (Hp-WT, obtained from ATCC), cagA-knockout H. pylori with NCTC11637 background (Hp-1cagA, kindly provided by Dr. Sasakawa (Asahi et al., 2000; Suzuki et al., 2009) and H. pylori cagA-knockout complementation mutant (Hp-c-cagA, constructed by our group), were cultured on brain-heart infusion medium (10% rabbit blood) under microaerophilic conditions (5% O2, 10% CO2, and 85% N2) at 37◦C. Hp-c-cagA mutant was obtained by amplifying the cDNA fragments of cagA gene from the gene of NCTC11637 by polymerase chain reaction (PCR), and the primers of PCR is following: forward: 5′ -G CGCTCGAGATGACTAACGAACC-3′ ; reverse: 5′ -GCGCTGC AGTTAAGATTTTTGG-3′ . The product of PCR was digested with XhoI and PstI, and then ligating the cDNA fragments of cagA gene between cagA-upstream and -downstream sequences cloned on the pHel3 shuttle vector. The pHel3 shuttle vector with cagA gene was electroporated into Hp-1cagA cells carrying the kanamycin resistance. Hp-c-cagA clones were cultured in brain-heart infusion medium as previous described.

AGS cells transfected with plasmids and/or siRNAs were infected with Hp-WT, Hp-1cagA, or Hp-c-cagA, respectively, with different multiplicity of infection (MOI = 10, 50, 100, 200), for 6 h. Cells without infection served as controls.

#### Reagents and Antibodies

Rapamycin (Rapa, R8781), 3-methyladenine (3-MA, M9281), bafilomycin A1 (Baf-A1, B1793), antibodies against ATG12 (WH0009140M1), MAP1LC3B (L7543), and ATG5 (WH0009474M1) were purchased from Sigma-Aldrich (Shanghai, CHINA), and MK-2206 2HCL (S1078) purchased from Selleckchem (Houston, TX, USA). Antibodies against Akt (9272), mTOR (2972), AMPK (2532), phospho-Akt (Ser473) (4060), LAMP1 (9091), phospho-mTOR (Ser2448) (5536), phospho-S6 ribosomal protein (Ser235/236) (2211), ribosomal protein (2217), phospho-c-Met (Y1234/Y1235) (4033), and c-Met (4560) were obtained from Cell Signaling Technology (Beverly, MA, USA), whereas antibodies against β-actin (sc-10731), VacA (sc-25790), CagA (sc-17450), phospho-tyrosine (PY99) (sc-7020) and SQSTM1 (sc-28359), and siRNAs specific for SQSTM1 (human, sc-29679), ATG12 (human, sc-72578), c-Met (human, sc-29397), and ATG5 (human, sc-41445), along with a control siRNA (sc-44230) were obtained from Santa Cruz Biotechnology (CA, USA).

#### Immunohistochemistry for SQSTM1

Gastric biopsy sections from patients infected with different genotypes of H. pylori were stained using SQSTM1 antibody (Enzo Life Sciences, BML-PW9860-0025) by the DAB reagent technique. SQSTM1 staining was assessed by three pathologists with no prior knowledge of the different groups, and scored as 0 (no staining), 1 (<10% of SQSTM1 staining), 2 (10–50% of SQSTM1 staining), or 3 (>50% of SQSTM1 staining).

#### Western Blotting, Quantitative RT-PCR, NF-κB Activity, and ELISA

The protein level of SQSTM1, LAMP1, phospho-mTOR (Ser2448), MAP1LC3B, ATG12, ATG5, AMPK, phospho-Akt (Ser473), Akt, mTOR, ribosomal protein, phospho-c-Met (Y1234/Y1235), c-Met, phospho-S6 ribosomal protein (Ser235/236), VacA, CagA, and phospho-tyrosine (PY99) were performed to determine by Western blotting in gastric tissues or AGS cells as described previously (Tang et al., 2015).

Trizol reagent (Invitrogen, 15596-026) was used to extract total RNA from gastric biopsy sections. qRT-PCR analysis of gastric biopsy sections from patients for the mRNA of SQSTM1, BECN1, IL-8, TNF-α, β-actin, and IL-1β were performed by using PrimeScript RT-PCR kits (Takara, Tokyo, Japan, DRR037), and run on a Bio-Rad IQ5 thermocycler (Bio-Rad Laboratories, Inc., Hercules, USA), β-actin as an internal control. The conditions for PCR were as follows: 1 cycle of 95◦C for 30 s, 40 cycles of 95◦C for 6 s, 60◦C for 6 s, and 72◦C for 31 s, and the primer sequences used are shown in Supplementary Table 1.

AGS cells were cotransfected with Renilla control vector (pRL-TK, Promega, #2241) and luciferase reporter vector pNF-κB-TA-Luc (Clontech, #631904) with lipofectamine 2000 (the ratio of 20:1) for 1 day, followed with H. pylori infection. The Dual-Luciferase Reporter Assay System (Promega, E1910) detected firefly luciferase activity and Renilla luciferase activity according to the manufacture's protocol.

The supernatant of AGS cells with different treatment were detected by DuoSet ELISA Development System (IL-8, IL-1β, and TNF-α; R&D, Minneapolis, USA) as our previous study (Tang et al., 2016).

#### Transfection of AGS With Plasmids and/or siRNAs

The GFP-MAP1LC3B plasmid and RFP-MAP1LC3B expression plasmid were kindly provided by Dr. Tamotsu Yoshimori (Department of Cell Biology, National Institute for Basic Biology, Presto, Japan) and Dr. Maria Colombo (Universidad Nacional de Cuyo, Mendoza, Argentina), respectively. The CagA expression plasmid, pEGFP-C1-CagA (GFP-CagA) (Asahi et al., 2000; Suzuki et al., 2009), was kindly provided by Dr. Chihiro Sasakawa. The cagA mutant plasmid, pEGFP-C1-CagA-Mut (GFP-CagA-Mut) was constructed by Life Technologies, Shanghai, China, and a series of CagA mutants with the Tyr residues of 899, 918, and 972 being substituted by Ala were generated from a plasmid-encoding fragment of cagA gene of H. pylori ATCC 26695 on pBluescript (Promega, Madison, USA) using a Gene Editor in vitro Site-Directed Mutagenesis System (Promega). These mutants were at the sites 2,695– 2,697, 2,752–2,754, and 2,914–2,915 bp, respectively, started from the sequence ATG. Lipofectamine 2000 (Invitrogen, #11668019) was used to transfect plasmids and/or siRNAs into AGS cells. 3 × 10<sup>6</sup> AGS cells were seeded into a 100-mm dish and incubated with transfection complexes containing 100 nM siRNA for 24 h.

#### Immunoprecipitation Assays

AGS cells were harvested in RIPA buffer on crushed ice, and centrifuged at 6,000 rpm for 5 min, and commercial Lowry Assay (Bio-Rad DC) detected the concentration of protein. Five milligrams per milliliter of protein and 1 µg/µL anti-CagA (sc-17450, Santa Cruz Biotechnology) was incubated overnight at 4◦C, then added 50 µL 50% protein A/G-sepharose bead suspension for 2 h, and washed three times with pre-cold RIPA buffer, and added 50 µL protein sample buffer to collect in each tube, and proteins detected by western blotting analysis.

#### Puncta Formation Assays

AGS cells were transfected with GFP-MAP1LC3B or RFP-MAP1LC3B plasmid for 24 h, following H. pylori infection for another 24 h. Radiance 2000 laser scanning confocal microscope detected the images of the cells, and image analysis with LaserSharp 2000 software (Bio-Rad, San Francisco, CA) as our previous study (Tang et al., 2016). According to methods for monitoring GFP-LC3 and mRFP-GFP-LC3 puncta formation assays (Klionsky et al., 2007; Mizushima et al., 2010), the average number of MAP1LC3B puncta per cell in GFP-MAP1LC3B or RFP-MAP1LC3B-positive cells (200 cells per sample) was determined (Pattingre et al., 2005).

#### Cell Viability

AGS cell viability was assessed using an MTT assay (Sigma) according to the manufacturer's instructions. Five milligrams per liter of MTT was added to each wells of AGS cells for 1–2 h, and dissolved in MTT solubilization solution. The absorbance at 590 nm (A590) was determined for each well using a microplate reader (Bio-Rad). After subtracting the background absorbance, the A590 value of the treated cells was divided by that of the untreated cells to determine the percentage of viable cells.

## MDC and AO Staining Assays

Monodansylcadaverine (MDC) and acridine orange (AO) staining was used to quantify the number of autolysosomes in AGS cells. Following treatment with H. pylori or transfection with plasmids/siRNAs, cells were stained with 10 mM MDC (sigma, 30432) and 1 mg/mL AO solution (sigma, A8097) at 37◦C for 10 min, and fixed in 3% paraformaldehyde in PBS for 30 min. Photographs were obtained with a Radiance 2000 laser scanning confocal microscope (MDC, excitation wave length about 380 nm and emission filter 525 nm; AO, emission peak at 650 nm). The cells were then trypsinised and quantified by flow cytometry using a FACScan cytometer and CellQuest software (BD, New Jersey, USA). The percentage of cells with characteristic MDC or AO staining over the total cells was assessed.

## Transmission Electron Microscopy

AGS cells or gastric biopsy sections were collected and fixed in 2% paraformaldehyde, 0.1% glutaraldehyde and 0.1 M sodium cacodylate buffer (pH 7.4) for 2 h, then post-fixed in 1% OsO4, 0.5% potassium ferricyanide in cacodylate buffer for 1.5 h, then dehydrated with graded alcohol, and embedded in straight resin. Ultrathin sections were counterstained with 0.3% lead citrate and detected by Philips EM420 electron microscope. The method of counting autophagosomes' numbers was followed as described previously by Yla-Anttila et al. (2009). Data obtained by scoring for the presence of autophagic vacuoles (autophagosomes, autolysosomes) profiles per cell profile on the sections, and a total of 35 cells were recorded for triplicate samples per condition per experiment.

## Statistical Analyses

The Student t-test was used to analyze between two groups, and one-way analysis of variance (ANOVA) was used to analyze among multiple group data, and expressed as mean ± standard error (SEM). GraphPad Prism software (GraphPad, San Diego, CA) was used for all statistical analyses. For all inferential statistics a P < 0.05 was considered significant.

## RESULTS

## Autophagy Is Down-Regulated in Human Gastric Mucosa With CagA Positive *H. pylori*

The clinical characteristics of 117 patients with (106) and those without (11) H. pylori infection are shown in Supplementary Table 2. H. pylori was successfully isolated from 106 patients, and genotyping for cagA and vacA. All H. pylori strains carry the vacA gene. To exclude the effect of VacA, the toxigenic vacA genotype (vacAs1m1), expressing a functional VacA toxic, were excluded from the study. In order to ensure that approximately equal numbers of each group, three equal groups were created for analyzing via random sampling methods, including normal control (8 cases), cagA−/vacAs1m2 (7 cases), cagA+/vacAs1m2 (8 cases).

To verify the effect of CagA in severe tissue inflammation, we evaluated the level of inflammation in gastric mucosa. Firstly, the degree of gastric inflammation was higher in patients infected with cagA+/vacAs1m2 strains than in those infected with cagA−/vacAs1m2 strains (**Figure 1A**). Notably, the mRNA levels of IL-8, TNF-α, and IL-1β in the gastric epithelial cells were significantly higher in patients infected with cagA+/vacAs1m2 strains than in patients without H. pylori infection or those infected with cag−/vacAs1m2 strains (**Figure 1B**).

Furthermore, we evaluate the autophagic activity in gastric mucosal tissues from patients infected with different genotypes H. pylori. The SQSTM1/p62 (sequestosome1) protein serves as a link between LC3 and ubiquitinated substrates (Wang et al., 2010). Dysfunctional autophagy could result in an accumulation of SQSTM1, which has been involved in promoting inflammation (Raju et al., 2012). Therefore, we detected the levels of SQSTM1 in human gastric biopsies. As shown in **Figure 2A**, accumulation of SQSTM1 in the gastric biospy with H. pylori infection was significantly higher than those uninfected normal control, and the accumulation of SQSTM1 was significantly higher in the gastric epithelium cells in patients infected with cagA+/vacAs1m2 strains than in those infected with cagA−/vacAs1m2 strains (P < 0.001) as determined by immunohistochemistry. Moreover, in patients infected with cagA−/vacAs1m2 strains, the ratio of microtubule-associated protein 1 light chain 3 beta-II (MAP1LC3B-II) to β-actin and the lysosomal-associated membrane protein 1 (LAMP1, the late endosomal lysosomal marker) protein levels was higher than that in patients infected with cagA+/vacAs1m2 strains, and the SQSTM1 protein levels increased in patients infected with CagA-positive H. pylori (**Figure 2B**). In addition, infection of cagA+/vacAs1m2 strains was significantly associated with increased levels of mRNA expression of SQSTM1 (Supplementary Figure 1A), but not of BECN1 (Supplementary Figure 1B). Furthermore, it was also revealed an increase in the number of autophagosomes in patients infected with cagA−/vacAs1m2 strains compared with that in cagA+/vacAs1m2 group in TEM analysis (**Figure 2C**). Both cagA+/vacAs1m2 and cagA−/vacAs1m2 groups displayed high autophagy activity than the normal control group. These findings indicate that H. pylori infection could induce inflammation response and autophagy activity in the gastric epithelium cells in

vivo, but CagA-positive H. pylori are associated with more severe inflammation, and down-regulates autophagic response in vivo.

## CagA Could Inhibit the Generation of Autophagosomes in AGS Cells

To further validate the role of CagA in autophagy regulation, we next infected AGS cells with the H. pylori wide-type (Hp-WT), H. pylori cagA-knockout mutant (Hp-1cagA) or H. pylori cagA-knockout complementation (Hp-c-cagA) (MOI = 100:1), which strains the expression of VacA is similar during infection (Supplementary Figure 2A), and evaluated the kinetics of autophagosome formation by a GFP-MAP1LC3B puncta formation assay. Formation of MAP1LC3B puncta, peaked at 12 h and decreased at 24 h (**Figure 3A** and Supplementary Figure 2B). And compared with cells infected with Hp-WT or Hp-c-cagA, there was a significantly increased percentage of cells with formation of MAP1LC3B puncta for cells infected with Hp-1cagA (**Figure 3A**). TEM revealed an increase in the number of autophagic vacuoles (autophagosomes and autolysosomes) in AGS cells infected with Hp-1cagAinfected cells, compared with cells infected with Hp-WT or Hp-c-cagA (**Figure 3B**). Similar results were obtained in MDC (**Figure 3C** and Supplementary Figure 2D) and AO (**Figure 3D** and Supplementary Figure 2D) staining. Additionally, Hp-1cagA induced MAP1LC3B-II formation, and decreased SQSTM1 protein expression at a higher level, compared with Hp-WT or Hp-c-cagA, at 6, 12, and 24 h (**Figure 3E** and Supplementary Figure 2C). Furthermore, inhibition of autophagy by Baf-A1 challenge resulted in further accumulation of both MAP1LC3B-II and SQSTM1 in AGS cells after 6 h of Hp-WT or Hp-1cagA infection (**Figure 3F**), suggesting that H. pylori CagA did not inhibit the fusion of autophagosomes with lysosomes. Furthermore, under Hp-WT or Hp-1cagA infection, the levels of MAP1LC3B-II in AGS cells transfected with the CagA expression plasmid (GFP-CagA) were decreased in comparison to that in transfected-control cells (**Figure 3G**), suggesting that overexpression of CagA lead to further reduction of autophagic flux. Collectively, these data suggest that H. pylori CagA may inhibit the generation of autophagosomes in AGS cells.

## CagA Down-Regulates Starvation-Induced Autophagy in AGS Cells

In order to eliminate the influence of H. pylori itself on autophagy, starvation-triggered autophagy was performed in AGS cells after transfecting the CagA expression plasmid (GFP-CagA) or tyrosine phosphorylation point mutant of CagA plasmid (GFP-CagA-Mut). At least 50% transfection efficiency was achieved for transfection of GFP-CagA and GFP-CagA-Mut in AGS (Supplementary Figure 3A). Although cell viability was influenced by starvation to a certain extent during the first 4 h, it appears not to be significantly influenced afterwards (Supplementary Figure 3B). During nutrient starvation, in the

FIGURE 2 | Autophagy is down-regulated in human gastric mucosa of patients infected with CagA positive *H. pylori* strains. (A) Immunohistochemistry showing SQSTM1 expression in the gastric mucosa of patients without *H. pylori* infection and those infected with *cagA*−/*vacAs1m2* or *cagA*+/*vacAs1m2* strains of *H. pylori*. The intensity of staining is shown in the right graph and the data are expressed as mean ± SEM. (B) Western blot assay showing the protein levels of MAP1LC3B-II, SQSTM1 and LAMP1 in the gastric mucosa of patients of normal control (patients 1–4), *cagA*−/*vacAs1m2*(patients 5–8), and *cagA*+/*vacAs1m2* (patients 9–12) with the rates to β-actin being illustrated in the graphs in which the data are expressed as mean ± SEM. (C) Transmission electron microscopy showing autophagosomes in gastric biopsy sections of patients without *H. pylori* infection and those infected with *cagA*−/*vacAs1m2* or *cagA*+/*vacAs1m2* strains of *H. pylori*. Normal controls are patients without *H. pylori* infection. The white arrows indicate the autophagosomes. The numbers of autophagic vacuoles per cell in each TEM section (*n* = 35 cells) are shown in the right graph and the data are expressed as mean ± SEM. Experiments performed in triplicate showed consistent results. \**P* < 0.05, or \*\**P* < 0.01.

without *H. pylori* infection (UI), and transfected AGS cells with the wild type *H. pylori* (*Hp*-WT), the *cagA*-knockout *H. pylori* (*Hp*-1*cagA*) or the *cagA*-knockout complementation mutant *H. pylori* (*Hp*-*c-cagA*) (MOI = 100:1) infection for 6 h (left) and the indicated periods of time (right). Scale bars: 10 µm. The number of *GFP-MAP1LC3B* puncta in each cell (*n* ≥ 200 cells) was counted. (B) Representative transmission electron microscopy showing AGS cells without *H. pylori* infection and those infected with *Hp*-WT, *Hp*-1*cagA,* or *Hp*-*c-cagA* (MOI = 100:1) for 6 h. The white arrows indicate autophagosomes, and the black arrows indicate autolysosomes, and the white triangle indicate *H. pylori*. The numbers of autophagic vacuoles per cell in each TEM section (*n* = 35 cells) are shown in the lower left graph and the data are expressed as mean ± SEM. (C,D) Flow cytometry showing MDC and AO staining of AGS cells 6 h after infection with *Hp*-WT, *Hp*-1*cagA,* or *Hp*-*c-cagA* (MOI = 100:1). (E) Western blotting showing the protein levels of CagA, SQSTM1, and MAP1LC3B-II with the rates of SQSTM1 and MAP1LC3B-II to β-actin in AGS cells infected with *Hp*-WT, *Hp*-1*cagA,* or *Hp*-*c-cagA* (MOI = 100:1) for 6 h. (F) Measurement of MAP1LC3B-II conversion and SQSTM1 in AGS cells infected with *Hp*-WT or *Hp*-1*cagA* (MOI = 100:1) for 6 h in the presence of Baf-A1 (10 nM). (G) AGS cells were transfected with *GFP-CagA,* and then infected with *Hp*-WT or *Hp*-1*cagA* (MOI = 100:1) for 6 h in the presence of Baf-A1 (10 nM). Results shown are representative of three independent experiments. \**P* < 0.05, \*\**P* < 0.01.

AGS cells transfected with GFP-CagA or GFP-CagA-Mut, there was a significant decrease in the number of autophagosomes as determined by TEM, compared with cells transfected with control plasmid (P < 0.05, **Figure 4A**). The ratio of MAP1LC3B-II to β-actin was also significantly decreased in cells transfected with GFP-CagA or GFP-CagA-Mut following

and SQSTM1 in AGS cells transfected with *GFP-CagA*, *GFP-CagA-Mut*, or a control (*pEGFP-C1*) plasmid in the nutrient rich medium or 4 h starvation. (C) Confocal microscopy showing AGS cells co-transfected with *RFP-MAP1LC3B* and *GFP-CagA*, *GFP-CagA-Mut*, or a control (*pEGFP-C1*) plasmid in the nutrient rich medium or 4 h starvation. Scale bars: 5 or 10 µm. The number of *RFP-MAP1LC3B* puncta in each cell (*n* ≥ 200 cells) was counted. Experiments were performed in triplicate, and all replicates showed similar results. \**P* < 0.05, \*\**P* < 0.01.

starvation treatment (P < 0.05, **Figure 4B**). Similarly, SQSTM1 expression was increased in cells transfected with GFP-CagA or GFP-CagA-Mut following starvation treatment (P < 0.05, **Figure 4B**). Interestingly, we observed that transfection with GFP-CagA or GFP-CagA-Mut had no effect on the expression of the AMP activated protein kinase (AMPK, an energy sensor; **Figure 4B**), indicating that CagA suppressed starvation-induced autophagy may not via the AMPK signal pathway. Furthermore, as shown in **Figure 4C**, the number of RFP-MAP1LC3B puncta in AGS cells co-transfected with RFP-MAP1LC3B and GFP-CagA was decreased after starvation treatment (P < 0.05). Taken together, these results suggest that CagA suppressed starvationinduced autophagy, which may not be dependent on tyrosine phosphorylation of CagA.

## Autophagy Inhibition Increases Cytokines Production

To clarify the role of CagA in the inflammation, the expression of proinflammatory cytokines (IL-8, TNF-α, and IL-1β), which are involved in gastritis during H. pylori infection (Nakachi et al., 2000), was examined by ELISA assay. These cytokines were significantly higher in AGS cells infected with Hp-WT or Hp-c-cagA than in those infected with Hp-1cagA at different time points (**Figure 5A**). Moreover, These cytokines in AGS cells infected with Hp-WT and Hp-1cagA was increased following the increase of the bacterial load. AGS cells infected with Hp-WT produced greater amounts of the cytokines than cells infected with Hp-1cagA (**Figure 5B**).

We also examined the production of the cytokines with the autophagy enhancer (Rapa, Rapamycin) or inhibitors (3-MA or Baf-A1) treatment during Hp-WT and Hp-1cagA infection. The effects of two autophagy inhibitors, and one enhancer, are shown in Supplementary Figures 4A–C. Autophagy inhibitors significantly increased the cytokines and activated NF-κB, and enhancer Rapa decreased the ones in AGS cells infected Hp-1cagA infection (**Figure 5C** and Supplementary Figure 4E). After 24 h infection, the three proinflammatory cytokines were increased with the inhibitors in cells infected with Hp-WT and Hp-1cagA (Supplementary Figure 4F). Moreover, the production of proinflammatory cytokines and activity of NF-κB were significantly increased in AGS cells transfected with siRNAs for ATG5 or ATG12 upon H. pylori infection (**Figure 5D** and Supplementary Figure 4G). These data suggested that autophagy plays a critical role in the inflammation induced by H. pylori.

## c-Met Is an Important Adaptor in CagA-mediated Autophagy Pathway

The previous study reported that CagA has been known to activate c-Met and the PI3K/AKT pathway (Churin et al., 2003). However, it is not clear whether c-Met could regulate autophagy. The wild type H. pylori infection activated c-Met in AGS cells (**Figure 6A**). CagA was coimmunoprecipitated with c-Met in AGS cells infection with Hp-WT (**Figure 6B**). This result was consistent with previous study (Oliveira et al., 2009). The effects of c-Met depletion through siRNA interference are shown in Supplementary Figure 4D. The number of GFP-MAP1LC3B puncta in c-Met siRNA group was higher than that of control group upon infection with the Hp-WT (P < 0.05, **Figure 6C**). It was a significant increase in the ratio of MAP1LC3B-II to β-actin in c-Met siRNA group than in the control siRNA upon infection with the Hp-WT (P < 0.05, **Figure 6D**). Furthermore, MDC and AO staining showed that c-Met siRNA group induced the formation of autophagolysosomes in AGS cells at a significantly higher level, compared with control siRNA group in AGS cells infected with Hp-WT (P = 0.008 and 0.018, respectively, **Figures 6E,F** and Supplementary Figure 5A). Moreover, in CagA-expressing AGS cells, the ratio of MAP1LC3B-II to β-actin significantly increased by c-Met siRNA regardless of infection status (**Figure 6G**). These results demonstrate that c-Met may be an important adaptor in CagAmediated autophagy pathway.

## CagA Regulates Autophagy through c-Met/Akt Signaling Pathway

Given that c-Met could activate PI3K/AKT/mTOR pathway (Lim and Walikonis, 2008; Tang et al., 2015), we hypothesized that PI3K/AKT/mTOR pathway might play an important role in the process of CagA-mediated autophagy. We analyzed the activation status of the key members of autophagy-related PI3K/Akt/mTOR pathways. As shown in **Figure 7A**, Hp-WT activated Akt kinase at Ser-473 site at a significantly higher level than did Hp-1cagA (P = 0.018), which was consistent with a previous report (Tabassam et al., 2009). Both Hp-WT and Hp-1cagA increased MAP1LC3B-II expression, but Hp-1cagA did at a significantly higher level than did Hp-WT (**Figure 7A**). There was a significant increase in the levels of phosphorylated mTOR (pmTOR) and phosphorylated S6 ribosomal protein (p-S6) upon Hp-WT vs. Hp-1cagA (**Figure 7A**). Tyrosine phosphorylation of CagA did not affect the expression levels of proteins related to PI3K/Akt/mTOR pathway and autophagy in AGS cells (**Figure 7B**), whereas c-Met siRNA significantly decreased the level of p-Akt, p-mTOR, and p-S6, and increased MAP1LC3B-II levels (**Figure 7C**). Moreover, treatment of MK-2206, a specific inhibitor of Akt, abrogated Akt activation, and reversed the ratio of MAP1LC3B-II/β-action, and decreased the level of p-Akt, pmTOR, and p-S6 (**Figure 7D**). Then, to investigate whether c-Met siRNA or MK-2206 reverse inflammatory response during H. pylori infection, we detected the expression of inflammatory cytokines. As shown in **Figure 7E**, there was a significant decrease in the production of proinflammatory cytokines in cells transfected with siRNA specific for c-Met upon infection. Similarly, the expression of inflammatory cytokines significantly decreased in AGS cells treated with MK2206 during infection (**Figure 7F**). Together, we concluded that the CagA-mediated autophagy pathway may be dependent on the c-Met/Akt signaling pathway, which could regulate the expression of inflammatory cytokines.

## DISCUSSION

In the present study, we observed that (i) autophagy was downregulated in gastric mucosal tissues infected with cagA+ positive H. pylori strains, with increased gastric inflammation; (ii) CagA inhibited autophagy and induced production of proinflammatory cytokines in AGS cells; (iii) CagA downregulated starvationinduced autophagy; (iv) Inhibition of autophagy enhanced H. pylori-induced cytokine production; (v) c-Met siRNA significantly affected CagA-mediated autophagy; and (vi) CagA regulates autophagy through c-Met/Akt signaling pathway. These findings indicate that CagA may act as a negative regulator of autophagy in H. pylori-induced inflammatory response. Specifically, given that inflammation and autophagy are major determinants of gastric malignancy (Mohri et al., 2012), it also opens a new avenue of research on gastric malignancies, especially prophylaxis and treatment.

Autophagy, as the quality control of the cellular environment, plays an important role in the protective response during infection (Deretic, 2010). However, a number of pathogens

enzyme-linked immunosorbent assay (ELISA). (C) After pretreatment of SC (solvent control, 0.1% DMSO), 3-MA (2 mM), Baf-A1 (10 nM) or Rapa (100 nM), AGS cells were infected with *Hp*-WT or *Hp*-1*cagA* (MOI = 100:1) for 6 h. Supernatants were assessed by ELISA for levels of IL-8, IL-1β, and TNF-α. (D) Production of IL-8, IL-1β, and TNF-α in AGS cells transfected with siRNA specific for ATG5 or ATG12 (50 nM) for 24 h and infected with *Hp*-WT or *Hp*-1*cagA* (MOI = 100) for 6 h, as assessed by ELISA. Data are presented as the mean ± SEM of three experiments. \**P* < 0.05, \*\**P* < 0.01.

could subvert autophagy to promote inflammation generation, the occurrence and promotion of tumor, and genetic instability (Deretic and Levine, 2009). Previous studies have reported that autophagosome formation was induced by VacA of H. pylori in vitro (Terebiznik et al., 2009), but VacA could also disrupt autophagic flux to promote the infection (Raju et al., 2012). In the present study, we demonstrated that CagA could inhibit autophagy, increased the production of proinflammatory cytokines and facilitated gastric inflammation. In gastric mucosal tissues, autophagy was downregulated in patients infected with CagA positive H. pylori strains, which was accompanied with an increased production of cytokines. To rule out the effect of

control siRNA for 24 h, and then infected with *Hp*-WT or *Hp*-1*cagA* for 6 h. The percentages of cells with MAP1LC3B punctas are shown in the right graph with data being expressed as means ± SEM of three experiments (*n* ≥ 200 cells). (D) Western blot analysis of p-c-Met, MAP1LC3B-II conversion and β-actin in AGS cells transfected with c-Met siRNA or control siRNA and infected with *Hp*-WT or *Hp*-1*cagA* for 6 h. p-c-Met and MAP1LC3B-II band intensity was normalized to β-actin. (E,F) Flow cytometry showing MDC (upper panel) and AO (lower panel) staining of AGS cells transfected with c-Met siRNA or control siRNA and then infected with *Hp*-WT or *Hp*-1*cagA* for 6 h. (G) Western blot analysis of p-c-Met, MAP1LC3B-II conversion and β-actin in CagA-expressing AGS cells (AGS cells after transfecting the CagA expression plasmid, *GFP-CagA*) after transfected with c-Met siRNA or control siRNA and infected with *H. pylori* as described above. Experiments performed in triplicate showed consistent results. \**P* < 0.05, \*\**P* < 0.01.

VacA on autophagy, the toxigenic vacA genotype (vacAs1m1), expressing a functional VacA toxic, were excluded from the study. We selectively recruited patients with H. pylori negative infection, ones infected with H. pylori cagA−/vacAs1m2 strains and ones infected with cagA+/vacAs1m2 strains in the present study. As shown in **Figures 2A–C**, the signaling molecules such as, lower MAP1LC3B-II conversion, SQSTM1 accumulation and decreased LAMP1 expression (late endosomal/lysosomal marker;

cells were infected with *Hp*-WT or *Hp*-1*cagA* (MOI = 100:1) for 6 h. Supernatants were assessed by ELISA for levels of IL-8, IL-1β, and TNF-α. Experiments performed in triplicate showed consistent results. \**P* < 0.05.

Yu et al., 2010) in gastric mucosal tissues infected with cagA<sup>+</sup> H. pylori strains compared with cagA<sup>−</sup> H. pylori strains, which indicated that autophagic activity was decreased with increased gastric inflammation in patients infected with cagA+/vacAs1m2 strains. These results suggest that H. pylori CagA might induce inflammation by inhibiting autophagy. The intracellular CagA could be degraded by autophagy and short lived in AGS cells (Tsugawa et al., 2012). These finding suggest that induction of autophagy by H. pylori infection can degrade CagA by host cell defenses. Our observations indicate that persistent infection of bacterial exerts CagA to inhibit autophagy and induce inflammation.

Our observation that CagA is a negative regulatory factor for autophagy induced by H. pylori infection is consistent with findings of Deen's study (Deen et al., 2015), which showed that cagPAI of H. pylori has an inhibitory role in autophagy in macrophages. In addition, our results are also consistent with another study in which gastric biopsies from patients infected with cagA+/vacAs1m1 strains showed a significantly higher accumulation of SQSTM1 in the gastric epithelium compared with patients infected with a nonfunctional cagA−/vacAs2m<sup>2</sup> strains (Raju et al., 2012). Given that tyrosine phosphorylation of CagA plays critical roles in the activation of many pathways (Moss et al., 2001; Boonyanugomol et al., 2011; Wandler and Guillemin, 2012), we constructed the corresponding tyrosine phosphorylation mutants of the parent CagA (GFP-CagA-Mut). Our results demonstrated that tyrosine phosphorylation of CagA did not affect the PI3K/Akt/mTOR pathway, autophagy, and inflammation, suggesting that inhibition of autophagy is not dependent on tyrosine phosphorylation of CagA. Thus, the more specific mechanism of autophagy inhibited by CagA needs to be further investigated in the future.

It is well-established that autophagy plays critical roles in innate and adaptive immunity (Deretic et al., 2013), and disrupted autophagy is involved in secreting the proinflammatory cytokines, such as: IL-1α, IL-8, and IL-18 (Martins et al., 2015). Several studies have reported that autophagy may be an important mechanism for controlling inflammation in patients with Crohn's disease (Hampe et al., 2007; Rioux et al., 2007). Here, we demonstrated that autophagy inhibition enhanced the production of proinflammatory cytokines in H. pylori infection. SQSTM1, which is a major cargo ubiquitin-binding receptor in cells, is degraded by autolysosomes, and deficiencies of autophagy leads to accumulation of SQSTM1 (Wang et al., 2006). Moreover, SQSTM1 has further beneficial effects in NF-κB dependent cytokine production (Dupont et al., 2009). In the present study, there was a significant accumulation of SQSTM1 in the gastric mucosa of patients infected with CagA-positive H. pylori strains. When autophagy was inhibited, the activity of NF-κB was enhanced in AGS cells infected with mutant H. pylori strains (i.e., Hp-1cagA). These results suggested that autophagy inhibited by CagA leads to accumulation of SQSTM1, resulting in NF-κB dependent cytokine production.

CagA activates c-Met through its CRPIA (i.e., conserved repeat responsible for phosphorylation-independent activity) motif, which is critical for activation of PI3K/Akt signaling pathway and the pleiotropic transcriptional responses in H. pylori infection, including activation of NF-κB and βcatenin (Suzuki et al., 2009). Our data showed that, CagA was coimmunoprecipitated with c-Met in AGS cells during H. pylori infection, and siRNA silencing mediated c-Met knockdown in AGS enhanced the autophagy significantly in cells infected with wide-type cagA<sup>+</sup> H. pylori strain (i.e., Hp-WT). The PI3K/Akt signaling pathway participates in autophagy via mTOR, an autophagic regulators, resulting in autophagy suppression (Harashima et al., 2012). In the present study, we showed

#### REFERENCES


that CagA-positive H. pylori significantly increased the level of phosphorylated Akt at Ser473 and the levels of p-mTOR and p-S6 in AGS cells. The Akt inhibitor reversed the ratio of MAP1LC3B-II/β-actin in CagA-positive H. pylori infection, and blocked the level of phosphorylated Akt at Ser473. These findings clearly indicate that CagA inhibits autophagy via the c-Met-PI3K/AktmTOR signaling pathway.

Although CagA has already been reported to be a virulent factor in the inflammation induced by H. pylori infection, this is a new study demonstrating that CagA negatively regulates autophagy through c-Met-PI3K/Akt-mTOR signaling pathway, which is associated with increased expression of proinflammatory cytokines. Therefore, we postulate that inhibition of autophagy by CagA promotes gastric inflammation, which, in turn, initiates the multistep of gastric carcinogenesis (Correa, 1992). Moreover, given the pleiotropic actions of CagA, the interplay between CagA and autophagy regulation mechanism, which needs to be further investigated. A better understanding of the molecular mechanisms by which H. pylori infection modulates and interplays with autophagy will shed new insight into the development of more effective therapeutic strategies for H. pylori infection.

## AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: NL and BT. Performed the experiments: YJ, PZ, YZ, and YF. Analyzed the data: QL, KW, WZ, GG, and TW. Wrote the paper: YjF, BQ, XM, and QZ.

### ACKNOWLEDGMENTS

Thanks Dr. Ji-qin Lian and Dr. Shi-ming Yang (Third Military Medical University, Chongqing, China) for editing of the manuscript, and Medjaden Bioscience Limited for assisting in the preparation of this manuscript. This work was supported by grants from National Natural Science Foundation of China (NSFC, 81301482 and 81501721).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00417/full#supplementary-material

proliferation, apoptosis, and inflammation in biliary cells. Dig. Dis. Sci. 56, 1682–1692. doi: 10.1007/s10620-010-1512-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 © 2017 Li, Tang, Jia, Zhu, Zhuang, Fang, Li, Wang, Zhang, Guo, Wang, Feng, Qiao, Mao and Zou. 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.

# Probiotics for the Treatment of Atopic Dermatitis in Children: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Ruixue Huang<sup>1</sup> , Huacheng Ning<sup>1</sup> , Minxue Shen2, 3, Jie Li 2, 3, Jianglin Zhang2, 3 and Xiang Chen2, 3 \*

<sup>1</sup> Department of Occupational and Environmental Health, Xiangya School of Public Health, Central South University, Changsha, China, <sup>2</sup> Department of Dermatology, Xiangya Hospital, Central South University, Changsha, China, <sup>3</sup> Hunan Key Laboratory of Skin Cancer and Psoriasis, Xiangya Hospital, Central South University, Changsha, China

Objective: Atopic dermatitis (AD) is a prevalent, burdensome, and psychologically important pediatric concern. Probiotics have been suggested as a treatment for AD. Some reports have explored this topic; however, the utility of probiotics for AD remains to be firmly established.

Methods: To assess the effects of probiotics on AD in children, the PubMed/Medline, Cochrane Library Scopus, and OVID databases were searched for reports published in the English language.

#### Edited by:

Pascale Alard, University of Louisville, United States

#### Reviewed by:

Valerio Iebba, Sapienza Università di Roma, Italy Arianna Aceti, Università di Bologna, Italy

> \*Correspondence: Xiang Chen chenxiang\_xy@126.com

Received: 19 May 2017 Accepted: 22 August 2017 Published: 06 September 2017

#### Citation:

Huang R, Ning H, Shen M, Li J, Zhang J and Chen X (2017) Probiotics for the Treatment of Atopic Dermatitis in Children: A Systematic Review and Meta-Analysis of Randomized Controlled Trials. Front. Cell. Infect. Microbiol. 7:392. doi: 10.3389/fcimb.2017.00392 Results: Thirteen studies were identified. Significantly higher SCORAD values favoring probiotics over controls were observed (mean difference [MD], −3.07; 95% confidence interval [CI], −6.12 to −0.03; P < 0.001). The reported efficacy of probiotics in children < 1 year old was −1.03 (95%CI, −7.05 to 4.99) and that in children 1–18 years old was −4.50 (95%CI, −7.45 to −1.54; P < 0.001). Subgroup analyses showed that in Europe, SCORAD revealed no effect of probiotics, whereas significantly lower SCORAD values were reported in Asia (MD, −5.39; 95%CI, −8.91 to −1.87). Lactobacillus rhamnosus GG (MD, 3.29; 95%CI, −0.30 to 6.88; P = 0.07) and Lactobacillus plantarum (MD, −0.70; 95%CI, −2.30 to 0.90; P = 0.39) showed no significant effect on SCORAD values in children with AD. However, Lactobacillus fermentum (MD, −11.42; 95%CI, −13.81 to −9.04), Lactobacillus salivarius (MD, −7.21; 95%CI, −9.63 to −4.78), and a mixture of different strains (MD, −3.52; 95%CI, −5.61 to −1.44) showed significant effects on SCORAD values in children with AD.

Conclusions: Our meta-analysis indicated that the research to date has not robustly shown that probiotics are beneficial for children with AD. However, caution is needed when generalizing our results, as the populations evaluated were heterogeneous. Randomized controlled trials with larger samples and greater power are necessary to identify the species, dose, and treatment duration of probiotics that are most efficacious for treating AD in children.

Keywords: probiotics, constipation, children, meta-analysis, randomized controlled trial

## INTRODUCTION

Atopic dermatitis (AD), is one of the most common chronic inflammatory skin disorders among infants and children. AD is characterized by itching and recurrent eczematous lesions, and its incidence has increased worldwide over the past several decades. The current prevalence rate is 10–20% in infants and children (Weidinger and Novak, 2016). As the leading non-fatal medical skin disorder, AD imposes severe psychosocial burdens on pediatric patients and their families (Chamlin and Chren, 2010; Silverberg, 2016; Sidbury and Khorsand, 2017). AD is associated with high risks of allergy, asthma, and mental health issues (Sung et al., 2017). Infants and children with AD are typically treated with topical corticosteroids (TCS), antihistamines, and even antibiotics (Totri et al., 2017). However, these medications exert several adverse side effects, and AD symptoms may recur rapidly after treatment is stopped. Furthermore, long-term TCS use may trigger new-onset AD.

Probiotics is becoming increasingly attractive as a treatment option for some illnesses in children (Fuchs-Tarlovsky et al., 2016). Probiotics (live bacteria or yeasts) are not necessarily harmless, but they help to protect hosts from harmful bacteria (Mizock, 2015). When administered in adequate amounts, probiotics may play beneficial roles not only in the gastrointestinal tract but also in the gut–brain–skin axis (Ogden and Bielory, 2005; Dehingia et al., 2015; Huang et al., 2016; Huang and Hu, 2017). Several studies on the benefits of probiotics for pediatric AD patients have appeared over the past decades. In 2000, Pessi et al. reported that oral probiotics alleviated the clinical symptoms of gastrointestinal inflammation and AD (Pessi et al., 2000). Kirjavainen et al. (2003) reported lower Bacteroides counts in the fecal microflora of children with atopic eczema than in healthy infants and suggested that probiotics can be used to treat AD in children (Kirjavainen et al., 2003); however, some reports yielded contrasting results (Licari et al., 2015). For instance, Gruber et al. found that Lactobacillus rhamnosus strain GG (LGG) exerted no therapeutic effects in infants with mild-to-moderate AD (Gruber et al., 2007). Therefore, we systematically evaluated the effects of probiotics used to treat AD in children.

## METHODS

#### Inclusion Criteria

The inclusion criteria for the meta-analysis were (1) RCTs of children aged ≤18 years in whom AD severity was graded by experienced dermatologists using the Severity scoring of atopic dermatitis: the SCORAD index (1993); Yoon et al. (2015) (2) that evaluated the use of any probiotic culture/strain/dose/therapy regimen (including studies on fermented yogurt; all dosage forms including tablets, powders, oil suspensions, and capsules were included). All results are presented as means ± standard deviation. However, if multiple reports evaluated the same group of patients, we selected only the most recent complete report. SCORAD, developed by the European Task Force on AD in 1993 (1993), assesses the AD area, clinical features, visual analog scale data, and clinical symptoms, and it is widely used to evaluate AD severity in children (Machura et al., 2008).

#### Exclusion Criteria

Studies that did not meet the inclusion criteria or that were published in languages other than English were excluded.

#### Search Process

Two individuals of our team searched the following databases from the times of the earliest records in 2000 to April 12, 2017: PubMed (https://www.ncbi.nlm.nih.gov/pubmed), Embase (https://www.embase.com/login), Cochrane Library (http:// www.cochranelibrary.com/) and Scopus (https://www.elsevier. com/solutions/scopus) (available on the internet); and Ovid, Orbis, and the Web of Science (available at our university library with free downloads). The following search string was used in searching: [(infant OR infants) OR (neonate OR neonates) OR (newborn OR newborns) OR (toddler OR toddlers)] AND (probiotic OR probiotics OR pro-biotics OR probio<sup>∗</sup> ) AND (atopic dermatitis OR atopic eczema) OR (SCORAD) OR (atopic OR atopy) NOT (animals) NOT (adult). The references listed in each report were examined to allow us to retrieve additional information. We only reviewed works in the English language, thus not those in (for example) Korean or Chinese. Furthermore, conference abstracts were excluded, because they lacked detailed data.

### Data Collection

The two individuals collected all data independently. The eligibility of studies was confirmed by both reviewers. A tabulation of study author(s), publication date, recruited numbers, probiotic strain(s), dosage, treatment duration, and treatment results was prepared (**Table 1**). If the study data were unclear, we attempted to contact the corresponding author via email to obtain further information.

#### Statistical Analysis

RevMan 5.3 software (Cochrane Collaboration, Nordic Cochrane Center, Copenhagen, Denmark; http://community.cochrane. org/tools/review-production-tools/revman-5/) was accessed to conduct the meta-analysis. SCORAD was commonly used to measure the efficacy of probiotics in children with AD. As the results were continuous data, the mean difference (MD) and 95%CI were calculated for statistical analyses, and either a randomized-effects model or fixed-effects model was used depending on whether heterogeneity was apparent. Subgroup assessment was performed with regard to different geographical status, infants aged <1 year, children aged between 1 and 18 years, different strains, and LGG. The c<sup>2</sup> test was used to identify statistical heterogeneity (Margolis and Mitra, 2017). The I 2 statistic was calculated to identify and quantify inconsistency. When I<sup>2</sup> was ≥ 50%, indicating significant heterogeneity, we used a random-effects model for meta-analysis. When I<sup>2</sup> was < 50% indicating no heterogeneity, we employed a fixed-effects model. Publication bias was assessed by constructing funnel plots. A two-tailed P < 0.05 was used to reflect statistical significance. Sensitivity analyses, also termed uncertainty analyses, were used to explore the extent to which our results and conclusions were altered by changes in the data or analysis approach (Alexander et al., 2016). If the conclusions did not change upon application of the sensitivity analysis, those conclusions were considered robust.

#### TABLE 1 | Characteristics of included RCTs for meta-analysis.


In meta-analyses, sensitivity analyses are conducted by excluding studies one-by-one to identify those studies that materially affect the results (Copas and Shi, 2000). The risk of bias in each RCT was explored using the "risk of risk" tool in Revman software. The PRISMA statement published in 2009 aimed to improve the reporting of systematic reviews and meta-analyses. PRISMA defines an evidence-based minimum set of items to employ, and we followed this guideline (http://www.prisma-statement.org/). PRISMA features both a checklist and a flow diagram. We used the checklist to ensure that our study structure was appropriate and the flow diagram to map the numbers of records identified, included, and excluded, as well as the reasons for exclusion (Zhang et al., 2017). Publication bias was checked by drawing funnel plots, which are commonly used in systematic reviews and meta-analyses. Publication bias is considered absent if the study results are distributed in close proximity to the averages.

#### RESULTS

#### Included Studies

The PRISMA flow diagram (**Figure 1**) shows how we selected the relevant reports. We initially screened 392 articles, excluded those that did not meet our inclusion criteria, and finally retained 26 articles. As some reports did not report data as means ± SD,

we contacted the corresponding authors by email. Unfortunately, we sent 13 emails and didn't receive any data suitable for inclusion in the meta-analysis. Ultimately, 13 studies involving 1,070 children fulfilled our selection criteria (**Table 1**).

#### Quality Assessment

**Figure 2A** shows the risk of bias within all enrolled RCTs, as adjudged by the two reviewers. **Figure 2B** presents the individual risks of bias, again as perceived by the reviewers. Both figures show that the risks of bias were rather low, because all were RCTs that adhered to high standards. Four studies divided children into probiotic intervention and control groups; two studies created three groups (probiotics, a placebo, and another intervention). Twelve studies were of double-blind design. All 13 studies reported baseline data including socioeconomic status and mean age; these did not differ significantly among the groups.

## Probiotics and Children with AD

Data from 1,070 children (intervention group, 553; control group, 517) were assessed. The outcome of a random-effects

meta-analysis model involving all 13 trials is shown in **Figure 3**. Significant differences in SCORAD values favoring probiotics over the control were observed overall (MD, −3.07; 95%CI, −6.12 to −0.03; P < 0.00001). However, a high degree of heterogeneity was observed across these 14 trials (I<sup>2</sup> = 87%).

#### Subgroup Analysis of Probiotics Efficacy by Age

All 13 trials involved children aged 0–18 years. We categorized the children into two groups: infants <1 year old and children 1–18 years old. Accordingly, five trials were included in the <1 year subgroup, and nine trials were included in the 1–18 years subgroup (**Figure 4**). The efficacy of probiotics in the former subgroup was −1.03 (95%CI, −7.05 to 4.99) and that in the latter subgroup was −4.50 (95%CI, −7.45 to −1.54; P < 0.001). However, a high degree of heterogeneity was observed among the <1 year subgroup (I<sup>2</sup> = 94%).

#### Subgroup Assessment by Continent

Subgroup assessment by continent showed different effects. In Europe, probiotics showed no effect on SCORAD, whereas significantly lower SCORAD values were reported in Asia (MD, −5.39; 95%CI, −8.91 to −1.87). In Australia, the MD was −11.20 (95%CI, −13.76 to −8.64). However, there was heterogeneity among these trials (**Figure 5**).

#### Subgroup Assessment of Different Cultured Organisms

MD scoring compared to control and placebo interventions was performed by cultured organism group. LGG (MD, 3.29; 95%CI, −0.30 to 6.88; P = 0.07) and LP (MD, −0.70; 95%CI, −2.30 to 0.90; P = 0.39) showed no significant effects on SCORAD values in children. However, LF (MD, −11.42; 95%CI, −13.81 to −9.04), LS (MD, −7.21; 95%CI, −9.63 to −4.78), and a mixture of different strains (MD, −3.52; 95%CI, −5.61 to −1.44) showed significant effects on SCORAD values in children (**Figure 6**).

## Publication Bias

We used RevMan software to draw funnel plots (**Figure 7**), wherein each dot represents data from a single RCT. A randomeffects model was used to this end. The funnel plots were somewhat asymmetrical, thus indicating potential publication bias, perhaps attributable in part to the fact that we included only English-language publications and excluded conference abstracts. However, studies with positive outcomes are more

likely to be published than are those with negative outcomes, thus creating bias.

## Sensitivity Testing

We performed sensitivity analyses to assess the relative influence of each study by excluding the studies one by one, and the results suggested no significant changes in effects with regard to subgroups.

## DISCUSSION

Overall, the data suggested an overall benefit of probiotics supplementation in children with AD, and age-specific subanalyses showed that probiotics effectively reduce SCORAD values in children aged 1–18 years. Geography-specific subanalyses showed that probiotics effectively reduced SCORAD values in Asia, while no effect was observed for Europe. Strain-specific sub-analyses indicated that Lactobacillus (LS), Lactobacillus fermentum (LF), and a probiotic mixture reduced SCORAD values in children with AD, while LGG and Lactobacillus plantarum (LP)showed no effect in children with AD.

Hippocrates (460–370) stated that "All diseases begin in the gut", which is the earliest suggestion that bacteria affect health (Hippocrates, 2002). Metchnikoff, known as the father of probiotics (Gordon, 2016), proposed that colonic bacteria afforded health benefits in aging adults. In recent decades, probiotics that aid in the resolution of pediatric atopic eczema have been investigated. Viljanen et al. explored probiotic effects on pediatric atopic eczema/dermatitis syndrome but found no significant difference between the treatment and control groups (Viljanen et al., 2005). Passeron et al. compared probiotics and prebiotics and found that both significantly improved AD manifestations in children (Passeron et al., 2006). Brouwer et al. evaluated the clinical and immunological effects of Lactobacillus rhamnosus (LR) supplementation in a hydrolyzed formula given to children with AD but found no significant effect (Brouwer et al., 2006). The cited authors suggested that the discrepancies between their results and those of other trials were likely attributable to differences in treatment timing and the strains

used. Sistek et al. conducted a 12-week trial in the UK and found that a combination of LR and Bifidobacteria lactis (BL) improved AD symptoms in food-sensitive children (Sistek et al., 2006). At roughly the same time, a prospective German study by Folster-Holst et al. yielded insufficient evidence to make the conclusion that LGG is an effective treatment for moderate-to-severe AD in infants (Folster-Holst, 2010). Gruber et al. also found that LGG had no therapeutic effect in such patients (Gruber et al., 2007). Despite these discouraging findings, Gerasimov et al. reported that Lactobacillus acidophilus DDS-1 and Bifidobacterium lactis UABLA-12 afforded significant clinical improvements in children with moderate-to-severe AD (Gerasimov et al., 2010). Similarly, Wu et al. showed that Lactobacillus salivarius (LS) exerted shortterm beneficial effects in patients with moderate-to-severe AD (Wu et al., 2012). Drago et al. suggested that such effects may be attributable to restoration of the altered intestinal microbiota (Drago et al., 2011). In contrast, Gore et al. found that LS exerted no beneficial effects on eczema when given as an adjunct to basic topical treatment (Gore et al., 2012). Several reports have examined the effects of other bacterial strains on AD in children. Supplementation with LPCJLP 133, Lactobacillus paracasei, and LF was reported to be effective. The discrepancies described above may be attributable to differences in the strains used, the study areas, and/or the ethnicities of the subjects. Several groups have performed meta-analyses to evaluate the effectiveness of probiotics on AD. Da Costa Baptista et al. reviewed all published trials and reported that the biological effects observed in most trials suggest that probiotic adjuvant treatments are of benefit for AD (da et al., 2013). The cited review, although comprehensive, did not report total MDs or 95%CIs. Chang performed a

meta-analysis of studies in which either prebiotics or probiotics were given and reported that synbiotics may be useful to treat AD (Chang et al., 2017). However, the focus was on synbiotics rather than probiotics. Szajewska et al. stressed the need for data on individual probiotic strains rather than on probiotics in general (Szajewska and Mrukowicz, 2003; Szajewska et al., 2015). Ogden et al. suggested probiotics as a complementary approach to the treatment and prevention of pediatric AD (Ogden and Bielory, 2005). They concluded that probiotics should be an active area of investigation, considering the role of gut microbiota in altered immune responses in atopic patients. However, the authors did not perform a meta-analysis to obtain further details about the treatment effects of probiotics. Kim et al. reviewed 25 RCTs on the effects of probiotics in the treatment of AD in patients of all ages. They observed significant differences in SCORAD values favoring probiotics over the control group in children 1–18 years old and in adults, whereas no favorable effects were seen in infants <1 year old (Kim et al., 2014). We found that probiotics were efficacious in children aged 1–18 years (MD, −4.50; 95%CI, −7.45 to −1.54) and showed strong efficacy in Asia but not in Europe; furthermore, LGG had no effects on AD whereas LS, LF, LP, and a mixture of strains showed beneficial effects. Our findings are in agreement with those of Lee et al., who concluded that the evidence for probiotics as a useful treatment of AD in children is convincing. However, the cited authors reviewed only trials published before 2008, whereas we included later

trials to afford greater insight. The differences may be because we included only RCTs involving children under the age of 18 years and those that reported MDs. Some RCTs presented values other than MDs, including the study by Kim et al. (2010), and some presented the results as figures, rendering calculations impossible. We contacted the corresponding authors but did not receive useful replies. Thus, we excluded those studies. Third, some of the included studies had small sample sizes, which may affect the reliability and validity of the conclusions. Thus, our overall results are affected by these issues, and the data were highly heterogeneous. These topics require further attention. Also, in the subgroup analyses, children with AD may have different gut microbiota profiles from those of normal children. Thus, probiotics supplementation in children < 1 year old and 1–18 years old may promote a healthier gut microbiota profile, boosting their immune response. People from different areas have different dietary structures and gut bacterial compositions. Dehingia et al. compared gut bacterial diversity between Indian populations and worldwide data (Dehingia et al., 2015). Zhang et al. also suggested that a phylogenetically diverse gut microbiota at the genus level may be commonly shared by distinctive healthy populations, which may explain the diversity of the effects of probiotics across people from different countries (Zhang et al., 2015). The above discussion is of importance to physicians, dermatologists, and other public healthcare workers who deal with diverse ethnic populations.

To the best of our knowledge, there are no previous reports on the effects of different probiotic strains on AD in children. In our meta-analysis, all trials involving LGG and one trial involving LP showed no effects, while two studies confirmed the beneficial

#### REFERENCES

(1993). Severity scoring of atopic dermatitis: the SCORAD index. Consensus Report of the European Task Force on Atopic Dermatitis. Dermatology 186, 23–31. doi: 10.1159/000247298

effects of LP on AD (MD, −0.70; 95%CI, −2.30 to 0.90; P = 0.39). This discrepancy may be associated with differences in dosages, the timing and duration of intervention, and sample sizes, and further trials are required to clarify this point.

Our meta-analysis had certain limitations. First, we attempted to minimize heterogeneity and publication bias, but significant heterogeneity among trials remained evident. Differences in study samples, study populations, and intervention methods contributed to the heterogeneity. For example, some of the included studies had small sample sizes, compromising the reliability and validity of the conclusions. In addition, the RCTs were performed in various countries, thus, the subjects differed among RCTs in terms of their genetic make-up and microbial exposure, which in turn are associated with varying responses to the same probiotic. Also, we excluded some RCTs from this meta-analysis, and fewer studies included will reduce the confidence associated with the data interpretation and increase heterogeneity and publication bias. Finally, we cannot draw robust conclusions as to which probiotic strain/mixture should be given to children with AD and which population(s) would receive maximum benefit from such treatment.

#### CONCLUSION

Our present work demonstrated that probiotics may have the potential to decrease SCORAD values in children with AD. However, the findings presented here must be generalized with caution because of heterogeneity. The results are a source of optimism with regard to the management of AD in children. More adequately powered RCTs using standardized measurements are necessary to assess which species of probiotics and dosages and what treatment periods are most efficacious for children with AD.

#### AUTHOR CONTRIBUTIONS

RH, MS, and XC proposed the idea of this study and designed the study; RH, HN, MS, and JL conducted data screening and performed quality assessment; RH and MS used RevMan software to assess the data and performed the statistical analysis and gave the explanations of the statistical results. RH drafted the initial manuscript. JL, JZ, and XC critically reviewed and revised the manuscript.

#### FUNDING

This work was supported by China's National Basic Work of Science and Technology (Grant# 2015FY111100).

Alexander, P. E., Bonner, A. J., Agarwal, A., Li, S. A., Hariharan, A. T., Izhar, Z., et al. (2016). Sensitivity subgroup analysis based on single-center vs. multi-center trial status when interpreting meta-analyses pooled estimates: the logical way forward. J. Clin. Epidemiol. 74, 80–92. doi: 10.1016/j.jclinepi.2015. 08.027


importance of viability. J. Pediatr. Gastroenterol. Nutr. 36, 223–227. doi: 10.1097/00005176-200302000-00012


cohorts across lifestyles, geography and ethnicities. ISME J. 9, 1979–1990. doi: 10.1038/ismej.2015.11

Zhang, Y. M., Chu, P., and Wang, W. J. (2017). PRISMA-combined alphablockers and antimuscarinics for ureteral stent-related symptoms: a meta-analysis. Medicine 96:e6098. doi: 10.1097/MD.00000000000 06098

**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 Huang, Ning, Shen, Li, Zhang 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.

# Interactions of Intestinal Bacteria with Components of the Intestinal Mucus

Jean-Félix Sicard<sup>1</sup> , Guillaume Le Bihan<sup>1</sup> , Philippe Vogeleer <sup>1</sup> , Mario Jacques <sup>2</sup> and Josée Harel <sup>1</sup> \*

<sup>1</sup> Centre de Recherche en Infectiologie Porcine et Aviaire, Faculté de Médecine Vétérinaire, Université de Montréal, Saint-Hyacinthe, QC, Canada, <sup>2</sup> Regroupement de Recherche Pour un Lait de Qualité Optimale (Op+Lait), Faculté de Médecine Vétérinaire, Université de Montréal, Saint-Hyacinthe, QC, Canada

The human gut is colonized by a variety of large amounts of microbes that are collectively called intestinal microbiota. Most of these microbial residents will grow within the mucus layer that overlies the gut epithelium and will act as the first line of defense against both commensal and invading microbes. This mucus is essentially formed by mucins, a family of highly glycosylated protein that are secreted by specialize cells in the gut. In this Review, we examine how commensal members of the microbiota and pathogenic bacteria use mucus to their advantage to promote their growth, develop biofilms and colonize the intestine. We also discuss how mucus-derived components act as nutrient and chemical cues for adaptation and pathogenesis of bacteria and how bacteria can influence the composition of the mucus layer.

Keywords: mucus, commensals, pathogens, biofilm, microbiota, microflora, goblet cells

## INTRODUCTION

#### Edited by:

Pascale Alard, University of Louisville, United States

Reviewed by: Valerio Iebba, Sapienza Università di Roma, Italy Bruce Vallance, University of British Columbia, Canada

> \*Correspondence: Josée Harel josee.harel@umontreal.ca

Received: 02 May 2017 Accepted: 18 August 2017 Published: 05 September 2017

#### Citation:

Sicard J-F, Le Bihan G, Vogeleer P, Jacques M and Harel J (2017) Interactions of Intestinal Bacteria with Components of the Intestinal Mucus. Front. Cell. Infect. Microbiol. 7:387. doi: 10.3389/fcimb.2017.00387 The gastrointestinal tract harbors a complex bacterial community called the intestinal microbiota that, in healthy conditions, maintains a commensal relationship with our body. Various mechanisms are used by the host to keep intestinal homeostasis and to prevent aberrant immune responses directed against the microbiota. One of these is the production of a mucus layer that covers the epithelial cells of the gut. This mucus is synthesized and secreted by host goblet cells and form an integral structural component of the mammal intestine. Its major function is to protect the intestinal epithelium from damage caused by food and digestive secretions (Deplancke and Gaskins, 2001). The mucus layer provides a niche for bacterial colonization because it contains attachment sites and is also a carbon source (Harel et al., 1993). Effectively, the mucus is a direct source of carbohydrates that are released in the lumen. Therefore, several bacterial species of the microbiota can use mucus glycan as a carbon source (Ouwerkerk et al., 2013). An alteration in glycan availability modifies the composition of the microbiota (Martens et al., 2008). The mucus layer also prevents pathogens from reaching and persisting on the intestinal epithelial surfaces and thereby is a major component of innate immunity. It is constantly renewed and acts as a trap for commensal residents, but also for pathogens, preventing their access to the epithelia (Johansson et al., 2008; Bertin et al., 2013). Although its composition and thickness vary along the gut, the mucus layer is mainly

**Abbreviations:** MLG, Mucus gel layer; A/E, Attaching and effacing; NAG, N-acetyl-D-glucosamine; NANA, N-acetylneuraminic acid; EHEC, Enterohemorrhagic E. coli; MUB, Mucus-binding proteins; LAB, Lactic acid bacteria; HBGA, Histoblood group antigen; SIgA, Secretory IgA; AIEC, Adherent invasive E. coli; LPB, LPS-binding protein; TLR, Toll-like receptor; VPI, Vibrio pathogenicity island.

formed of glycoproteins containing different glycans; nonspecific antimicrobial molecules, such as antimicrobial peptides (AMP); secreted antibodies targeting specific microbial antigens; and other intestinal proteins (McGuckin et al., 2011; Antoni et al., 2014). Interaction with the mucus layer is important for the colonization of gut commensals as well as some pathogens that have evolved to adhere to mucus and exploit it (Juge, 2012). Some pathogens also use mucus components as a cue to modulate the expression of virulence genes and thereby adapt to the host environment. In this Review, we describe the interactions between bacteria and components of the human mucus layer: their use as carbon sources, adhesion sites and their genetic adaptation (**Figure 1**).

## THE GASTROINTESTINAL MUCUS

#### Mucus Composition

The intestinal mucus is composed mainly of mucins that are complex agglomerates of structural glycoproteins with specific O-linked glycans (O-glycans) produced by specialized cells of the host called goblet cells (Forstner, 1995). Mucins can either be secreted and form a gel, or be produced as membrane-bound glycoproteins that are part of the epithelial glycocalyx (Johansson et al., 2008, 2011; Jonckheere et al., 2013; Nilsson et al., 2014). These glycoproteins share a common structure made of tandem repeated amino acids rich in proline, threonine and serine and are call PTS domains. These sequences of amino acid provide sites for the covalent attachment of the polysaccharides and are widely O-glycosylated (Moran et al., 2011). Four different types of polysaccharide core structures are commonly found in mucin glycoproteins. These cores are formed by a combination of three polysaccharides, galactose, N-acetyl-galactosamine and N-acetyl-glucosamine (Larsson et al., 2009; Juge, 2012). Different chains of glycan will be attached to the core. The terminal monosaccharide is usually a fucose or a sialic acid (Larsson et al., 2009; Juge, 2012). Oligosaccharide chains are also sulfated, especially in colonic regions (Rho et al., 2005). The mucin proteins MUC1, MUC5AC, and MUC6 mainly form the mucus layer in the stomach, whereas MUC2 is the most abundant mucin in the small intestine and the colon (Johansson et al., 2009; Moran et al., 2011). The thickness of the mucus layer varies through the gut. The colon, which harbors the highest density of microorganisms, is covered by the thickest mucus layer (Gum et al., 1994). It is composed of an inner layer that is dense and firmly attached to the epithelium and an outer loose layer that is exposed to bacterial proteolytic activity. The numerous O-glycans of the outer layer can serve as adhesion sites and as nutrients for bacteria while the inner layer is less permissive to bacterial penetration in healthy individuals (Johansson et al., 2008, 2011). Most bacterial residents are present in the outer mucus layer and the competition for survival in this niche shapes the composition of the microbiota. The differential resource utilization of bacterial species participates to the establishment of distinct communities that includes non-mucolytic bacteria (Li et al., 2015).

## Role of the Mucus Layer

The mucus barrier has an important role in regulating the severity of infectious diseases. It provides protection against many intestinal pathogens, including Yersinia enterocolitica, Shigella flexneri, Salmonella, and Citrobacter rodentium (Mantle and Rombough, 1993; Bergstrom et al., 2010; Arike and Hansson, 2016). MUC2 (Mouse, Muc2) plays a crucial role

dense inner layer, firmly attached to cells, that does not allow bacteria to penetrate. Further from the epithelium, the outer layer becomes loose and permissive, providing a niche for intestinal bacteria. (B) Mucus oligosaccharides can act as adhesion sites for bacteria, facilitating their colonization. Some bacteria are able to form microcolonies and biofilms. (C) Bacteria with mucolytic activity can release monosaccharides from mucin O-glycans and metabolize them. These sugars can also be utilized by nearby bacteria. (D) Mucus components can influence the behavior of pathogenic bacteria by increasing or decreasing their virulence expression, adhesion, motility, proliferation, or growth.

during infection. Using Muc2-deficient mice, it was shown that the glycoprotein is critical in controlling Salmonella infection (Zarepour et al., 2013). Moreover, Muc2−/<sup>−</sup> mice revealed higher susceptibility to attaching and effacing (A/E) Citrobacter rodentium infections (Bergstrom et al., 2010).

An alteration of mucosal integrity is generally associated with health problems, such as inflammatory bowel diseases, including ulcerative colitis and Crohn's disease (Trabucchi et al., 1986; Hanski et al., 1999). During ulcerative colitis, alteration of mucus integrity results in a thinner mucus layer due to goblet cell depletion (Pullan et al., 1994) and a reduced O-glycosylation and sulfation of mucins (Raouf et al., 1992; Larsson et al., 2011). During Crohn's disease, the mucus layer is essentially continuous and comparable to healthy mucosa (Strugala et al., 2008) although there is evidence of abnormal expression and glycosylation of the mucin (Buisine et al., 2001; Moehle et al., 2006; Dorofeyev et al., 2013). These changes in the mucosal environment could also be linked to dysbiosis, an abnormal change in the composition of the intestinal microbiota due to Crohn's disease. Once impaired, the mucus barrier becomes permeable to bacteria that are able to access the epithelium and therefore cause inflammation (Antoni et al., 2014; Johansson et al., 2014), which is why the integrity of the mucus layer is critical for the upkeep of a homeostatic relationship between the intestinal microbiota and its host.

## MUCIN AS A GROWTH SUBSTRATE

Mucin proteins are highly glycosylated and therefore constitute a carbon and energy source for intestinal microbiota. A key nutritional aspect of the mucus layer for gut bacteria is its high polysaccharide content with up to 80% of the mucin biomass being composed of mostly O-linked glycans (Johansson et al., 2009, 2011; Marcobal et al., 2013).

## Mucolytic Bacteria

A distinct subset of intestinal bacteria possesses the enzymatic activity, such as glycosidases, necessary for the degradation of mucin oligosaccharides, which can be further metabolized by resident microbiota (Koropatkin et al., 2012; Ouwerkerk et al., 2013). Indeed, various anaerobic bacteria species of gut microbiota, such as Akkermansia muciniphila (Derrien et al., 2004; Png et al., 2010), Bacteroides thetaiotaomicron (Xu et al., 2003; Sonnenburg et al., 2005), Bifidobacterium bifidum (Crociani et al., 1994; Png et al., 2010; Garrido et al., 2011), Bacteroides fragilis (Macfarlane and Gibson, 1991; Swidsinski et al., 2005a; Huang et al., 2011), Ruminococcus gnavus (Png et al., 2010; Crost et al., 2013), and Ruminococcus torques (Hoskins et al., 1985; Png et al., 2010) are now known as mucin-degrading specialists. These bacteria will use their specific enzymatic activities to release monosaccharides attached to the mucin glycoproteins. Some mucolytic bacteria, such as B. thetaiotaomicron, that possess an important variety of glycosidases, are better suited for the utilization of a wide range of glycans (Xu et al., 2003; Marcobal et al., 2013). To complete the degradation of mucins, a combination of enzymatic activity of several mucolytic bacteria is needed (Derrien et al., 2010; Marcobal et al., 2013). Therefore, MUC2 glycans act as nutritional sources for bacteria that can utilize the mucus-derived sugars, but lack the enzymes necessary for cleaving sugar linkages (Johansson et al., 2015; Arike and Hansson, 2016). Commonly, several bacteria collaborate in a community and it has been shown that the sulfatase activity of some commensal bacteria on sulfomucin allows glycosidases to access and act on mucins (Rho et al., 2005). Released saccharides, such as N-acetyl-Dglucosamine (GlcNAc also called NAG), N-acetylgalactosamine (GalNAc), galactose, fucose and sialic acid (N-acetylneuraminic acid also called NANA) can then be used by the degrader itself or by other resident bacteria (Bjursell et al., 2006; Martens et al., 2008; Sonnenburg et al., 2010). As example, commensal E. coli that are limited to growth on mono- or disaccharides, are unable to degrade the complex polysaccharides that constitute mucin (Hoskins et al., 1985) and therefore use such carbohydrate sources (Chang et al., 2004; Png et al., 2010; Bertin et al., 2013). Another example is vancomycin-resistant Enterococcus that can grow on mucin pre-digested with extracts from human stools, but not on purified mucin. This suggests that Enterococcus can benefit of the microbiota activity on mucin and uses released mucusderived products (Pultz et al., 2006). In this way, mucolytic bacteria make mucus O-glycan derived products also available for other bacterial residents.

## Use of Mucus-Derived Nutrients by Pathogens

Intestinal pathogens have developed strategies to compete with commensal microflora for nutrients, such as carbohydrates and these strategies have been reviewed in Conway and Cohen (2015), Vogt et al. (2015), and Baumler and Sperandio (2016). Pathogenic and commensal E. coli strains displayed considerable catabolic diversity when colonizing streptomycin-treated mice, indicating that nutrient availability can influence their colonization success and their niche adaptation (Maltby et al., 2013). For example, pathogenic E. coli such as enterohemorrhagic E. coli (EHEC) strain EDL933 efficiently utilizes some mucusderived monosaccharides. This can provide competitive growth compared to that of commensal E. coli (Fabich et al., 2008). Moreover, the metabolic flexibility of some pathogenic strains to use both glycolytic and gluconeogenic nutrients may be advantageous (Bertin et al., 2013). The pathogen Vibrio cholerae's preferential use of mucus-derived monosaccharides, such as GlcNAc and sialic acid confers an advantage in the infant mouse model of infection (Almagro-Moreno et al., 2015). C. jejuni also possess the ability to metabolize fucose. Its growth is enhanced in culture medium supplemented with it (Alemka et al., 2012). In addition, antibiotic treatment also perturbs the microbiota and therefore affects the availability of mucin carbohydrates. The concentration of free fucose and sialic acid reaching high levels during antibiotic treatment facilitates expansion of pathogens such as Salmonella enterica serotype Typhimurium and Clostridium difficile (Ng et al., 2013). In addition, Salmonella serotype Typhimurium is known both to bind glycoprotein containing sialic acids (Vimal et al., 2000) and to have the ability to release the carbohydrate using its sialidase (Hoyer et al., 1992). Thereby, to colonize specific niches, many pathogens have evolved in a way to use mucus-derived sugars as a carbon source.

## BACTERIAL ADHESION TO MUCINS

Mucins proteins are highly glycosylated. Their O-glycans are used as ligands for bacterial adhesins (Juge, 2012). It can be speculated that adhesion to mucins may initiate colonization of the intestine. The carbohydrate structures on mucins can provide initial attachment site to bacteria including specialized pathogens and could facilitate the invasion of epithelial cells (Derrien et al., 2010). As example, pathogenic microorganisms, such as Campylobacter and enterotoxinogenic E. coli (ETEC) are known to adhere to the glycoprotein MUC1 that is present in human breast milk. This interferes with colonization of these pathogens in the infant GI tract (Martin-Sosa et al., 2002; Ruiz-Palacios et al., 2003). Although no specific mucus-adherent microflora was identified (van der Waaij et al., 2005), there are evidence that bacteria can bind directly to mucins by expressing specific proteins, pili, fimbriae and flagella (**Table 1**).

#### Interactions between Mucin and Surface Proteins

To adhere to mucus, commensal and pathogenic bacteria use different strategies. First, they can produce proteins that specifically bind the mucus. Mucus-binding proteins (MUB) are cell-surface proteins mainly described in lactic acid bacteria (LAB) (Boekhorst et al., 2006), especially in Lactobacillus reuteri (Roos and Jonsson, 2002; MacKenzie et al., 2009). MUB contain domains that are similar to the model mucin-binding protein (MucBP) from the Pfam database (Boekhorst et al., 2006). The MucBP domain is found in a variety of bacterial proteins that are known for their capacity to adhere to mucus (Juge, 2012). MUB also share structural and functional homology with pathogenic Gram-positive adhesins that have specificity to sialylated mucin glycans (Etzold et al., 2014). For example, some surface proteins of Listeria monocytogenes contain a MucBP domain similar to those found in Lactobacillus, allowing them to adhere to mucin (Bierne et al., 2007; Mariscotti et al., 2014). The causative agent of cholera, V. cholerae, can also bind to mucin using surface protein called GbpA (chitin-binding protein) that binds specifically to N-acetyl D-glucosamine residues of intestinal mucins (Bhowmick et al., 2008). In addition, C. jejuni is wellknown for its ability to interact with different human histoblood group antigens (HBGAs) expressed in mucosa (Naughton et al., 2013). The major outer membrane protein (MOMP) of C. jejuni is involved in these interactions (Mahdavi et al., 2014). This way, C. jejuni can interact with intestinal mucin MUC2 in the intestine (Tu et al., 2008). Furthermore, Bifidobacterium spp. is also known for its specific adhesion to mucus. For example, in a B. bifidum mucin-binding assay, the expression of an extracellular transaldolase correlated with a positive


mucin-binding phenotype (Gonzalez-Rodriguez et al., 2012). B. longum subsp. infantis is another species that binds specifically to mucin using family-1 solute binding proteins (Kankainen et al., 2009). Interestingly, a study using gnotobiotic mice colonized by B. fragilis and E. coli revealed that the commensal bacterium B. fragilis was found in the mucus layer while E. coli was only found in the lumen. Further analysis showed that B. fragilis specifically binds to highly purified mucins. This indicated that a direct bond with intestinal mucus could be a mechanism used by B. fragilis for gut colonization (Huang et al., 2011).

#### Interactions between Mucin and Pili/Fimbriae

In addition to produce specific mucus binding proteins, some bacteria can also use cell-surface appendix, such as pili or fimbriae to bind the mucus. For example, production of pili by LAB was shown to be implicated in mucus-binding activity (Douillard et al., 2013) and moreover, the SpaC pilus protein of L. rhamnosus GG was shown to strongly binds the human mucins (Kankainen et al., 2009). An in vitro study using mucussecreting HT29-MTX intestinal epithelial cell model showed that the adhesion of Salmonellae enterica serotype Typhimurium to mucus-secreting intestinal epithelial cells was higher than in non- and low-mucus producing cells (Gagnon et al., 2013). Moreover, virulent strains seem to bind more efficiently to mucus than avirulent strains and the binding that preferentially targets the neutral mucin is mannose-dependant (Vimal et al., 2000). As with some uropathogenic E. coli (Wurpel et al., 2014), the adhesion of S. enterica serotype Typhimurium could be the result of interaction between fimbrial adhesin and mucin glycans, more specifically terminal fucose residues (Chessa et al., 2009). The E. coli K88 (F4) fimbriae is also able to bind mucus from the small intestines of 35-day-old piglets with a specificity to the glycolipid galactosylceramide (Blomberg et al., 1993). Hence, pili and fimbriae are involved in specific adhesion to mucus.

#### Interactions between Mucin and Flagella

Many enteric bacteria also produce flagellum. In addition to their role in motility, flagella are also involved in adhesion. As example, the E. coli probiotic strain Nissle 1917 was shown to be able to interact, via its flagella, with human and porcine mucus but not with murine mucus. Furthermore, the mucus component gluconate has been identified as one receptor for the adhesion of these flagella (Troge et al., 2012). Other studies have revealed the role of the flagella for the binding of mucin glycoproteins by C. difficile (Tasteyre et al., 2001) and pathogenic E. coli (Erdem et al., 2007). Indeed, a mutation of the flagellum element fliC prevents the adhesion of EPEC and EHEC to mucins (Erdem et al., 2007). More recently, the flagella of EPEC (O127:H6) and EHEC (O157:H7) were shown to adhere to mucin-type core 2 O-glycan in MUC2. C. jejuni is another pathogen that uses its flagella to bind mucin. It was showed that the major flagella subunit protein (FlaA) is also involved in the adhesion to HBGA in the mucus. Therefore, flagella can be used in attachment strategies by gut residents.

## BACTERIAL BIOFILM AND MUCUS

There are more mucus-associated bacteria in the proximal region of the colon than in distal colonic sites. Among the complex microbial communities within the gut, some are believed to form mucosal biofilm, that is a complex and self-produced polymeric matrix where microorganisms can attach to each other and be attached to the mucosal surface (de Vos, 2015). The rapid growth of the intestinal mucus and the lack of effective preservation techniques complicated the study investigating biofilms in healthy individuals (Bollinger et al., 2007; de Vos, 2015). However, biofilms were observed in artificial mucin gels that simulate the proximal and distal colon (Macfarlane et al., 2005), and also by electron microscopy in uninflamed proximal large bowel of mice (Swidsinski et al., 2005a), rat, baboon, and humans (Palestrant et al., 2004). Some evidence, such as the rates of plasmids transfer and the expression of colonization factors by gut bacteria, plead for the presence of biofilms in the gut (Macfarlane et al., 1997; Licht et al., 1999; Hooper and Gordon, 2001). In addition, components of the mucus layer, such as secretory IgA (SIgA) and mucins are likely to play a role in biofilm formation as they have been shown to modulate biofilm production in vitro (Bollinger et al., 2003, 2006; Slizova et al., 2015). Moreover, adherence of bacteria to mucin proteins could lead to growth of microcolonies that could further develop into biofilms (Kleessen and Blaut, 2007). Biofilms could also be formed on the surface of intestinal or gastric epithelia and interact with the secreted or membrane-bound mucins.

Alteration of the mucus layer occurs in cases of inflammatory bowel diseases (Bodger et al., 2006; Baumgart et al., 2007; Sheng et al., 2012). The increased presence of B. fragilis group and Enterobacteriaceae and their ability to form biofilms could play a role in these diseases (Swidsinski et al., 2005b, 2009). Within the Enterobacteriaceae family, the adherent-invasive E. coli (AIEC) strains associated with Crohn's disease (Masseret et al., 2001; Darfeuille-Michaud et al., 2004; Eaves-Pyles et al., 2008; Martinez-Medina et al., 2009a), are shown to be higher biofilm producers than non-AIEC strains (Martinez-Medina et al., 2009b). As with inflammatory bowel diseases, impaired mucin production is related to colorectal cancer (Weiss et al., 1996; Kim and Ho, 2010) that is also linked to the presence of bacterial biofilms (Dejea et al., 2014). Altogether, these studies show that biofilms could play a key role in bacterial colonization of the healthy gut and in intestinal diseases.

## ROLE OF MUCIN COMPONENTS IN MODULATION OF BACTERIAL VIRULENCE

In addition to acting as a carbon source or as receptors, mucin glycoprotein can influence the expression of different genes implicated in colonization and pathogenicity (Vogt et al., 2015). As example, MUC2 in the mucus layer can play a modulatory role in the pathogenesis of pathogens. Indeed, the ability of S. enterica serotype Typhimurium to cause cecal pathology in muc2−/<sup>−</sup> mice is more dependent on its invA gene, coding a Salmonella inner membrane protein component of the SPI-1 type

3 secretion system, than it is in wild-type mice (Zarepour et al., 2013). C. jejuni can also utilize mucin proteins as a signal to modulate the expression of its virulence factors. Many virulence genes of this pathogen are upregulated in the presence of MUC2 glycoprotein (Tu et al., 2008). Another example is the ability of V. cholerae to downregulate the expression of vps, coding for its polysaccharide, in response to mucosal signaling and inversely promoting motility in the mucus (Liu et al., 2015). Mucin also activates the two-component sensor histidine kinase ChiS in V. cholera. ChiS is the regulator of the chitinases and the chitin utilization pathway, but also plays a role in the virulence of the bacteria since the mutant strain is hypovirulent (Chourashi et al., 2016). Released monosaccharides from mucin O-glycans degradation can also act as a chemical cue to help pathogens to sense their environment and adapt accordingly. As such, sialic acid and GlcNAc are signals that regulate type 1 fimbriae gene expression and curli activity in E. coli (Barnhart et al., 2006; Konopka, 2012). GlcNAc and sialic acid also play roles in the virulence of EHEC. In aerobic condition, these mucinderived sugars inhibit EHEC adhesion to epithelial cells. These amino sugars also repress the expression of genes of the locus

TABLE 2 | Effects of bacterial effectors on mucin.

of enterocyte effacement (LEE) via the transcriptional regulator NagC involved in the regulation of NAG catabolism (Le Bihan et al., 2017). In contrast, as the sole carbon sources under microaerobic conditions, sialic acid and NAG were shown to stimulate the production of EspB, an effector of the LEE (Carlson-Banning and Sperandio, 2016). EHEC and C. rodentium also sense fucose by a two-component system FusKR. It represses the expression of virulence genes while promoting growth (Pacheco et al., 2012; Keeney and Finlay, 2013). Moreover, it was also shown that fucose influences chemotaxis and biofilm formation of C. jejuni that are important during infection (Dwivedi et al., 2016). Thus, mucus and its derived sugars can play a role in the expression of virulence genes by pathogens.

#### MODULATION OF MUCIN COMPOSITION BY BACTERIA

Microbial molecular exchange with the host influences mucin composition. Several bacterial effectors can modulate the expression of mucin by mucus-producing cells (**Table 2**). Studies


using germ-free rats revealed that the presence of microflora through the gastro intestinal tract has a strong and positive influence on the thickness and composition of the mucin (Szentkuti et al., 1990; Enss et al., 1992; Sharma et al., 1995). Different probiotic agents, such as Lactobacillus species, can stimulate the production of MUC2 and thereby the secretion of mucin in the intestine, improving pathogen resistance (Mack et al., 1999; Mattar et al., 2002; Caballero-Franco et al., 2007). Other commensal bacteria, such as B. thetaiotaomicron can increase the differentiation of goblet cells and their mucus-related gene expression (Wrzosek et al., 2013). Moreover, bacterial fermentation products, such as short-chain fatty acids (SCFAs) like butyrate and propionate enhance the production of MUC2 by the goblet cell in the gut (Barcelo et al., 2000; Burger-van Paassen et al., 2009). This could explain the therapeutic effect of butyrate in colitis where the mucin layer is altered (Finnie et al., 1995). Therefore, commensal residents are important in the maintenance of the mucus layer integrity.

#### Modulation of Mucin by Pathogens

Pathogens have also adapted mechanisms to modulate mucin secretion to enhance pathogenesis by acting on the mucinsecreting cells, altering or inhibiting mucin production (**Table 2**). One of them is S. flexneri that alters the mucus layer through a type III secretion system-dependent manner. This pathogen will act on different elements, such as gene expression, mucin glycosylation and secretion, leading to a less effective mucus barrier (Sperandio et al., 2013). C. difficile produces a toxin, ToxA that is responsible for barrier dysfunction and causes severe inflammatory enteritis. ToxA also decreases the mucin exocytosis of colonic mucus-producing cells (Kelly et al., 1994; Branka et al., 1997). The recognition of bacterial components by these cells can also lead to an increased production and secretion of mucin in order to harm the present pathogen. As example, the adhesion of the EHEC O157:H7 to human colon cells HT-29 leads to an increased expression of MUC2 (Xue et al., 2014). Moreover, the cholera toxin of V. cholerae and lysteriolysin O of L. monocytogenes enhance the secretion of mucin by goblet cells and HT29-MTX cells, respectively (Lencer et al., 1990; Epple et al., 1997; Coconnier et al., 1998; Lievin-Le Moal et al., 2002, 2005). Surprisingly, the Pic protein secreted by S. flexneri and enteroaggregative E. coli (Henderson et al., 1999; Harrington et al., 2009) is known for its mucolytic activity, but is also a potent mucus secretagogue that induced hypersecretion of mucus by goblet cells (Navarro-Garcia et al., 2010). These studies show how pathogens can affect the behavior of mucus-producing cells in their advantage.

## Mucin Degradation by Pathogens

Pathogens also developed specific mechanisms to subvert and penetrate the mucus barrier. Some bacteria can directly act on the mucin through a mucinase activity. During enterotoxigenic E. coli infections, the autotransporter A (EatA) is involve in mucin degradation and this participate to the delivery of E. coli toxins to the cell surface (Kumar et al., 2014). Another example is the adherent and invasive E. coli strain LF82, associated with Crohn's disease. LF82 possesses a protease called Vat-AIEC that is implicated in the degradation of mucins and therefore decreases mucus viscosity (Gibold et al., 2016). The Pic autotransporter found in enteroaggregative E. coli and Shigella flexneri can also degrade various glycoproteins including mucins (Henderson et al., 1999; Harrington et al., 2009). Moreover, the plasmid-bearing Yersinia enterocolitica, which contain mucindegrading enzyme(s), will increase the permeability of the mucus gel layer, allowing the bacteria to move more easily through the mucin (Mantle and Rombough, 1993). V. cholerae also produces a secreted protease called TagA that is encoded by the Vibrio pathogenicity island (VPI). TagA specifically cleaves mucin glycoproteins and may directly modify host cell surface molecules during V. cholerae infection (Szabady et al., 2011). Therefore, to facilitate their infection process, pathogens can directly modify the mucus.

### Inflammation and Mucins

Pathogens associated molecular patterns, such as lipopolysaccharide (LPS) and peptidoglycan are also known to stimulate mucin production (Petersson et al., 2011). This stimulation can occur directly on secreting cells, but also be through proinflammatory cytokine production. Recognition of LPS by LPS-binding protein (LBP), CD14, and TLR4 (Toll-Like Receptor) leads to a strong pro-inflammatory response in mammalian cells. LPS has been shown to induce mucin gene expression by binding to TLR4 and LBP (Dohrman et al., 1998; Smirnova et al., 2003). LPS and flagellin from Gram-negative bacteria as well as lipoteichoic acid, a component of the cell wall of Gram-positive bacteria, induce mucin upregulation through the Ras pathway (McNamara and Basbaum, 2001; Theodoropoulos and Carraway, 2007). LPS also increases the production of IL-8 by goblet cells, which leads to secretion of mucin (Smirnova et al., 2003). In addition, pro-inflammatory cytokine IL-6 and TNF-α increase secretion of MUC2, MUC5A, MUC5B, and MUC6 by the intestinal cell line LS180 despite a reduced glycosylation (Enss et al., 2000). Inflammation could be one of the aspects affecting the integrity of the mucus layer in inflammatory bowel diseases. Furthermore, the AIEC strain LF82 is able to alter the expression of the mucin gene and IL-8 of colonic cells T84 that could also lead to a defective mucus layer (Elatrech et al., 2015). Thus, pathogens can also alter the mucus production indirectly, through inflammation.

## CONCLUSION

Intestinal bacteria have adapted to colonize the mucus layer by adhering to intestinal mucus components, using mucusderived nutrients and sensing chemical cues for adaptation. In many ways, pathogenic bacteria have used these strategies for successful infection. There has been growing recognition of the important role played by the mucus barrier and microbiota and their interaction with the pathogens in regulating the severity of infectious diseases. But, the precise mechanisms by which enteric bacterial pathogens interact with mucus components in combination with the microbiota activity are being investigated. As the mucus layer acts as a first line of defense against enteric bacteria, further investigations are needed to understand the interactions between pathogens, microbiota and the mucus layer, in order to develop efficient therapeutic strategies. Identifying and characterizing specific mucin signal(s) and corresponding regulatory adaptation and virulence responses could contribute to the development of new anti-infective strategies. In doing so, other weapons could be added to the arsenal against intestinal pathogens.

## AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication. The manuscript was written by J-FS and JH and was duly revised by GLB, PV and MJ.

#### REFERENCES


### ACKNOWLEDGMENTS

We thank Judith Kashul for editing the manuscript. This research was supported by a Team grant from the Fonds de Recherche du Québec, Nature et Technologies (FRQNT PT165375), to JH and MJ and by the Discovery grant program of the Natural Sciences and Engineering Research Council of Canada (RGPIN-2015- 05373 to JH and RGPIN-2016-04203 to MJ). J-FS is a recipient of a scholarship from the NSERC Collaborative Research and Training Experience Program in Milk Quality; and PV is a recipient of a scholarship from the FRQNT Québec Wallonie program.

monozygotic twins of inflammatory bowel disease patients. Gut 55, 973–977. doi: 10.1136/gut.2005.086413


the cell-entry of Listeria monocytogenes. Cell. Microbiol. 7, 1035–1048. doi: 10.1111/j.1462-5822.2005.00532.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 Sicard, Le Bihan, Vogeleer, Jacques and Harel. 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.

# Alterations of the Gut Microbiome in Hypertension

Qiulong Yan1, 2, Yifang Gu<sup>3</sup> , Xiangchun Li <sup>4</sup> , Wei Yang<sup>5</sup> , Liqiu Jia<sup>1</sup> , Changming Chen<sup>1</sup> , Xiuyan Han<sup>1</sup> , Yukun Huang<sup>1</sup> , Lizhe Zhao<sup>1</sup> , Peng Li <sup>3</sup> , Zhiwei Fang<sup>3</sup> , Junpeng Zhou<sup>3</sup> , Xiuru Guan<sup>5</sup> , Yanchun Ding<sup>6</sup> , Shaopeng Wang<sup>7</sup> , Muhammad Khan<sup>8</sup> , Yi Xin<sup>9</sup> , Shenghui Li <sup>3</sup> \* and Yufang Ma<sup>1</sup> \*

*<sup>1</sup> Department of Biochemistry and Molecular Biology, Dalian Medical University, Dalian, China, <sup>2</sup> Department of Microbiology, Dalian Medical University, Dalian, China, <sup>3</sup> Shenzhen Puensum Genetech Institute, Shenzhen, China, <sup>4</sup> Beijing Genomics Institute, Shenzhen, China, <sup>5</sup> Department of Laboratory Diagnostics, The First Affiliated Hospital of Harbin Medical University, Harbin, China, <sup>6</sup> Department of Cardiology V, The Second Affiliated Hospital of Dalian Medical University, Dalian, China, <sup>7</sup> Department of Cardiology, The First Affiliated Hospital of Dalian Medical University, Dalian, China, <sup>8</sup> College of Basic Medical Sciences, Dalian Medical University, Dalian, China, <sup>9</sup> Department of Biotechnology, Dalian Medical University, Dalian, China*

Introduction: Human gut microbiota is believed to be directly or indirectly involved in cardiovascular diseases and hypertension. However, the identification and functional status of the hypertension-related gut microbe(s) have not yet been surveyed in a comprehensive manner.

Methods: Here we characterized the gut microbiome in hypertension status by comparing fecal samples of 60 patients with primary hypertension and 60 gender-, age-, and body weight-matched healthy controls based on whole-metagenome shotgun sequencing.

#### Edited by:

*Venkatakrishna Rao Jala, University of Louisville, United States*

#### Reviewed by:

*Bina Joe, University of Toledo, United States Morgan Langille, Dalhousie University, Canada*

#### \*Correspondence:

*Shenghui Li lishenghui@puensum.com Yufang Ma yufang\_ma@hotmail.com*

Received: *16 May 2017* Accepted: *09 August 2017* Published: *24 August 2017*

#### Citation:

*Yan Q, Gu Y, Li X, Yang W, Jia L, Chen C, Han X, Huang Y, Zhao L, Li P, Fang Z, Zhou J, Guan X, Ding Y, Wang S, Khan M, Xin Y, Li S and Ma Y (2017) Alterations of the Gut Microbiome in Hypertension. Front. Cell. Infect. Microbiol. 7:381. doi: 10.3389/fcimb.2017.00381* Results: Hypertension implicated a remarkable gut dysbiosis with significant reduction in within-sample diversity and shift in microbial composition. Metagenome-wide association study (MGWAS) revealed 53,953 microbial genes that differ in distribution between the patients and healthy controls (false discovery rate, 0.05) and can be grouped into 68 clusters representing bacterial species. Opportunistic pathogenic taxa, such as, *Klebsiella* spp., *Streptococcus* spp., and *Parabacteroides merdae* were frequently distributed in hypertensive gut microbiome, whereas the short-chain fatty acid producer, such as, *Roseburia* spp. and *Faecalibacterium prausnitzii*, were higher in controls. The number of hypertension-associated species also showed stronger correlation to the severity of disease. Functionally, the hypertensive gut microbiome exhibited higher membrane transport, lipopolysaccharide biosynthesis and steroid degradation, while in controls the metabolism of amino acid, cofactors and vitamins was found to be higher. We further provided the microbial markers for disease discrimination and achieved an area under the receiver operator characteristic curve (AUC) of 0.78, demonstrating the potential of gut microbiota in prediction of hypertension.

Conclusion: These findings represent specific alterations in microbial diversity, genes, species and functions of the hypertensive gut microbiome. Further studies on the causality relationship between hypertension and gut microbiota will offer new prospects for treating and preventing the hypertension and its associated diseases.

Keywords: hypertension, gut microbiome, microbial dysbiosis, metagenome-wide association study

### INTRODUCTION

Hypertension is a global public health problem. In 2010, about 31% of the world's population has been estimated to suffer from hypertension and over 1 billon of this population is living in low- and middle- income countries(Mittal and Singh, 2010; Mills et al., 2016). Hypertension is one of the major risk factors for cardiovascular diseases, such as, stroke and heart failure (Lim et al., 2012; Faraco and Iadecola, 2013). Moreover, it is believed to be one of the most common comorbidities associated with chronic renal disease (Lash et al., 2009), obesity and type 2 diabetes (Landsberg and Molitch, 2004; Kotchen, 2010). Presently, genome-wide association studies (GWAS) have identified a series of genetic loci and pathways associated with blood pressure (Xu et al., 2015; Liu et al., 2016).The environmental factors, such as, dietary salt intake, alcohol consumption and lack of exercise, are also linked to the occurrence of hypertension (Fuchs et al., 2001; Karppanen and Mervaala, 2006). Recent practice of metabolomics also identified new pathogenic pathways involved in blood pressure regulation (Menni et al., 2015; Galla et al., 2017). Nevertheless, due to the complexity and heterogeneity of hypertension, identification of the causes of this disease is still challenging.

Recent studies have demonstrated that the gut microflora plays an essential role in development of cardiovascular diseases, via metabolizing dietary choline, phosphatidylcholine and L-carnitine to produce trimethylamine (TMA), which is further oxidized into TMA N-oxide (TMAO, a metabolite that enhances atherosclerosis; Wang et al., 2011; Koeth et al., 2013; Tang et al., 2013). Even though the direct link between hypertension and TMAO has not been established currently, TMAO's role to prolong the hypertensive effect of angiotensin II were reported (Ufnal et al., 2014). Inhibition of gut microbiota-mediated TMAO production may serve as a potential therapeutic approach for the treatment of cardiometabolic diseases (Wang Z. et al., 2015). These findings suggest an intricate and predictable correlation between hypertension and gut microbiota. To validate this, a recent study based on metagenomic analyses of the fecal samples of 41 healthy controls, 56 pre-hypertension subjects, and 99 hypertension individuals described a novel causal role of aberrant gut microbiota in contributing to the pathogenesis of hypertension, and emphasized the significance of early intervention for pre-hypertension (Li et al., 2017). Moreover, rat experiments have linked gut microbial dysbiosis with hypertension (Mell et al., 2015; Yang et al., 2015; Adnan et al., 2017; Santisteban et al., 2017). The causal role of gut microbiome in obstructive sleep apnea-induced hypertension have been reported (Durgan et al., 2016). Here, to investigate the alteration of the human gut microbiome underlying hypertension, we compared the microbial composition of fecal samples obtained from 60 patients with primary hypertension and 60 healthy counterparts of Chinese origin. We used quantitative metagenomic analysis to identify genic, microbial, and functional characteristics underlying hypertension.

## METHODS

#### Subjects and Sample Collection

Sixty primary hypertensive patients (current blood pressure ≥140/90 mm Hg) and sixty gender-, age-, and body weightmatched healthy controls (current blood pressure ≤ 120/80 mm Hg) were recruited for this study. Other than systolic blood pressure (SBP) and diastolic blood pressure (DBP), the other clinical parameters have no significant differences in the two groups of populations, except for triglyceride (TG). The characteristics of the subjects are summarized in **Table 1**, and detailed information is given in Table S1. Subjects were excluded if they had symptoms of respiratory infection or digestive tract disease, or if they were treated with antibiotics or anti-inflammatory agents in recent 2 months before sampling. Subjects with hypertension or severe cardiovascular diseases (such as, coronary artery disease or stroke) history in previous 5 years were also excluded from healthy controls. Fresh fecal samples were collected from each subject and were stored at a −80◦C freezer immediately.

#### Ethics Statement

This study received approval from the ethics committee of The First Affiliated Hospital of Harbin Medical University, and written informed consent was obtained from each participant. The methods were carried out in accordance with the approved guidelines.

#### DNA Preparation and Sequencing

Genomic DNA was extracted from all samples according to a modified protocol provided in the QIAamp DNA mini kit (Qiagen, Manchester, UK; Yan et al., 2016). Briefly, ASL buffer (1.4 ml) was added to 220 mg of fecal sample and the pellets were homogenized in a 2 ml screw cap tubes

#### TABLE 1 | Characteristics of subjects.


*The data for age, BMI, SBP, DBP, FGB, HDL, LDL, TG, and TC were presented as mean* ± *SD. P-values for gender and smoke were calculated by Fisher's exact test. P-values for age, BMI, SBP, DBP, FGB, HDL, LDL, TG, and TC were calculated using Student's t-test. BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; HDL, high density lipoprotein; LDL, low density lipoprotein; TG, total triglyceride; and TG, total cholesterol.*

(Axygen) by vortex. The suspension was incubated at 95◦C for 5 min to lyse bacterial cells. After centrifugation (13,000 × g, 1 min) and incubation with an InhibitEx tablet, the supernatant was treated with 15 µl proteinase K and 200 µl Buffer AL at 70◦C for 10 min. The extracted DNA was dissolved in 100 µl sterile water. Paired-end DNA libraries (insert size 350 bp, read length 150 bp) were constructed according to the manufacturer's instructions (Illumina, USA). Wholemetagenome shotgun sequencing was performed on the Illumina HiSeq3000 platform. Further methodological detail is available in the Supplementary Methods.

### Bioinformatic Analysis

#### Quantification of Metagenomic Genes and Species

High-quality reads from each sample were aligned to the integrated non-redundant human gut gene catalog (IGC; Li et al., 2014) using SOAP2 (Li et al., 2009; >90% similarity). The relative abundance of a gene in a sample was estimated by dividing the number of reads uniquely mapped to that gene by the length of gene region and by the total number of reads from the sample. We also aligned the sequencing reads against the available microbial genomes (bacteria, archaea, and virus) from the National Center for Biotechnology Information (NCBI) database and generated the taxonomic compositions (i.e., phylum and genus composition) for all samples.

#### Alpha Diversity

The gene count (Le Chatelier et al., 2013) of a metagenomic sample were calculated based on their mapped reads number on the gene catalog (to eliminate the influence of sequencing amount fluctuation, 10 million reads were randomly extracted from each sample for mapping). The Shannon index (within-sample diversity) was calculated based on the gene relative abundance profiles, using the method described previously (Qin et al., 2012).

#### Functional Annotation and Profiling

The Kyoto Encyclopedia of Genes and Genomes (KEGG, downloaded Jan-2016) database (Kanehisa et al., 2014) was used for functional annotation of genes. Amino acid sequences were searched against the databases using USEARCH v8.1 (Edgar, 2010) with a minimum similarity of 30%. Each gene was assigned a KEGG ortholog (KO) based on the best hit protein. The abundance profiles of KO were calculated by summing the relative abundance of its genes. The choline-trimethylamine lyase (cutC, KO: K20038; Craciun and Balskus, 2012) was used to evaluate the gut microbiota-mediated TMA production in subjects, and the short-chain fatty acid (SCFA)-producing enzymes were represented by acetyl-CoA decarbonylase/synthase (K00193, K00194, K00197, K14138, which are key enzymes of the acetate biosynthesis pathways: KEGG modules M00377 and M00422; Koh et al., 2016), propionyl-CoA:succinate-CoA transferase (Reichardt et al., 2014), butyryl-CoA:acetate CoAtransferase (K01034, K01035), and butyrate kinase (K00929) (Pryde et al., 2002; Louis et al., 2010).

#### Metagenome-Wide Association Study

We used the metagenome-wide association study (MGWAS) method to identify gene markers that showed significant abundance differences between hypertensive patients and control subjects. The MGWAS was performed using the methodology developed by Qin et al. (2012). Co-abundance genes were clustered into metagenomic linkage groups (MLGs) based on the previous methods (Qin et al., 2012). Taxonomic assignment and abundance profiling of the MLGs were performed according to the taxonomy and the relative abundance of their constituent genes (see Supplementary Methods for detail). MLGs were considered to be interacted if absolute value of Spearman's correlation coefficient between them is greater than 0.4, and the co-occurrence network of MLGs was visualized by Cytoscape (Shannon et al., 2003).

#### Statistical Analyses

Statistical analyses were implemented using the R platform. Distance-based redundancy analysis (dbRDA) was performed on normalized taxa abundance matrices with R vegan package (Dixon, 2003) according to Bray-Curtis distance, then visualized with R ggplot2 package. Random forest models were trained with R randomForest package (10,000 trees) to predict hypertension status according to MLG abundance profiles. The performance of the predictive model was evaluated with cross-validation error. Receiver operator characteristic (ROC) analysis was performed using R pROC package. P-value < 0.05 was considered statistical significance, and the q-value was calculated to evaluate the false discovery rate (FDR) for correction of multiple comparisons.

#### Data Availability

The raw whole-metagenomic shotgun sequencing data acquired in this study have been deposited to the European Bioinformatics Institute (EBI) database under the accession code ERP023883.

## RESULTS

#### Comparison of the Gut Microbiota between Hypertensive Patients and Controls

To investigate the gut microbial composition of 60 hypertensive patients and 60 healthy controls, we obtained 652.9 Gbp high-quality data (5.4 ± 1.1 Gbp per sample) via wholemetagenome shotgun sequencing on their fecal samples. When we quantified the microbial (alpha) diversity within each subject, the patients showed significantly lower gene count and Shannon index compared with the controls (**Figure 1A**). Multivariate analysis based on Bray-Curtis distance between microbial genera revealed remarkable differences between patients and controls (**Figure 1B**). At the phylum level, patients had higher levels of Proteobacteria (p < 0.01), but fewer Actinobacteria (p = 0.02). At the genus level, Klebsiella, Clostridium, Streptococcus, Parabacteroides, Eggerthella, and Salmonella were frequently distributed in hypertensive gut compared to normal controls while Faecalibacterium, Roseburia, and Synergistetes were found to be higher in control group compared to hypertensive patients (**Figure 1C**). These findings demonstrated considerable gut microbial dysbiosis in hypertensive patients.

### Identification of Hypertension-Associated Markers from Gut Microbiome

To explore signatures of the gut microbiome in hypertensive patients and controls, we integrated the sequencing data into an existing gut microbial reference gene catalog to obtain a set of 5.3 million genes, which allowed for saturation mapping of the reads (80.3%). Using the MGWAS methods, we identified 53,953 genes that showed a significant difference between two groups (FDR corrected q < 0.05). Approximately, 69% of these genes were clustered into 68 metagenome linkage groups (MLGs, Table S2), that allowed to species level description for the microbiome differences. Thirty-one MLGs were higher in patients while 37 in controls. Consistent with the genus level observations, MLGs of Klebsiella (mainly consisting of K. pneumoniae and K. variicola), Streptococcus (S. infantarius, S. pasteurianus and S. salivarius), and Parabacteroides merdae were found to be higher in hypertensive samples, whereas MLGs of Roseburia (mainly consisting of R. intestinalis and R. hominis) and Faecalibacterium prausnitzii were higher in controls. Moreover, the MLGs enriched in hypertensive patients also contain several Bacteroides spp. (including B. eggerthii and B. cellulosilyticus), Sutterella wadsworthensis and Pyramidobacter piscolens, and the MLGs enriched in controls include several other Bacteroides spp. (including B. uniformis, B. nordii and B. dorei), Megasphaera spp. (M. micronuciformis), and Aeromicrobium massiliense. A co-occurrence network on these MLGs revealed a large number of interconnections within hypertension-enriched and controlenriched MLGs (**Figure 2A**), as well as some MLGs derived from two groups negatively correlated. This result suggested that the MLGs did not occur independently and interacted with the taxa in its environment.

We next found that the gross abundances of hypertensionand control-enriched MLGs are correlated to the severity of hypertension (**Figure 2B**), suggesting that the bacterial relative abundance of these MLGs could be related to the development and disease progress of hypertension.

## Functional Characterization of Gut Microbiota

Based on the KEGG pathway comparison, we revealed that the hypertensive gut microbiomes were more abundant in membrane transport, lipopolysaccharide (LPS) biosynthesis, and steroid degradation (**Figure 3A** and Table S3), while the controls were enriched in metabolism of "other amino acids," cofactors and vitamins (including folate biosynthesis and metabolism, riboflavin metabolism, and ubiquinone biosynthesis). In addition, the gut microbial enzymes involved in TMA production were enriched in the hypertensive patients compared to controls, whereas the SCFA-producing enzymes were depleted (**Figure 3B**).

#### Gut Microbiota-Based Classification of Hypertension

We evaluated the performance of gut microbiota composition to identify hypertension status in the MLG profiles using the Random Forest model, and obtained the discriminatory power of the area under the ROC curve (AUC) of 0.78 (95% CI 0.73–0.82; **Figure 4A**). Several control-enriched MLGs (including Clostridiales, Blautia hansenii, Megasphaera) and two hypertension-enriched members of Streptococcus (S. salivarius and S. infantarius) featured the highest score for the discrimination of hypertensive patients and healthy controls (**Figure 4B**).

## DISCUSSION

To identify and analyze the differences of the gut microbiota in hypertension, we characterized the genic, microbial, and functional repertoire of the microbiomes of 60 hypertensive patients and 60 gender-, age-, and body weight-matched controls. Our study strengthened previous metagenomic study on gut microbiome of hypertension (Li et al., 2017) by adding more information. Furthermore, we observed significant differences in microbial community dysbiosis, taxonomic shifts, and functional changes between hypertensive- and control-gut microbiome.

Previous studies showed that the gut microbes participate in choline and phosphatidylcholine metabolism to form circulating and urinary TMAO, while high levels of plasma TMAO promote accelerated atherosclerosis and increase the risk of cardiovascular disorders (Tang et al., 2013; Wang Z. et al., 2015). The choline utilization (cutC) gene, a critical gene that coverts the choline to trimethylamine, was identified in a variety of human gut commensals belonging to Firmicutes, Proteobacteria, and Actinobacteria (Craciun and Balskus, 2012). Notably, several genera such as, Klebsiella, Clostridium, and Streptococcus, which are highly distributed in hypertensive patients are choline degraders (Hakenbeck et al., 2009; Craciun and Balskus, 2012; Kalnins et al., 2015). Functional analysis also showed that the abundance of cutC gene was enriched in the gut microbiota of the hypertensive patients. These findings suggested that the dietary choline intake and TMAO production via gut microflora would be a probable pathway for hypertensive pathogenesis.

Klebsiella, is a pathogen routinely found in human gut that causes pneumonia, diarrhea, and urinary tract infection. The distribution of Klebsiella was found to be significantly higher in hypertensive patients compared to healthy controls as evident from **Figure 1C**. Overgrowth of Klebsiella usually foreshadows gut flora dysbiosis, which leads to a variety of serious chronic disease, such as, colitis (Garrett et al., 2010), Crohn's disease and ankylosing spondylitis (Ebringer et al., 2007).The present study demonstrates that Klebsiella species which are highly distributed in hypertensive patients are K. pneumoniae (the main component of Klebsiella that associated with nosocomial infection and multiple diseases), K. variicola [a human and animal opportunistic pathogen that is associated with bovine mastitis (Brisse and Duijkeren, 2005)], and four unclassified MLGs. Based on these information, however, the potential correlation between Klebsiella and hypertension is still unclear.

Streptococcus, the dominant species of human oral microbiome (Wade, 2013) that causes upper respiratory tract infection, were also found highly distributed in gut microbiota of hypertensive patients as compared to the controls. Gut streptococci is also associated with diseases, such as, inflammatory bowel disease (Conte et al., 2006) and liver cirrhosis (Qin et al., 2014). It has been reported previously that oral cavity and/or gut might be the source of streptococci found in the majority of atherosclerotic plaque microbiota (Koren et al., 2011). These findings suggest that possible correlation of gut streptococci in hypertension.

F. prausnitzii and Roseburia spp., which were abundantly distributed in controls compared to hypertensive patients, were also distributed abundantly in the healthy control microbiomes of many chronic diseases, including type 2 diabetes (Qin et al., 2012), liver cirrhosis (Qin et al., 2014), Crohn's disease (Gevers et al., 2014), and ulcerative colitis (Machiels et al., 2014). F. prausnitzii and Roseburia (both R. intestinalis and R. hominis) are the major SCFA producer in human colon (Shoaie et al., 2015), which might explain the depletion of SCFA-producing enzymes in hypertensive gut microbiome. Functionally, SCFAs modulates the gut inflammation and metabolism via functioning as important colonocytes energy source and signaling molecules (Donohoe et al., 2011), suggesting that low level of SCFA production in gut microbiota may be a considerable risk factor of multiple metabolic syndromes and hypertension.

stage. NS, not significant; \*, *q* < 0.05; \*\*, *q* < 0.01; Mann-Whitney *U*-test corrected by FDR.

Several other bacteria also played important function in human gut and showed potential function in hypertension, such as, the patient-enriched Bacteroides (including B. eggerthii, B. cellulosilyticus, and 3 unclassified Bacteroides MLGs) and Parabacteroides (P. merdae) which are generally opportunistic pathogens in infectious diseases and are able to develop antimicrobial drug resistance (Boente et al., 2010), and the control-enriched Megasphaera spp. (M. micronuciformis and two unclassified MLGs) which are producer of SCFAs, vitamins and essential amino acids (Shetty et al., 2013). In addition, coabundance analysis (**Figure 2A**) generated a striking number of positive correlations within the patient/control-enriched MLGs and negative correlations between the two groups, revealing that a comprehensive bacterial synergism and antagonism existed in the human gut. In this case, the microbial dysbiosis of hypertensive gut microbiome would not be determined by independent pathogens (e.g., the patient-enriched MLGs), but more likely to be caused by a series of risk factors (e.g., improper diet or lifestyle that inhibit the growth of beneficial bacterium) that change the balance of ecosystem. Intriguingly, the severity of hypertension was positively correlated with the total abundance of patient-enriched MLGs and negatively correlated with those of control-enriched MLGs (**Figure 2B**), suggesting that the bacterial relative abundance may also be a potential risk factor of hypertension development. Such a "dose response" was also found in the gut microbiome of liver cirrhosis (Qin et al., 2014) and colorectal adenoma-carcinoma patients (Feng et al., 2015).

Our study further provided the microbial markers for hypertension discrimination, and achieved an AUC of 0.78 for identifying disease status based on 68 species-level MLGs. This discriminatory power was higher than that from the prediction models based on genomic markers identified by GWAS (Evans

FIGURE 3 | Functional comparison of the gut microbiomes between hypertensive patients and healthy controls. (A), Distributions of relative abundances of KEGG pathway categories in hypertensive patients and controls. \*, *q* < 0.05; \*\*, *q* < 0.01; Mann-Whitney *U*-test corrected by FDR. (B), Difference of the relative abundance of *cutC* (TMA-producing) and SCFA-producing enzymes between hypertensive (HT) patients and controls.

colors represent enrichment in patients (black) or controls (white).

et al., 2009; Fava et al., 2013), and was almost at same level with the phenotype-based models (AUC 0.71-0.81) (Echouffo-Tcheugui et al., 2013; Wang A. et al., 2015). Thus, the fecal microbiota showed a good potential on prediction and early diagnosis of hypertension, however, systematic investigations of key species and gene markers identified here might be helpful in the future.

Drug-induced gut microbiome shifts were observed during the treatment of multiple diseases, such as, the metformin therapy in type 2 diabetes (Forslund et al., 2015) and antirheumatic drugs therapy in rheumatoid arthritis (Zhang et al., 2015). In this study, a part of patients (∼35%) had taken antihypertensive drugs or specific nutritious supplementary, however, the relationship between drug treatment and gut microbiota is still unclear. Another significant limitation of this study is that the gut microbial community would be sensitive to environmental factors, such as, host race, geography, life and diet style, and so on. Although our samples were age-, gender-, BMI-matched, some phenotype differences were still unobservable. To avoid this, larger cohort containing multi-types of hypertensive patients are needed for further investigation. Generally, hypertension is a highly complex and heterogeneous disease, it is still infeasible at this moment to draw any conclusions about causal relationships of gut microbiota and hypertension, and direct experimental studies (e.g., the animal model studies) are needed to show causality of proposed microbes or pathways.

In summary, our finding extends previous knowledge of correlation between gut microbiota and hypertension in animal models (Yang et al., 2015; Durgan et al., 2016) and provides a range of signatures in metagenomic diversity, genes, species, and functions of the hypertensive gut microbiome. Further studies on the causality relationship between hypertension and gut microbiota will lead to a better understanding of the mutual interaction.

#### AUTHOR CONTRIBUTIONS

YM, SL, and QY designed experiments; QY, LJ, CC, XH, YH, LZ, PL, and MK carried out experiments; QY, WY,

#### REFERENCES


YD, SW, and YX analyzed experimental results. SL, YG, XL, ZF, and JZ analyzed sequencing data. WY, XG, YD, and SW collected the samples. YM, SL, and QY wrote the manuscript.

#### FUNDING

This study was supported by grants from the National Naturel Science Foundation of China (81573469) and the National Basic Research Program of China (2012CB518803).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00381/full#supplementary-material

of hypertension by a genetic risk score in Swedes. Hypertension 61, 319–326. doi: 10.1161/HYPERTENSIONAHA.112.202655


**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 Yan, Gu, Li, Yang, Jia, Chen, Han, Huang, Zhao, Li, Fang, Zhou, Guan, Ding, Wang, Khan, Xin, Li and Ma. 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.

# Secretory Products of the Human GI Tract Microbiome and Their Potential Impact on Alzheimer's Disease (AD): Detection of Lipopolysaccharide (LPS) in AD Hippocampus

#### Yuhai Zhao1, 2, Vivian Jaber <sup>1</sup> and Walter J. Lukiw1, 3, 4 \*

 LSU Neuroscience Center, Louisiana State University Health Science Center, New Orleans, LA, United States, <sup>2</sup> Department of Anatomy and Cell Biology, Louisiana State University Health Science Center, New Orleans, LA, United States, Department of Ophthalmology, Louisiana State University Health Science Center, New Orleans, LA, United States, Department of Neurology, Louisiana State University Health Science Center, New Orleans, LA, United States

Although the potential contribution of the human gastrointestinal (GI) tract microbiome to human health, aging, and disease is becoming increasingly acknowledged, the molecular mechanics and signaling pathways of just how this is accomplished is not wellunderstood. Major bacterial species of the GI tract, such as the abundant Gram-negative bacilli Bacteroides fragilis (B. fragilis) and Escherichia coli (E. coli), secrete a remarkably complex array of pro-inflammatory neurotoxins which, when released from the confines of the healthy GI tract, are pathogenic and highly detrimental to the homeostatic function of neurons in the central nervous system (CNS). For the first time here we report the presence of bacterial lipopolysaccharide (LPS) in brain lysates from the hippocampus and superior temporal lobe neocortex of Alzheimer's disease (AD) brains. Mean LPS levels varied from two-fold increases in the neocortex to three-fold increases in the hippocampus, AD over age-matched controls, however some samples from advanced AD hippocampal cases exhibited up to a 26-fold increase in LPS over age-matched controls. This "Perspectives" paper will further highlight some very recent research on GI tract microbiome signaling to the human CNS, and will update current findings that implicate GI tract microbiome-derived LPS as an important internal contributor to inflammatory degeneration in the CNS.

Keywords: 42 amino acid amyloid-beta (Aβ42) peptide, Alzheimer's disease (AD), *Bacteriodetes fragilis* (*B. fragilis*), *Escherichia coli* (*E. coli*), lipopolysaccharide (LPS), microbiome, small non-coding RNAs (sncRNAs), thanatomicrobiome

## INTRODUCTION: THE HUMAN GI TRACT MICROBIOME

The human GI tract is fundamentally a highly vascularized and extensively innervated, columnar epithelial-cell lined tube about 9 m (30 feet) in length that consists of the stomach, small intestine (duodenum, jejunum, and ileum) and large intestine (cecum, colon, rectum, and anal canal; Reinus and Simon, 2014). Each anatomical region of this tubular structure harbors a complex and dynamic microbiome, containing ∼1,000 different species of anaerobic or facultative anaerobic

#### *Edited by:*

Michele Marie Kosiewicz, University of Louisville, United States

## *Reviewed by:*

Rebecca Drummond, National Institutes of Health, United States Valerio Iebba, Sapienza Università di Roma, Italy

> *\*Correspondence:* Walter J. Lukiw wlukiw@lsuhsc.edu

*Received:* 24 March 2017 *Accepted:* 27 June 2017 *Published:* 11 July 2017

#### *Citation:*

Zhao Y, Jaber V and Lukiw WJ (2017) Secretory Products of the Human GI Tract Microbiome and Their Potential Impact on Alzheimer's Disease (AD): Detection of Lipopolysaccharide (LPS) in AD Hippocampus. Front. Cell. Infect. Microbiol. 7:318. doi: 10.3389/fcimb.2017.00318 bacteria that appear to be characteristic for that GI tract segment. Indeed, the dynamism of the GI tract microbiome along its length is in part reflected by the abundance, speciation, complexity and stoichiometry of individual resident bacterial species. In addition to the major bacterial component of the GI tract are microbial eukaryotes, archaea, fungi, protozoa, viruses, and other commensal microorganisms which make up the remainder. Together with host cells these jointly comprise the complete metaorganism: (i) whose symbiotic associations and interactions are indispensable for homeostatic physiological functions in human health; and (ii) which exhibit alterations in composition in response to dietary factors, developmental stage, GI tract disturbances, aging, and neurological disorders, including AD (Bhattacharjee and Lukiw, 2013; Hill et al., 2014; Perez et al., 2014; Potgieter et al., 2015; Zhao and Lukiw, 2015; Alkasir et al., 2016; Ghaisas et al., 2016; Hu et al., 2016; Lukiw, 2016; Pistollato et al., 2016; Scheperjans, 2016).

## GI TRACT BACTERIAL MICROBIOME—EXUDATES AND SECRETORY PRODUCTS

Two large prokaryotic classes of Bacteria (or "Eubacteria") and Archaea (or "Archaeobacteria") have been recently reclassified (as of 10/2016) into 35 phyla (http://www.bacterio.net/-classifphyla.html) or major bacterial divisions. Interestingly the GI tract microbiome of Homo sapiens has co-evolved with just two major phyla: Bacteriodetes, which make up ∼20% of all GI tract bacteria, and Firmicutes, which make up ∼80% of all GI tract bacteria; with Actinobacteria (∼3%), Proteobacteria (∼1%), and Verrumicrobia (∼0.1%) making up significantly smaller fractions. These five bacterial groups appear to constitute the essential "core" of the human GI tract microbiome (http://www.bacterio.net/-classifphyla.html; Zhao et al., 2015; Hug et al., 2016; Lloyd-Price et al., 2016; Sender et al., 2016). The vast proportion of all GI tract microbiota consists of anaerobic or facultative anaerobic bacteria (Bhattacharjee and Lukiw, 2013; Heintz and Mair, 2014; Köhler et al., 2016; Lloyd-Price et al., 2016). For example, although variable, the obligate anaerobe Bacteroides fragilis (B. fragilis; phyla Bacteroidetes) and the facultative anaerobe Escherichia coli (E. coli; phyla Proteobacteria): (i) together constitute ∼35–40 percent of all GI tract bacteria; (ii) are the most abundant Gramnegative bacilli of the middle and lower colon, respectively, of the human GI tract; and (iii) constitute about ∼30–50 percent of the dry weight of fecal matter. B. fragilis or E. coli require about 20 min to divide under optimal conditions of commensal bacterial growth, and unless special biophysical processes of growth dynamics are in operation (such as dormancy, hibernation, spore formation, etc.) have a life span of up to several hours (Choi and Cho, 2016; Pinti et al., 2016; Todar, 2016). Interestingly, species of the obligate anaerobe Bacteroides such as B. fragilis display remarkably diverse antibiotic resistance mechanisms and exhibit the highest resistance rates of any anaerobic pathogen. This includes an inherent high-level resistance to penicillin through their ability to produce beta-lactamase enzymes which endow them with multiple resistance to β-lactam antibiotics such as penicillin and cephamycin (Ayala et al., 2005; Bush and Bradford, 2016; Hu et al., 2016). Specific species of Bacteroidetes such as Bacteroides fragilis (B. fragilis), normally an abundant commensal microorganism of the middle GI tract, are known to be generally beneficial to human health through their ability to digest dietary fiber and related dietary fiber precursors containing substances such as cellulose, lignin, and pectin, which are normally resistant to the action of host digestive enzymes.

Dietary fibers are catabolized into digestible short-chain fatty acids (SCFAs), volatile fatty acids and polysaccharides in part through the biosynthetic capability of this GI tract abundant bacillus (Keenan et al., 2016; Scheperjans, 2016). When B. fragilis escapes the highly compartmentalized microbe-dense environment of the GI tract (10<sup>11</sup> microbes per gram of fecal matter), they can induce substantial systemic inflammatory pathology with significant sickness, morbidity and mortality (Choi et al., 2016; Fathi and Wu, 2016; Cattaneo et al., 2017; Shivaji, 2017). Enterotoxigenic strains of B. fragilis have been associated with bacteremia, colitis, diarrhea, sepsis, systemic infection, and the development of GI tract cancers and neurological disorders, including AD, that have an increased incidence with aging (Choi et al., 2016; Fathi and Wu, 2016; Keenan et al., 2016; Scheperjans, 2016). Interestingly, certain species of Bacteroidetes have been recently shown to propagate in animal models fed high fat-cholesterol (HFC) diets deprived of sufficient intake of dietary fiber; this suggests that sufficient dietary fiber may have a significant role in regulating the abundance, complexity and stoichiometry of certain species in the GI tract microbiome, including B. fragilis (Heinritz et al., 2016; Köhler et al., 2016; Pistollato et al., 2016; unpublished observations). In addition to these positive health benefits however, these vast numbers of human GI tract resident Gram-negative bacilli when stressed secrete prodigious quantities of endotoxins, exotoxins, endotoxins, exotoxins, lipooligosacahrides (LOSs) and lipopolysaccharides (LPSs), amyloids, and small non-coding RNAs (sncRNAs; see below and **Figure 1**).

## ENDOTOXINS AND EXOTOXINS

Generally, microbiome-derived endotoxins are heat-stable polypeptides associated with the outer membranes of the cell wall of Gram-negative bacteria. They may be composed in part by the Lipid A component of LPS, and once they diffuse into the local environment induce irritation of the GI tract epithelia, capillaries and blood vessels inducing hemorrhage and various pro-inflammatory effects. Endotoxins also induce fever, hemorrhagic shock, diarrhea, altered resistance to bacterial infection, leukopenia followed by leukocytosis, and numerous other systemic effects (Choi et al., 2016; Seong et al., 2016; Zhan and Davies, 2016). For example, in addition to their prodigious LPS generation (see below), B. fragilis endotoxins are a leading cause of anaerobic bacteremia, sepsis and systemic inflammatory distress through their generation of the highly pro-inflammatory zinc metalloproteinase fragilysin, also known as B. fragilis toxin

FIGURE 1 | Like other Gram-negative bacilli, the gastrointestinal (GI) tract abundant Bacteroides fragilis (micrograph of B. fragilis shown; original photo courtesy of Rosa Rubicondior; (http://rosarubi condior.blogspot.com/2014/11/evolving-cooperation-but-for-who-orwhat.html) is capable, when stressed, of releasing a broad spectrum of highly neurotoxic, pro-inflammatory and potentially pathogenic molecules; these comprise five major classes of secreted molecules and include endotoxins, exotoxins, lipooligosacahride (LOS) and lipopolysaccharide (LPS), amyloids, and small non-coding RNAs (sncRNA). For example, the human GI tract-abundant B. fragilis secretes the endotoxin fragilysin and B. fragilis LPS (BF-LPS) both of which have been shown recently to be strongly pro-inflammatory and extremely neurotoxic toward human CNS neurons in primary culture (Li et al., 2016; Lukiw, 2016). While the phyla Bacteriodetes (∼20% of all GI tract bacteria), Firmicutes (∼80% of all GI tract bacteria), Actinobacteria, Proteobacteria, and Verrumicrobia (together, typically ∼4% of all GI tract bacteria), are the most common microbes in the human GI tract microbiome it should be kept in mind that other microbes including fungus, protozoa, viruses, and other commensal microorganisms may also contribute neurotoxic exudates which are highly toxic and detrimental to the homeostasis of CNS neurons.

or BFT (Zhao and Lukiw, 2015; Choi et al., 2016; Fathi and Wu, 2016). BFT has recently been shown to effectively disrupt epithelial cells of GI tract barriers via cleavage of the synaptic type-1 transmembrane zonula adhesion calcium-dependent adhesion protein E-cadherin (Choi et al., 2016; Seong et al., 2016; Zhan and Davies, 2016). It is currently not understood if GI tract- or BBB-disrupting proteolytic endotoxins such as BFT are able to propagate their pathogenic activities via the systemic circulation to further disrupt the GI tract or BBB at distant sites, to ultimately transfer endotoxins, exotoxins, LPSs, amyloids and/or sncRNAs into the cerebrovascular circulation to target brain cells within the CNS. B. fragilis has been suggested to contribute to neurodevelopmental pathology in autism spectrum disorder (ASD; Hsiao et al., 2013; Hofer, 2014; Keaney and Campbell, 2015). It has also recently been reported that along with BFTs amyloid peptide-dependent changes in synaptic adhesion affect both the function and integrity of synapses, suggesting that the observed failure of synaptic adhesion in AD play key roles in the progressive disruption of functional signaling throughout neuronal networks, as is observed in AD brain (Lin et al., 2014; Seong et al., 2015; Leshchyns'ka and Sytnyk, 2016).

Exotoxins are generally complex soluble polypeptides produced on the inside of pathogenic bacteria as part of their normal growth and metabolism, and these are typically excreted by living cells or released during bacterial cell lysis into the surrounding medium. The relatively short lifespan of GI tract bacteria (see above) and their subsequent lysis indicate that lysed bacteria contents may be a relatively persistent source of exotoxins which may need to be either efficiently neutralized or eliminated by the GI tract. Interestingly, under some conditions in rodents certain endotoxins are so toxic that they may be lethal to the host before the innate immune system has a chance to mount immune defenses to promote their neutralization (Bhattacharjee and Lukiw, 2013; Asti and Gioglio, 2014; Hill et al., 2014; Hill and Lukiw, 2015).

## LIPOOLIGOSACAHRIDE (LOS) AND LIPOPOLYSACCHARIDE (LPS)

As an abundant Gram negative bacilli of the human GI tract microbiome both B. fragilis and E. coli secrete lipooligosacahrides (LOS) and lipopolysaccharides (LPS) that are strongly immunogenic and highly pro-inflammatory toward human neurons (Bian et al., 2011; Alkasir et al., 2016; Fathi and Wu, 2016; Foster et al., 2016; Ghaisas et al., 2016; Hug et al., 2016; Lukiw, 2016; Rogers and Aronoff, 2016; Sender et al., 2016; Sharon et al., 2016). LPSs, as characteristic components of the outer leaflet of the outer membrane of Gram-negative bacteria shed into the extracellular space, play key roles in host-pathogen interactions and the innate-immune system (Hill and Lukiw, 2015; Zhao et al., 2015; Maldonado et al., 2016). While LPSs contain large and hypervariable oligosaccharide/polysaccharide regions, the relatively conserved lipid region (lipid A) is the endotoxic and biologically active moiety that is largely responsible for septic shock (Jiang et al., 2016; Maldonado et al., 2016). A canonic LPS structure is represented by that of E. coli LPS, one of the most potent neurotoxic lipid A species known, consisting of a 1,4′ -biphosphorylated glucosamine disaccharide bearing six fatty acids which are unbranched chains 12–14 methyl(ene) units in length. Other "lipid A" species show variability in the number, length, and composition of the attached fatty acids, as well as variability in the degree of phosphorylation and number and types of substituted phosphate ligands. For instance, BF-LPS lipid A is penta-acylated and mono-phosphorylated, and contains branched fatty acids 15–17 methyl(ene) units in length; deviations from the canonical lipid A structure are known to have a profound impact on innate-immune responses. Gram-negative bacterial exudates such as BF-LPSs are hypervariable in composition, and different Bacteroidetes species appear to generate unique temporal patterns of LPS production. These exhibit rapid and remarkably adaptive changes in LPS structure and alterations in damageor pathogen-associated molecular patterns (DAMP/PAMP) as strategies for host immune evasion (Friedland, 2015; Land, 2015; Maldonado et al., 2016; Richards et al., 2016). Here, for the first time, we provide evidence that E. coli LPS is abundant in neocortical and hippocampal extracts from AD brain, regions of the human limbic system targeted by intense neuroinflammation characteristic of the AD process (see **Figure 1** and legend). Similarly the pathological actions of LPS on the induction of pro-inflammatory signaling in primary human neurons have recently been demonstrated, and additional studies are in progress (Lukiw, 2016).

## AMYLOIDS

Atypical amyloid generation, aggregation, folding, and impaired clearance are characteristic pathological features of human neuro-inflammatory and neurodegenerative disorders of the CNS that include AD (Calsolaro and Edison, 2016; Andreeva et al., 2017). What is generally not appreciated is that a major secretory product of the GI tract microbiome is amyloid, and that the life-long contribution of microbial amyloid to CNS pathophysiology can be very substantial. "Amyloid" is a generic term for any aggregated, insoluble, lipoprotein-enriched deposit that exhibits β-pleated sheet structures oriented perpendicular to the fibrillar axis (Lukiw, 2012; Clark and Vissel, 2015; Lim et al., 2015; Andreeva et al., 2017; Bolós et al., 2017). The potential for amyloid formation is surprisingly high in almost all proteins; a major factor for amyloid formation is the presence within proteins of primary amino acid sequences that can form a tight, self-complementary interface with an identical segment, thus permitting the cooperative formation of a steric zipper. Two self-complementary beta-sheets form the backbone of the amyloid fibril (Goldschmidt et al., 2010; Buxbaum and Linke, 2012; Andreeva et al., 2017). The characterization of the "amylome," a categorization of amino acid sequences that possess self-complementary interfaces and high fiber-forming propensity has improved our understanding of the capability of different proteins to generate amyloid (Goldschmidt et al., 2010; Lukiw, 2012; Andreeva et al., 2017). The progressive generation and aggregation of amyloids contribute to "dense-deposit" disease; the pathogenesis of diseases that accumulate amyloid, including AD, all involve prominent inflammatory responses at sites of amyloid deposition—these accumulations are often mediated by microglial cells, the "resident immune cells" of the CNS. Interestingly, most microbial species, including fungi and bacteria, secrete self-associating and strongly amyloidogenic lipoproteins (Hill et al., 2014; Syed and Boles, 2014; Schwartz et al., 2016). For instance, amyloids are associated with fungal surface-structures and the recent observation of amyloidogenic fungal proteins and diffuse mycoses in the blood of AD patients suggest that chronic fungal infection over the course of aging may increase AD risk (Alonso et al., 2014; Hill et al., 2014). Of further relevance is that: (i) Aβ42 peptide monomers, dimers, oligomers and fibrils each induce patterns of pro-inflammatory gene signaling typical of the classical microglial-mediated innateimmune and inflammatory response induced by infectious agents such as bacterial LPS (Ferrera et al., 2014; Calsolaro and Edison, 2016; Lukiw, 2016; Andreeva et al., 2017); (ii) the presence of bacterial LPS or endotoxin/exotoxin-mediated inflammatory signaling strongly contributes to amyloid neurotoxicity (Lee et al., 2008; Asti and Gioglio, 2014; Zhao and Lukiw, 2015; Zhao et al., 2016); (iii) AD amyloids, like prion amyloids, once formed, may induce a self-perpetuating process leading to amplification, aggregation, and spreading of pathological aggregates (Le et al., 2014); and (iv) recently it has been shown that Aβ42 peptide fibrillogenesis is strongly potentiated by soluble bacterial exudates and viruses such as HSV-1, suggesting the contribution of microbial-sourced factors and/or infectious events to amyloidogenesis, a distinguishing feature of the AD neuropathology (Hill et al., 2014; Stilling et al., 2014; Zhao et al., 2015; Russo et al., 2017).

## SMALL NON-CODING RNA (sncRNA)

While the secretion of proteins, lipids, and nucleic acids (both RNA and DNA) from neural cells into the extracellular space is a commonly recognized phenomenon in neurobiology, the secretion of small non-coding RNA (sncRNA) from microbial cells into the GI tract has only been very recently characterized (Ghosal et al., 2015; Lukiw, 2016; Ghosal, 2017). Employing multi-component secretion systems, sncRNAs may be exuded from bacteria as separate entities, or more commonly, contained within lipid spheres or outer membrane vesicles (OMVs; Ghosal et al., 2015; unpublished observations). A major fraction of all secreted extracellular RNAs are sncRNAs in the size range of 15–40 nucleotides derived from specific intracellular bacterial RNAs. These sncRNAs have been speculated to be involved in immune-evasion, intra-species communication, in inter-kingdom genetic exchanges, pathogenicity and/or microbiome-host signaling; indeed protein-, lipid-, and nucleic acid-containing OMVs released by GI tract Gram-negative bacteria can be intensely pro-inflammatory, pathogenic or even lethal to the host (Zhao and Lukiw, 2015; Ghosal, 2017; Lukiw and Rogaev, 2017; unpublished observations). Several important questions remain to be answered: (i) do secreted sncRNAs play any role in microbiome survival, immune evasion and/or antibiotic resistance? (ii) how do GI tract microbes promote and organize the regulation of sncRNA trafficking (iii) how are bacterial sncRNAs transported across bacterial membranes and subsequently released into the extracellular space? (iv) how are the sncRNAs selected and packaged for export? and (v) are there differences in secreted sncRNA profiles between pathogenic and non-pathogenic bacteria and/or between healthy and diseased states of the host? Further investigations in the field of extracellular bacterial sncRNAs are clearly needed to shed light on their potential role as mediators of microbiomehost signaling and intercellular communication. By studying bacterial secreted sncRNA patterns, we may be able to further advance our understanding of the complex interactions that exist between humans and their GI tract microbiome and design, perhaps through dietary manipulation, highly effective intervention strategies that could improve and optimize human neurological health.

## THANATOMICROBIOME

Evidence for the immense biophysiological efforts in keeping the GI tract microbiome contained within GI tract compartments and from expansion beyond its normal niche, comes from analysis of the human microbiome at the time of death. Very little data are available concerning what happens to the microbiome when a human host dies—in a healthy adult, most internal organs such as the spleen, liver, heart, and brain are generally devoid of microbes because the innate-immune system or other microbial components keeps them in check. After death, however, the generation of ATP ceases, the innate-immune system falters and microbes proliferate throughout the body; this has recently been shown to begin in the ileocecal area of the GI tract, spreading to the liver and spleen, and continuing to the heart and brain (Alan and Sarah, 2012; Can et al., 2014; Clement et al., 2016; Javan et al., 2016). Still evolving concepts of what happens to GI tract microbiome speciation and complexity at the time of death are currently being researched. Indeed the thanatomicrobiome (thanatos, Greek for death) is a relatively new designation defined as the composition and organization of the GI tract microbiome and other microbial communities following cessation of all life activities (Clement et al., 2016; Javan et al., 2016). Recent studies so far underscore the fact that in the GI tract microbiome there is a constant struggle to contain GI tract microbiome integrity and regulate specific bacterial abundance and complexity (Clement et al., 2016; Javan et al., 2016). Ongoing work from temporal studies on the thanatomicrobiome across defined post-mortem intervals (PMI) further indicate (i) that the majority of the microbes within the human body and those which propagate most rapidly at the time of death are the obligate anaerobes that begin to non-randomly proliferate from the GI tract continuing throughout the human organs over the PMI (Javan et al., 2016); and (ii) that comprehensive knowledge of the number and abundance of each organ's microbial signature could be useful to forensic microbiologists as a new source of data for estimating PMI. These data combined with nucleic acid sequencing and bioinformatics would also be invaluable in aiding researchers who use post-mortem tissues in their research work and in forensic criminology, microbial speciation and the study of microbiome-host genetics in the later stages of life.

#### CONCLUDING REMARKS

In summary, the human GI tract constitutes the largest repository of the human microbiome, and its impact on human neurological aging, health and disease is becoming increasingly appreciated. Consisting of about ∼4 × 10<sup>13</sup> microorganisms, the human GI tract microbiome forms a highly complex, symbiotic and dynamic ecosystem within the host and dietary factors and host genetics appear to have a strong influence on microbial abundance, speciation and complexity, and their ability to influence CNS functions (Foster et al., 2016; Li et al., 2016; Richards et al., 2016; Brandscheid et al., 2017; Tremlett et al., 2017). We sincerely hope that this "Perspectives" article has effectively highlighted recent findings on microbial-derived endotoxins, exotoxins, LOSs and LPSs, amyloids and sncRNAs and has stimulated interest in the potential contribution of these neurotoxic and pro-inflammatory microbial exudates to agerelated inflammatory neurodegeneration, amyloidogenesis, and AD-relevant pathology (**Figure 1**). Taken together, these current observations and recent data advance at least seven areas in our understanding of the role of the GI tract microbiome in age-related neurological diseases associated with progressive, inflammatory neurodegeneration of the human brain: (i) that the GI tract microbiome are a potent source of neurotoxic species that are abundantly secreted by multiple Gram-negative bacilli in the gut (B. fragilis, E. coli, and others); (ii) that bacterial LPS are readily detectable in the neocortex and hippocampus of the AD brain, and at significantly higher abundance in AD than controls, indicating that LPS may be able to transit physiological barriers to access CNS compartments (**Figure 2**); (iii) that the transit of highly pro-inflammatory neurotoxins such as LPS across compromised GI tract and blood-brain barriers underscore the critical roles of cellular adhesion structures in allowing passage of noxious molecules from the GI tract into the systemic circulation and CNS (Montagne et al., 2016; Soenen et al., 2016; van de Haar et al., 2016); (iv) that extremely complex mixtures of neurotoxins may be generated by either single microbes or by combinations of bacilli that constitute the GI tract microbiome (**Figure 1**); (v) that biophysical, gastrointestinal, and neurobiological barriers that may become more "leaky" with aging again underscore the important role of intact membrane barriers in moderating systemic and CNS inflammation and immune-mediated inflammatory disease (Hill and Lukiw, 2015; Keaney and Campbell, 2015; Montagne et al., 2015; Choi et al., 2016; Köhler et al., 2016; Minter et al., 2016a; Richards et al., 2016; van de Haar et al., 2016; Zhan and Davies, 2016; Varatharaj and Galea, 2017); (vi) that bacterial complexity, neurotoxin abundance, speciation, and complexity in the CSF, blood serum or in brain tissues may be useful for the diagnosis of AD (Zhao et al., 2015; Soenen et al., 2016); and (vii) that studies on the thanatomicrobiome should be useful for a clearer understanding of the neuro- and micro-biological processes in operation over the PMI that should be useful in scientific research that utilizes post-mortem tissues in basic research, in forensic applications, in criminology and in the more accurate diagnosis of neurological disease (Clement et al., 2016; Javan et al., 2016). While one other recent investigation reported the detection of LPS in gray matter (temporal lobe) and white matter (frontal lobe) in AD (Zhan et al., 2016), here for the first time we report the detection of bacterial LPS in brain lysates from AD hippocampus, an anatomical region of the AD brain that develops the earliest and most profound neuropathology. Some advanced AD hippocampal patients exhibited up to a 26-fold increase in LPS over age-matched controls. Lastly, more research into the intriguing field of human GI tract microbiome-host interaction and its potential contributory role to human aging, neurological health and disease is clearly needed. The study of these symbiotic prokaryotic and eukaryotic divisions, their evolution and their intriguing interrelationships, genetic interactions and associations in future work should be useful in expanding our understanding of microbiome-host interplay and control in the initiation, development, propagation, and diagnosis of human neurological disorders in which microbial involvement appears to play some contributory or even deterministic role.

FIGURE 2 | (A) human brain temporal lobe neocortex [N = 6 control and 6 sporadic AD cases; quantified in (B)]; and (C) hippocampus (N = 2 control and N = 4 AD cases; quantified in (D)] were analyzed for LPS against β-actin abundance in the same sample (using anti-E. coli LPS; cat # ab35654 from Abcam, Cambridge UK and anti-β-actin cat # 3700 from Cell Signaling, Danvers MA, USA) using Western analysis as previously described by our group (Bhattacharjee et al., 2016; Zhao et al., 2016); all AD and control tissues were analyzed in a RNA-analysis clean room facility; all control and AD tissues were age- and gender-matched; there were no significant differences between the age (control 72.9 ± 8.1 years, AD 74.2 ± 9.1 years), gender (all female), PMI (all tissues 3.5 h post-mortem or less), RNA quality or RNA yield between each of the two groups; LPS abundance was found to be on average over two-fold as abundant in AD when compared to age-, gender, and PMI-matched control neocortex in 6 of 6 cases; LPS was found to be on average three-fold as abundant in AD when compared to age-, gender, and PMI-matched control hippocampus in 3 of 4 cases; some advanced AD hippocampal samples exhibited up to a 26-fold increase in LPS over age-matched controls (C, LPS in control lane 2 vs. AD lane 5); because one major source of LPS are Gram-negative bacteria of the human GI tract (predominantly B. fragilis and E. coli), this suggests that in vivo intensely pro-inflammatory LPS species may be able to "leak" through at least two major biophysiological barriers—the GI tract barrier and the BBB—to access brain compartments (see Devier et al., 2015; Halmer et al., 2015; Choi et al., 2016; Minter et al., 2016b; Montagne et al., 2016; Richards et al., 2016; Soenen et al., 2016; van de Haar et al., 2016; Zhan and Davies, 2016; Zhao et al., 2016; Varatharaj and Galea, 2017). Unpublished work from this laboratory further indicates the positive detection of LPS in 36 of 36 AD tissues sampled from the superior temporal lobe neocortex in aged individuals (age range 66–79 yr; see Table 1 in Cui et al., 2010). Another recent investigation reports the finding of LPS in gray matter (temporal lobe) and white matter (frontal lobe) of the AD brain (Zhan et al., 2016). Together these data also suggest that neurotoxic cocktails secreted by multiple GI tract microbes or other microbial species (Figure 1) may have considerable potential to support intense pro-inflammatory signaling within the CNS especially over the course of aging when barriers become more "leaky" (Hill and Lukiw, 2015; Keaney and Campbell, 2015; Montagne et al., 2015; Choi et al., 2016; Köhler et al., 2016; Minter et al., 2016a,b; Richards et al., 2016; van de Haar et al., 2016; Zhan and Davies, 2016; Varatharaj and Galea, 2017); (B) and (D) represent the mean plus one standard deviation of that mean; \*p < 0.05, \*\*p < 0.01 ANOVA; NC, negative control using a control murine brain extract (strain C57BL/6J); in (B) and (D) a dashed horizontal line at 100 is included for ease of comparison.

## AUTHOR CONTRIBUTIONS

YZ and VJ analyzed brains for LPS content; WL compiled and analyzed the data and wrote the paper.

## ACKNOWLEDGMENTS

This research work was presented in part at the Society for Neuroscience (SFN) Annual Meeting 12–16 November 2016, San Diego CA, USA and at the Alzheimer Association International Congress 2016 (AAIC 2016) Annual conference 21–27 July 2016 in Toronto, Canada. These studies utilized total nucleic acid and/or cytoplasmic fractions extracted from primary human neuronal-glial (HNG) co-cultures; sincere thanks are extended to Drs. P. N. Alexandrov, J. G. Cui, F. Culicchia, W. Poon, K. Navel, C. Hebel, and C. Eicken for short PMI human brain tissues or extracts, unpublished Western data and immunochemistry, HNG tissue culture and NF-kB-DNA binding assay, initial bioinformatics and data interpretation, and to D. Guillot and A. I. Pogue for expert technical assistance and medical artwork. All human tissues were used in strict accordance with ethical compliance procedures and protocols followed by donor institutions; thanks are also extended to the Institute for Memory Impairments and Neurological Disorders (MIND), to the University of California at Irvine (UCI) and to the many neuropathologists, physicians and researchers of the US, Canada and Europe who have provided high quality, short PMI human CNS or extracted tissue fractions for scientific study. Research

#### REFERENCES


on the human microbiome, pro-inflammatory and pathogenic signaling in the Lukiw laboratory involving the innate-immune response, neuroinflammation and amyloidogenesis in AD 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.

protease contributes to anaerobic sepsis in mice. Nat. Med. 22, 563–567. doi: 10.1038/nm.4077


microbial diversity influences neuro-inflammation and amyloidosis in a murine model of Alzheimer's disease. Sci Rep. 6:30028. doi: 10.1038/srep 30028


**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, 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.

# Molecular Characterization of the Human Stomach Microbiota in Gastric Cancer Patients

Guoqin Yu<sup>1</sup> \*, Javier Torres <sup>2</sup> , Nan Hu<sup>3</sup> , Rafael Medrano-Guzman<sup>4</sup> , Roberto Herrera-Goepfert <sup>5</sup> , Michael S. Humphrys <sup>6</sup> , Lemin Wang<sup>3</sup> , Chaoyu Wang<sup>3</sup> , Ti Ding<sup>7</sup> , Jacques Ravel <sup>6</sup> , Philip R. Taylor <sup>3</sup> , Christian C. Abnet <sup>3</sup> and Alisa M. Goldstein<sup>8</sup> \*

1 Integrative Tumor Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, United States, <sup>2</sup> Unidad de Investigacion en Enfermedades Infecciosas, Unidad Medica de Alta Especialidad Pediatria, Centro Medico Nacional SXXI, Instituto Mexicano del Seguro Social, Mexico City, Mexico, <sup>3</sup> Metabolic Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, United States, <sup>4</sup> Unidad Medica de Alta Especialidad Oncología, Centro Medico Nacional SXXI, Instituto Mexicano del Seguro Social, Mexico City, Mexico, <sup>5</sup> Instituto Nacional de Cancerología, Secretaria de Salúd, Mexico City, Mexico, <sup>6</sup> Institute for Genome Sciences, University of Maryland School of Medicine, Baltimore, MD, United States, <sup>7</sup> Shanxi Cancer Hospital, Taiyuan, China, <sup>8</sup> Clinical Genetics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, United States

#### Edited by:

Lorenza Putignani, Bambino Gesù Ospedale Pediatrico (IRCCS), Italy

#### Reviewed by:

Jeong-Heon Cha, Yonsei University, South Korea Valerio Iebba, Sapienza Università di Roma, Italy

#### \*Correspondence:

Guoqin Yu yug3@mail.nih.gov Alisa M. Goldstein goldstea@mail.nih.gov

Received: 15 March 2017 Accepted: 20 June 2017 Published: 06 July 2017

#### Citation:

Yu G, Torres J, Hu N, Medrano-Guzman R, Herrera-Goepfert R, Humphrys MS, Wang L, Wang C, Ding T, Ravel J, Taylor PR, Abnet CC and Goldstein AM (2017) Molecular Characterization of the Human Stomach Microbiota in Gastric Cancer Patients. Front. Cell. Infect. Microbiol. 7:302. doi: 10.3389/fcimb.2017.00302 Helicobacter pylori (Hp) is the primary cause of gastric cancer but we know little of its relative abundance and other microbes in the stomach, especially at the time of gastric cancer diagnosis. Here we characterized the taxonomic and derived functional profiles of gastric microbiota in two different sets of gastric cancer patients, and compared them with microbial profiles in other body sites. Paired non-malignant and tumor tissues were sampled from 160 gastric cancer patients with 80 from China and 80 from Mexico. The 16S rRNA gene V3–V4 region was sequenced using MiSeq platform for taxonomic profiles. PICRUSt was used to predict functional profiles. Human Microbiome Project was used for comparison. We showed that Hp is the most abundant member of gastric microbiota in both Chinese and Mexican samples (51 and 24%, respectively), followed by oral-associated bacteria. Taxonomic (phylum-level) profiles of stomach microbiota resembled oral microbiota, especially when the Helicobacter reads were removed. The functional profiles of stomach microbiota, however, were distinct from those found in other body sites and had higher inter-subject dissimilarity. Gastric microbiota composition did not differ by Hp colonization status or stomach anatomic sites, but did differ between paired non-malignant and tumor tissues in either Chinese or Mexican samples. Our study showed that Hp is the dominant member of the non-malignant gastric tissue microbiota in many gastric cancer patients. Our results provide insights on the gastric microbiota composition and function in gastric cancer patients, which may have important clinical implications.

Keywords: Helicobacter pylori, 16S rRNA, KEGG modules, microbiome, gastric cancer

**Abbreviations:** GC, Gastric Cancer; Hp, Helicobacter pylori; HMP, Human Microbiome Project; NCI, National Cancer Institute; OTUs, Operational Taxonomy Units; QIIME, Quantitative Insights into Microbial Ecology; PD\_whole\_tree, Phylogenetic diversity; PICRUSt, Phylogenetic Investigation of Communities by Reconstruction of Unobserved States; PERMANOVA, Permutational Multivariate Analysis of Variance.

## BACKGROUND

Gastric cancer (GC) is the fifth most common cancer in the world and the third leading cause of cancer death (Ferlay et al., 2013). GC incidence varies widely with high rates in Asia, Eastern Europe, and Central and South America, and low rates in North America and Africa (Carneiro, 2014). GC may arise in cardia or in non-cardia (the fundus, body, or pylorus section). Chronic colonization of Helicobacter pylori (Hp) is known to increase the risk of non-cardia cancer (Cavaleiro-Pinto et al., 2011). The association between Hp colonization and gastric cardia cancer varies by populations. The studies in Western countries tend to show a neutral or even negative association while in Eastern populations namely China, Japan, and Korea, there is strong evidence of a higher risk of cardia cancer among subjects with Hp colonization (Cavaleiro-Pinto et al., 2011).

Chronic inflammation of the stomach may progress through a series of steps including atrophic gastritis, intestinal metaplasia, dysplasia, and gastric adenocarcinoma (Correa, 2013). Atrophic gastritis, the loss of specialized glandular tissue with impaired acid secretion and differentiation of gastric progenitor cells, results in hypochlorhydria in the stomach. It is generally believed that Hp prefers a healthy gastric mucosa and that as the steps to GC progress, Hp is also gradually fading, until it disappears. Therefore, the stomachs of patients with GC should facilitate the colonization of the gastric mucosa by bacteria other than Hp (Sheh and Fox, 2013). Studies of gastric microbiota are sparse. Previous studies were often small, not in GC patients or used biopsy samples collected during endoscopy, which may have led to contamination from the oral cavity (see Supplementary Table 1 for summary of previous studies). Therefore, gastric microbiota in GC patients remains largely unknown.

Chronic colonization of Hp is the major risk factor for GC in both Chinese and Mexican populations (Kamangar et al., 2007; Ayala et al., 2011). However, GC occurs mainly at the cardia of the stomach in Shanxi, China, but in the noncardia of the stomach in Mexico. In this study, we profiled the taxonomic and functional profiles in non-malignant gastric tissue from two collections of GC patients separately, one from China (cardia cancer cases) and the second from Mexico (noncardia cancer cases). We compared gastric non-malignant tissue with paired tumor tissues and with other body sites including oral, nasal cavity, stool, vagina, and skin using data from the Human Microbiome Project (HMP) (Human Microbiome Project Consortium, 2012). We also evaluated differences in the gastric microbiota by Hp colonization status, anatomical sites within the stomach for the non-cardia cancer samples, and tissue type (non-malignant and tumor) separately for the two sample populations.

#### MATERIALS AND METHODS

#### Study Subjects and Sample Collection

The Chinese gastric tissue samples were from 80 gastric cardia cancer patients recruited at the Shanxi Cancer Hospital in Taiyuan, Shanxi Province, China, between 1998 and 2001. This study was approved by the Institutional Review Boards of the Shanxi Cancer Hospital and the National Cancer Institute (NCI). All subjects provided written informed consent prior to participation. Cases were histologically confirmed as adenocarcinomas by pathologists at both the Shanxi Cancer Hospital and the NCI. Clinical data was collected by review of medical records. Patients who were <18 years old, with cancer other than GC or with previous treatment for GC were excluded. Tumor tissues and matched non-malignant tissues distant to the tumor were obtained from surgical resections, snap frozen in liquid nitrogen, and stored at −130◦C until used. H&E slides were used to determine the percentage of tumor cells in the tissues. Total DNA was extracted using the Allprep RNA/DNA/Protein mini kit (QIAGEN) following the protocol provided by the manufacturer.

The Mexican gastric tissue samples were from 80 gastric noncardia cancer patients recruited at the Oncology Hospital, Centro Médico Nacional Siglo XXI, Instituto Mexicano del Seguro Social, and the Instituto Nacional de Cancerología, Secretaria de Salud in Mexico City, Mexico, between 2008 and 2013. The study was approved by the ethics committee of each hospital and written informed consent was obtained from all patients prior to enrollment in the study. Cases were histologically confirmed by the pathologist. The clinical and pathological data were recorded in questionnaires. Patients who were <18 years old, with any autoimmune disease, diabetes, or cancer other than GC, and with a previous treatment for GC were excluded from the study. Tumor tissue and matched non-malignant tissue distant to the tumor were obtained from surgical resection specimens, placed immediately in microfuge tubes and submerged in a container with liquid nitrogen, and stored at −70◦C until tested. H&E slides were used to determine the percentage of tumor cells in the tissues. Total DNA was extracted by QIAamp DNA mini kit (QIAGEN) using the protocol provided by the manufacturer.

All the non-malignant tissue samples were verified with absence of tumor cells. Tumor tissue samples without tumor cells were excluded from all analyses. The percentage of tumor cells were 70–80% in the Chinese tumor samples and 30–50% in Mexican tumor samples. Examples of H& E slides are shown in Supplementary Figure 1.

#### 16S rRNA Gene Sequence Analysis

The V3–V4 region of the 16S rRNA gene was amplified and sequenced on the Illumina MiSeq platform using the 300 pairedend protocol at the Institute of Genome Sciences, University of Maryland School of Medicine as described previously (Fadrosh et al., 2014).

Sequence reads were processed to remove low quality, short, or chimera reads (Yu et al., 2015). We removed low quality reads (reads with average quality <20 over 30 bp window based on Phred algorithm; paired reads which have at least one read with length <75% of its original length) and chimera reads (by UCHIME). The remaining reads with at least 97% sequence identity were clustered into species-level Operational Taxonomy Units (OTUs) in the software package Quantitative Insights into Microbial Ecology (QIIME 1.8.0) (Caporaso et al., 2010) by using command pick\_open\_reference\_otus.py with usearch61 clustering algorithm and other default settings. The OTUs were assigned to taxa (e.g., genus, family, phylum) using the Greengenes database as reference (version 13\_8; DeSantis et al., 2006). OTUs with only one read were excluded from analysis. Samples with <1,000 reads were excluded from analysis. The sequence data were submitted to BioProject database (accession number of 310127) at the National Center for Biotechnology Information website.

Alpha diversity was estimated as number of OTUs, Shannon's Index (Shannon, 1997), and Phylogenetic diversity (PD\_whole\_tree) (Faith and Baker, 2006) by averaging over 20 rarefied tables of 1,000 reads/sample. Alpha diversity was used to measure the species diversity of each sample. The number of OTUs, also known as richness, is a measure of diversity that does not consider the frequency of OTUs. Shannon's index is estimated by both the number and frequency of the OTUs. PD\_whole\_tree further takes account of the phylogenetic relationship of OTUs. The phylogenetic tree of OTUs used for PD\_whole-tree estimates was prepared in QIIME based on neighbor-joining method. The alpha diversity increased with number of sequence reads sampled (Supplementary Figure 2). The alpha diversity showed differences by sample groups with 1,000 reads/sample; the order of sample groups based on alpha diversity did not change by number of sequence reads.

Beta diversity was measured as unweighted (presence/absence of taxa) and weighted (using taxa relative abundance information) UniFrac distance (Lozupone et al., 2011). Beta diversity measures dissimilarities of two samples in microbial profiles. We calculated both alpha and beta diversity based on rarefied tables of 1,000 reads/sample.

The relative abundance of taxa at different levels (phylum, class, order, family, and genus) was calculated based on the unrarefied table. The taxa relative abundance was estimated as the proportion of OTUs assigned to a taxon.

The Human Microbiome Project (HMP) 16S rRNA V3– V5 data were downloaded for comparison (http://hmpdacc.org/ HMQCP/; Human Microbiome Project Consortium, 2012). The HMP sequence reads were processed in the same manner as described above. A total of 2,579 samples from 5 body sites (including oral, nasal cavity, stool, vagina, and skin) of 242 healthy US adults in HMP phase 1 were used for comparison (Aagaard et al., 2013). The study by Lozupone et al. (2013) showed that the difference in population or technologies used should not affect the comparison by body sites. In addition, we limited the comparison of HMP and stomach microbiota data to the highest and least variable taxonomic (phylum)/functional (module) level so that the population/technology differences between these two studies would have limited effect on the comparisons.

#### Metagenomic Prediction

We used Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt)1.0.0 (Langille et al., 2013) to predict virtual metagenomes for each sample from the 16S rRNA gene sequence data and used the KEGG database as a reference (Kanehisa et al., 2014) to determine the relative abundance of metabolic pathways and modules within the virtual metagenomes. PICRUSt requires the use of Greengenes reference, version 13\_5 to cluster reads into OTUs (DeSantis et al., 2006). Therefore, we re-clustered all the sequence data (including HMP) in QIIME with the command parallel\_pick\_otus\_usearch61\_ref.py and the Greengenes reference version 13\_5.

#### Statistical Analysis

The Wilcoxon rank-sum test was used to examine gastric microbiota alpha diversity and taxa relative abundance differences between antrum and corpus in Mexican samples or between Hp+ and Hp− samples. Wilcoxon signed-rank test was used for the differences between non-malignant and matched tumor samples in each population. When examining taxa relative abundance, Bonferroni correction was used to adjust for tests of multiple taxa. Permutational Multivariate Analysis of Variance (PERMANOVA, adonis) was used to compare sample groups by unweighted/weighted UniFrac distance matrix. P < 0.05 were considered significant after adjustment for multiple tests.

In order to compare gastric microbiota with the oral, nasal, stool, skin, and vagina microbiota from the HMP study, we calculated Euclidean/Bray-Curtis distances and generated a matrix from the phylum-level/KEGG module profiles, performed principal coordinate analysis (PCoA) on the Euclidean/Bray-Curtis distance matrix, and then plotted the figure based on the first three principal coordinates to visualize the similarities and differences among different body sites. All statistical analyses were performed in R. Bray-Curtis and Euclidean distance showed similar results, therefore only Bray-Curtis distance is shown.

## Quality Control

To address the concerns about possible contamination, we included 2 blank samples as negative controls. We also included 1 vaginal and 1 stool sample as positive controls to evaluate the performance of DNA amplification and sequencing. The two positive controls generated 2,703 and 58,201 reads, respectively, suggesting good performance of DNA amplification and sequencing. The blanks had extremely low number of reads (41 and 43 reads/sample, respectively). Furthermore, the OTUs found in both blanks were extremely rare in the gastric samples with accumulated relative abundance range of 0–0.006. Therefore, our results were unlikely to have been affected by contamination.

The conventional DNA extraction method for microbiome studies often includes an extra cell lysis step (bead-beating) to break the hard-to-break cell membranes of some species. To examine whether some taxa were missed due to the lack of a bead-beating step in our DNA extraction protocols, we evaluated 2 Chinese tissue samples using two different DNA extraction methods: a DNA extraction method with a bead-beating step and commonly used for microbiome study (Flores et al., 2012) and the method used for our Chinese samples in the current study. We found 14 genus-level taxa discovered by the extraction method with the bead-beating step that were not discovered by our DNA extraction method (Supplementary Table 2). However, these taxa were extremely rare with total cumulated relative

abundance of 0.007 and 0.038 for two samples, respectively. Therefore, the DNA extraction method should not have adversely affected our findings, although we cannot exclude missing some rare taxa.

## RESULTS

#### Characteristics of the Study Subjects

After excluding samples with <1,000 reads per sample, 77 non-malignant gastric tissue samples from China and 80 from Mexico were included for analysis and the median (interquartile range) was 10,460 (5,454–19,980) reads per sample. The raw and qualified number of reads for each sample group are shown in Supplementary Table 3. The average age of these Chinese cases was 60.8 years old, 83% were male, and all were diagnosed with gastric cardia adenocarcinoma. The average age of Mexican cases was 64.5 years old, 54% were male, and all tumors were located in the non-cardia regions of the stomach (21 antrum, 24 corpus, 35 unspecified). In addition, 80 tumor samples from China and 54 from Mexico were also included for comparison [median (interquartile range): 9,406 (4,228–15,330) reads/sample] after excluding samples with <1,000 reads per sample or no tumor cells.

## Taxonomic and Functional Profiles of Non-malignant Gastric Microbiota

The taxonomic and functional profiles are shown in **Figure 1** for non-malignant tissue samples. According to the non-malignant gastric tissues, the gastric microbiota for both sample sets was mainly composed of Proteobacteria, followed by Bacteroidetes in Chinese samples or Firmicutes in Mexican samples (**Figure 1A**). The majority of samples from China (78%) and Mexico (50%) were dominated by Proteobacteria (relative abundance >50%). Nineteen Mexican samples were dominated by Firmicutes and one Chinese sample was dominated by Bacteroidetes. The remaining samples did not have a dominant phylum.

The most abundant genus in the non-malignant microbiota of both Chinese and Mexican gastric cancer patients was Helicobacter (**Figure 1B**), and 99% of the Helicobacter reads [median (interquartile range): 98.8% (98.7–99.3%)] were classified as Hp. As shown in **Table 1**, the majority of GC patients' stomachs (94% Chinese and 55% Mexican) were colonized by Hp, and 53% Chinese and 28% Mexican gastric microbiota were dominated by Hp (Hp relative abundance >50%).

The virtual reconstructed functional profiles (KEGG modules) of non-malignant gastric tissue samples predicted by PICRUSt are shown in **Figure 1C**. The most abundant module functions in gastric microbiota were membrane transport, amino acid metabolism, carbohydrate metabolism, replication and repair, and energy metabolism in both Chinese and Mexican samples. Compared to the variation in taxonomic profiles, the variation in functional profiles among non-malignant gastric tissue samples was more limited (**Figure 1A** vs. **Figure 1C**).

Neither Chinese nor Mexican samples showed an association between gastric microbial features and age or gender (data not shown). Within Mexican samples, no significant difference in microbial alpha diversity, beta diversity and taxa relative abundance for the antrum and corpus non-malignant samples was observed (Supplementary Table 4).

## Comparison of Non-malignant Gastric Tissue to Matched Tumor Tissue

After excluding samples with <1,000 reads, 80 tumor tissue samples from China and 54 from Mexico remained for comparison with their matched non-malignant tissues. The taxonomic and functional profiles for these samples are shown in Supplementary Figure 3. The average profiles for both nonmalignant and tumor sample groups are shown in **Figure 2**. Similar to the profiles in non-malignant tissues, the tumor gastric microbiota for both sample sets was also mainly composed of Proteobacteria, followed by Bacteroidetes in Chinese tumor samples or Firmicutes in Mexican tumor samples (Supplementary Figure 3A vs. **Figure 2A**). The genus with the most abundance in both tumor sample sets was Helicobacter (Supplementary Figure 3B vs. **Figure 2B**) with average relative abundance of 21% in Chinese samples and 18% in Mexican samples. Compared to non-malignant tissues, tumor tissue had less Proteobacteria, and higher Bacteriodetes, Firmicutes, Fusobacteria, and Spirochaetes in Chinese samples. There was no significant change in Mexican samples in phylum-level taxa (Supplementary Table 5). At the genus level, tumor tissue had lower Helicobacter abundance relative to non-malignant tissue in both Chinese and Mexican samples. Chinese samples showed substantial differences in alpha diversity as well as several other genus taxa. Mexican samples showed differences in Clostridia relative abundance, but did not display differences in alpha diversity measures (Supplementary Table 5). Hp relative abundance was also lower in tumor tissues compared to matched non-malignant tissues in both sample sets (Supplementary Table 5). However, the majority of tumor tissues (94% Chinese and 56% Mexican) were colonized by Hp, and many tumor samples (20% Chinese and 17% Mexican) were dominated by Hp (Hp relative abundance >50%; **Table 1**).

The most abundant module functions in tumor tissues were membrane transport, amino acid metabolism, carbohydrate metabolism, replication and repair, translation, and energy metabolism in both Chinese and Mexican samples (Supplementary Figure 3C, **Figure 2C**). In both Chinese and Mexican samples, the functional module of infectious disease was higher in non-malignant than in tumor tissues (Supplementary Table 6). Chinese samples showed substantial differences in other functional modules after Bonferroni correction for multiple comparisons (Supplementary Table 6). Mexican samples did not display differences in relative abundance for other functional modules between tumors and non-malignant tissues (Supplementary Table 6). Functional and taxonomic profiles were correlated (Supplementary Figure 4). For example, high infectious disease function was mainly contributed by Helicobacter as these factors were positively correlated in relative abundance.

We made PCoA plots based on both unweighted or weighted UniFrac distance matrix to visualize similarities and differences among gastric samples. Both plots suggested that gastric samples

respectively. On average, 99% of Helicobacter sequence reads were classified as Hp. Only the most abundant phyla/genera/modules in Chinese or Mexican samples are shown. All the samples from (A–C) are in the same order. The anatomical location and source of the samples are shown at the bottom of the figure.

were primarily clustered by geographic location, rather than by tissue types (**Figure 3**).

### Comparison of Non-malignant Gastric Tissue to the Other Body Sites

The average relative abundance of the top abundant genera by body sites are shown in **Table 2**. The top abundant genera are the genera with average relative abundance >0.05 in at least one body site. The top abundant genera in stomach includes Helicobacter and an unknown Enterobacteriaceae genus in either Chinese or Mexico samples, and two additional genera Streptococcus and Lactobacillus in Mexico samples. The top abundant genera in other body sites included Streptococcus, Prevotella, Haemophilus, Veillonella, and Neisseria in oral cavity, Lactobacillus in vagina, Bacteroides, Faecalibacterium, and Alistipes in stool, Staphylococcus, Corynebacterium, and Propionibacterium in both skin and nasal cavity. These top genera in each body site were considered as the genera associated with their corresponding body site (e.g., vagina\_associated genus refers to Lactobacillus). We found that the stomach microbiota was enriched with the genera associated with the oral cavity (combined relative abundance of 17.6 and 11.6% in Mexico and China samples, respectively).

Similarities and differences of taxonomic/functional profiles by body sites are shown in **Figure 4**. The principal coordinates plots based on taxonomic profiles (phylum-level) demonstrated the primary clustering of samples by body sites (**Figures 4A,B**). The stomach samples, Chinse or Mexico, largely overlapping with oral sample cluster, which was clearer when Helicobacter reads were removed (**Figures 4A,B**). Compared to the principal

TABLE 1 | Hp in GC patients for both nonmalignant and tumor gastric tissue microbiota.


coordinates plots based on taxonomic profiles, the plots based on functional profiles (KEGG modules) showed a much clearer pattern of clustering by body sites (**Figure 4A** vs. **Figure 4C**, **Figure 4B** vs. **Figure 4D**). The stomach samples, with or without Helicobacter reads removed, either Mexico or Chinese samples, did not cluster with the other body sites, but they also did not cluster with each other as closely as the samples in the other body sites. It suggested higher inter-subject dissimilarity in stomach samples than in the other body sites in functional profiles. The inter-subject dissimilarity by body sites based on Bray-Curtis distance of phylum/KEGG module profiles were then evaluated (Supplementary Figure 5). Mexican stomach samples had the highest inter-subject dissimilarity in phylum profiles. The Chinese samples, however, had inter-subject dissimilarity higher than the other body sites only when Helicobacter reads were removed (Supplementary Figure 5A). This might be due to the fact that almost all Chinese samples (94%) had Helicobacter. The inter-subject dissimilarity in functional profiles was much higher in the stomachs of both sample sets than in other body sites (Supplementary Figure 5B).

## Comparison of Gastric Tissue Microbiota Features by Hp Colonization Status

To further evaluate the gastric tissue microbiota by Hp colonization status, we removed the Helicobacter reads from the Hp+ samples (with Hp) and then compared them to the

FIGURE 3 | Comparison of gastric microbiota among different sample groups by PCoA plots based on unweighted (A) and weighted UniFrac distance (B). The color represents different sample groups as shown in the legend (non-malignant, China\_N, Mexico\_N; and tumor, China\_T, Mexico\_T).


TABLE 2 | The average relative abundance of top abundant genera by body sites and their comparison.

The stomach microbiota in this table were based on non-malignant tissue samples only. Top abundant genera refer to genera with relative abundance >0.05 in at least one body site (bolded and highlighted). These top genera in each body site were considered as genera associated with their corresponding body site. For example, oral\_associated genera include Streptococcus, Prevotella, Haemophilus, Veillonella, and Neisseria, and their combined relative abundance in stomach samples is the relative abundance of oral\_associated genera in the stomach.

Hp− samples (without Hp) for alpha diversity, beta diversity, and taxa relative abundance separately for non-malignant and tumor samples. No significant differences were observed among Mexican samples (Supplementary Table 7). A similar comparison could not be performed in the Chinese samples because too few samples were Hp− (n = 5).

In the largest study of gastric tissue to date, we investigated the gastric microbiota in sets of patients from Mexico and China. In both sets, we showed that Hp is the most abundant member of the stomach microbiota, followed by the genera that are commonly seen in the oral microbiota. The principal coordinates plots of Bray–Curtis distance matrix based on phylum-level taxonomic profiles suggested that stomach samples largely overlapped with oral samples. The principal coordinates plots based on functional profiles, however, suggested that stomach microbiota was distinct from the microbiota of other body sites, and had higher intersubject dissimilarity. We found no differences in microbiota composition by anatomic site or Hp status, although we had only limited sample size to detect differences. We did find that the relative abundance of Hp was higher in non-malignant than in tumor tissues for both Chinese and Mexican samples.

Gastric microbiota was dominated by phyla Proteobacteria in Chinese samples, and by Proteobacteria and Firmicutes in Mexican samples. This result is consistent with most previous studies based on gastric biopsy, fluid, or tissue in either healthy or cancer cases (Supplementary Table 1). Overall, in cases with high Hp relative abundance, the most abundant gastric phylum is Proteobacteria, otherwise it is Firmicutes.

GC patients, both non-cardia and cardia in Asian populations have been proposed to have chronic gastritis that leads to hypochlorhydria in the stomach (Cavaleiro-Pinto et al., 2011; Sheh and Fox, 2013). Therefore, GC patients have been hypothesized to have diminished or no colonization of Hp in the stomach (Sheh and Fox, 2013). Previous studies of gastric microbiota in GC patients were limited, small, and found inconsistent results (summarized in Supplementary Table 1). Studies from Sweden (Eun et al., 2014) and Mexico (Aviles-Jimenez et al., 2014) did not identify Hp as the dominant species in any of the samples evaluated, while four other studies in Korea, Taiwan and USA showed results consistent with our finding with Hp as the dominant species in GC samples (Dicksved et al., 2009; Eun et al., 2014; Zhang et al., 2015; Tseng et al., 2016). This difference between studies might be due to Hp prevalence heterogeneity across study samples. A recent study of 212 chronic gastritis and 103 GC patients in China that used quantitative PCR showed that the bacteria load in the gastric mucosa was increased in cancer patients compared to gastritis patients, and the bacterial load was positively correlated with Hp quantity (R = 0.38, P < 0.001), suggesting Hp colonization in GC patients (Wang et al., 2016). In the current study, we found that many Chinese and Mexican GC patients had stomachs dominated by Hp. This finding may be relevant to the decision-making of GC treatment. Endoscopic resection has been considered the first line of treatment for early GC in Korea and Japan because it is minimally invasive and effective (Chung et al., 2009; Isomoto et al., 2009). Several studies have shown a benefit for Hp eradication in reducing metachronous tumors after resection for early gastric cancer (Fukase et al., 2008; Bang et al., 2015). Our data suggest that the majority of patients diagnosed with GC in these populations have current Hp colonization and this may explain why eradication therapy at time of diagnosis may be beneficial.

Our study of gastric microbiota showed less between-sample variation in the functional profiles than in the taxonomic profiles, which is similar to a previous study in other body sites (Human Microbiome Project Consortium, 2012). This finding of less variation in function than in taxa is consistent with functional redundancy across taxa and suggests that taxonomically distinct microbes may have similar functions. Therefore, analysis of functional modules appears to provide insights that analysis of phyla alone may not be able to identify. For example, we showed that stomach microbiota was distinct from microbiotas in other body sites in functional profiles, but not in taxonomical profiles. In addition, inter-subject dissimilarity in functional profiles is much higher in stomach than in other body sites. However, it is important to note that the functional profiles were based on prediction only. Therefore, it is possible that prediction-based biases toward well-documented microbial genomes resulted from exclusion of unknown or poorly documented taxa. Further studies are needed to validate these findings.

Consistent with a study in a United States population (Bik et al., 2006), we did not observe differences in the microbiota between the antrum and corpus in the Mexican GC samples. In contrast, a Chinese study reported that gastritis patients without Hp infection had decreased Prevotella in the antrum compared to the corpus (Li et al., 2009). Larger studies in subjects without GC are needed to further compare the gastric microbiota by different anatomical sites and also control for population and health conditions.

Hp colonization may impact gastric microbiota by induction of host antimicrobial peptides (Hornsby et al., 2008), by directly killing other bacteria through the activity of its own cecropinlike peptide (Putsep et al., 1999), or by inducing physiological changes in host stomach such as pH alteration (Smolka and Backert, 2012), epithelial surface (Wroblewski et al., 2016), gastric hormones and immunologic state (Blaser and Atherton, 2004). However, the difference in the gastric microbiota by host Hp colonization status is not fully understood. Consistent with our findings, a study in the United States showed that the relative abundance of non-Hp bacteria in Hp+ subjects was not altered compared to Hp− subjects when Hp sequences were eliminated from the analysis (Bik et al., 2006). A study in China however suggested that the major influence of Hp on microbiota is the increased bacterial load in the stomach, not the relative abundance of non-Hp bacteria groups (Wang et al., 2015). In contrast, a small study of 10 Amerindians and 2 non-Amerindians using the PhyloChip reported marked differences in relative abundance of non-Hp bacteria by Hp status (Maldonado-Contreras et al., 2011). Our study examined cancer patients, and thus Hp− GC patients may have a prior history of Hp colonization. Larger studies in subjects without GC using advanced sequencing technology are needed.

As has been previously shown for tumor and matched non-malignant samples from colorectal cancer patients (Burns et al., 2015), we also found taxonomical and functional composition differences between non-malignant and tumor tissues in both Chinese and Mexican cases. This observation might suggest the change of local environment in tumor (e.g., reduction of acid secretion) compared to non-malignant tissues, which leads to Hp diminution and corresponding microbial functional changes in tumor. Whether these changes contribute to gastric carcinogenesis or tumor progression require further investigation. Our recent study of the Chinese sample set suggested that the changes in the gastric microbiota including Hp relative abundance in non-malignant tissue were associated with cancer risk factors and clinical outcomes including family history of upper gastrointestinal cancer and tumor grades (Yu et al., 2017). Similar associations were, however, not found in the tumor tissues or in the Mexican sample set (data not shown).

While this study includes noteworthy strengths, it also includes limitations. Although, it is the largest study of the gastric non-malignant tissue microbiota from GC patients to date, it includes samples from two different populations with different rates and types of GC. The non-malignant tissue samples were obtained distant but unmeasured from the tumor lesion under sterile conditions and were frozen immediately. Also, unlike most studies of gastric microbiota, we analyzed not only the taxonomic profiles and Hp relative abundance, but also virtual reconstructed functional profiles. We compared our gastric tissue data to the HMP data, which included different populations, DNA extraction techniques, and sequencing platforms. We tried to minimize the effects of these differences by restricting our comparisons to the highest and least variable taxonomic level (phylum-level) and functional entity (KEGG module). In addition, a meta-analysis of microbiota studies suggested that differences in microbial populations across body sites are larger than those driven by the experimental protocols, age, geography, and other population characteristics (Lozupone et al., 2013). Another limitation was the use of a DNA extraction method without a bead-beating step. However, we showed that although we may have missed certain bacteria with potentially hard-tobreak cell membranes, these bacteria were rare and should not have adversely affected our conclusions. Finally, our study was restricted to cancer patients and we did not have gastric samples from subjects without GC for comparison. Therefore, we cannot evaluate whether results found in GC patients generalize to subjects without GC.

## CONCLUSIONS

By analyzing the gastric tissues of two different populations with different types of GC separately, we showed that Hp was the dominant taxa in the stomach of many subjects with GC, followed by oral-associated bacteria. Comparison with other body sites suggested that stomach microbiota resembled oral microbiota in phylum-level taxonomical profiles, but not in functional profiles. Our study provided insights of gastric microbiota composition and function in GC patients.

## REFERENCES


## AUTHOR CONTRIBUTIONS

GY designed the study, analyzed the data and wrote the initial manuscript. JT, NH, RM, RH, MH, LW, CW, TD, PT, and AG performed sample collection and laboratory experiments. GY, JT, JR, PT, CA, and AG contributed to the data interpretation and manuscript revision.

## FUNDING

This work was supported by Intramural Research Program of Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health. Work in Mexico was supported by Coordinacion de Investigacion, Instituto Mexicano del Seguro Social. (Grant number FIS/IMSS/PRIOR/PROT/13/027).

### ACKNOWLEDGMENTS

We thank all the study participants. We thank B. Ma from Ravel lab for delivering the sequence data to us.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00302/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 Yu, Torres, Hu, Medrano-Guzman, Herrera-Goepfert, Humphrys, Wang, Wang, Ding, Ravel, Taylor, Abnet and Goldstein. 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.

# Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis

Gregory R. Young<sup>1</sup> \*, Darren L. Smith<sup>1</sup> , Nicholas D. Embleton<sup>2</sup> , Janet E. Berrington<sup>2</sup> , Edward C. Schwalbe<sup>1</sup> , Stephen P. Cummings <sup>3</sup> , Christopher J. van der Gast <sup>4</sup> and Clare Lanyon<sup>1</sup> \*

<sup>1</sup> Faculty of Health and Life Sciences, University of Northumbria, Newcastle upon Tyne, United Kingdom, <sup>2</sup> Newcastle Neonatal Service, Newcastle upon Tyne Hospitals NHS Foundation Trust, Newcastle upon Tyne, United Kingdom, <sup>3</sup> School of Science and Engineering, Teesside University, Middlesbrough, United Kingdom, <sup>4</sup> School of Healthcare Science, Manchester Metropolitan University, Manchester, United Kingdom

Necrotising enterocolitis (NEC) and sepsis are serious diseases of preterm infants that can result in feeding intolerance, the need for bowel resection, impaired physiological and neurological development, and high mortality rates. Neonatal healthcare improvements have allowed greater survival rates in preterm infants leading to increased numbers at risk of developing NEC and sepsis. Gut bacteria play a role in protection from or propensity to these conditions and have therefore, been studied extensively using targeted 16S rRNA gene sequencing methods. However, exact epidemiology of these conditions remain unknown and the role of the gut microbiota in NEC remains enigmatic. Many studies have confounding variables such as differing clinical intervention strategies or major methodological issues such as the inability of 16S rRNA gene sequencing methods to determine viable from non-viable taxa. Identification of viable community members is important to identify links between the microbiota and disease in the highly unstable preterm infant gut. This is especially important as remnant DNA is robust and persists in the sampling environment following cell death. Chelation of such DNA prevents downstream amplification and inclusion in microbiota characterisation. This study validates use of propidium monoazide (PMA), a DNA chelating agent that is excluded by an undamaged bacterial membrane, to reduce bias associated with 16S rRNA gene analysis of clinical stool samples. We aim to improve identification of the viable microbiota in order to increase the accuracy of clinical inferences made regarding the impact of the preterm gut microbiota on health and disease. Gut microbiota analysis was completed on stools from matched twins (n = 16) that received probiotics. Samples were treated with PMA, prior to bacterial DNA extraction. Meta-analysis highlighted a significant reduction in bacterial diversity in 68.8% of PMA treated samples as well as significantly reduced overall rare taxa abundance. Importantly, overall abundances of genera associated with protection from and propensity to NEC and sepsis such as: Bifidobacterium; Clostridium, and Staphylococcus sp. were significantly different following PMA-treatment. These results suggest non-viable cell exclusion by PMA-treatment reduces bias in gut microbiota analysis from which clinical inferences regarding patient susceptibility to NEC and sepsis are made.

Keywords: preterm, neonate, stool, microbiota, viability, propidium monoazide

#### Edited by:

Nathan W. Schmidt, University of Louisville, United States

#### Reviewed by:

Valerio Iebba, Sapienza Università di Roma, Italy Renate Lux, UCLA School of Dentistry, United States

#### \*Correspondence:

Gregory R. Young greg.young@northumbria.ac.uk Clare Lanyon clare.lanyon@northumbria.ac.uk

> Received: 17 March 2017 Accepted: 22 May 2017 Published: 06 June 2017

#### Citation:

Young GR, Smith DL, Embleton ND, Berrington JE, Schwalbe EC, Cummings SP, van der Gast CJ and Lanyon C (2017) Reducing Viability Bias in Analysis of Gut Microbiota in Preterm Infants at Risk of NEC and Sepsis. Front. Cell. Infect. Microbiol. 7:237. doi: 10.3389/fcimb.2017.00237

## INTRODUCTION

Severely preterm infants (<32 weeks) have immature immune systems (Levy, 2007; Strunk et al., 2011), improperly formed intestinal lumen (Halpern and Denning, 2015), and often feeding intolerance (Fanaro, 2013). All such characteristics increase the risk of onset of nosocomial infection, necrotising enterocolitis (NEC), and sepsis (Gregory et al., 2011). Outbreaks of NEC within the neonatal intensive care unit (NICU) (Boccia et al., 2001) and the absence of such diseases prior to bacterial colonization at birth suggest a key role of gut bacterial dysbiosis in these conditions. True causation is, however, very difficult to identify and compacted by complex to understand, highly turbulent community characteristics in the gut, probably in part affected by "routine" interventions of neonatal intensive care (antibiotic administration, feeding strategies, etc.) (Stoll et al., 1996; Hoy et al., 2000).

Targeted 16S rRNA gene sequencing technologies are used to produce microbial metadata for entire populations of a biotope with superior depth, specificity and, most importantly, in significantly less time than previous culture-based or molecular methods (Weinstock, 2012). In addition, the price of sequencing continues to decline (Caporaso et al., 2012), providing further incentives for microbiologists to employ this technique. However, 16S rRNA gene sequencing does introduce inherent biases (von Wintzingerode et al., 1997), including enrichment of particular bacterial groups before storage (Rochelle et al., 1994), insufficient or preferential disruption of certain bacterial cells (Leff et al., 1995; Schneegurt et al., 2003), introduction of sequencing artefacts such as chimeras (Wang and Wang, 1996) and inability to exclude DNA from non-viable sources (Nocker et al., 2006; Nocker and Camper, 2009; Rogers et al., 2013). Persistence of non-viable DNA is due to the stability of the molecule, which enables DNA to remain in an environment long after the originating organism has died. DNA from non-viable bacterial cells (NVBCs) can persist in the lumen of the GI tract, resulting in identification during targeted 16S rRNA gene sequencing analyses. Such bias is especially important whilst studying the highly unstable (Koenig et al., 2011; Bergstrom et al., 2014), low diversity (Tuddenham and Sears, 2015) gut bacterial communities of severely preterm infants. A technique to enable non-viable cell exclusion (NVCE), from such analyses is, therefore, an important and necessary requirement in order to reduce bias and improve the current understanding of bacterial taxa associated with NEC and sepsis.

PMA is a DNA chelating compound that cannot translocate across a viable cellular membrane (Nocker et al., 2007). Nocker et al. (2006), developed the use of Propidium Monoazide (PMA), for differentiation between viable and non-viable bacterial cells during targeted 16S rRNA gene sequencing microbiota analyses (Nocker et al., 2006; Nocker and Camper, 2009; Rogers et al., 2013). This process has been applied in microbial ecology studies of other environments, including wastewater samples (Nocker et al., 2007, 2010), human oral cavities (Sanchez et al., 2013), human adult faeces (Bae and Wuertz, 2009; Fujimoto and Watanabe, 2013), the cystic fibrosis lung (Rogers et al., 2010; Nguyen et al., 2016), and other lower lung respiratory infections (Rogers et al., 2013). The technique, however, has not yet been validated for use in the unique biotope of preterm infant stool despite vast quantities of research being published regarding this microbiota (Mshvildadze et al., 2010; Mai et al., 2011; Torrazza et al., 2013; McMurty et al., 2015). Furthermore, no studies so far have validated combining PMA treatment of this sample type in conjunction with the Schloss method for paired end targeted 16S rRNA gene sequencing (Kozich et al., 2013).

This study aims to identify and alleviate the bias associated with non-viable bacterial DNA inclusion in studies of the gut microbiota of significantly preterm infants at risk of NEC and sepsis. In doing so we hope to increase the accuracy of microbiota characterization in patients at risk of NEC and sepsis, therefore improving the quality of clinical inferences made in relation to the conditions.

The effects of PMA treatment were assessed by comparing bacterial richness, diversity, and community structure as well as individual taxa abundances within PMA-treated and untreated frozen stool samples (n = 16) when assessed using targeted paired end sequencing of the 16S rRNA gene.

### MATERIALS AND METHODS

Faecal samples were collected when available from day of life 43–81 from a set of significantly preterm twins born 25(+2) weeks gestation and at ≤710 g, enrolled on the SERVIS study at the Royal Victoria Infirmary NICU, Newcastle upon Tyne, England, with ethical permission (NRES Committee North East—Newcastle & North Tyneside 2). Both patients were administered Infloran <sup>R</sup> (Laboratorio Farmaceutico SIT, Mede, PV, ITA) probiotic supplements throughout the course of the sampling period (Bifidobacterium bifidum, Lactobacillus acidophilus). Stool was collected in sterile glass pots with sealed lids and frozen immediately on the ward. Batch collection and transportation to freezers at Northumbria University followed. Samples were stored at −80◦C until PMA treatment and DNA extraction for analysis.

## PMA Treatment and DNA Extraction

PMA was supplied by Biotium (Hayward, CA, USA), and dissolved in dimethyl sulfoxide to a stock concentration of 20 mM. Faecal samples were homogenised in 2.5 ml PBS per 0.1 g of stool (≤0.5 g), and centrifuged. The centrifuged pellet was resusupended in 2 ml PBS and split evenly to facilitate PMA-treated and untreated conditions per sample. PMA stock solution was added to a final concentration of 50µM in treated samples and the equivalent volume of PBS was added to untreated samples. PMA cross-linking was initiated by 30 min incubation on ice, in the dark with occasional mixing. Following this, samples were exposed to blue LED light at 464 nm during 30-s intervals for a total of 2 min. After light exposure, samples were centrifuged at 10,000 × g for 5 min. The supernatant was discarded and DNA extracted from the cellular pellet using MoBio PowerLyzer PowerSoil DNA Isolation Kit (Carlsbad, CA, USA), as per manufacturer's instructions.

## Nested PCR Protocol and MiSeq Analysis

Prior to paired end targeted 16S rRNA gene analysis, extracted viable DNA was amplified by PCR. Nested PCR was employed in this scenario not to increase copy number prior to sequencing but to increase impact of PMA-intercalation of DNA by blocking amplification of the whole 16S rRNA gene sequence prior to targeted sequencing of the shorter V4 region. Banihashemi et al. (2012) showed that amplification of a 200 bp fragment failed to omit dead cell signals fully from DNA based community analyses. Universal bacterial 16S rRNA gene specific primers 27f (Lane, 1991), and 1,492r (Turner et al., 1999) were used under the following conditions: initial denaturation at 95◦C for 5 min then 25 cycles of 30 s denaturation at 95◦C; primer annealing at 44.5◦C for 30 s; elongation at 72◦C for 30 s then a final elongation at 72◦C for 10 min.

PCR products were serially diluted 1:10 and paired end targeted analysis of V4 regions of the 16S rRNA gene was performed as described by Kozich et al. (2013), on the Illumina MiSeq using primers described by Caporaso et al. (2011). MiSeq 250 × 2 chemistry was used to perform the targeted 16S rRNA sequencing.

## Analysis

Sequence reads with phred-score ≥Q30 were trimmed, merged and processed in Mothur (Schloss et al., 2009), following the MiSeq SOP. Number of sequences passing Q30 in each sample are illustrated in Figure S1. Reads with phred-score <Q30 were not included in analysis. Uncorrected pairwise distances were calculated before clustering sequences in to OTUs using average neighbor joining, as recommended by Schloss and Westcott (2011). The same sequence reads were also submitted to the EBI ENA database for analysis (study accession PRJEB10326; http://www.ebi.ac.uk/ena/data/view/PRJEB10326).

Singletons were not removed from analysis to allow identification of PMA-treatment on all rare taxa identified by targeted sequencing. Normalization was not performed by rarefaction or subsampling due to the nature of the investigation. Instead relative abundances of individual taxa per sample were calculated. This is because the impact of PMA NVCE was assessed by omission of sequence reads from the community, therefore the absence of any sequence read was as informative as the presence of the same.

Per sample richness and beta-diversity was calculated using R statistical software (R\_Core\_Team., 2014) and the vegan package for community ecology (Oksanen et al., 2015). Metaanalysis (Borenstein et al., 2009) was used to compare results by treatment condition. Meta-analysis has previously been used to quantify the effect of PMA-treatment on bacterial communities of expectorated CF sputum samples (Rogers et al., 2013), allowing direct comparison of the effect of PMA-treatment between paired and unpaired samples by comparing effect size, rather than comparing means of highly variable individual samples by t-test. Each microbiota was randomly sub-sampled with bootstrapping n = 1,000 times. Standard error was reported.

SIMPER comparison of individual taxa relative abundance per treatment condition was performed using PAST (Hammer et al., 2001). Significance of results was calculated and plotted using R statistical software.

Comparison of non-frozen and frozen stool microbiotas was performed using ANOSIM and unconstrained Morisita–Horn cluster analysis.

## RESULTS

Stool samples from a set of significantly preterm twins (25+2 weeks gestation) (n = 16) receiving Infloran <sup>R</sup> probiotic supplements were subjected to PMA-treatment for comparison to an untreated control of each sample. 16S rRNA gene sequencing identified a total of 161 individual taxa producing 4.72 × 10<sup>6</sup> total reads from 16 × 2 samples.

### Identification of Common and Rare Taxa

To identify differences between common and rare taxa in PMA-treated and untreated conditions distribution abundance relationship plots were produced (**Figure 1**).

Significant positive distribution abundance relationships were observed between taxa abundance and persistence of taxa across samples in both treatment conditions (untreated: r <sup>2</sup> = 0.58, n = 120, P = <0.001; PMA-treated: r <sup>2</sup> = 0.72, n = 97, P = <0.001). Using this relationship, taxa in the upper quartile of occupancy (≥75% samples), in each treatment condition were classified as common, the remaining taxa were classified as rare.

Distribution of taxa appeared more even in the PMA-treated condition: fewer ubiquitous taxa dominate the communities in the PMA-treated condition (2 taxa); compared to the untreated condition (6 taxa).

In untreated sample conditions 6 taxa were identified as common, all of which were observed in every sample. Bifidobacterium, Enterococcus, 2 Clostridia spp., a Veillonella and an unclassified Enterobacteriaceae accounted for 77.9% of the total community member sequences. In PMA-treated samples, 8 common taxa were identified, comprising 82.2% of total community member sequences however of these, only 2 (Bifidobacterium and Enterococcus) were found in all samples. Anaerococcus and Finegoldia sp. were identified as common in PMA-treated samples but not in untreated samples.

## Effect of PMA Treatment on Bacterial Richness and Diversity

Due to the large coverage variability between the stool sample communities (m = 1.47 × 10<sup>5</sup> , SD = 1.26 × 10<sup>5</sup> ), metaanalysis was used to identify the effect size of PMA-treatment on microbiota composition by bacterial OTU richness (O∗); Shannon diversity index (H′ ); and Inverse Simpson's diversity index (1/D).

Bacterial OTU richness was variable between stool samples of the same treatment condition (untreated m = 9.1 ± 2.7, PMAtreated m = 8.8 ± 1.9). The effect of PMA-treatment on O∗ was only once greater than the significance threshold (0.2), and showed no directional consistency (**Figure 2**).

Like richness, bacterial diversity also varied between individual stool samples in the same treatment condition (**Table 1**).

Meta-analysis showed negative effect sizes of PMA-treatment on bacterial diversity in 71.9% of samples, of which 73.9% were highly significant (>0.8) (**Figures 3**, **4**). Significant negative mean

FIGURE 1 | The number of samples in which taxa were observed plotted against mean taxa abundance (log10 scale) in untreated sample (A), and PMA-treated sample (B), conditions [(A: r <sup>2</sup> = 0.58; P = <0.001), (B: r <sup>2</sup> = 0.72; P = < 0.001)]. Common taxa (>75% sample occupancy) are plotted square, rare samples (<75% sample occupancy) are plotted circle.

FIGURE 2 | (A) Bootstrapped (n = 1,000) O\* values for untreated (hollow points), and PMA-treated (solid points) conditions of all samples (1–16). Error bars are included. (B) Hedges' d effect size of PMA-treatment on O\* of all samples (1–16). Positive effect was observed in 50% of samples. Sample 6 was the only observed effect size >0.2 (small effect).

TABLE 1 | Mean and standard deviation for Shannon and inverse Simpson diversity indices for PMA-treated and untreated conditions.


overall effect sizes on both measures of diversity were observed following PMA-treatment (m: H′ = −0.95; 1/D = −1.23).

## Effect of PMA-Treatment on Individual Bacterial Taxa Abundance

To investigate the effect of PMA-treatment on observable abundance values of individual taxa relative sequence abundance was calculated. Initial analysis of the PMA-treatment effect on individual taxa abundance within the bacterial communities of stools was performed by SIMPER (**Table 2**). SIMPER provides an insight in to the variance, expressed as a percentage, between abundance of taxa from the untreated group and the PMAtreated group.

**Table 2** illustrates which taxa contributed greatest to dissimilarity of common and rare community structures between untreated and PMA-treated conditions.

Greater average dissimilarity is observed in rare (83.21%), than common (42.87), taxa. The taxon labelled Escherichia\_shigella by the SILVA database (Quast et al., 2013) appears to contribute to the average dissimilarity between non-PMA and PMA treated conditions most (63.45%), in spite of a mean abundance difference of only 1.3%. This contrasts with other taxa such as Bifidobacterium and Enterococcus, for which lower average dissimilarities of 15.20 and 12.25 are observed, however much greater mean abundance differences of 14.6 and 14.8 are found, respectively. This incongruence could be explained by variation in abundance of individual taxa between samples within the same condition. The SD of Escherichia

FIGURE 3 | (A) Bootstrapped (n = 1,000) Shannon diversity index values for untreated (hollow points), and PMA-treated (solid points), conditions for all samples (1–16). Standard error bars are included. (B) Hedges' d effect size of PMA-treatment on Shannon diversity index values of all samples (1–16). Seventy-five percentage of samples exhibit significant (>0.8) effect size, 75% of which are negative.

samples (1–16). Standard error bars are included but error is too insignificant to be visible. (B) Hedges' d effect size of PMA-treatment on inverse Simpson diversity index values of all samples (1–16). 68.8% samples exhibit significant (>0.8) effect size. 72.3% of which are negative rare (B) taxa.

abundance is much greater (untreated: m = 16.2, SD = 18.2; PMA-treated: m = 14.9, SD = 24.3), than that of Bifidobacterium (untreated: m = 41.1, SD = 15.2; PMA-treated: m = 26.5, SD = 19.5), or Enterococcus (untreated: m = 25.0, SD = 10.2; PMA-treated: m = 39.8, SD = 14.0).

To normalise for this variance Wilcoxon rank sum tests were performed to assess the similarities of mean abundance for each taxon in both conditions as identified by SIMPER analysis. The same means were used to calculate a fold change in taxa abundance between the two conditions and both parameters were plotted on a volcano plot (**Figure 5**).

The majority of taxa showed substantial fold changes in abundance following PMA-treatment; however only 6 of these fold changes pass the significance threshold (P < 0.05): Bifidobacterium; Enterobacteriaceae; Enterococcus; Clostridium; Actinomyces; and Peptoniphilus sp.

In untreated conditions Bifidobacterium and Enterobacteriaceae sp. abundances are significantly greater while Enterococcus, Clostridium, Peptoniphilus, and Actinomyces sp. abundances are significantly lower. This suggests that the presence of non-viable DNA originating from highly abundant species such as Bifidobacteria and Enterobacteriaceae could potentially mask that of less abundant species such as Enterococcus, Clostridium, Peptoniphilus, and Actinomyces sp.

As volcano plots (**Figure 5**) only represent fold change in abundance for taxa present in both sample conditions, rank abundance plots (**Figure 6**) were generated to illustrate abundance of taxa identifiable in only one treatment condition.

Fewer taxa were observed PMA-treated than untreated sample conditions. Of 111 total taxa present in only one condition 68 (61.3%), were present in untreated samples while only 43 (38.7%), were present in PMA-treated samples, representing a 22.6% reduction in presence of taxa measurable in only one condition. Levels of 2 rare taxa (Staphylococcus and Phenylobacterium), were observable at levels > 0.09% sequence abundance (almost 10 fold more than all other taxa observable in only one treatment condition), following PMA-treatment. These taxa were completely masked in the untreated condition. All other

#### TABLE 2 | SIMPER analysis of common (A) and rare (B) taxa.


Cut-off value set at taxa contributing <1% to dissimilarity between treatment conditions.

\*Denotes taxa only attributed common status in PMA-treated condition.

Cut-off set at taxa contributing <1% to dissimilarity between treatment conditions.

taxa are blue and labelled. Rare taxa are red.

taxa abundances were reduced to <0.009% following PMAtreatment. A significant difference between mean abundances of rare taxa presence between untreated (m = 0.0052), and PMAtreated (m = 0.0012), sample conditions (P = <0.001), was observed. This suggests that performing gut microbiota analysis of frozen stool by paired end targeted 16S rRNA gene sequencing without PMA-treatment could fail to identify presence of rare community members due to significant background sequence noise origniating from non-living taxa. This could explain the significant overall reduction in mean bacterial diversity following PMA-treatment observed in meta-analysis (**Figures 3**, **4**). Principle coordinate analysis of sample communities was also performed based on Bray-Curtis dissimilarity of taxa abundances (Figure S2).

To confirm the differences observed were due to genuine viable differences in the sample microbiota rather than bacterial cell death during freezing (−80◦C) one further stool sample was split (n = 10). The microbiota of frozen and non-frozen samples in untreated and PMA-treated conditions were compared. Overall 4.04 × 10<sup>6</sup> reads were recorded from 10 × 2 samples (m = 2.02 × 10<sup>5</sup> ). Sample storage had an insignificant effect on observed Bray–Curtis community similarity between treatment conditions (**Table 3**), and no separation by PMA-treatment within storage groups was observed (Figure S3). This data further supports Shaw et al. (2016), findings suggesting freeze storage of severely preterm stool samples does not significantly impact the gut microbiota observed with or without PMA-treatment.

#### DISCUSSION

The gut microbiota of significantly preterm infants held within the neonatal intensive care unit has been previously identified to be extremely changeable (Koenig et al., 2011; Bergstrom et al., 2014). Microbial communities colonising this biotope are challenged by frequent antibiotic intervention (Craft et al., 2000), administration of probiotics (AlFaleh and Anabrees, 2014),

TABLE 3 | P-values associated with ANOSIM analysis comparing Bray Curtis dissimilarity between frozen and non-frozen samples in control and PMA-treated sample conditions.


fluctuating pH due to the use of proton pump inhibitors (Omari et al., 2007, 2009), and gut lumen and immune system maturation (Israel, 1994; Levy, 2007; Strunk et al., 2011). All of these factors complex the first months of a newborn infant's life, thereby it is considered the most unstable with respect to microbiota composition. In order for clinicians to accurately assess the requirement for, and effect of, intervention strategies on infant microbial populations the analysis techniques used must be able to reliably quantify unbiased and viable microbiotas.

Currently techniques either cannot provide results within a short turnaround time at a sufficient phylogenetic resolution to assess the diversity in the gut (bacterial culture), or fail to differentiate viable from non-viable community members (Q-PCR). While RNA sequencing enables exclusive identification of genes actively transcribed by viable cells there are downstream issues regarding storage and contaminating RNAses (Zheng et al., 1996) RNA samples require collection in an RNA preservative (Mutter et al., 2004) which is not always possible in the clinic. Furthermore, use of DNA in combination with PMA eliminates the need for reverse transcription of sequences prior to analysis.

Nocker and Camper (2009), have previously shown PMAtreatment excludes DNA from non-viable cells. This study builds on those results by illustrating PMA-treatment of frozen preterm infant stool alters observable microbiota structure and diversity following paired end targeted 16S rRNA gene sequencing. This would suggest inclusion of non-viable community members during preterm infant stool microbiota analysis introduces a bias. Additionally, DNA from non-viable cells can have significant impact on individual taxa quantification. We propose it may be necessary to employ the use of PMA as a tool for NVCE in 16S rRNA gene sequencing based microbiota analysis. Effects of PMA NVCE should not be attributed to cell death during storage as no difference in PMA effect was observed between frozen and fresh stool samples. It is probable that the changes in abundance illustrated by PMA NVCE are caused by antibiotic, probiotic, or other clinical interventions however further study is required to confirm this.

Importantly, this study illustrates that the presence of particular, clinically relevant taxa may be either over-represented (Bifidobacterium), or under-represented (Clostridium, Staphylococcus), in the absence of PMA-treatment. This is most probably due to the suppression of DNA sequence reads from rare taxa by dominant taxa as illustrated by the reduced bacterial diversity and presence of rare taxa observed in PMA-treated samples.

These findings are of particular relevance in the gut microbiota of the preterm infants analysed in this study due to administration of probiotic supplements. While Bifidobacterium remained ubiquitous and abundant across samples in both treatment conditions it has been shown in several studies (Alander et al., 2001; Charbonneau et al., 2013; Rattanaprasert et al., 2014) that administered probiotic strains often fail to engraft long-term. This would make PMA treatment extremely important for analysis of future intervention trials of this manner. Maldonado-Gómez María et al. (2016), showed that presence of phylogenetically or functionally similar keystone species can prevent engraftment of probiotic strains. The results of this study suggest Bifidobacteria within the probiotics may not maintain viability throughout the entire GI tract; a further possible reason for this failed engraftment. We demonstrate persistent DNA from non-viable Bifidobacteria may conceal the presence of less abundant, transient colonisers with the potential to confound clinical inferences drawn from 16S rRNA gene sequencing data. Further work should compare the functional profiles of the probiotic Bifidobacterium strain in Infloran <sup>R</sup> and the bacterial metagenomes of patients administered the supplement as well as investigating the community engraftment potential of the specific probiotic strains administered to patients enrolled in this study.

This study has deliberately selected a pair of twins with good longitudinal sampling to evaluate the effect of PMA treatment on observable microbiota members, however the number of actual samples (n = 16), is relatively small and recruitment purely convenience based. In lieu of the individual taxa for which significant changes in abundance are observed in this study may not be replicated in repeated studies, dependant on viable and non-viable taxa abundances. Specifically, Bifidobacterium may not necessarily be observed at significantly altered abundances in microbiota of patients not receiving probiotics or in patients with a greater engraftment potential. Given the key role of Bifidobacteria in preterm gut health, this requires further exploration. We stress that use of PMA need not be limited to that of preterm infant stool but could be applied to any unstable environment where clinical microbiota intervention is employed or in which abundance of community members may be regularly changeable. Further studies may wish to explore the use of such techniques in these environments.

Consideration should be granted that the use of a nested PCR technique represents a potential source of amplification bias in populations of low bacterial load (Yu et al., 2015). In contrast, Fan et al. (2009), have demonstrated that use of a 25 cycle nested PCR does not significantly affect observable bacterial communities. Moreover, nested PCR was employed in this study to increase the inhibitory capacity of PMA on chelated DNA, rather than increase identifiable sequences.

One possible reason for the widespread disregard of PMA use for NVCE could be the specification of non-viable cells solely as membrane-compromised cells using this method. Contreras et al. (2011), describe membrane integrity as a "conservative parameter" for viability identification, explaining inability to culture bacteria occurs sooner than membrane denaturation in heat-killed cells. We propose conservative NVCE is more appealing in this clinical context than nihil NVCE, in which nonviable DNA persists and can bias results or exaggerated NVCE, where community members may be excluded from analysis while still viable.

This study represents the first time PMA-treatment has been combined with paired end targeted 16S rRNA gene sequence analysis of a gut microbiota using the methods described by Kozich et al. (2013). Future research should focus on validation of this method of analysis in a larger sample cohort to include greater inter-sample microbiota variation. Analysis of probiotic and commensal bacterial viability throughout the preterm infant GI tract would be another logical progression from this work.

## ETHICS STATEMENT

This study was carried out in accordance with the recommendations of "NRES Committee North East – Newcastle & North Tyneside 2" 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 "NRES Committee North East – Newcastle & North Tyneside 2."

## AUTHOR CONTRIBUTIONS

GY, Cv, and CL conceived the study. NE and JB collected samples and clinical data. GY designed the study and performed the experiments. DS ran the sequencing. GY, Cv, ES, and SC analysed the data. GY, DS, and Cv wrote the paper. All authors proof read and approved the paper prior to submission.

## FUNDING

This work was supported by Northumbria University [grant number 10031605/2 awarded to GY].

## ACKNOWLEDGMENTS

The authors would like to acknowledge the work of the clinical staff of the neonatal intensive care unit at the Royal Victoria Infirmary, Newcastle for collection and storage of samples. Particular thanks go to Julie Groombridge, the research nurse at the RVI NICU for organisation and sample labelling. Further personal acknowledgement must be given to Dr. Andrew Nelson, Northumbria University, for being a great antagonist: challenging and questioning the findings; to help improve the validity of the research. Sequencing of the 16S rRNA gene was performed by the NUOMICS sequencing service.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00237/full#supplementary-material

REFERENCES


Young et al. Preterm Stool Bacterial Viability Determination

infants with pathological acid reflux. J. Pediatr. Gastroenterol. Nutr. 44, 41–44. doi: 10.1097/01.mpg.0000252190.97545.07


**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 Young, Smith, Embleton, Berrington, Schwalbe, Cummings, van der Gast and Lanyon. 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 NAG Sensor NagC Regulates LEE Gene Expression and Contributes to Gut Colonization by Escherichia coli O157:H7

Guillaume Le Bihan<sup>1</sup> , Jean-Félix Sicard<sup>1</sup> , Philippe Garneau<sup>1</sup> , Annick Bernalier-Donadille<sup>2</sup> , Alain P. Gobert <sup>2</sup> , Annie Garrivier <sup>2</sup> , Christine Martin<sup>2</sup> , Anthony G. Hay <sup>3</sup> , Francis Beaudry <sup>4</sup> , Josée Harel <sup>1</sup> \* and Grégory Jubelin<sup>2</sup> \*

<sup>1</sup> Faculté de Médecine Vétérinaire, Centre de Recherche en Infectiologie Porcine et Aviaire, Université de Montréal, Saint-Hyacinthe, QC, Canada, <sup>2</sup> INRA, Université Clermont Auvergne, MEDIS, Clermont-Ferrand, France, <sup>3</sup> Department of Microbiology, Cornell University, Ithaca, NY, USA, <sup>4</sup> Groupe de Recherche en Pharmacologie Animal du Québec, Département de Biomédecine Vétérinaire, Faculté de Médecine Vétérinaire, Université de Montréal, Saint-Hyacinthe, QC, Canada

Edited by:

Pascale Alard, University of Louisville, USA

#### Reviewed by:

Fernando Navarro-Garcia, Center for Advanced Research, The National Polytechnic Institute, Cinvestav-IPN, Mexico Mauricio J. Farfan, Universidad de Chile, Chile

#### \*Correspondence:

Josée Harel josee.harel@umontreal.ca Grégory Jubelin gregory.jubelin@inra.fr

Received: 27 January 2017 Accepted: 31 March 2017 Published: 24 April 2017

#### Citation:

Le Bihan G, Sicard J-F, Garneau P, Bernalier-Donadille A, Gobert AP, Garrivier A, Martin C, Hay AG, Beaudry F, Harel J and Jubelin G (2017) The NAG Sensor NagC Regulates LEE Gene Expression and Contributes to Gut Colonization by Escherichia coli O157:H7. Front. Cell. Infect. Microbiol. 7:134. doi: 10.3389/fcimb.2017.00134 Enterohemorrhagic Escherichia coli (EHEC) O157:H7 are human pathogens responsible for bloody diarrhea and renal failures. EHEC employ a type 3 secretion system to attach directly to the human colonic epithelium. This structure is encoded by the locus of enterocyte effacement (LEE) whose expression is regulated in response to specific nutrients. In this study, we show that the mucin-derived sugars N-acetylglucosamine (NAG) and N-acetylneuraminic acid (NANA) inhibit EHEC adhesion to epithelial cells through down-regulation of LEE expression. The effect of NAG and NANA is dependent on NagC, a transcriptional repressor of the NAG catabolism in E. coli. We show that NagC is an activator of the LEE1 operon and a critical regulator for the colonization of mice intestine by EHEC. Finally, we demonstrate that NAG and NANA as well as the metabolic activity of Bacteroides thetaiotaomicron affect the in vivo fitness of EHEC in a NagC-dependent manner. This study highlights the role of NagC in coordinating metabolism and LEE expression in EHEC and in promoting EHEC colonization in vivo.

Keywords: NagC, LEE, EHEC, N-acetylglucosamine (or eventually NAG), N-acetylneuraminic acid (or eventually NANA)

## INTRODUCTION

Escherichia coli O157:H7 are human foodborne pathogens responsible for outbreaks mostly in developed countries. Infections by EHEC occur following ingestion of contaminated food and provoke symptoms ranging from watery or bloody diarrhea to hemolytic and uremic syndrome (HUS). A range of virulence factors are involved in EHEC O157:H7 pathogenicity including the Shiga-toxin which is associated with development of HUS, and the T3SS which enables the pathogen to attach to the intestinal epithelium and cause diarrhea (Kaper et al., 2004).

T3SS-encoding genes are gathered into the locus of enterocyte effacement (LEE) that is composed of five operons (LEE1 to LEE5) which encode for structural proteins, regulators, chaperones and effectors that are secreted into the host cells (Kaper et al., 2004). The first gene of the LEE, ler, encodes an activator that regulates the five major LEE operons. Expression of ler is controlled by several regulators in response to intestinal metabolites, such as bacterial waste products (Nakanishi et al., 2009), quorum-sensing molecules (Sircili et al., 2004), hormones (Walters and Sperandio, 2006), biotin (Yang et al., 2015), fucose (Pacheco et al., 2012), and ethanolamine (Kendall et al., 2012).

During its infectious cycle, EHEC O157:H7 encounters a large amount of mucin-derived sugars (Fabich et al., 2008; Bertin et al., 2013). Mucin is part of the mucous layer covering the intestinal epithelium and is heavily O-glycosylated. The mucous layer is a physical barrier that limits contact between bacteria and host epithelial cells (McGuckin et al., 2011). By producing specific glycosidases, several species of the gut microbiota release sugars from O-glycans into the intestinal lumen (Bertin et al., 2013; Ng et al., 2013; Elhenawy et al., 2014). Released mucin sugars, including N-acetylglucosamine (NAG), N-acetylneuraminic acid (NANA), galactose, fucose, mannose and N-acetylgalactosamine, represent an important reservoir of nutrients that promotes the growth of commensal and pathogenic bacteria including E. coli (Fabich et al., 2008; Bertin et al., 2013; Conway and Cohen, 2015). Escherichia coli and more particularly EHEC O157:H7 are able to concomitantly metabolize in vitro up to nine mucin sugars at a time, and preferentially use NAG and galactose (Fabich et al., 2008; Bertin et al., 2013; Conway and Cohen, 2015).

Genes involved in the catabolism of sugars are often regulated by proteins responding to the presence of their cognate sugar. For example, the regulator NagC, known as a repressor of NAG and galactose catabolism, is a NAG-6 phosphate (NAG-6P) sensing protein, NAG-6P being produced during the catabolism of NAG and NANA (Plumbridge, 1991; El Qaidi et al., 2009). When NAG-6P concentrations are low, NagC acts as a DNA binding protein, an activity that is lost with high intracellular NAG-6P concentration (Plumbridge and Kolb, 1991; Sohanpal et al., 2004). In addition to the role they play as nutrients, some mucin sugars can act as regulatory signals that influence bacterial colonization and adherence to cells (Sohanpal et al., 2004; Barnhart et al., 2006; Pacheco et al., 2012).

Previously, we have shown that EHEC O157:H7 respond to the metabolic activity of the human gut microbiota by activating the expression of genes required for NANA utilization and by down-regulating the expression of the LEE genes (Le Bihan et al., 2015). In this study, the effect of NANA and NAG on the adhesion phenotype of EHEC O157:H7 was examined. We found that NANA and NAG are inhibitors of EHEC O157:H7 adhesion to epithelial cells. We demonstrated that NANA and NAG reduce the expression of the five LEE operons in a NagCdependent way. Mutation in nagC diminished the expression of LEE genes. In addition, NagC was shown to bind directly to the LEE1 promoter region, thereby could influence expression of ler gene, which encodes the LEE master regulator. We also show that NagC promotes EHEC colonization of mouse intestine. Further, we demonstrate that exogenous addition of NAG into the intestine or gavage with the mucin degrading commensal Bacteroides thetaiotaomicron modulates the fitness of EHEC in vivo in a NagC-dependent manner. Taken together, our data indicate that NagC coordinates the catabolism of mucus-derived sugars and T3SS production, and promotes EHEC intestinal colonization.

## MATERIALS AND METHODS

#### Bacteria, Mutagenesis, and Growth Conditions

Strains and plasmids are listed in Table EV1. The EHEC O157:H7 strain EDL933 (O'Brien et al., 1983) was used in this study. B. thetaiotamicron strain VPI-5482 was grown anaerobically at 37◦C in a complex medium containing clarified rumen fluid (Leedle and Hespell, 1980). The medium was prepared, dispensed and inoculated by using strictly anaerobic techniques in Balch tubes. The EDL933 1nagC and 1nanR mutants were generated by allelic exchange using a suicide vector as described in EV Methods. When required, the growth medium was supplemented with kanamycin (25 mg/ml), ampicillin (50 mg/ml), NANA (0.1 mM or 1 mM), or NAG (0.1 mM or 1 mM) (Sigma Aldrich).

#### Beta-Galactosidase Assays

The entire intergenic region between ler (LEE1) and espG (bp −1,225 to +19) containing two ler promoters (Sperandio et al., 2002; Porter et al., 2005) was inserted upstream of lacZ in pRS551 (see EV Methods). The resulting plasmid pGLB was introduced into EDL933 or its isogenic mutants. After growth in DMEM with or without NANA or NAG until OD<sup>600</sup> of 0.6, β-galactosidase assays were performed as described previously (Miller, 1972). Student t-tests were performed to determine statistical significance.

### Quantitative Real Time PCR (qRT-PCR)

Bacteria were harvested at OD<sup>600</sup> of 0.6 and RNA was extracted as previously described (Le Bihan et al., 2015). cDNAs were synthesized from 10 µg RNA using reverse transcriptase. The concentration of cDNA samples was then adjusted to 25 ng/µL. A standard curve was performed for genes of interest to determine the copy number of targeted transcripts in 50 ng of cDNA. Primers used in this study are listed in Table EV2. Results are presented as the ratios between the cDNA copy number of the gene of interest and the cDNA copy number of the housekeeping gene. Student t-tests were performed to calculate p-values.

#### Western Blotting

Western blot analyses were performed with slight modifications to those previously described (Chekabab et al., 2014). Culture supernatants (8 ml) were harvested and supplemented with 1µg Bovine Serum Albumin (BSA). Proteins were precipitated overnight at 4◦C using 10% trichloroacetic acid and sodium deoxycholate, pelleted by centrifugation, washed with acetone, resuspended in SDS sample buffer and boiled. Proteins were then run on 14% SDS-PAGE gels and transferred to nitrocellulose membranes. Protein transfer including BSA was assessed using Ponceau S dye. The EspB protein was revealed using a rabbit EspB specific polyclonal antiserum (1:2,000) and a goat antirabbit HRP-linked secondary antibody (Bio-Rad Laboratories, Hercules, CA).

#### Cell Culture and Bacterial Infections

Epithelial cell lines HeLa, HCT-8, and HCT-116 were maintained in MEM with 10% FBS, 100 U/mL penicillin and 100 mg/mL streptomycin at 37◦C under 5% CO2(Branchu et al., 2014). Cells were seeded into 6-well plates (5 × 10<sup>5</sup> cells/well) and grown for 24 h without antibiotics. Bacteria were pre-grown in DMEM in the presence or absence of NANA or NAG before the infection assays. Cells were then washed and infected with bacteria with an MOI of 10 for 90 min, in the presence or absence of NANA or NAG. Following infection, cells were washed 3 times with DPBS, trypsinized for 5 min at 37◦C, pelleted by a low-speed centrifugation (100 g; 3 min). The pellet was washed once to ensure the plating of adherent bacteria only. Results are presented as the percentage of adherent bacteria as compared to the wildtype strain incubated without NANA or NAG. The EDL933 1escN strain that does not produce the T3SS (Deng et al., 2004) was used as a negative control. Student t-tests were performed to determine the significance.

#### Electrophoretic Mobility Shift Assays (EMSA)

The EMSA was adapted from previous report (Chekabab et al., 2014). The EMSA reaction mix consisted of purified NagC at the desired concentration (0.5–2.5 µM), 50 nM of 5′ 6-FAM labeled pLEE1 probe, 0.1 mg/mL calf thymus DNA and 0.1 mg/mL BSA in EMSA buffer (50 mM NaCl, 20 mM Tris, pH 7.4, 0.02% v/v sodium azide). Reactions were incubated for 30 min at 25◦C, and then loaded onto a 12% native polyacrylamide gel running at 120 V in 1x TBE buffer. The forward primer contained a 5′ 6- FAM fluorescein tag. Competitive EMSA assay was done with 50 nM of 6-FAM labeled pLEE1 probe and unlabeled probes corresponding to PLEE1, Pkan (negative control) or PnagB−nagE (positive control). The ratio "cold probe/labeled probe" was 10/1.

#### DNAse Footprinting

DNase I footprinting of free DNA and DNA–protein complexes was performed as described (El Qaidi et al., 2009; Graveline et al., 2011). The DNA fragment corresponding to the ler regulatory region (259 bp) was amplified using primers listed in Table EV2, alternately end labeled with <sup>32</sup>P (140,000 cpm, 0.6 nM). Each endlabeled amplicon was subsequently incubated in a total volume of 80 µl in binding buffer (25 mM Hepes (pH 8.0), 100 mM K glutamate (pH 8.0), 0.5 mg/ml BSA). After incubation with purified NagC for 10 min at room temperature, 2 µl of DNase I (1.3 U/ml; New England BioLabs) containing 5 mM CaCl<sup>2</sup> and 25 mM MgCl<sup>2</sup> was added for 5 min. The reaction was stopped by the addition of 150 µL of phenol/chloroform/isoamyl alcohol and 350 µl of stop buffer (0.5 M Na acetate pH 5.0, 2.5 mM EDTA, 10µg/ml Salmon sperm DNA) to each sample. DNA fragments were precipitated in ethanol, and amounts with equivalent cpm (5.10<sup>4</sup> ) from each reaction were loaded onto 6% polyacrylamide– 7 M urea gels. Maxam-Gilbert A+G reactions were carried out on the appropriate 32 P-labeled DNA fragments, and the products loaded alongside the DNase I footprinting reaction mixtures. The gels were dried and analyzed by autoradiography. A control footprinting experiment realized with nagE-nagB regulatory region and with NagC (Figure EV5).

#### Mice Infection

BALB/c mice were purchased from Janvier Labs (Le-Genest-St-Isle, France). Sets of 5 mice aged 5 weeks were given drinking water containing streptomycin sulfate (5 g/l) throughout the experiment. On day 1 following the addition of streptomycin, each mouse was infected intragastrically with 100µl of a mix containing 10<sup>7</sup> each of EDL933 Sm<sup>R</sup> and EDL933 Sm<sup>R</sup> 1nagC strains. Mice treated with B. thetaiotaomicron were gavaged daily with 5 × 10<sup>9</sup> cells of B. thetaiotaomicron strain VPI-5482, starting from day 1 before EHEC infection to day 7. At indicated time points, fecal or tissue samples were collected, homogenized in PBS and subsequently diluted before plating on LB-streptomycin agar plates and LB-streptomycin-kanamycin agar plates. Output ratios were calculated for each time point and competitive indices were obtained by dividing the output ratio by the input ratio. A One-way ANOVA was performed to determine the significance. For NANA and NAG quantification in intestinal contents, see EV Methods.

#### Ethics Statement

All animal experiments were reviewed and approved by the Auvergne Committee for Animal Experimentation (C2E2A). All procedures were carried out according to the European directives for the protection of animals used for scientific purposes, 2010/63/EU, and to the guidelines of the local ethics committee.

## RESULTS

#### NANA and NAG Inhibit EHEC O157:H7 Adhesion to Epithelial Cells by Repressing LEE Genes

The effect of NANA and NAG on EHEC adhesion was examined by measuring the ability of the EHEC O157:H7 EDL933 strain to adhere to cultured epithelial cells in the presence or absence of NANA or NAG. Our data showed that EDL933 adhesion to HeLa cells was significantly decreased by 40 ± 21 or 23 ± 11% in presence of 1 mM NAG or NANA, respectively (**Figure 1A**; Figure EV1). At 0.1 mM, only NAG significantly decreased the number of cell-attached bacteria (53 ± 22%).

The adhesion of EHEC O157:H7 to HeLa cells is mainly driven by the production of a T3SS, as previously demonstrated (Branchu et al., 2014) and as verified using the 1escN strain which is defective in the production of the T3SS (Figure EV1). Thus, we investigated the effect of NANA and NAG on LEE gene expression. As a first step, strain EDL933 carrying a PLEE1:lacZ fusion was cultured in the presence of different concentrations of the sugars. The expression of LEE1 was significantly repressed in the presence of either 1 mM NANA or 0.1 and 1 mM NAG (**Figure 1B**), but not at 0.01 mM of either sugar. Next, we examined the expression of genes from the five LEE operons and observed that expression of ler (LEE1), sepZ (LEE2), escV (LEE3), espB (LEE4), tir, and eae (LEE5) was significantly lower in presence of 1 mM NANA or NAG (**Figure 1C**). Consistent with the decreased expression of the LEE4 gene, the secretion of the effector EspB, encoded by the LEE4, was dose dependently reduced when EDL933 was incubated with 0.1 and 1 mM of

NANA or NAG. (B) β-galactosidase assays using the PLEE1:lacZ transcriptional fusion integrated into EDL933. EDL933 was grown in DMEM with or without NANA or NAG 0.1 or 1 mM and cells were harvested at OD600 = 0.6. Results are presented as Miller Units. (C) qRT-PCR measurement of LEE gene expression in DMEM with or without NANA or NAG. Results are shown as the ratio copy number of the LEE transcripts/copy number of rpoA transcripts. (D) Western blot analysis of the EspB secretion by EDL933 grown in DMEM with or without NANA or NAG. BSA was used as a loading control. n ≥ 3, ns for non-significant, \*p < 0.05, \*\*p < 0.01, and \*\*\*p < 0.001.

either sugar (**Figure 1D**). Taken together, these data indicate that NANA and NAG inhibit the adhesion of EDL933 to epithelial cells and repress T3SS encoding genes. In addition no significant difference was observed upon addition of other mucin sugars, such as mannose, galactose, N-acetylgalactosamine and glucuronate on the expression of ler (Figure EV2).

### Repression of LEE Gene Transcription by NANA and NAG Is NagC-Dependent

Activation of the metabolic pathways required for the catabolism of NANA and NAG influence the activity of two transcriptional regulators, NanR and NagC (**Figure 2A**). Intracellular NANA inactivates NanR while NAG-6P derived from the catabolic conversion of both NANA and NAG inactivates NagC. To evaluate the role of NanR and NagC on LEE1 promoter activity, the PLEE1:lacZ fusion was introduced into 1nanR and 1nagC mutants. The nagC deletion led to a significant decrease of the activity of PLEE1 whereas the deletion of nanR had no significant effect (**Figure 2B**). Moreover, the addition of NANA or NAG still repressed the activity of PLEE1 in the 1nanR mutant whereas it did not in the 1nagC mutant. These results indicate that LEE1 repression by NANA and NAG was NagCdependent but NanR-independent. We further demonstrated that a nagC deletion also impairs the expression of escV, sepZ, espB, tir, and eae genes with a fold change ranging from −2.9 to −5.5 (**Figure 2C**), as well as secretion of EspB (**Figure 2D**). Consequently, the 1nagC mutant adhered less to epithelial cells, using HeLa and the two human intestinal cell lines HCT-8 and HCT-116 (**Figure 2E**; Figures EV1, EV3). Importantly, LEE gene expression, EspB secretion and adhesion levels were restored to a wild-type status in a 1nagC-complemented strain. Additionally, we assessed the potential involvement of NagC in the transcription of other genes encoding adhesins in EDL933. Neither fliC (flagellin), ycbQ (fimbriae) (Samadder et al., 2009), hcpA (hemorrhagic coli pilus), espP (autotransporter), lpfA (long polar fimbriae), csgA (curli) nor pgaB (poly-β-1,6- N-acetyl-D-glucosamine) were differentially expressed between wild type and 1nagC strains (Supplementary Figure EV4). Since NagC does not affect the synthesis of these adhesins, it suggests that the defect in cell adhesion observed with the nagC mutant is mainly driven by a reduced production of the T3SS.

#### NagC Interacts with LEE1 Promoter Region In vitro

Using a NagC consensus DNA binding site generated from seven NagC binding sequences, we identified a single putative NagC binding site in the LEE1 promoter region (5′ -GTATTTTACACATTAGAAAAAAG-3′ ) located at a position that overlaps the −10 box of the distal promoter (**Figure 3A**). The binding of NagC to the LEE1 promoter was first investigated by EMSA that showed that NagC forms a specific low-mobility complex with the LEE1 promoter as previously observed with NagC interacting with type 1 fimbriae fim intergenic region upstream of the fimB promoter (Sohanpal et al., 2004). Competition EMSA using an unrelated probe (Pkan) demonstrated that NagC binding to the LEE1 promoter is specific (**Figure 3B**). Further, DNase footprinting experiments confirmed that NagC bound to the predicted NagC binding site (**Figure 3C**). Consistent with the expected specificity, a single base substitution at position 18 of the putative NagC binding site prevented DNAse protection by NagC (**Figure 3D**). An additional DNase footprinting control experiment using nagBE intergenic region as expected showed clear protection zones indicating the functional activity of NagC (Figure EV5). These findings demonstrate that NagC interacts with the LEE1 promoter region in a sequence specific manner. Interestingly, the NagC-binding sequence in the promoter of ler is conserved among other EHEC O157:H7 strains and this is correlated with ler repression in the presence of 1 mM of NAG (Figure EV6). EHEC, EPEC,

or C. rodentium strains with degenerated NagC binding site in the LEE1 promoter region were insensitive to NAG exposure.

## Mucin-Derived Sugars Sensing by NagC Is Important for Successful Colonization in Mice

To assess if NagC regulates the gut colonization process, we co-infected mice with an equal mixture of wild-type EDL933 and the 1nagC mutant and followed the outcome of each strain overtime. We observed a marked increase in the wildtype strain over the 1nagC mutant in the feces at days 6 and 8 post-infection (competitive index (CI) of 13 ± 5 and 20 ± 6, respectively), as well as in the cecal content at day 8 (CI of 270 ± 76) (**Figures 4A,B**). The competitive advantage of the wild-type strain was also recorded for bacteria adhering to cecal and colonic mucosa (Figure EV7). These data demonstrate that the deletion of nagC greatly impaired the ability of EDL933 to colonize the intestinal tract of mice.

We next sought to determine if the concentrations of NANA and NAG may alter EDL933 fitness in vivo through the modulation of NagC activity. For that, the drinking water of WT/1nagC-infected mice was supplemented with purified NANA or NAG. Supplementation led to increased sugar concentrations in the cecal content of uninfected mice but not in the cecal content of infected mice (**Figure 4C**). Interestingly, NANA and NAG concentrations also significantly decreased in supplemented mice upon infection, with fold-change of 4.9- and 1.9, respectively, suggesting that EDL933 consumes NANA and NAG in the intestine of mice. In these conditions, NAG supplementation significantly decreased the competitive advantage of the wild-type strain over the 1nagC mutant by a factor 7.1 (**Figure 4B**). Co-infected mice were also subjected to a daily gavage with B. thetaiotaomicron to see if the behavior of EHEC is modulated by the population level of a mucin degrading bacterium. Gavage of mice did not change NANA concentration in the gut of infected mice but led to a 1.9-fold increase of NAG concentration (**Figure 5**). Importantly, we observed that the competitive index between the wild-type strain and the 1nagC mutant significantly dropped from 297 in control mice to 54 in B. thetaiotaomicron–treated animals (**Figure 5**). Overall, our findings indicate that NAG concentration in the intestine, derived notably from activity of mucin degrading commensals, such as B. thetaiotaomicron, affects the fitness of EHEC in vivo in a NagC-dependent manner.

## DISCUSSION

This work demonstrates that the host mucin-derived sugars NAG and NANA inhibit the expression level of LEE genes in EHEC O157:H7 strain EDL933 and, consequently, inhibit the ability of the pathogen to adhere to epithelial cells in vitro. NAG and NANA are known to be used as carbon sources by commensal E. coli and EHEC O157:H7 in the gut (Fabich et al., 2008; Bertin et al., 2013, 2014; Conway and Cohen, 2015). Their catabolism induces transcriptional responses mediated by NanR and/or NagC regulatory proteins with NanR controlling NANA catabolism and NagC controlling both NAG and galactose catabolism (Plumbridge, 1991; El Qaidi et al., 2009). The role of NagC as a repressor of the expression of nagE, nagB and galP encoding the NAG PTS permease, the glucosamine-6-P deaminase, and the major galactose transporter respectively, was confirmed in EDL933 (Figure EV8). We also demonstrated that NagC controls the expression level of LEE genes in EDL933 through a direct activation of LEE1 gene transcription. NagC has been shown to modulate the expression of distinct adhesins in other E. coli strains. Indeed, NagC, together with NanR, activates the expression of fimB encoding for a recombinase required for the expression of the type I fimbriae in K-12 strain MG1655 (McClain et al., 1991; Sohanpal et al., 2004). This is not the case in EDL933 since this strain does not produce type 1 fimbriae due to a 16-bp deletion in the regulatory switch region of fimA (Vogeleer et al., 2015). While we observed no change in EDL933, a deletion of nagC has been shown to decrease expression of csgAB and csgDEFG genes and curli production in strain C600, though the mechanism remains unknown (Barnhart et al., 2006). Overall these observations suggest that NagC has served a common role as a regulator of adhesin expression during the evolution of several E. coli strains.

Most operons known to be controlled by NagC require two sites for NagC to function (nagE-B, chb, glmU, fimB) so that cooperative binding to two sites through DNA looping is necessary for regulation (Plumbridge, 1996; Sohanpal et al., 2004; El Qaidi et al., 2009; Brechemier-Baey et al., 2015). However, like the galP promoter, another target of NagC (El Qaidi et al., 2009), only one potential NagC operator is found in LEE1 promoter region of EHEC strain EDL933. Interestingly, this NagC sequence was conserved in other O157:H7 strains but not in other LEE encoding pathogens, such as EPEC E2348/69 or C. rodentium ICC168. Yet it is not clear how NagC activates the LEE1 promoter. As proposed by authors working on fimB and galP, NagC could contact RNA polymerase directly (or another regulatory protein bound closer to the LEE1 promoter region) to enhance transcription activation, or that the nucleoprotein complex that includes NagC and other regulators alters the DNA structure nearer the promoter in such a way as to facilitate transcription initiation (Sohanpal et al., 2004, 2007; El Qaidi and Plumbridge, 2008; El Qaidi et al., 2009).

By regulating genes involved in sugar catabolism and genes involved in T3SS production, NagC is likely to influence the behavior of EHEC during an infection. Indeed, we demonstrated by co-infection experiments that deletion of nagC strongly affect the fitness of EHEC in the digestive tract of infected mice. Moreover, the addition of NAG in the drinking water of infected mice reduced the competitive advantage of the wild type strain over the 1nagC mutant. This suggests that intestinal concentration of NAG modulates NagC activation and therefore expression level of NagC-dependent genes, affecting the fitness of EHEC in vivo. In the gut, some commensal species expressing mucinolytic enzymes can degrade mucins from the outer layer of mucus and release free carbohydrates into the intestinal lumen (Xu et al., 2003; Elhenawy et al., 2014; Tailford et al., 2015). These sugars can then be consumed by members of the gut

in red.

microbiota (Derrien et al., 2010). By affecting the concentration of free NAG and NANA available in the digestive tract, gut bacterial species expressing sialidase or N-acetylglucosaminidase might therefore affect the fitness of EHEC through a modulation of NagC activity. Indeed, we showed that the metabolic activity of the mucin degrader B. thetaiotaomicron when gavaged to coinfected mice increased the concentration of NAG and reduced the competitive advantage of the wild-type strain over the nagC mutant. To our knowledge, no information is available on NAG and NANA concentration in human intestine. However, its concentration probably fluctuates within the gastrointestinal tract since there is high variations in terms of (i) abundance and types of mucins; (ii) patterns of O-glycosylation and (iii) bioavailability of carbohydrates (free or mucin-linked forms). In **Figures 4**, **5**, the concentration of NANA and NAG in the cecal content of uninfected mouse was 0.27 mM (500 nmol/g) and 0.12 mM (225 nmol/g), respectively. In another study, NANA and NAG were quantified in the bovine small intestine content to be 0.1 and 0.45 mM (Bertin et al., 2013). Altogether, it gives information on physiological NAG and NANA concentrations in

indicated time points. (B) Competitive indices WT/1nagC obtained at day 8 in the cecal contents of mice provided with water with or without NANA 0.05% or NAG 0.5%. (C) Concentration of NANA and NAG in the cecal contents of non-infected mice or EHEC-infected mice provided with or without either NANA 0.05% or NAG 0.5%. \*p < 0.05, \*\*p < 0.01, \*\*\*p < 0.001, and \*\*\*\*p < 0.0001.

digestive tracts and indicates that concentrations used in our in vitro studies were relevant.

If we determined that NagC is essential for the fitness of EHEC in the digestive tract of mice, we have no information about the NagC-regulated genes involved in fitness alteration. Fabich et al. demonstrated that a mutation in gene nagE encoding the NAG transporter, causes a colonization defect for EDL933 in infected mice, indicating that NAG is utilized by the pathogen in the digestive tract of mice. In contrast, a mutation in nanAT where nanT encodes the NANA transporter, has no impact on colonization efficiency (Fabich et al., 2008), indicating that NANA catabolism is not essential for a good colonization of mice gut. In contrast to nagE and nanAT mutants that are unable to internalize NAG and NANA respectively, nagC mutant is still able to uptake and catabolize both sugars since gene deletion leads to an upregulation of nagE and nagBACD gene expression (Figure EV8). In addition to sugar catabolism and T3SS related genes, NagC also probably controls the expression of other genes in EHEC, that might be involved in EHEC fitness. One example is the gene z2210 (Figure EV9), which encodes a putative sulfatase that might be involved in mucus degradation since secreted mucins are heavily sulfated (Nieuw Amerongen et al., 1998). NagC of Vibrio fischeri was shown to facilitate colonization of the light organ of the squid Euprymna scolopes (Miyashiro et al., 2011). Sun Y et al. proposed that in V. fischeri NagC ability to regulate gene expression contributes to its overall fitness in environments that vary in levels of GlcNAc (Sun et al., 2015).

In many pathogens, relationship between metabolism and virulence has been determined (Wilharm and Heider, 2014). We propose that NagC is part of the regulatory circuit controlling the infectious process of EHEC by coordinating mucin-derived sugar metabolism and T3SS production (Figure EV10). At the level of the colonic intestine within the mucus there is a gradient of NANA and NAG due to their release from the intestinal mucin by mucinolytic bacteria, such as B. thetaiotaomicron (Derrien et al., 2010). When the concentration of NANA and/or NAG is high, their catabolism by E. coli O157:H7 produces high amount of intracellular NAG-6P which inactivates the transcriptional regulator NagC. In such case, the expression of NAG catabolic genes is induced while that of the LEE genes is reduced and thus adherence is prevented. This suggestion is supported by our observation that expression of LEE operons decreased after EDL933 growth in cecal content of gnotobiotic rats inoculated with human cecal content (Le Bihan et al., 2015). In reaching deeper mucus layer toward the intestinal epithelium, NANA and NAG are less found as free forms in the inner layer of mucus but are rather complexed to mucins (Derrien et al., 2010). In consequence, the amount of intracellular NAG-6P is low and the protein NagC is active allowing the repression of nagB, nagE and galP and activation of the LEE genes and thus adherence is promoted. Such a mechanism could contribute to the relocation of the pathogen from the intestinal lumen to the surface of intestinal epithelial cells, as previously suggested by others (Kamada et al., 2012; Pacheco et al., 2012; Cameron and Sperandio, 2015).

In this study, we described a novel mechanism by which EHEC O157:H7 regulate the expression of its T3SS-encoding genes in response to sugars derived from intestinal mucin. The NAG-6P sensor NagC was shown to promote the adherence of EHEC O157:H7 to intestinal cells in vitro through a direct regulation of ler and to be an important regulator for the fitness

#### REFERENCES


of EHEC in vivo. This work sheds further light on the link between the nutrient availability and EHEC O157:H7 adaptation and virulence gene expression.

#### AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: GL, CM, JH, and GJ. Performed the experiments: GL, JS, PG, AG, FB, and GJ. Analyzed the data: GL, AB, APG, CM, JH, and GJ. Wrote the paper: GL, CM, JH, and GJ.

## FUNDING

GL was supported by a scholarship from the Institut de recherche en santé publique de l'Université de Montréal-Fonds de recherche du Québec-santé (Project 40148). This work was also supported in part by the 61st Session de la Commission permanente de coopération franco-québécoise (Project 61. 116 to CM and JH), by the Natural Sciences and Engineering Research Council of Canada (NSERC) (Strategic and Discovery Grants to JH, RGPIN STP 307430 and RGPIN-2015-05373, respectively) and by EADGENE, N FOOD-CT-2004-506416, Network of Excellence under the 6th Research Framework Program of the European Union (CM).

### ACKNOWLEDGMENTS

We thank C. Del'Homme, E. Delmas, G. Lopes and J. Daniel for excellent technical assistance. We are grateful to Dr. J. Plumbridge (Institut de Biologie Physico-Chimique, Paris) her helpful comments and sharing information on NagC and help in identifying putative NagC binding DNA sequences and to Judith Kashul, for editing the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00134/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 Le Bihan, Sicard, Garneau, Bernalier-Donadille, Gobert, Garrivier, Martin, Hay, Beaudry, Harel and Jubelin. 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.

# Sampling Strategies for Three-Dimensional Spatial Community Structures in IBD Microbiota Research

Shaocun Zhang1, 2, 3 †, Xiaocang Cao4 † and He Huang1, 2, 3 \*

*<sup>1</sup> Department of Biochemical Engineering, School of Chemical Engineering and Technology, Tianjin University, Tianjin, China, <sup>2</sup> Key Laboratory of Systems Bioengineering, Ministry of Education, Tianjin University, Tianjin, China, <sup>3</sup> Collaborative Innovation Center of Chemical Science and Engineering, Tianjin, China, <sup>4</sup> Department of Gastroenterology and Hepatology, Tianjin Medical University General Hospital; Tianjin Medical University, Tianjin, China*

Identifying intestinal microbiota is arguably an important task that is performed to determine the pathogenesis of inflammatory bowel diseases (IBD); thus, it is crucial to collect and analyze intestinally-associated microbiota. Analyzing a single niche to categorize individuals does not enable researchers to comprehensively study the spatial variations of the microbiota. Therefore, characterizing the spatial community structures of the inflammatory bowel disease microbiome is critical for advancing our understanding of the inflammatory landscape of IBD. However, at present there is no universally accepted consensus regarding the use of specific sampling strategies in different biogeographic locations. In this review, we discuss the spatial distribution when screening sample collections in IBD microbiota research. Here, we propose a novel model, a three-dimensional spatial community structure, which encompasses the x-, y-, and z-axis distributions; it can be used in some sampling sites, such as feces, colonoscopic biopsy, the mucus gel layer, and oral cavity. On the basis of this spatial model, this article also summarizes various sampling and processing strategies prior to and after DNA extraction and recommends guidelines for practical application in future research.

#### Edited by:

*Nathan W. Schmidt, University of Louisville, USA*

#### Reviewed by:

*Thomas Thurnheer, University of Zurich, Switzerland Venkatakrishna Rao Jala, University of Louisville, USA*

#### \*Correspondence:

*He Huang huang@tju.edu.cn*

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

Received: *25 November 2016* Accepted: *10 February 2017* Published: *24 February 2017*

#### Citation:

*Zhang S, Cao X and Huang H (2017) Sampling Strategies for Three-Dimensional Spatial Community Structures in IBD Microbiota Research. Front. Cell. Infect. Microbiol. 7:51. doi: 10.3389/fcimb.2017.00051* Keywords: sampling strategies, community structure, IBD microbiota research, feces, colonoscopic biopsy, mucus gel layer, oral cavity

## INTRODUCTION

Inflammatory bowel diseases (IBDs), including Crohn's disease (CD) and ulcerative colitis (UC), are emerging as a part of a worldwide epidemic. CD was first diagnosed by Dr Burril B. Crohn (Crohn et al., 1932), in New York, in 1932, and UC was first described by White (1888), in Europe, in 1888. The former condition can cause inflammation in any digestive tracts, while the latter

**Abbreviations:** IBD, Inflammatory bowel disease; CD, Crohn's disease; UC, Ulcerative colitis; NGS, Next-generation sequencing technologiesl; HMP, International Human Microbiome Project; IBS, Irritable bowel syndrome; FMT, Fecal microbiota transplantation; VOC, Volatile organic compound; SOP, Standard operating procedures; IHMS, International Human Microbiome Standards; OUT, Operational taxonomic units; PBS, Phosphate buffered saline; ADD, Abundance–distance dispersion; MGL, Mucus gel layer; MUP, Mucus-binding protein; PCR, Polymerase chain reaction; PSB, Protected specimen brush; LCM, Laser capture microdissection; ANOVA, Analysis of variance; DSS, Dextran sulfate sodium.

invariably affects the mucosa of the large intestine and rectum. Previous studies revealed that the prevalence of IBDs were greatly related to time (Molodecky et al., 2012), regions (Reinberg, 2015), age (Choi et al., 2015; Connelly et al., 2015), genes (Sharp et al., 2015; Wang and Achkar, 2015; Yang et al., 2015), stress (Gray et al., 2015), diet (Vagianos et al., 2016), etc., Some of these factors, including diet, were thought to be crucially connected to the genetic imbalance of the intestinal microbiota (Kosiewicz et al., 2011; Manichanh et al., 2012; Gevers et al., 2014; Kostic et al., 2014; Munyaka et al., 2016). Several studies have shown dysbiosis of the gut microbiome between patients with IBD and healthy individuals (Sokol et al., 2006; Andoh et al., 2012; Ottman et al., 2012). Owing to the decreasing cost and rapid development of next-generation sequencing (NGS) technologies (Zoetendal et al., 2008; Sheridan, 2014), the advancement of bioinformatics tools (Schloss et al., 2009; Caporaso et al., 2010; Glass et al., 2010), and the updating of online databases (DeSantis et al., 2006; Quast et al., 2013), 16S rRNA gene amplicon sequencing (Minamoto et al., 2015; Scher et al., 2015) and metagenomics analysis (Pérezcobas et al., 2014; Wang et al., 2015) have opened new frontiers to identify the variability of IBD microbiota research, which simultaneously characterizes multiple samples; it can also enable subsequent studies of microbial communities, both structurally, and functionally, while determining their interactions with the habitats they occupy.

Besides IBD, intestinal dysbiosis also plays a profound role in multiple chronic and metabolic diseases, including diabetes (Heintz-Buschart et al., 2016), obesity (Greenhill, 2015), irritable bowel syndrome (IBS) (Bennet et al., 2015), and so forth. Similar to IBD research; many studies conducted on the intestinal microbiota in relation to diabetes mellitus have predominantly used feces samples (Qin et al., 2012; Heintz-Buschart et al., 2016; Knip and Siljander, 2016). Additionally, in view of the connections between the periodontitis and diabetes mellitus, some studies have explored the diversity of subgingival microbiota between healthy controls and diabetics (Demmer et al., 2016). When investigating the relationship between intestinal microbiota and obesity, plenty of studies targeted the fecal microbiota for the reason that it is easily obtainable (Aguirre and Venema, 2015). Even though the small intestine is much more difficult to acquire than feces specimens, some researchers believed that sampling site should focus on the small intestinal microbiota, because it is where the calories are absorbed (Angelakis and Lagier, 2016). Moreover, a recent work showed that the obesity affected the subgingival microbial composition (Maciel et al., 2016). In IBS studies, the prevalently obtainable materials when sampling intestinal microbiota are feces and mucosal biopsies (Rangel et al., 2015; Parthasarathy et al., 2016). Accordingly, each disease has suitable sampling methods depending on pathophysiology and feasibility of the operation. Compared with other diseases, spatial ecological patterns are evident in common diseases of the colon, including the distribution of UC, and CD, which make the sampling sources diversified in IBD research (Lavelle et al., 2015). Meanwhile, understanding how the potentially complex pathogenesis of IBD occurs requires the integration of tools from spatial ecology with comprehensive sampling sources to define microbial dysbiosis in various niches (Lavelle et al., 2013).

The human body is composed of many niches. Biogeography studies the patterns of biological diversity in different niches, varying in both time and space (Fierer, 2008). The selection pressures of biology and the environment, elucidated by biogeography, are thought to be responsible for shaping the various habitats in the body (Lavelle et al., 2016). The community structure of microbiota across spatial niches might be disturbed to different degrees and in association with various disease states. Without cooperation among the other dimensions of microbial ecology, it may be difficult to investigate subjective signals from disturbances in a single niche (Jeffery et al., 2012; Lozupone et al., 2012). The International Human Microbiome Project (HMP)<sup>1</sup> , with its sum total funding of \$115 million, has showcased the distinct variations of the human microbiota in different community structures (Group et al., 2009). Other studies of the human microbiome have also characterized the bacterial biogeography of different habitats (Costello et al., 2009; Grice et al., 2009; Zhou et al., 2013). Numerous research initiatives have shown interpersonal variation in humanassociated microbiota in IBD (Lavelle et al., 2015, 2016). Likewise, intrapersonal variability has been discovered between different niches. Currently, the bacterial diversity in IBD research is determined by analyzing different community structures, and following the various aspects of feces (Kolho et al., 2015; Norman et al., 2015), colonoscopic biopsy samples (De Cruz et al., 2015; Rossen et al., 2015), and the mucus gel layer (MGL) (Johansson, 2014; Johansson et al., 2014). To obtain the MGL, researchers often use rectal swabs (Araújopérez et al., 2012), microbiological protected specimen brushes (PSBs) (Lavelle et al., 2013), and laser capture microdissection (LCM) (Lavelle et al., 2015). Recent research studies have indicated that oral microbiota will be used in clinical and diagnostic utilities (Yoshizawa et al., 2013; Said et al., 2014). Despite very promising prospects in the future, there is still no clear guidance identifying those methodologies that can be accurately used to systematically collect and process the samples. Some highly complex biological samples are often difficult to process, which can introduce much bias. These drawbacks can potentially influence the final result; yet, to comprehensively study the microbial diversity in IBDs, more information is indispensable in the design of spatial sampling strategies.

In this review, we focus on discussing the different sampling strategies used in IBD microbiota research from the perspective of three planes. Y-axis distribution includes the oral cavity and feces. X-axis gradients are distributed in intestinal biopsies, with sampling levels varying in the ileum, colon (ascending colon, transverse colon, and descending colon), rectum, and caecum. Z-axis distribution involves collecting luminal, mucosal, and mucous communities in a specific and regional manner, and it includes the feces, colonoscopy biopsy samples, and the MGL. Starting with a description of the y-axis distribution, we discuss the classic sampling sites—feces and the oral cavity. We

<sup>1</sup> International Human Microbiome Standards (IHMS) project http://www. microbiome-standards.org/ [Online]. [Accessed].

then describe the x-axis distributions of colonoscopy biopsy. Ultimately, we will concentrate on the different sampling methods used for the MGLs, which are located on the z-axis. We herein provide an overview of the most crucial sampling strategies to help researchers make informed decisions.

### SAMPLING SITES DISTRIBUTED ALONG THE Y-AXIS

#### Feces

In the 1680s, Leeuwenhoek first described fecal bacteria using homemade microscopes (Egerton, 2006). With the rapidly evolving research on IBD in the nineteenth century, fecal flora was frequently used to represent intestinal microflora, as it was easily collected in patients. Firmicutes and Bacteroidetes phyla constitute the majority of dominant fecal microbiota using 16S rRNA amplicon sequencing, and with Bacteroides being the most abundant (Arumugam et al., 2011). Some work suggested that fecal bacterial communities could be divided into three enterotypes (Bacteroides, Prevotella, and Ruminococcus; Arumugam et al., 2011; Wu et al., 2011). Nowadays, fecal microbiota transplantation (FMT) has been widely used in the treatment of patients with IBD, which was found to be an effective therapy for some recipients (Kelly et al., 2015; Ince et al., 2016; Vermeire et al., 2016); thus, it was concluded that there should be some close connections between fecal microbiota and IBD. Probert et al. (2014) compared IBD patients and animal models of colitis with healthy individuals, and they found that the volatile organic compound (VOC) in feces held a potential role in identifying a novel diagnostic method for IBD. With a high sensitivity to inflammatory states, bacterial biomarkers in stool may therefore constitute a promising noninvasive source to diagnose IBD (Berry et al., 2015). In IBDs, the pH progressively increases along the duodenum to the terminal ileum; it decreases in the caecum, and then slowly rises from the colon to the rectum (Nugent et al., 2001). Such changes in colonic physiology are possibly reflected in the microbiota. Additionally, important factors such as diet (Lee et al., 2016), physical exercise (Queipoortuño et al., 2013), smoking habits (Biedermann et al., 2013), and antibiotic use (Pérezcobas et al., 2013) should exert subtle differences on fecal microbiota composition; of these, antibiotic use has a strong impact on one's initial microbiota composition (Macfarlane, 2014; Zhang et al., 2015b). Consequently, all of these issues shall be considered prior to sampling.

#### Sampling Operating Procedures

In view of the importance of the fecal sampling method, the study of the standard operating procedures (SOP) used to collect the fecal specimens has been, and still is, crucial for identifying pathogens. In the early stages, Moore (Moore and Holdeman, 1974) pointed out that some unique problems may arise with respect to the isolation and identification of intestinal bacteria in fecal flora studies, including collection, shipping, and isolation. Some experiments confirmed that the collection procedures and storage conditions did influence the diversity and integrity of the microbial flora (Cardona et al., 2012; Gorzelak et al., 2015; Boers et al., 2016; Nishimoto et al., 2016). It has been suggested that stool consistency is strongly associated with gut microbiota diversity (Vandeputte et al., 2016).

Swidsinski et al. (2008a,b) developed a new method using a punched-out freshstool cylinder; they demonstrated that the fecal flora were highly structured and spatially organized. The homogenization step in this procedure significantly reduced the intra-individual variation in the detected bacteria (Hsieh et al., 2016). Specifically, the results indicated that the relative abundance of Firmicutes to Bacteroidetes was significantly higher when snap-freezing fecal samples were compared with fresh samples (Bahl et al., 2012). Meanwhile, a study recommended that stool should be frozen within 15 min of being defecated, and it should be stored in a domestic, frost-free freezer for <3 days before DNA extraction (Carroll et al., 2012). During storage and processing, freeze–thaw cycles were detrimental to microbial cell integrity (Cardona et al., 2012). Conventionally, samples can be stored at −80◦C in the long term until DNA extraction (for no longer than 6 months; Carroll et al., 2012). Based on the above, the ideal storing procedure might be as follows: homogenizing prior to sampling, sampling aliquot fresh stool to avoid subsampling; and then freezing at 80◦C as soon as possible. If the laboratory has difficulty snap freezing, some researchers believe that RNAlater <sup>R</sup> (Life Technologies) might be selected to maintain DNA stabilization at +4 ◦C, or even at room temperature, for several days without affecting the 16S rRNA repertoire (for specific treatments, see **Figure 1**). However, a new study suggested that RNAlater should be avoided due to its ability to degrade the yield of DNA and bacterial taxa (Gorzelak et al., 2015). Otherwise, a guanidine thiocyanate solution might ensure the high stability of fecal microbiota at room temperature (Nishimoto et al., 2016). Despite this, there are still no universally accepted standards in the field of feces sampling.

#### Sample Extraction

According to the instructions and manual operation, 100 or 200 mg were the most frequently used dosages. One study showed that a 200 mg starting weight produced significantly higher DNA yields than 100 mg (Claassen et al., 2013); however, there was no similarity with respect to DNA purity. Conversely, Ariefdjohan (Ariefdjohan et al., 2010) tested 10–50 mg fecal samples and found that these weights, and not 100 mg or 200 mg, could result in maximum DNA yields. The phenol: chloroform-based DNA isolation method was illustrated to effectively obtain the requisite DNA yield (Mackenzie et al., 2015); however, this method is not suitable for clinical or large-scale studies. Owing to the bead-beating step, hot phenol with bead beating resulted in a proportional increase in Firmicutes (Wu et al., 2010; Mackenzie et al., 2015).

With respect to DNA extraction kits, those associated with the HMP view the MoBio PowerSoil <sup>R</sup> DNA Isolation Kit as the most effective microbial DNA extraction method. Moreover, some researchers involved in the International Human Microbiome Standards (IHMS; http://www.microbiome.standard.org/) prefer to use the QIAamp DNA Stool Mini Kit. Some researchers have conducted several studies on different extraction methods. As a result, the combination of mechanical cell disruption by

repeated bead-beating (Yu and Morrison first described the repeated bead-beating and column purification method, Yu and Morrison, 2004) for 6 min, (Salonen et al., 2010) and with a 95◦C heating step, showed greater bacterial diversity; it resulted in the significantly improved DNA extraction abundance of archaea and some bacteria, especially for bacteria in the phylum Firmicutes, including Clostridium cluster IV (Salonen et al., 2010; Thomas et al., 2015). However, bead-beating for long periods of time had a negative effect on DNA yield, and zirconium–silica beads were considered to be the best choice (Salonen et al., 2010). Due to the aromatic acids that exist in stool, some inhibition removal technology or substances were utilized to prevent interference—such as the inhibitEX tablets in the QIAamp DNA Stool Mini Kit (Thomas et al., 2015). Additionally, the size of the spin columns may also influence filter efficiency; for instance, sizes smaller than 0.45 µm would hold back some larger fragments (Thomas et al., 2015). Several studies have compared various DNA extraction kits and methods to assess the bacterial diversity in stool samples (Wu et al., 2010; Claassen et al., 2013; Kennedy et al., 2014; Mackenzie et al., 2015; see **Table 1**). It was found that finding a protocol to extract DNA without bias is a challenging task.

#### Sample Sequencing

Two methods are frequently used for taxonomic classification of organisms that are found in microbiomes: 16S rRNA gene amplicon sequencing and metagenomic sequencing. 16S rRNA gene amplicon sequencing is increasingly being used to provide information about the compositions and the relative abundance of microorganisms and classify microbial communities based on amplification of 16S rRNA gene, both taxonomically and phylogenetically (Clarridge, 2004). To analyze 16S rRNA gene sequences from microbial communities, QIIME, Mothur, and


TABLE

1


Overview

of

different

processing

methods

or

commercial

DNA

extraction

kits

that

were

compared

in

some

studies

to

extract

DNA

from

stool

samples

for

further

bioinformatics

analysis.

February 2017 | Volume 7 | Article 51

*A–C stands for the performance*

 *rank: A (best performance)*

 *to C (worst performance)*

LotuS have been widely used to process data from highthroughput sequencing (Schloss et al., 2009; Kuczynski et al., 2011; Hildebrand et al., 2014). Additionally, PICRUSt (http:// picrust.github.com/) has been developed to predict metabolic pathways based on 16S data and a reference genome database (Langille et al., 2013). Although this approach is unable to outperform metagenomic sequencing, it can predict and compare probable functions across a large amount of samples from different niches. Meanwhile, it can reproduce functional information that shows highly similar to the metagenomic sequencing in the HMP and other data sets (Anonymous, 2013). Compared with 16S rRNA gene amplicon sequencing, metagenomic approach is able to identify some of the distinctive functional attributes encoded in intestinal microbiota and comprehensively characterize metabolic capabilities of the microorganisms (Gill et al., 2006). Several tools have been developed to process the metagenomic data, such as MetaPhlAn (Segata et al., 2012), HUMAnN (Abubucker et al., 2012), and TruSPADES (Hildebrand et al., 2014). All approaches have merits and drawbacks. 16S rRNA gene sequencing is more costeffective and less time consuming than metagenomic sequencing. However, metagenome approaches enable the analyses of all kingdoms as well as viral sequences. The 16S rRNA gene captures broader range of microbiome diversity, but with a lower resolution and sensitivity compared with metagenomic (Poretsky et al., 2014). Limitations withstanding, 16S rRNA is limited by the biases inherent to PCR amplification, which results from the lack of truly universal primers and different copy numbers of 16S rRNA gene (Vallescolomer et al., 2016). As for metagenomic sequencing, it could be less efficient at detecting rare species in a microbial community compared with 16S rRNA. Metagenomic sequencing also requires advanced bioinformatics skills to process and analyze the data (Shakya et al., 2013).

Theoretically, the best analysis method currently available is metagenomics; however, its associated costly budget is not suitable for clinic settings or large cohorts, and it faces some limitations with respect to environmental interactions. As a result, it was found that until recently, 16S rRNA gene amplicon sequencing is often used as an exploratory step before metagenomic research. With respect to the sequencing, the 16S rRNA database only includes bacteria and archaea; yet, the absence of viruses and eukaryotes misses many pathogenic factors, which may bias the analysis. The smallest units of operational taxonomic units (OTUs) are species, so the strains resulting in antibiotic resistance, as well as mobile elements cannot be identified (Thomas et al., 2015). Besides, Bifidobacteriaceae are not well represented in some 16S V1–V3 analyses (Jumpstart Consortium Human Microbiome Project Data Generation Working, 2012). According to some investigations, the optimal choice for the variable regions in the 16S rRNA approach were V1–V3 and V3–V5, as the choice of a V6–V9 primer did not appear to efficiently cover the V6–V9 regions (Wu et al., 2010; Jumpstart Consortium Human Microbiome Project Data Generation Working, 2012). Otherwise, the amount of chimera increased and amplified the polymerase chain reaction (PCR) bias (Schloss et al., 2011). To reduce the bias of the PCR methods, and to minimize the errors introduced during sequencing, some researchers developed a method known as Low-Error Amplicon Sequencing (LEA-Seq) (Faith et al., 2013), which has been applied to QIIME. Next, for high-throughput sequencing, both 454 GS FLX and 454 Titanium sequencing methods can be used, depending on convenience (Wu et al., 2010). With read lengths of currently up to 2 × 300 bp and low sequencing costs, Illumina's MiSeq (Solexa) is increasingly becoming one of the most potential sequencing platforms worldly used in IBD research (Quince et al., 2015; Chung et al., 2016). It gathers the integration of cluster generation, sequencing, and data analysis in a single instrument and can analyze data within 24 h (as few as 8 h; Liu et al., 2012). For sequencing technology, instead of pyrosequencing technology applied to 454 sequencer, MiSeq leverages sequencing by synthesis. Compared with 454 platforms, the MiSeq has a higher throughput per run and a lower error rate but a shorter reads (Liu et al., 2012; Loman et al., 2012). At the start of the IHMS project, the SOPs of fecal sample self-collection, conservation practice, and formulated sequencing standards are crucial for better understanding the fecal microbiome and for optimizing data comparisons in clinical settings.

## Oral Cavity

While feces are frequently used in IBD research, there are certain limitations associated with outpatient distaste for handling these samples. Yet, researchers seek a simpler, more efficient, and more acceptable method. Oral samples are an important option. The oral cavity is a complex environment that includes the saliva, the tongue, teeth, tonsils, the buccal mucosa, and gingival sulci, which are colonized by a number of molecular and microbial analytes and bacteria (Human Microbiome Project, 2012). The microbiota in the oral cavity has a multitude of opportunities to reach the gut (Rochet et al., 2007). Pittock et al. (2001) reported oral lesion in nearly half of children that were newly diagnosed with CD. Similarly, one prospective study found that more than 30% of children with CD had involvement of the mouth (Harty et al., 2005). Another study noted a significant decrease in the overall diversity in the oral microbiota of pediatric CD patients (Docktor et al., 2012). Some bacteria in the oral cavity have recently been investigated for their association with IBD (Yoneda et al., 2016); these bacteria can be analyzed as microbial biomarkers for evaluating pathologies of the oral cavity, such as Campylobacter concisus (Ismail et al., 2012) and Fusobacterium nucleatum (Swidsinski et al., 2009). Thus, using oral microbial diagnostics is not a novel concept. Nowadays, scientists pursue a timely, accurate, cost-effective, and non-invasive diagnostic method to detect IBD. In view of these, further research on the oral microbiota in IBD might hold potential clinical and diagnostic utility in the future (Docktor et al., 2012). In this review, two frequently used sampling origins are primarily discussed: saliva and subgingival plaques.

#### Saliva

The average adult produces more than 1,000 mL of saliva per day, which always flows into the gastrointestinal tract. Thus, it can be stated that the salivary microbiota affects the development of gut microbiota in some respects. The composition of salivary microbiota was found to be different between CD patients, UC patients, and healthy controls (Said et al., 2014). Furthermore, when analyzing the composition of the tongue, buccal mucosa, saliva, and stool microbiota in colitis patients, the saliva microbiota exhibited the most alterations in terms of abundance (Rautava et al., 2015). The dominant genera, Veillonella and Haemophilus were recommended to largely contribute to dysbiosis of salivary microbiota in IBD patients (Said et al., 2014). At the species level, C. concisus (Ismail et al., 2012; Mahendran et al., 2013) and Mycobacterium avium Paratuberculosis (Bruno and Isabelle, 2015) have been investigated for its role in saliva dysbiosis of IBD patients.

For sample processing, DNA yield and quality, as well as 16S rRNA/DNA products and representations of the microbial community from oral wash samples, were investigated by six commonly used commercial DNA extraction kits, utilizing either mechanical bead-beating or enzymatic methods for cell lysis (Wu et al., 2014). Researchers discovered that mechanical beadbeating extraction kits produced less total DNA when compared with the enzymatic methods. On the other hand, microbial diversity showed no difference by either mechanical beadbeating or enzymatic extraction methods. As non-invasive and informative as saliva sampling is, but now there are currently no universally accepted techniques for sample collection. Prior to sampling the saliva, one must clean the oral cavity by rinsing it with water; this is imperative to avoid the presence of contaminants (Yoshizawa et al., 2013).

#### Subgingival Plaques

As a human microbiome community, dental plaques were initially observed by Leeuwenhoek (Dobell, 1932) over 300 years ago. Using combinatorial labeling and spectral imaging fluorescent in situ hybridization (FISH) to differentiate up to 15 fluorescent probes, Welch and colleagues (Mark Welch et al., 2016) showed, for the first time, the informative value of the oral microbiota biogeography at the micron scale. The fantastic color images that they created showed that the oral cavity acted as a "coaggregation." Similar to the role of canopies in hedgehog structures, Corynebacterium primarily gathered in subgingival plaques and supragingival dental plaques. Zhang et al. (2015a) first combined subgingival plaques and feces to analyze the microbiota perturbed in disease, and they partly normalized after treatment; at the same time, the researchers strongly confirmed the overlap in the abundance and function of species at different body sites. This will lead to potential ways to use the supragingival microbiota community for diagnosis and prognosis. Several recent studies have demonstrated connections between the composition of IBD and periodontitis (Kelsen et al., 2013; Elburki, 2015; Agossa et al., 2016). Meanwhile, additional studies have illustrated the associations between the composition of the subgingival microbiota and IBD (Brito et al., 2013; Kelsen et al., 2015). By analyzing inflamed subgingival sites, which depends on the checkerboard DNA– DNA hybridization technique, researchers found that the levels of Prevotella melaninogenica, Staphylococcus aureus, Streptococcus anginosus, and Streptococcus mutans are higher in CD patients than in controls. Furthermore, UC patients harbored a greater abundance of Staphylococcus aureus and Peptostreptococcus anaerobius than controls (Brito et al., 2013).

Thus, it is essential to study and collect subgingival plaques. To do so, place cotton balls in such a way that they can clean out residual supragingival plaques, prior to the collection of subgingival samples. Collect the subgingival plaque in a tube with buffer, using a sterile Gracey curette to gather the targeted teeth of the mesio-buccal surface. Then, firmly close the cap on the tube and shake the tube for 5 s to entirely homogenize the sample distribution in the buffer. Finally, place the sample on ice and send it to the biology lab within 4 h (McInnes and Cutting, 2010). The HMP method uses the MoBio PowerSoil <sup>R</sup> DNA Isolation Kit; other researchers have used the MasterPure DNA Extraction Kit (Moutsopoulos et al., 2015), the FastDNA spin Kit (Kuehbacher et al., 2008), the PSP Spin Stool DNA Plus Kit (Kelsen et al., 2015), and others. Optimal methods for DNA extraction are still under development.

#### SAMPLING SITES DISTRIBUTED ALONG THE X-AXIS

#### Colonoscopy Biopsy

Accordingly, luminal microbiota and mucosa-associated microbiota have been reported to be different in IBD (Lepage et al., 2005; Morgan et al., 2012; Gevers et al., 2014). Fecal microbiota might not adequately represent bacterial communities at the epithelial interface. Colonoscopy biopsy is the most common sampling technique used to assess microbial niches associated with the intestinal mucosa; it was shown to play a crucial role in diagnosis, and it can distinguish between disease types in IBD (Salvatori et al., 2012). Mucosal biopsies sample multiple amounts of the submucosa, epithelium, and MGL. The most comprehensive method to analyze the mucosa-associated microbiota may be proctocolectomy. In fact, Chiodini et al. (2013) were the first to examine the microbial populations of submucosal tissues using proctocolectomy during active disease; they also discussed the submucosal microbiota and biotypes within CD. Some other works also elected to use tissue sections of the terminal ileum and colon, obtained during surgery, for this process (Kleessen et al., 2002; Neut et al., 2002). As accurate as proctocolectomy is, this method cannot be applied to most of IBDs, except on rare occasions. Therefore, a more suitable method to obtain the tissue should be colonoscopy.

#### Sampling Spatial Distribution and Processing

It has been said that diverse bacteria distribute heterogeneously along the small bowel to the colon (Eckburg et al., 2005). Biopsy specimens can be taken from different gut locations, such as the ileum, colon (ascending colon, transverse colon, and descending colon), rectum, and caecum. In addition, the intestinal tract contains a variety of distinct microbial communities along the ileum (around 155 cm from the anus), caecum (around 150 cm from the anus), ascending colon (around 142 cm from the anus), transverse colon (around 109 cm from the anus), descending colon (around 64 cm from the anus), and rectum (around 10 cm from the anus; Zhang et al., 2014), and the difference between longitudinal regions in the intestinal tract should be positioned to select the target regions for sampling (**Figure 2A**). Comparing the microbial diversity of samples obtained with sheathed forceps with those obtained with standard unsheathed forceps, biopsies from the specific sites were not contaminated with the work channel (Dave et al., 2011). Additionally, a novel biopsy technique (Brisbane Aseptic Biopsy Device) has been developed to prevent cross-contamination from intestinal luminal contents (Shanahan et al., 2016). To avoid the influence of biopsy specimen sizes of colonoscopic tissue, researchers quantified tissue cell numbers using primers of the β-globin gene to determine the total amount of mucosa-associated microbiota in the biopsy specimens (Wang et al., 2014b). Previous studies revealed that bowel preparation (PEG electrolyte solution) before endoscopy affected the composition and diversity of the tissue and stool samples (Harrell et al., 2012; Jalanka et al., 2015; Shobar et al., 2016). Dividing a single dose into two separate dosages may introduce fewer alterations to the intestinal microbiota, which is preferred in clinical practice (Jalanka et al., 2015). Still, bowel preparation may have little effect on the next sampling procedure, as it has a short-term effect on the composition of the intestinal microbiota (O'Brien et al., 2013). Once taken, some works suggested that biopsy samples were placed in a cryovial with a lid, immediately snap-frozen in liquid nitrogen, and then stored at −80◦C until further analysis (van den Heuvel et al., 2015; Hedin et al., 2016; Munyaka et al., 2016). However, other mucosal biopsy specimens were harvested and then washed twice in 500 mL of phosphate buffered saline (PBS; pH 7–8) to ensure

that there was no fecal contamination prior to being snap–frozen in liquid nitrogen (Shen et al., 2010; Sanapareddy et al., 2012; Budding et al., 2014; Berry et al., 2015). Considering the actual process, a protective solution can maintain the sample at −20◦C for a few weeks, or at 4◦C for 24 h (Zoetendal et al., 2006). Despite this, it is recommended that biopsy samples be processed as soon as possible to avoid the lysis of microbial cells.

#### Sample Extraction and Analysis

Quantities of bacterial cells in biopsy samples are 1% less than in feces samples (Lyra et al., 2012). DNA extraction procedures should be more carefully conducted in order to better represent the microbial community. A study that compared some DNA extraction methods, drew the conclusion that the bead-beating and column method, as well as high molecular weight methods, were likely to result in the increased production of DNA yield, which primarily included the Firmicutes bacteria (Ó Cuív et al., 2011). Nowadays, a large number of studies have preferred to use the QIAamp DNA Mini Kit for IBD biopsy DNA extraction (Hansen et al., 2013; Chen et al., 2014; Wang et al., 2014a; Lavelle et al., 2015). The positive effect of bead-beating on mechanical cell lysis has been discussed for fecal samples, which are sometimes also used in DNA isolation from biopsy samples (Chen et al., 2014). However, it appears that bead-beating may not require efficient microbial DNA extraction from biopsy specimens due to the fact that mechanical cell lysis of the biopsy specimens might increase the concentration of eukaryotic DNA,

which may bias 16S rRNA gene sequencing analysis (Carbonero et al., 2011). A microbiome DNA enrichment method might potentially yield a higher fraction of microbial production, which methylated the human genomic DNA to selectively separate from microbial DNA (Yigit et al., 2016).

As for the spatial community structures (ileum, ascending colon, transverse colon, descending colon, and rectum) of human mucosal-associated intestinal microbiota, spatial variations of mucosa-associated microbiota have not provided feasible explanations to account for the observed longitudinal variations along the intestine, despite the previously observed spatial heterogeneity of mucosa microbiota (Aguirre de Carcer et al., 2011; Hong et al., 2011). Single-species abundance– distance dispersion (ADD) modeling results indicated that it was impossible to use conventional multivariate analysis methods to describe spatial heterogeneity and co-relationships across the multiple loci of microbial communities. The cooccurrence network analysis (Barberan et al., 2012) revealed a huge specialization among vertical and lateral gradients, and it addressed how interpersonal variation was a significant constituent of variance, particularly in light of the fact that the microbiota remains stable (Faust et al., 2012; Zhang et al., 2014). To reveal the longitudinal gradients in the microbiota along the x-axis distribution, studies may need to develop suitable statistical models and bioinformatics software.

## SAMPLING SITES DISTRIBUTED ALONG THE Z-AXIS

#### Mucus Gel Layer

Secreted by goblet cells that reside in intestinal crypts, the colonic MGL partially or entirely covers the epithelium and creates a boundary between the lumen and the host mucosa. Mucus is subsequently secreted and the layers fall off, generating a "district" that is carried into the fecal stream (Swidsinski et al., 2008b). The mucus is continuously secreted and can be divided into two layers: an outer, loosely adherent layer that can be removed by suction or gentle scraping; and an inner, firmly stratified layer that adheres to the epithelial cells (Atuma et al., 2001). In mouse models, the thickness of both MGL layers is appropriately estimated at 150µm, with the outer layer measured at 100µm and the inner layer at 50µm (Johansson et al., 2008). The thickness of the human MGL is thought to be between 107 and 155µm, depending on the loci (Pullan et al., 1994). Both layers are made up of MUC2-type mucin (Johansson et al., 2008). In healthy individuals, the inner layer is devoid of bacteria, while the outer layer serves as a habitat for the commensal microbiota (Hansson and Johansson, 2010; Johansson et al., 2011). The architecture of MGL exhibits a diverse range of polymers, including the mucus-binding protein (MUP), which offers numerous binding locations for both pathogenic and commensal bacteria (MacKenzie et al., 2009; Alemka et al., 2012). Some commensal bacteria are able to bind to and degrade the MUP, and they can be utilized as a barrier to pathogen binding. Mucin degradation of the MLG provides nutrients for some commensals, and it may initiate the initiation of pathogen invasion (Lennon et al., 2014b). As a result, the MGL plays a double role, providing a mutually beneficial environment for the host cells and resident microbiota, while serving as the first line of defense against pathogen bacteria translocating into the mucosa (see **Figure 3**). In IBD, bacteria are allowed to penetrate the inner MGL and reach the epithelium, triggering an inflammatory response; this suggests that the barriers of MUC2, with the absence of the MUC2 mucin polymer constituent, are disturbed, resulting in inflammatory responses (Schultsz et al., 1999; Swidsinski et al., 2007; Johansson et al., 2014).

On the basis of the aforementioned biological mechanism, identification of the mucus-degrading bacteria in the MGL is crucial. Conventionally, the MGL isolated from the precise fixation of intestinal biopsies or tissues, where dehydrating aldehyde fixatives are used, can result in loss and detachment of the mucus. Matsuo (Matsuo et al., 1997) demonstrated that using Carnoy's solution can preserve the integrity of surface mucus in paraffin sections of human colon specimens. Recent developments in overcoming this experimental limitation have achieved great success. Here, we describe three main sampling methods: rectal swab, the microbiologically protected specimen brush, and LCM. The vivid cross-sectional organization of each sampling method can be seen in **Figure 2B**.

#### Rectal Swab

As a simple, standardized, non-invasive, and inexpensive method, rectal swab represents an important contribution when the patient does not wish to handle feces or undergo the discomfort and inconvenience of colonoscopy. A swab-sucked microbiota is reproducible, and the procedure can be performed by either the patient at home or by medical professionals in clinical settings; thus, this method may be suitable for clinical diagnostic purposes and clinical studies (Budding et al., 2014). Rectal swabs aim at collecting the colorectal mucus (Braun et al., 2009). Rectal swab specimens can be easily handled and stored immediately without perturbation of the microbiota. Swab specimens are obtained about 1–2 cm from the anal verge and collected by inserting a sterile cotton-tipped swab. This pioneering work suggested that swab sampling, without previous bowel preparation, harvested undisturbed microbiota (Budding et al., 2014). The swab was inserted into sterile PBS shaken for at least 2 min to ensure the sufficient release of microbiota, and the samples were then stored at −80◦C until DNA isolation (Araújopérez et al., 2012); conversely, the samples could also be placed in tubes containing 500 mL of Reduced Transport Fluid buffer and maintained at room temperature for 2 h prior to storage at −20◦C until DNA isolation (Syed and Loesche, 1972; Budding et al., 2014). For DNA isolation, the bead-beating step may have a negative effect on the estimated abundance of Bacteroidetes (Budding et al., 2014). DNA extraction kits can use the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) or Qiagen's DNeasy Blood and Tissue Kit (Araújopérez et al., 2012; Budding et al., 2014).

Previous work that has analyzed T-RFLP profiles and quantitative PCR (qPCR) has highlighted the differences in community diversity between samples obtained by biopsy or swab, and it was found that a higher abundance of Lactobacillus

and Eubacteria were present in the swab specimens when compared with biopsies (Araújopérez et al., 2012). It was also previously demonstrated that Staphylococcus aureus, a dominant skin bacteria, could be used to assess the level of skin contamination between swabs and biopsies (Araújopérez et al., 2012). With respect to spatial organization, the fecal samples and swabs seemed to harbor more or less distinct diversity (Budding et al., 2014). One study revealed that the microbiota obtained by rectal biopsy and swab showed a greater similarity to one another than to feces (Glover et al., 2013). The diagnoses that are usually based on culture or NAAT on rectal swabs are widely utilized to distinguish between Chlamydia proctitis and CD (Hoentjen and Rubin, 2012). To prevent disturbances, from occurring, harvesting samples through a sheathed swab might lower the level of contamination by the skin and luminal microbiota in further studies.

#### Microbiological Protected Specimen Brush

In recent research, a specimen brush was often applied to sample the human lung microbiota (Dickson et al., 2015; Schmidlin et al., 2015; Hogan et al., 2016; Sibila et al., 2016). Inspired by these investigations, Lavelle and colleagues (Lavelle et al., 2013) developed a novel sampling technique using the microbiological PSB for spatial microbial assessment; they targeted the superficial MGL from the luminal side, as it can fold over the light mucosa and avoid pools of fluid. Structurally, when compared with rectal swabs, this brush also targets an outer, colonized mucus layer that becomes separated from the epithelium via a dense layer of removable mucus. As a sterile, single–use sampling method, the brush is covered with a sheath, which consists of a distal plug at the tip to seal the brush when introducing and retracting the brush through the colonoscopy channel. After collecting the specimen, a sterile wire cutter is used to separate the tip of the wire and the plug, and the sample is then placed in a sterile, nuclease–free container until DNA extraction. The Qiagen DNA Mini Kit is frequently employed to extract DNA. The qPCR confirmed that the increased proportion of microbial DNA is sampled in the brush when compared with biopsy samples. Based on the 16S rRNA gene, the analysis of similarity analyses illustrated that there was a similar and highly significant difference between the PSB and biopsy samples, as well as between the Shannon Diversity Index values for reduced diversity in brush samples when compared to the biopsy samples (Lavelle et al., 2013).

#### Laser Capture Microdissection

Developed at the National Institutes of Health (Emmert-Buck et al., 1996), LCM is a systemic technique whereby individual DNA, RNA, and proteins can be sampled from the gut tissue by fixing targeted cells to an adhesive film with a laser beam; they are then observed under the microscope (Zhang et al., 2016b). LCM is a powerful method used to directly isolate pure sections from complex tissues with greater rapidity, specificity, and precision. This method does not require specific markers for identification, either prior to or after isolation, which is in contrast to rectal swabs and the microbiological PSB. To get at the MGL, researchers used LCM in healthy subjects undergoing a clinical routine colonoscopy, as well as in UC patients undergoing proctocolectomy for sampling (Lavelle et al., 2015), as based on the PALM MicroBeam system (Rowan et al., 2010a). Specifically, some researchers combined LCM and PCR to isolate and count the total amount of some mucosaadherent bacteria, such as Desulfovibrio copies in the mucous gel of UC patients (Rowan et al., 2010a; Lennon et al., 2014a), as well as adherent–invasive E. coli from the macrophages of CD patients (Elliott et al., 2015). Given that Mycobacterium avium subsp paratuberculosis micro-organisms are few in number when present in CD patients, LCM was used to overcome this issue by accurately isolating subepithelial tissue, thus preventing contamination from the lumen (Ryan et al., 2002). Significant variations were observed between the colonic crypts and the central luminal compartment in mouse models, which used LCM to specifically profile the composition of the microbial communities in a discontinuous locus (Nava et al., 2011; Pedron et al., 2012). As a result, the study of colonic crypt mucus in UC patients, using LCM-harvested specimens, found that these patients had a lower abundance of crypt-associated bacteria than controls (Rowan et al., 2010b). Studies using LCM have placed standard and systemic histological sections of stained tissue under a microscope, and subsequently visualized the MGL of interest (Lennon et al., 2014a; Lavelle et al., 2015). Using a joystick to navigate around the image, researchers simply pushed a button to transfer the desired pure cells of the heterogeneous tissue to each slide to yield an average sample area of 175 mm<sup>2</sup> . Then, the LCM-harvested productions were catapulted onto an inverted opaque AdhesiveCap. As a targeted and specific quantified sampling method, LCM is suitable for research in precision medicine.

## CONCLUSION

As is well-known, suitable sampling strategies play an important role when studying the full landscape of intestinal microbiota. Here, this review highlighted the biogeographically stratified sampling strategies used in IBD, and it simultaneously proposed a novel three-dimensional spatial model of different community structures. Across these sampling sites, the non-invasive nature of fecal sampling can be implemented on a large scale as a screening or follow-up tool. However, feces are comprised of a mixture of products from all intestinal regions, which may not reflect the true nature of host–bacterial interactions in different biogeographic locations (Swidsinski et al., 2008b). Compared with fecal sampling, standard colonoscopy biopsy sample is sufficient to assess mucosal microbiota, which might affect mucosal and epithelial function to a greater degree than fecal sampling, as mucosal microbiota has a closer contact with immune cells and epithelial cells (Sartor, 2015). Furthermore, biopsy samples can be captured from specific regions ranging from the caecum to the rectum. These deep strengths notwithstanding, biopsy collection requires streamlining the logistics for sampling with nurses, physicians, and endoscopy technicians in advance to decrease the patients' time under sedation (Tong et al., 2014). The microbial profiles have indicated that at the early stage of disease, assessing rectal biopsy microbiota offered particular potential for convenient and early diagnosis of CD (Gevers et al., 2014). Particularly, in mouse studies, both tissue and feces sampling allowed targeted analyses of microbial under tractable and reproducible conditions. Fecal samplings could timely process feces to study the diversity of intestinal microbiota, varying in time (Zackular et al., 2013; Zhang et al., 2016a). Meanwhile, fecal pellets could also be collected from sacrificed mouse across different anatomical sites which often utilized caecal and colon contents (Bibiloni et al., 2005; Gaudier et al., 2005; Mishiro et al., 2013). Sometimes, the luminal content were flushed together by injecting PBS and then collected (Berry et al., 2012). The mucosa-associated microbiome is sampled by washing with PBS to remove the fecal contents then releasing epithelial cells (containing mucosal microbes) from the intestine tissue with mechanical means (Nagalingam et al., 2011; Tong et al., 2014). Specifically, LCM could specifically sample microbes that were located in the particular parts of mucosa (Nava et al., 2011). Evaluation of microbial community composition revealed striking differences between feces and tissues. The comparison between dextran sulfate sodium (DSS) colitis mouse and controls showed that the 16S rDNA content (bacterial) was significantly decreased in feces but increased in mucosa, exhibiting the same trend as 18S rDNA (fungal; Qiu et al., 2015).

Coupled with the luminal microbiota, researchers have demonstrated that when using the MGL and entire mucosal biopsies, there is spatial variation in the intestinal microbiota, particularly among different community niches in UC patients (Lavelle et al., 2015). Moreover, human swab and colon biopsy samples have revealed that the mucosal diversity is prominent and enriched, particularly among the species from the phyla Proteobacteria and Actinobacteria, and when compared with the fecal microbiota (Albenberg et al., 2014). Zhou (Zhou et al., 2013) characterized the microbial variation between different community niches using a Dirichlet–Multinomial Distribution model, which concluded that feces and oral samples had the lowest interpersonal variability across the studied body sites studied in terms of community structure. To further illustrate this point, it has been reported that the numbers of bacteria in the Clostridium coccoides group remained stable in both feces and saliva over time (Singhal et al., 2011). Stearns et al. (2011) sampled species across the human digestive tract, including from feces, the stomach, colon, duodenum, and oral cavity, and illustrated that the oral cavity harbored the greatest phylogenetic diversity. Predictably, the oral microbiota holds great potential with respect to clinical and diagnostic utility.

Specific to mucosal biopsies and the MGL, there should be heterogeneity in the mucosal species that exist along crosssectional and longitudinal axes of the bowel within specific individuals. However, due to the masking of a high level of individual variation, significant differences across longitudinal variations were not discovered by analysis of variance (ANOVA) (Zhang et al., 2014). Employing a multidisciplinary approach (such as by investigating ecological relationships and performing co-occurrence network analysis) may lift this mask of spatial variation to uncover the truth in prospective studies (Zhang et al., 2014). Specific to our study, we are devoted to developing statistical models to show the informative value of microbial biogeography in IBD research.

Traditional protocols are currently limited by the present difficulties associated with comprehensively evaluating the microbiota in IBD research. Such difficulties include fastidious experimental requirements and sampling errors. Therefore, it is critical that risk-free, standardized, simpler, and inexpensive sampling strategies be formulated in the future. To study potential contributions of the microbiota in IBD research, we should standardize the SOPs and reach a consensus that better

#### REFERENCES


facilitates our understanding of these methods in subsequent studies. Moreover, data should be exchanged and further studies should be designed in which we evaluate the microbiota within those individuals at the early stages of IBD. To construct a full picture of the microbial diversity in IBD research, synergistic profiles, combined with a co-culture consortium that can study bacteria, will be necessary. Comprehensively, it should be stated that a mutually beneficial cooperative effort can be achieved, but only if data on these methods are shared all over the world.

#### AUTHOR CONTRIBUTIONS

SZ wrote the paper; XC and HH performed the collected the data. All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.

#### ACKNOWLEDGMENTS

This work was supported by National High Technology Research and Development Program of China, No. 2015AA020701 and National Natural Science Foundation of China, No. 31470967. China Alliance of Inflammatory Bowel Disease, Wu Jie Ping Medical Foundation, No. 2017001.


with untreated periodontitis. Eur. J. Gastroenterol. Hepatol. 25, 239–245. doi: 10.1097/MEG.0b013e32835a2b70


resistance: the oral infections, glucose intolerance and insulin resistance study. J. Clin. Periodontol. doi: 10.1111/jcpe.12664.[Epub ahead of print].


multi-omic approach. Gut Microbes 62, 1591–1601. doi: 10.1136/gutjnl-2012- 303184


**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 VJ and handling Editor declared their shared affiliation and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Zhang, Cao and Huang. 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.

# Enterotype May Drive the Dietary-Associated Cardiometabolic Risk Factors

Ana C. F. de Moraes <sup>1</sup> , Gabriel R. Fernandes <sup>2</sup> , Isis T. da Silva<sup>1</sup> , Bianca Almeida-Pititto<sup>3</sup> , Everton P. Gomes <sup>4</sup> , Alexandre da Costa Pereira<sup>4</sup> and Sandra R. G. Ferreira<sup>1</sup> \*

<sup>1</sup> Department of Epidemiology, School of Public Health, University of São Paulo, São Paulo, Brazil, <sup>2</sup> René Rachou Research Center, Oswaldo Cruz Foundation, Belo Horizonte, Brazil, <sup>3</sup> Department of Preventive Medicine, Federal University of São Paulo, São Paulo, Brazil, <sup>4</sup> Laboratory of Genetics and Molecular Cardiology, Heart Institute (Incor), University of São Paulo Medical School, São Paulo, Brazil

Analyses of typical bacterial clusters in humans named enterotypes may facilitate understanding the host differences in the cardiometabolic profile. It stills unknown whether the three previously described enterotypes were present in populations living below the equator. We examined how the identification of enterotypes could be useful to explain the dietary associations with cardiometabolic risk factors in Brazilian subjects. In this cross-sectional study, a convenience sample of 268 adults (54.2% women) reported their dietary habits and had clinical and biological samples collected. In this study, we analyzed biochemical data and metagenomics of fecal microbiota (16SrRNA sequencing, V4 region). Continuous variables were compared using ANOVA, and categorical variables using chi-square test. Vsearch clustered the operational taxonomic units, and Silva Database provided the taxonomic signatures. Spearman coefficient was used to verify the correlation between bacteria abundances within each enterotype. One hundred subjects were classified as omnivore, 102 lacto-ovo-vegetarians, and 66 strict vegetarians. We found the same structure as the three previously described enterotypes: 111 participants were assigned to Bacteroides, 55 to Prevotella, and 102 to Ruminococcaceae enterotype. The Prevotella cluster contained higher amount of strict vegetarians individuals than the other enterotypes (40.0 vs. 20.7 and 20.6, p = 0.04). Subjects in this enterotype had a similar anthropometric profile but a lower mean LDL-c concentration than the Bacteroides enterotype (96 ± 23 vs. 109 ± 32 mg/dL, p = 0.04). We observed significant correlations between bacterial abundances and cardiometabolic risk factors, but coefficients differed depending on the enterotype. In Prevotella enterotype, Eubacterium ventriosum (r BMI = −0.33, p = 0.03, and r HDL-c = 0.33, p = 0.04), Akkermansia (r 2h glucose = −0.35, p = 0.02), Roseburia (r BMI = −0.36, p = 0.02 and r waist = −0.36, p = 0.02), and Faecalibacterium (r insulin = −0.35, p = 0.02) abundances were associated to better cardiometabolic profile. The three enterotypes previously described are present in Brazilians, supporting that those bacterial clusters are not population-specific. Diet-independent lower LDL-c levels in

#### Edited by:

Michele Marie Kosiewicz, University of Louisville, USA

#### Reviewed by:

Valerio Iebba, Sapienza University of Rome, Italy Gena D. Tribble, University of Texas Health Science Center at Houston, USA

\*Correspondence:

Sandra R. G. Ferreira sandrafv@usp.br

Received: 30 October 2016 Accepted: 07 February 2017 Published: 23 February 2017

#### Citation:

de Moraes ACF, Fernandes GR, da Silva IT, Almeida-Pititto B, Gomes EP, Pereira AdC and Ferreira SRG (2017) Enterotype May Drive the Dietary-Associated Cardiometabolic Risk Factors. Front. Cell. Infect. Microbiol. 7:47. doi: 10.3389/fcimb.2017.00047 subjects from Prevotella than in other enterotypes suggest that a protective bacterial cluster in the former should be driving this association. Enterotypes seem to be useful to understand the impact of daily diet exposure on cardiometabolic risk factors. Prospective studies are needed to confirm their utility for predicting phenotypes in humans.

Keywords: gut microbiota, enterotype, cardiometabolic risk, diet, lipid profile

## INTRODUCTION

Cardiometabolic diseases are among the leading causes of mortality, and an unhealthy diet plays a significant etiopathogenetic role (World Health Organization, 2011; Laslett et al., 2012). Pieces of evidence indicate that the gut microbiota mediates the relationship between dietary habits and cardiometabolic abnormalities (Koeth et al., 2013; Yin et al., 2015). The vast number of intestinal bacteria, and the large intra- and inter-individual variability has limited the understanding of such relationship. The observation of bacterial clusters in human gut has represented a way to reduce the complexity of these analyses. Arumugam et al. (2011) found three bacteria groups in humans, driven by high proportions of one of three taxa: Bacteroides (enterotype 1), Prevotella (enterotype 2), and Ruminococcus (enterotype 3). The bacterial communities play an important role driving diverse pathophysiological processes (Arumugam et al., 2011). Another study discussed the associations of dietary habits with two enterotypes, distinct from this seminal study since the Bacteroides enterotype was fused with the less distinct Ruminococcus enterotype (Wu et al., 2011). Animal protein and fat intake were associated with Bacteroides cluster, while Prevotella with a carbohydrate-enriched diet.

Populations are exposed to different dietary habits, and it is unknown how the enterotypes are distributed worldwide. Most studies that describe the enterotypes involves European, North American, and Asian (Arumugam et al., 2011; Wu et al., 2011; Lim et al., 2014; Roager et al., 2014) populations. Only a few scientific publications analyzed the clusters in South American or African individuals (Yatsunenko et al., 2012; Ou et al., 2013). The knowledge on the distribution of enterotypes, in populations with different genetic backgrounds and lifestyle, could be useful to understand underlying mechanisms linking dietary habits with the risk of cardiometabolic diseases (Zupancic et al., 2012; Koeth et al., 2013; Kelder et al., 2014).

Brazilian population offers an opportunity to investigate the presence of enterotypes in a high-food variety environment, and to deepen knowledge on the role of the gut microbiota mediating the impact of diet on metabolic disturbances. We hypothesized that enterotypes might participate in underlying mechanisms linking dietary habits to cardiometabolic diseases. We investigated whether enterotypes could be identified in a sample of Brazilians and examined the impact of this categorization of the gut microbiota on the association with the cardiometabolic profile.

#### MATERIALS AND METHODS

#### Subjects

In this cross-sectional analysis, we included a convenience sample of 268 participants from the major study named ADVENTO—Analysis of Diet and Lifestyle for Cardiovascular Prevention in Seventh-Day Adventists (http://www.estudoadvento.org). The ethical committee of the School of Public Health, Univesity of São Paulo, approved this study; all individuals provided written consent. Inclusion criteria were age from 35 to 65 years and body mass index (BMI) <40 kg/m2. Diabetes mellitus, history of inflammatory bowel diseases, persistent diarrhea, and use of antibiotics or probiotic or prebiotic supplements within the 2 months before the data collection were exclusion criteria. Dietary data was obtained using a validated food frequency questionnaire from the ADVENTO. Subjects were classified according to the dietary habit adopted for at least 12 months, in strict vegetarian (no consumption of animal products), lacto-ovo-vegetarian (consumption of dairy products and/or eggs), and omnivore (consumption of animal products more than once a month; Tonstad et al., 2009).

#### Clinical Data

Weight was measured using a digital scale with 200 kg capacity, height using a fixed stadiometer and BMI was calculated as weight in kilograms divided by height in meters squared. Blood pressure (BP) was measured with a standard oscillometric device (Omron HEM 705CPINT, Omron Health Care, Lake Forest, IL, USA). Blood samples were taken after an overnight fasting. Plasma glucose was measured by the hexokinase method (ADVIA Chemistry; Siemens, Deerfield, IL, USA). Measurements of total cholesterol, triglyceride, and high-density lipoprotein (HDL-c) were assessed by enzymatic methods. Lowdensity lipoprotein cholesterol (LDL-c) was calculated by the Friedewald equation.

#### Gut Microbiota

The analysis of the 16S rRNA gene (V4 region) was performed by Illumina <sup>R</sup> MiSeq platform using 200 mg of fecal samples maintained under refrigeration (6◦C) within a maximum of 24 h after collection, and the aliquots stored at −80◦C until analysis. The Maxwell <sup>R</sup> 16 DNA purification kit was used to extract DNA, and the manufacturer's instruction was used to carry out the protocol in the Maxwell <sup>R</sup> 16 Instrument (Promega, Madison, WI, USA). The DNA was amplified by a PCR assay using the 515F and 806R primers, as described by Caporaso et al. (2012), and sequenced in Illumina Miseq platform generating paired reads of 250 bp. 16S ribosomal DNA sequences are available under study accession PRJEB19103.

The paired reads were trimmed to remove bases with Phred score lower than five at the 5′ and 3′ extremities. These procedures also trimmed sequences with an average quality <15 in a sliding window of 4 bases. The software Trimmomatic (Bolger et al., 2014) performed this quality filter. Paired reads were merged using the FLASh tool (Magoc and Salzberg, 2011 ˇ ), requiring a minimum overlap of 20 nucleotides.

The redundancy among the sequences was removed using the dereplication step from Vsearch (Rognes et al., 2016), and filtered to remove the unique entries. The dereplicated reads with 97% identity were clustered, using the same tool, to create the OTUs. Taxonomical assignment to the OTUs was performed by the assign\_taxonomy script from Qiime (Caporaso et al., 2010) and Silva database, version 123 (Quast et al., 2013).

#### Enterotype Clustering

The enterotypes were identified by the methods previously described (Arumugam et al., 2011, 2014) and available in http://enterotype.embl.de/enterotypes.html. The Calinski-Harabasz (CH) index suggested the optimal number of clusters. A silhouette analysis and elbow plot evaluated the groups' robustness (Supplementary Figure S1).

#### Statistical Analysis

The descriptive statistical analysis calculated means, standard deviations, medians, and interquartile ranges. Variables with skewed distributions were log-transformed before analysis to achieve normality. ANOVA with Bonferroni post-hoc test was used to compare variables according to enterotypes and diet. Chisquare test was employed to compare proportions. The Spearman correlation coefficient pointed associations between metadata and most common genera or species (present in at least 80% of subjects). The most abundant genera were shown in the figures. Statistical analyses were performed using Statistical Package for the Social Sciences (SPSS), version 23 (IBM, Armonk, NY, USA), and R for enterotype analyses (cluster package). Beta diversity comparisons were computed as Principal Coordinate Analyses generated from Jensen-Shannon divergence matrices. A p < 0.05 was considered to identify important correlations.

#### RESULTS

The mean age of participants was 49.4 ± 8.4 years, 54.2% were women and 41.4% and had increased BMI (≥25 kg/m2). Sixtysix subjects were considered strict vegetarians, 102 lacto-ovovegetarians, and 100 omnivores. Strict and lacto-ovo-vegetarians had lower BMI (23.1 ± 4.1 and 24.4 ± 3.9 vs. 26.4 ± 4.7 kg/m2, respectively, p < 0.001) and LDL-c values (99 ± 31 and 101 ± 27 vs. 112 ± 29 mg/dL, respectively, p = 0.005) than omnivores (Supplementary Table S1).

Taxonomical distribution of fecal samples showed the predominance of Firmicutes and Bacteroidetes (**Figure 1**). The 10 most abundant phyla and 20 genera according to enterotypes and dietary habits were depicted in Supplementary Figure S2.

Three bacterial clusters were identified; 111 participants were assigned to Bacteroides, 55 to Prevotella, and 102 to Ruminococcaceae enterotype (**Figure 2A**). Relative abundances in each enterotype confirmed the expected predominance of genera Bacteroides, Prevotella, and Ruminococcaceae, respectively (**Figure 2B**). Subjects in each enterotype did not differ according to sex distribution, mean age, and BMI. The frequency of strict vegetarians was greater in Prevotella than in the Bacteroides and Ruminococacceae enterotypes (40.0 vs. 20.7 and 20.6%, p = 0.04, respectively), but frequencies of lacto-ovo-vegetarians and omnivores did not differ (Supplementary Figure S3).

Comparisons of clinical variables among enterotypes showed lower mean LDL-c values in Prevotella compared to Bacteroides (96 ± 23 vs. 109 ± 32 mg/dL, p = 0.04), despite similar measurements of body adiposity (**Table 1**). When a substratification of Supplementary Table S1 comparing enterotypes according to dietary habits (Supplementary Table S2), the lowest LDL-c levels were invariably observed in the Prevotella enterotype independently of the dietary pattern. Within the Prevotella enterotype, the strict vegetarian and lacto-ovovegetarian showed mean LDL-c values significantly lower than omnivores (92 ± 23 and 88 ± 18 vs. 107 ± 25 mg/dL, respectively, p = 0.04). Strict vegetarians belonging to the Ruminococcaceae cluster had the greatest mean value of HDLc that was significantly higher than subjects from the same enterotype but consumers of other dietary habits (59 ± 1 and 47 ± 1 vs. 51 ± 1 mg/dL, respectively, p = 0.004).

Correlations of clinical variables to bacteria abundances considering the entire sample ranged from −0.23 to 0.21. When stratified by enterotypes (**Figure 3**), the coefficients changed. In Bacteroides cluster, the abundance of Streptococcus was correlated to body adiposity (r BMI = 0.25, p = 0.02) and Blautia to systolic (r = 0.22, p = 0.04) and diastolic BP (r = 0.26, p = 0.01), while abundances of Desulfovibrio were inversely correlated to BMI (r = −0.22, p = 0.04) and Haemophilus to triglyceride levels (r = −0.22, p = 0.04).

The strongest correlations coefficients to cardiometabolic risk factors were detected in the Prevotella enterotype. Blautia (r BMI = −0.34, p = 0.03 and r waist = −0.37, p = 0.02), Coprococcus (r BMI = −0.45, p < 0.01; r waist = −0.41, p < 0.01; r glucose = −0.37, p = 0.02; r insulin = −0.31, p = 0.04, and r triglyceride = −0.37, p = 0.02), Roseburia (r BMI = −0.36, p = 0.02 and r waist = −0.36, p = 0.02), Faecalibacterium (r insulin = −0.35, p = 0.02), Eubacterium ventriosum (r BMI = −0.33, p = 0.03 and r HDL-c = 0.33, p = 0.04), and Akkermansia (r 2h glucose = −0.35, p = 0.02) abundances were correlated to a better cardiometabolic profile, while Streptococcus (r systolic BP = 0.44, p = 0.004, r diastolic BP = 0.51, p < 0.001, r insulin = 0.33, p = 0.04, and r LDL-c = 0.40, p = 0.009) and Desulfovibrio (r BMI = 0.42, p = 0.006, r diastolic BP = 0.37, p = 0.02, r insulin = 0.32, p = 0.04, and r LDL-c = 0.40, p = 0.01) abundances to a worse profile.

In Ruminococcaceae enterotype, Blautia abundance was directly correlated to systolic (r = 0.20, p = 0.02) and diastolic BP (r = 0.22, p =0.008) and inversely to HDL-c levels (r = −0.20, p = 0.02). Roseburia was correlated to unfavorable lipid profile (r LDL-c = 0.24, p < 0.01 and r HDL-c = −0.28,

p < 0.001) and Eubacterium hallii to BMI (r = 0.21, p = 0.02), while Bifidobacterium (r total cholesterol = −0.21, p = 0.01) and Haemophilus (r diastolic BP = −0.23, p < 0.01) to better cardiometabolic parameters.

#### DISCUSSION

The three enterotypes, described in populations from the North hemisphere, were found in the Brazilian population in a similar structure as previously described. Our observation of increased proportion of strict vegetarians in the Prevotella enterotype supports that dietary habits are important determinants of commensal bacteria clustering. Additionally, vegetarian diet associated with lower LDL-c levels suggest that the presence of a protective bacterial cluster in this enterotype could be driving this association. Such consistency of findings in the Prevotella cluster was not seen in the other enterotypes, in which we observed different correlations between bacterial abundances and cardiometabolic parameters. A broader variety of the dietary components of the subjects from the Bacteroides and Ruminococcaceae enterotypes could have limited identifying the relationship between bacteria and risk factors.

The main phyla, Firmicutes and Bacteroidetes, as well as the most common commensal genera that usually dominate the human gut microbiota, were observed in our sample. Cluster analyses clearly identified the three bacterial groups previously described (Arumugam et al., 2011). Other studies conducted in American, Korean, and Danish populations failed to demonstrate them (Wu et al., 2011; Lim et al., 2014; Roager et al., 2014), which may be attributed, in parts, to differences in methodological approaches to clustering data (Koren et al., 2013).

An opportunist characteristic of our sample was the diversity of dietary patterns, which allowed investigating how the participants were distributed among the bacterial clusters. The diversity facilitated our interpretation of possible physiological roles of bacteria present in the enterotypes. Our findings suggest that different diet-dependent combinations of bacteria should result in different effects on the cardiometabolic profile. Apparently, the importance of genetic factors, breastfeeding, and other early life events for the cardiometabolic risk cannot be neglected.

Lower LDL-c levels were found in subjects belonging to Prevotella enterotype, which is consonant with the greater number of strict vegetarians in this enterotype. We speculated that the absence of animal food-derived saturated fatty acids could account for this result (Bradbury et al., 2014; Le and Sabaté, 2014), although our methods are unable to confirm such assumption. Participants classified in this enterotype were not leaner or had lower plasma glucose levels, as previously reported in subjects consuming plant-based diet (Le and Sabaté, 2014; Sabaté and Wien, 2015). Few studies examined the association of enterotypes with cardiometabolic risk factors (Zupancic et al., 2012; Lim et al., 2014); one conducted in Koreans reported increased uric acid concentration in Bacteroides cluster compared to the other enterotypes (Lim et al., 2014). As far as we know, this is the first study that detects differences in lipid metabolism using bacterial clustering. Furthermore, when consumers of diverse diets were stratified according to enterotypes, the lowest LDL-c values seen in Prevotella enterotype seems to be independent of the dietary habit. This finding suggests that bacteria associated with Prevotella may be important drivers of the effect in lipid metabolism.

We tested the correlations of bacteria with cardiometabolic variables within each enterotype to clarify the pathophysiological relationship. We found diverging relationships between a given genus and metabolic parameter when compared one enterotype to another. Blautia abundance was favorably correlated to anthropometric measurement in Prevotella cluster, but in Bacteroides and Ruminococcaceae enterotypes showed an unfavorable relationship with cardiometabolic parameters.

This genus belonging to Lachnospiraceae family was more commonly found in animals consuming herbivore diet and is known due to its capacity to degrade complex polysaccharides to short-chain fatty acids, such as butyrate, acetate, and propionate (Furet et al., 2009; Biddle et al., 2013; Eren et al., 2015). Several fermentation-dependent metabolic benefits have been described. Butyrate stimulates enteroendocrine cells to secrete incretins (Kasubuchi et al., 2015; Woting and Blaut, 2016), inhibits of pro-inflammatory cytokines production (Miquel et al., 2013; Hippe et al., 2016) and enhances expression of tight-junction proteins (Cani et al., 2009; Peng et al., 2009) that improve gut barrier and reduce metabolic endotoxemia. Such effects have a protective impact on obesity and insulin resistance (Cani et al., 2009; Brahe et al., 2015; Kasubuchi et al., 2015; Hippe et al., 2016). Our findings suggest that this could be occurring in subjects belonging to the Prevotella enterotype. Since a high number of vegetarians was present in this enterotype, we suggest that butyrate-producing bacteria should contribute inducing several metabolic benefits.

Only in Prevotella enterotype, abundances of other butyrateproducing bacteria, E. ventriosum, Roseburia, Coprococcus, and Faecalibacterium (Barcenilla et al., 2000; Pryde et al., 2002; Brahe et al., 2015), showed correlations that are suggestive of a protective role of increased body adiposity and metabolic disturbances. Interestingly, the positive relationship between E. ventriosum abundance and HDL-c had not been described. This correlation was not an unexpected finding since another butyrate property is the capacity of activating the GPR109A, which in turn


TABLE 1 | Mean values (±standard deviation) of clinical and biochemical data of 268 participants according to their enterotypes.

BP, blood pressure.#Log-transformed values for analysis and were back-transformed to return to the natural scale. ANOVA followed by Bonferroni post hoc test. vs. Bacteroides.

regulates lipid homeostasis (Elangovan et al., 2014). Also, this effect is coherent with lower LDL-c levels observed in participants belonging to Prevotella enterotype. Coprococcus was previously associated with adequate bacterial richness in healthy lean adults (Furet et al., 2010) and high abundance of Faecalibacterium in subjects consuming fiber-enriched diets (Canani et al., 2011; Matijašic´ et al., 2014) and low in those with obesity and type 2 diabetes (Furet et al., 2010; Zhang et al., 2013). Our findings are in agreement with the majority of investigators who suggested that these genera abundances are markers of gut health (Miquel et al., 2013; Martín et al., 2015; Hippe et al., 2016), but not all (Balamurugan et al., 2010; Feng et al., 2014). Additionally, the correlation of Akkermansia abundance and plasma glucose is consistent with previously reported benefits of this genus in inflammatory status and glucose metabolism (Everard et al., 2013; Schneeberger et al., 2015; Greer et al., 2016).

We speculate that the enterotype-mediated risk pattern is dependent of the local microenvironment, and the combination of abundant bacteria in each enterotype would drive the pathophysiological outcomes. The fiber-rich diet of vegetarians included in the Prevotella enterotype could have triggered beneficial effects at the intestinal and systemic levels. Therefore, our findings are consistent reports of favorable cardiometabolic risk profile in subjects consuming diets rich in fruits and vegetables like the Adventists (Pettersen et al., 2012; Sabaté and Wien, 2015).

Interestingly, in Ruminococacceae enterotype, Eubacterium hallii, and Roseburia were unfavorably associated with metabolic parameters, while Desulfovibrio and Haemophilus, from the Proteobacteria phylum, with a protective relationship. It is well-known that the latter are gram-negative bacteria with lipopolysaccharide on its surface. This endotoxin is an important ligand for toll-like receptor 4 that activates the innate immune system, which could result in a pro-inflammatory condition (Cani et al., 2009; Velloso et al., 2015). Considering that Proteobacteria preferentially metabolize proteins (Ferrocino et al., 2015), higher abundance of bacteria from this phylum could be expected in Bacteroides and Ruminococacceae enterotypes, in which lacto-ovo-vegetarians and omnivores were more commonly present. This agrees with a report of high abundance of Proteobacteria in children consuming a protein-fat based diet (De Filippo et al., 2010). However, some inverse correlations with cardiometabolic factors were unexpectedly detected in both enterotypes. Only in the Prevotella enterotype, Desulfovibrio abundance was directly correlated to BMI, BP, insulin, and LDLc, in line with previous animal and human studies. In db/db mice (Geurts et al., 2011) and humans with cardiovascular diseases (Yin et al., 2015) compared to respective controls, Proteobacteria was more abundant. Such results may reinforce that the resulting balance of a great variety of bacteria present in gut drives metabolic processes in the host. Therefore, different diet-dependent combinations of bacteria would be related to distinct cardiometabolic risk profile. Comparisons of clinical data of subjects within each diet stratified by enterotype reinforced our assumption that enterotype may be driving the dietary lipidassociated risk since the LDL-c values were invariably lower in the subsets of participants from the Prevotella enterotype.

In all enterotypes, the abundance of Streptococcus was correlated to unfavorable cardiometabolic risk profile (increased adiposity, BP, and lipids), although correlation coefficients in the Ruminococcaceae enterotype were weak (data not shown). This genus belongs to Firmicutes phylum, which was originally described as the predominant in animal obesity (Ley et al., 2005; Turnbaugh et al., 2006). We have reported a greater abundance of Streptococcus alactolyticus in obese animals compared to hypertensive and Winstar rat (Petriz et al., 2014). Our correlations might be in part due to its proinflammatory role previously described (Al-Jashamy et al., 2010; Jiang et al., 2015).

Our study has limitations. Regarding the dietary intake assessment, raw data were not available impeding to establish associations of nutrients and the microbiota. Determination of fecal supernatants would be desirable to support the assumption of a lower content of fat among strict vegetarian subjects. Fecal consistency was not systematically obtained and was not considered as a confounder in our analyses. Recently, the influence of fecal consistency with gut microbiota richness and composition and bacterial growth rates has been raised (Vandeputte et al., 2016). Our sample from the ADVENTO study is not representative of the general population living in Brazil. As a matter of fact, smoking and drinking habits are known to be less frequent among Adventists. On the other hand, such characteristics should have contributed to minimizing confounders in our analyses.

In conclusion, the three enterotypes previously described are present in Brazilians, supporting that those bacterial clusters are not population-specific. Diet-independent lower LDL-c levels in subjects from Prevotella than in other enterotypes suggest that a

protective bacterial group in the former should be driving this association. Enterotypes seem to be useful to understand the impact of daily diet exposure on cardiometabolic risk factors. Prospective studies are needed to confirm their utility for predicting phenotypes in humans.

#### AUTHOR CONTRIBUTIONS

Ad, BA, SF had substantial contributions to the conception and design of the work. Ad, EG, AP, SF to the acquisition of data. Ad, GF, Id, SF to the analysis and interpretation of data for the work. Ad, Id, SF drafted the work and GF, BP, EG, AP revised it critically for important intellectual content. Ad, GF, Id, BP, EG, AP, SF participated of the final approval of the version to be published. Ad, GF, Id, BP, EG, AP, SF 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.

### FUNDING

The present study was supported by FAPESP (2012/12626-9 and 2012/03880-9).

#### ACKNOWLEDGMENTS

Authors thank the FAPESP, Advento Study Group<sup>∗</sup> and participants. <sup>∗</sup>Members of the Advento Study Group: I. J. M. Bensenor, P. A. Lotufo, K. R. M. Gomes, L. C. B. Soares, V. Kunz, N. V. Silva, L. A. Portes, D. T. Kanno, L. F. Sella, R. França, M. C. Teixeira, S. Gasparini, E. O. L. Ferreira, B. Bonifácio, T. C. Souza, F. M. Diaz, S. C. C. Dammann, I.R. Pinheiro, W. F. S. Costa, D. M. S. Larchert, D. F. Nunes, J. S. Amorim, E. M. Reis, I. P. Manfrim, N. V. Ferreira, J. L. V. Passos, E. Barreto.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00047/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 de Moraes, Fernandes, da Silva, Almeida-Pititto, Gomes, Pereira 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.

# Human Enterovirus 68 Interferes with the Host Cell Cycle to Facilitate Viral Production

Zeng-yan Wang<sup>1</sup> , Ting Zhong<sup>2</sup> , Yue Wang<sup>3</sup> , Feng-mei Song<sup>4</sup> , Xiao-feng Yu<sup>5</sup> , Li-ping Xing<sup>1</sup> , Wen-yan Zhang<sup>5</sup> , Jing-hua Yu<sup>5</sup> \*, Shu-cheng Hua<sup>1</sup> \* and Xiao-fang Yu<sup>5</sup> \*

*<sup>1</sup> Department of Internal Medicine, The First Hospital of Jilin University, Jilin University, Changchun, China, <sup>2</sup> Medicinal Chemistry, College of Pharmacy, Changchun University of Chinese Medicine, Changchun, China, <sup>3</sup> Chemistry of Traditional Chinese Medicine, College of Pharmacy, Changchun University of Chinese Medicine, Changchun, China, <sup>4</sup> Department of Experimental Pharmacology and Toxicology, School of Pharmacy, Jilin Univrsity, Changchun, China, <sup>5</sup> Institute of Virology and AIDS Research, The First Hospital of Jilin University, Jilin University, Changchun, China*

Enterovirus D68 (EV-D68) is an emerging pathogen that recently caused a large outbreak of severe respiratory disease in the United States and other countries. Little is known about the relationship between EV-D68 virus and host cells. In this study, we assessed the effect of the host cell cycle on EV-D68 viral production, as well as the ability of EV-D68 to manipulate host cell cycle progression. The results suggest that synchronization in G0/G1 phase, but not S phase, promotes viral production, while synchronization in G2/M inhibits viral production. Both an early EV-D68 isolate and currently circulating strains of EV-D68 can manipulate the host cell cycle to arrest cells in the G0/G1 phase, thus providing favorable conditions for virus production. Cell cycle regulation by EV-D68 was associated with corresponding effects on the expression of cyclins and CDKs, which were observed at the level of the protein and/or mRNA. Furthermore, the viral non-structural protein 3D of EV-D68 prevents progression from G0/G1 to S. Interestingly, another member of the *Picornaviridae* family, EV-A71, differs from EV-D68 in that G0/G1 synchronization inhibits, rather than promotes, EV-A71 viral replication. However, these viruses are similar in that G2/M synchronization inhibits the production and activity of both viruses, which is suggestive of a common therapeutic target for both types of enterovirus. These results further clarify the pathogenic mechanisms of enteroviruses and provide a potential strategy for the treatment and prevention of EV-D68-related disease.

Keywords: enterovirus 68 (EV-D68), cell cycle, G0/G1 arrest, viral replication, host-pathogen interaction

#### INTRODUCTION

Human enterovirus 68 (EV-D68) is an emerging pathogen that can cause severe respiratory disease and is associated with cases of paralysis, especially among children. It was first isolated from samples obtained in California in 1962 from four children with pneumonia and bronchiolitis (Schieble et al., 1967). Over the past 10 years, EV-D68 infection outbreaks have been reported in Italy, the United States, Germany, China, and several other countries (Esposito et al., 2015; Farrell et al., 2015; Reiche et al., 2015; Zhang et al., 2015), with a record number of confirmed cases in 2014 (http://www.cdc.gov/non-polio-enterovirus/about/ev-d68.html). Unfortunately, no vaccines for prevention or medicines for treatment are currently available for future outbreaks, mainly due to the fact that information on host factors required for EV-D68 replication is scarce.

#### Edited by:

*Chioma M. Okeoma, University of Iowa, USA*

#### Reviewed by:

*Jingwen Wang, Yale University, USA Xin Zhao, Institute of Microbiology (CAS), China*

\*Correspondence:

*Jing-hua Yu yjh-0-2002@163.com Shu-cheng Hua shuchenghua@eyou.com Xiao-fang Yu yuxiaofang@jlu.edu.cn*

Received: *08 October 2016* Accepted: *20 January 2017* Published: *08 February 2017*

#### Citation:

*Wang Z-y, Zhong T, Wang Y, Song F-m, Yu X-f, Xing L-p, Zhang W-y, Yu J-h, Hua S-c and Yu X-f (2017) Human Enterovirus 68 Interferes with the Host Cell Cycle to Facilitate Viral Production. Front. Cell. Infect. Microbiol. 7:29. doi: 10.3389/fcimb.2017.00029*

EV-D68 belong to enterovirus (family Picornaviridae, genus Enterovirus), which are non-enveloped, positive-sense singlestrand RNA viruses of approximately 7500 nt and contain a large open reading frame that encodes a polyprotein that is cleaved to yield corresponding viral proteins. Based on the molecular and biological characteristics, four human enterovirus (HEV) species are currently designated as HEV-A, -B, -C, and -D (Oberste et al., 1999a,b). The representative of HEV-A serotype is human enterovirus 71 (EV-A71), which is a primary causative agent for Hand, foot, and mouth disease (HFMD) that is associated with the recent outbreaks in Asia (Liu et al., 2011; Wang et al., 2012). EV-D68 is assigned to HEV-D serotype, but EV-D68 is unlike other enteroviruses in that it is acid labile and biologically more similar to human rhinoviruses that are associated with respiratory diseases (Smura et al., 2010).

As a feature of their pathogenic mechanism, many viruses facilitate their own replication by interacting with host factors that regulate cell cycle progression. Examples can be discovered in DNA viruses, retroviruses and RNA viruses. DNA viruses, which replicate in the nucleus, have been extensively investigated in regard to control the cell cycle of host cells. For example, some small DNA viruses including simian virus 40 (DeCaprio et al., 1988), adenovirus (Howe et al., 1990; Eckner et al., 1994), and human papillomavirus (Werness et al., 1990), which lack their own polymerases, use the host polymerase to promote the entry of cells into S phase from G0/G1 phase. For other large DNA viruses, for example, herpesviruses can induce G0/G1 arrest in order to avoid competing for cellular DNA replication resources (Flemington, 2001). Cell cycle regulation also has been observed for retroviruses, which, like DNA viruses, replicate in the nucleus. The Vpr protein of human immunodeficiency virus type 1 is responsible for eliciting cell cycle arrest in G2/M phase (He et al., 1995; Goh et al., 1998). Furthermore, RNA viruses, whose primary site of replication is normally the cytoplasm, have also been demonstrated to interfere with the host cell cycle. Infectious bronchitis virus (IBV) induces an S and G2/M-phase arrest to favor viral replication (Dove et al., 2006; Li et al., 2007), and mouse hepatitis virus (MHV) (Chen and Makino, 2004) and some severe acute respiratory syndrome coronavirus (SARS-CoV) proteins induce cell cycle arrest in G0/G1 phase (Yuan et al., 2005, 2006). In a previous study, we found that human EV-A71 and Coxsackievirus A16, manipulate the host cell cycle at S phase in order to promote their own viral replication (Yu et al., 2015); however, the potential manipulation of the host cell cycle by EV-D68, which is associated with higher lethality in recent large-scale outbreaks, has not been previously characterized.

In the current study, we examined the effects of the cell cycle status on EV-D68 viral replication, as well as the impact of EV-D68 virus on the host cell cycle. Our data show that EV-D68 replication is integrally associated with the host cell cycle, though the pattern of regulation differs distinctly from that of EV-A71. These results further increase the understanding of the pathogenic mechanisms of enteroviruses and provide a potential target for the treatment and prevention of enterovirus-related diseases.

## MATERIALS AND METHODS

#### Viruses and Cells

The Fermon (ATCC, VR-1826), US/KY/14-18953 (ATCC, VR-1825D), and US/MO/14-18947 (ATCC, VR-1823D) strains of EV-D68; and the Changchun077 strain of EV-A71 have been reported previously (Wang et al., 2012). Viruses were propagated in human rhabdomyosarcoma RD cells (No CCL-136), and the supernatants were harvested and stored at −80◦C. Human embryonic kidney cells (HEK 293T cells) (No CRL-11268) and RD cells were purchased from the ATCC (Manassas, VA, USA) and used according to a previous study (Wang et al., 2015). Cells were maintained in Dulbecco's modified Eagle's medium (DMEM) (Hyclone, Logan, UT, USA) supplemented with 10% fetal bovine serum (FBS) (GIBCO BRL, Grand Island, NY, USA).

### Viral Titer Determination

The viral titers were determined by measuring the 50% tissue culture infective dose (TCID50) in a microtitration assay using RD cells, as described (Gay et al., 2006). RD cells were seeded and incubated at 37◦C for 24 h in 96-well plates. Viruscontaining supernatant was serially diluted 10-fold, and 100µl of diluent virus was added per well in octuplicate. Until the experimental endpoint was reached the cytopathic effect was observed once per day. According to the Reed-Muench method (Reed, 1983) the viral titers of the TCID50 were determined, based on the assumption that material with 1 × 10<sup>5</sup> TCID50/ml will produce 0.7 × 10<sup>5</sup> plaque forming units/ml (www.protocol-online.org/biology-forums/posts/1664.html).

#### Infection

Cells were mock-infected or infected with EV-D68 or EV-A71 at a multiplicity of infection (MOI) of 0.8. After 2 h of virus adsorption, cells were washed with phosphate-buffered saline (PBS) one time, then added fresh culture medium.

#### Cell Cycle Release

Subconfluent cultures of RD cells were synchronized in G0/G1 phase by serum deprivation (He et al., 2010). Approximately 5 × 10<sup>5</sup> cells were plated in a 6-well plate and maintained in serumfree medium for 24 h. After EV-D68 virus infection, fresh 10% DMEM was added to release the cells from G0/G1.

## Synchronization of Cells

In order to observe the effects of the cell cycle on virus growth, subconfluent cultures of RD cells were synchronized in G0/G1 phase by serum deprivation for 24 h (He et al., 2010). For S-phase synchronization, a final concentration of 0.85 mM thymidine (Sigma) were added (Helt and Harris, 2005; Yu et al., 2015) for 24 h. For G2/M synchronization, 25 ng/ml of nocodazole (Sigma) was added (He et al., 1995; Yu et al., 2015) for 24 h. For sustained S and G2/M cell-cycle arrest after virus infection, cells were treated with fresh 0.85 mM thymidine and 25 ng/ml nocodazole for the indicated times.

## Cell Cycle Analysis by Flow Cytometry

Propidium iodide (PI) staining was used to measure the nuclear DNA content according to previous study (Yu et al., 2015). Firstly, the cells were collected and fixed with 1 ml of cold 70% ethanol at 4◦C overnight and then re-suspended in PI staining buffer (50µg/ml PI (Sigma), 20µg/ml RNase in PBS) for 2 h at 4◦C. Fluorescence-activated cell sorting (FACScan; BD) were used to analyze the PI-stained cells, and at least 10,000 cells were counted for each sample. ModFit LT, version 2.0 (Verity Software House) was performed for data analysis.

#### Western Blot Analysis

Virus-infected or mock-infected cells were collected at various times after EV-D68 infection and washed once with PBS as previously described (Yu et al.). The following antibodies were used in Western blot analyses: anti-CDK2 (Cell Signal), anti-cyclinE1 (Proteintech), anti-CDK4 (Cell Signal), anti-CDK6 (Cell Signal), anti-cyclinD (Cell Signal), anti-CDK1 (Boster), anti-cyclinB1 (Santa Cruz), and anti-histone (GenScript). Secondary antibodies from mouse or rabbit were obtained from Jackson Immuno Research.

### Quantitative Real-Time PCR

All work was carried out in a designated PCR-clean area as previously described (Yu et al.). RNA was extracted from infected and uninfected cells using Trizol reagent (Gibco-BRL, Rockville, Md.) and isolated as specified by the manufacturer. The RNA was DNAse-treated (DNase I-RNase-Free, Ambion) to remove any contaminating DNA; 200 ng of total RNA was reverse-transcribed with oligo dT primers using the High Capacity cDNA RT Kit (Applied Biosystems) in a 20 µl cDNA reaction, as specified by the manufacturer. For quantitative PCR, the template cDNA was added to a 20 µl reaction with SYBR GREEN PCR Master Mix (Applied Biosystems) and 0.2 µM of primer (**Table 1**). The amplification was carried out using an ABI Prism 7000 for 40 cycles under the following conditions: initial denaturation at 95◦C for 10 min; 40 cycles of 95◦C for 15 s and 60◦C for 1 min. The fold changes were calculated relative to GAPDH using the 11Ct method.

#### Enzyme-Linked Immunosorbent Assays

The cell lysates were examined for CDK4, CDK6, cyclinD1, CDK2, cyclinE1, CDK1, cyclinB1 and histone with ELISA kits (Meiyan, Shanghai, China) according to the manufacturer's instructions. The microplate was quantified using a microplate reader (Bio-Rad, Hercules, CA, USA). Target protein expression was normalized to the histone expression.

#### Statistical Analyses

Statistical differences were analyzed using the Student's t-test for all analysis, except of 3C and 3D dose-dependent test in **Figures 4B,D** with Pearson correlation coefficient. Data are presented as means and standard deviations (SD). <sup>∗</sup>P-values of < 0.05 were considered statistically significant.

primer for real time PCR and plasmid construct. Role Forward sequence5 ′–3 ′ (restriction enzyme) Reverse sequence5 ′–3 ′ (restriction enzyme) (Fermon) Real time CACCATACTCACAACTGTGGC AATGAAATGAATCCTGCTCCT Real time GCCCTTACTCCAGAAAAACA CAAAACCATCATAGAAAACT (US/MO/14-18947)Real time CGTGGGTCTTCCTGACTTGA GGGGGGTCGGAGATTTTAAA Real time AGCACCCACAGGCCAGAACACAC ATCCCGCCCTACTGAAGAAACTA Real time TCAGGGTATCAGTGGTGCGA CAAATCCAAGCTGTCTCTGTG Real time CTCCTGGGCTCGAAATATTATTCCACAG CCGGAAGAGCTGGTCAATCTCAGA Real time AAGCCGACCAGTTGGGCAAAAT GCTCCACGGGGCAGGGATACAT Real time GGTCAGGTTGTTTGATGTGTGC TATCCTTTATGGTTTCAGTGGG Real time CTACTACCGCCTCACACGCTTC TCCTCCTCCTCTTCCTCCTCCT Real time TCAAGTGGTAGCCATGAAAAAA TAACCTGGAATCCTGCATAAGC Real time TGGCCTCACAAAGCACATGA GCTGTGCCAGCGTGCTAATC Real time GCAAATTCCATGGCACCGT TCGCCCCACTTGATTTTGG Plasmid construct AACTGCAGACCATGTACCCTTACGACGTCCCAGATTACGCGGGTGAGATAGTTAGCAATGAGA (PST1) CGGGATCCCTAAAACGAATCTAACCATTTCCG (BamH1) Plasmid construct AACTGCAGACCATGTACCCTTACGACGTCCCAGATTACGCGGGACCAGGATTTGATTTT (PST1) CGGGATCCCTATTGTGTATCAGTAAAGTAAGAGT (BamH1)

TABLE 1 | The

Primer pair

EV-D68 VP1

EV-D68 VP1

(US/KY/14-18953)

EV-D68 VP1

EV-A71 VP1

CyclinE

CDK2 CDK4 CDK6 CylcinD

CDK1 CyclinB1

GAPDH

3D 3C

## RESULTS

## Synchronization at Different Cell Cycle Stages Has Profound Effects on EV-D68 Production

Viral replication often is integrally associated with the cell cycle status of host cells (Feuer et al., 2002). To explore the possible benefits of different cell cycle phases for EV-D68 viral replication, we synchronized cells in different phases and then assessed viral replication and virulence. First, we assessed the effects of G0/G1 synchronization by serum deprivation (Darzynkiewicz et al., 1980). RD cells were cultured in either serum medium (control) or serum-free medium (G0/G1 synchronization) for 24 h. Then the cells were infected with the same titer of 0.8 MOI of EV-D68 (Fermon strain) or were mock-infected for 2 h, and either serum medium or serum-free medium was added for another 24 h (**Figure 1A**). As previously reported (He et al., 2010), serum deprivation induced obvious G0/G1 arrest as assessed by flow cytometry (P < 0.001; **Figure 1B**). At 2 h post-infection (viral entry stage), the EV-D68 genomic RNA levels were not significantly different in the control and serumstarved cells (**Figure 1M**); however, at 18 h post infection (viral replication stage) 13.55 times more viral RNA was detected in the serum-starved cells than in the control cells (P < 0.01; **Figure 1C**). Furthermore, at 24 h (viral production stage) the TCID50/mL of infectious EV-D68 particles was 348.84 times higher for supernatant from G0/G1 phase-synchronized cells (202.17 ± 42.60 × 10<sup>5</sup> ) than for supernatant from control cells (0.59 ± 0.08 × 10<sup>5</sup> ) (P < 0.01; **Figure 1D**). These results suggest that G0/G1-phase arrest does not affect viral entry, but promotes EV-D68 viral replication and production.

To determine whether viral replication and production also is elevated at other phases of the cell cycle, the effect of S phase synchronization was assessed. The cells were cultured in medium or were synchronized in S phase by culture with 0.85 mM thymidine for 24 h. Then, the cells were mock infected or were infected with 0.8 MOI of EV-D68 for 2 h, and fresh culture medium or 0.85 mM thymidine was added for another 24 h (**Figure 1E**). Thymidine induced obvious S phase arrest (P < 0.001; **Figure 1F**). The genomic RNA level remained similar in S phase-synchronized cells and control non-synchronized cells at 2 h post-infection (**Figure 1M**) and at 24 h post-infection (P > 0.05; **Figure 1G**). Furthermore, the TCID50/mL values at 24 h post-infection were equivalent for the S phase-synchronized cell supernatant (2.59 ± 1.37 × 10<sup>5</sup> ) and the control cell supernatant (3.28 ± 1.80 × 10<sup>5</sup> ) (P > 0.05; **Figure 1H**). These results suggest that S-phase arrest does not affect EV-D68 viral entry, replication or production.

To assess the effects of G2/M phase synchronization, cells were cultured in medium or were treated with 25 ng/ml nocodazole for 24 h; then, the cells were mock infected or were infected with EV-D68 at 0.8 MOI for 2 h, and cultured in fresh medium or 25 ng/ml nocodazole for another 24 h (**Figure 1I**). Nododazole induced obvious G2/M arrest (P < 0.001; **Figure 1J**). At 2 h post-infection, there was no significant difference in the genomic RNA level in the control and G2/M phase-synchronized cells (**Figure 1M**); however, at 24 h post-infection the genomic level was lower in the synchronized cells than in the control cells (P < 0.05; **Figure 1K**). Furthermore, at 24 h post-infection the TCID50/mL for supernatant from the G2/M phase-synchronized cells (0.46 ± 0.29 × 10<sup>5</sup> ) was obviously lower than that from the control cells (1.40 ± 0.16 × 10<sup>5</sup> ) (P < 0.01; **Figure 1L**). Therefore, these results suggest that G2/M synchronization does not affect viral entry, but inhibits EV-D68 viral replication and production.

#### EV-D68 Infection Manipulates the Host Cell Cycle and Arrests Cells at G0/G1

Given that EV-D68 replication and production is dependent on the cell cycle, we next asked whether EV-D68 might have the ability to manipulate the host cell cycle to facilitate its own production. RD cells were infected with EV-D68 Fermon stain at an MOI of 0.8, and the cells were collected for cell cycle distribution analysis after 24 h. An obvious increase in the percentage of cells in G0/G1 was observed in EV-D68-infected cells (45.20 ± 0.14%) as compared to mock-infected cells (36.50 ± 0.76%) (23.84% increase; P < 0.01; **Figure 2A**). Therefore, EV-D68 itself can manipulate the host cell to accumulate preferentially at G0/G1 phase rather than at G2/M, which favors viral production.

Though the latter experiments were performed using the Fermon strain of EV-D68, which was isolated from 4 children with pneumonia and bronchiolitis in the United States in 1962 (Schieble et al., 1967; Zhang et al., 2015), several more recent strains of the virus have been isolated. The currently circulating EV-D68 US/MO/14-18947 and US/KY/14-18953 strains are similar to the Fermon strain in clinical characteristics and genome structure, but it is not known whether they are similar in virulence and ability to manipulate the cell cycle. Therefore, we compared the US/MO/14-18947 and US/KY/14-18953 strains to the Fermon strain. Under normal culture conditions, RD cells were infected with three strains virus at an MOI of 0.8 for 24 h, respectively, the TCID50/mL of the US/MO/14-18947 strain (43.07 ± 10.22 × 10<sup>5</sup> ) was 29.76 times higher and the TCID50/mL of the US/KY/14-18953 (195.00 ± 54.03 × 10<sup>5</sup> ) was 138.29 higher than the TCID50/mL of the Fermon strain (1.40 ± 0.16 × 10<sup>5</sup> ) (**Figure 2B**). These results suggest that the EV-D68 virulence has increased over time, which could explain the recent rise in the incidence of Enterovirus-related disease (Esposito et al., 2015; Farrell et al., 2015; Reiche et al., 2015; Zhang et al., 2015).

Next, we assessed the cell cycle distribution after RD cells were infected with the currently circulating strains of EV-D68 at an MOI of 0.8 for 24 h. An obvious increase in the percentage of cells in G0/G1 phase was observed for both US/MO/14- 18947 (25.83% increase; P < 0.01) (**Figure 2C**) and US/KY/14- 18953 (10.85% increase; P < 0.01) (**Figure 2D**). These strains also caused a corresponding increase in the ratio of cells in G0/G1–G2/M (53.54% increase for US/MO/14-18947; 85.91% increase US/KY/14-18953; P < 0.001; **Figures 2C,D**). Therefore, the currently circulating strains possess increased virulence and a similar ability as the Fermon strain to skew the cell cycle toward the G0/G1 phase, which facilitates viral production.

arrest (A–D), S phase arrest (E–H), and G2/M arrest (I–L). (A,E,I) Flow diagram of how RD cells were treated with serum starvation (starved) for G0/G1 synchronization (A), with thymidine (thymi) for S synchronization (E), or with nocodazole (noco) for G2/M synchronization (I). The top diagram in each panel shows the strategy for the control group, and the bottom panel shows the strategy for cell cycle synchronization. (B,F,J) Cell-cycle profiles were determined by flow cytometry after G0/G1, S, and G2/M synchronization with serum starvation, thymidine, and nocodazole treatment, respectively. Histograms below show the percentage of cells in each phase of the cell cycle as analyzed by the ModFit LT program. (C,G,K) Levels of intracellular EV-D68 Fermon strain RNA were detected in RD cells after cell cycle synchronization by quantitative real-time PCR. The results were standardized to GAPDH mRNA expression and normalized to 1.0 in mock-infected cells. (D,H,I) Progeny viruses in the supernatants were titrated using RD cells. A relative quantitative analysis of the TCID50/mL is shown. (M) Intracellular EV-D68 Fermon strain RNA levels were detected in RD cells with different cell cycle synchronization treatment by quantitative real-time PCR at post-infection 2 h. The results were standardized using GAPDH mRNA as a control and normalized to 1.0 in mock-infected cells. The results represent the mean ± S.D of three independent experiments. \**P* < 0.05, \*\**P* < 0.01, and \*\*\**P* < 0.001.

#### EV-D68 Infection Inhibits G0/G1 Exit

To further understand the mechanism of EV-D68 manipulation of the host cell cycle, we assessed whether EV-D68 could regulate cell cycle exit from G0/G1 into S phase. RD cells were synchronized in G0/G1 by serum starvation for 24 h, and then mock-infected or infected with EV-D68 Fermon strain for

of the progeny viruses in the supernatant. The results represent the mean ± S.D of three independent experiments. \*\**P* < 0.01 and \*\*\**P* < 0.001.

2 h. The cells were then stimulated with 10% FBS in order to trigger cell cycle re-entry into S phase from G0/G1. At 24 h of mitogenic stimulation with serum, the mock-infected cells progressed synchronously from G0/G1 into S phase. In contrast, the majority of the EV-D68-infected RD cells remained in G0/G1 phase over the 24 h time period without S entry (P < 0.001; **Figure 3A**). Therefore, these results support a model in which EV-D68 infection regulates the cell cycle by preventing entry into the S phase.

Cyclin/CDK complexes are known to regulate cell cycle progression (Sherr, 1994). To identify the key molecules and signaling pathways that may mediate the inhibition of cell entry into S phase by EV-D68, we examined the protein expression profiles of host G0/G1-phase and S-phase proteins by Western blotting of RD cells at 0, 16, 20, 24, and 28 h post-infection. Among the molecules CDK4, CDK6, and cyclinD (which mediate cell cycle progression in G0/G1; Massagué, 2004) and CDK2 and cyclinE1 (which mediate cell cycle transition from G0/G1 to S phase; Hinds et al., 1992), the expression of CDK6 was not changed, and the expression of cyclinE1 was increased at 24 h post-infecion (**Figure 3B**), while all of them were significantly decreased in virus-infected cells as compared to mock-infected cells at 28 h post-infection (**Figures 3B,G**). Furthermore, the CDK2 mRNA level was decreased by EV-D68 infection; however, there were no significant differences between the virus and mockinfected groups in CDK4, CDK6, cyclinD or cyclinE1 mRNA levels (**Figure 3C**). Therefore, EV-D68 infection inhibits host expression of several cell cycle proteins, which is consistent with its ability to inhibit G0/G1 to S phase entry, and the modulation is likely to occur transcriptionally for CDK2 and post-transcriptionally for CDK4, CDK6, cyclinD, and cyclinE1.

#### EV-D68 Infection Promotes G0/G1 Entry

To further examine the potential effect of virus infection on cell cycle transition from G2/M phase into G0/G1, RD cells were treated with 25 ng/ml nocodazole or medium for 24 h for G2/M phase synchronization and then the cells were mock infected or infected with EV-D68 at 0.8 MOI for 2 h. Next, the cells were treated for an additional 24 h with 25 ng/ml nocodazole or fresh medium. Nocodazole induced obvious G2/M cell cycle arrest (35.91 ± 1.44 vs. 18.77 ± 0.20%; P < 0.01); however, after EV-D68 infection for 24 h, the percentage of cells in G2/M was decreased (21.82 ± 1.07 vs. 35.91 ± 1.44; P < 0.001), and the percentage of cells in G0/G1 was increased (48.59 ± 1.22 vs. 32.51 ± 0.21; P < 0.01) (**Figure 3D**). Therefore, EV-D68 infection also regulates the cell cycle by promoting exit from G2/M phase. Consistent with these findings, the expression of cyclinB1 and CDK1 (which mediate G2/M progression; Coverley et al., 2002; Yam et al., 2002) was down regulated by EV-D68 infection (**Figures 3E,G**). CDK1 was decreased at the mRNA level upon EV-D68 infection (transition fromP < 0.01), but cyclinB1 mRNA expression was not significantly regulated (**Figure 3F**). Therefore, EV-D68 virus promotes cell cycle exit from G2/M and entry into G0/G1 by modifying the pathway of G0/G1 entry at the transcriptional level for CDK1 and at the post-translational level for cyclinB1.

## The Non-structural Proteins 3D and 3C of EV-D68 Mediate Cell Cycle Alterations

A previous study concluded that exogenous expression of EV-A71 viral non-structural 3D protein, an RNA-dependent RNA polymerase, mediates cell cycle arrest at S phase (Yu et al.). Given this finding, we examined whether non-structural 3D protein of EV-D68 had the same ability to mediate cell cycle alteration. Transfection of 3D expression vector (2µg) induced G0/G1 arrest, with an increase in the percentage of G0/G1 cells from 39.37 ± 0.52% to 44.76 ± 1.29% (13.69% increase; P < 0.01) and a corresponding increase in the G0/G1–G2/M ratio from 2.02 ± 0.04 to 2.62 ± 0.18 (29.70% increase; P < 0.01) (**Figure 4A**). Furthermore, the extent of the increase in the percentage of G0/G1 cells was dependent on the dose of 3D vector (0, 0.5, 1, 2µg; R = 0.932; P < 0.001), and the G0/G1 to G2/M ratio also depended on the dose of 3D vector (R = 0.827; P < 0.001; **Figure 4B**). These results suggest that the non-structural protein 3D of EV-D68 contributes to G0/G1 arrest.

We also examined the potential role of non-structural protein 3C in cell cycle regulation by EV-D68. Transfection of high dose 3C expression vector (2µg) decreased the percentage of cells in G2/M phase from 19.53 ± 0.26% to 13.19 ± 0.48% (32.46% decrease) and increased in the ratio of cells in G0/G1 to G2/M from 2.01 ± 0.04 to 2.75 ± 0.17 (36.82% increase; P < 0.01) (**Figure 4C**). High dose 3C transfection also increased the percentage of cells in S phase (from 41.10 ± 0.48% to 50.53 ± 0.73%). Although the change in the cell cycle profile was not dose-dependent, the G0/G1 to G2/M ratio was dose-dependent (R = 0.951; P < 0.001; **Figure 4D**). These results suggest that the non-structural protein 3C may contribute to the enhanced cell cycle exit from G2/M phase after EV-D68 infection.

To verify these findings and to evaluate the mechanism of cell cycle regulation after 3D and 3C transfection, we performed Western blotting assays to examine their effect on the

indicated. Left panel: The histograms show the percentage of each phase in the cell cycle. Right panel: The ratios of cells in G0/G1 to G2/M. (E,F) The expression of cell cycle-related proteins after transfection of 293T cells with 3D, 3C or corresponding vector (mock) was assessed by Western blot analysis at 36 h. Histone is shown as a loading control. (G) Model for the distinct effects of the non-structural proteins 3D and 3C in cell cycle arrest caused by EV-D68. The results indicate the mean ± S.D of three independent experiments. \*\**P* < 0.01 and \*\*\**P* < 0.001.

expression of cycle-related proteins. Consistent with the cell cycle analyses, 3D down-regulated the expression of cyclinD, CDK4, CDK2, and cyclinE1 (**Figure 4E**), while 3C down-regulated the expression of CDK1 and cyclinB1 (**Figure 4F**). Therefore the non-structural protein 3C facilitates exit from G2/M and the non-structural protein 3D mediates arrest in G0/G1 (**Figure 4G**).

### G0/G1-Phase Synchronization Has Distinct Effects on EV-D68 and EV-A71 Viral Replication

EV-D68 (serotype HEV-D) and EV-A71 (serotype HEV-A) are both enteroviruses. We have demonstrated that G0/G1 synchronization promotes EV-D68 viral replication (**Figures 1A–D**). However, in our previous study, we determined that G0/G1 synchronization inhibits EV-A71 viral replication. To exclude experimental variation as an explanation for the disparate responses of these viruses to G0/G1 synchronization, we performed a side-by-side comparison of the two enteroviruses. Our results confirm that no serum treatment induces G0/G1 synchronization (**Figures 5A,B**) and G0/G1 synchronization has opposite effects for the two viruses (**Figure 5**). After 18 h infection with EV-D68, the viral genomic mRNA level was 13.55 times higher in serum-starved cells than in control cells (P < 0.01; **Figure 5C**); however, after 18 h infection with EV-A71, the viral genomic mRNA level was 3.12 times lower in serum-starved cells (P < 0.05; **Figure 5D**). Furthermore, after 24 h infection, the TCID50/mL for EVD68 was 341.66 times higher for G0/G1 phase-synchronized cells

or EV-A71 (B,D,F) at an MOI of 0.8 and the effects of synchronization by starvation were assessed. (A,B) RD cells were serum-starved for 48 h to synchronize cells in G0/G1 phase. The cell-cycle distribution was then detected by flow cytometry. The histograms showed the percentage of each phase in the cell cycle. (C,D) At 18 h post-infection, intracellular EV-D68 and EV-A71 RNA levels were assessed in control medium (con+infected)-treated or no serum (starved+infected)-treated RD cells by quantitative real-time PCR. The results were standardized using GAPDH mRNA as a control and normalized to 1.0 in mock-infected cells. (E,F) At 24 h post-infection, The TCID50/mL was shown through titrating the progeny viruses of EV-D68 or EV-A71 in the supernatants with RD cells. The results represent the mean ± S.D of three independent experiments. \**P* < 0.05, \*\**P* < 0.01, and \*\*\**P* < 0.001.

than for control cells (202.17 ± 42.60 × 10<sup>5</sup> vs. 0.59 ± 0.08 × 10<sup>5</sup> ; P < 0.01; **Figure 5E**), while the TCID50/mL for EV-A71 was 489.37 times lower for G0/G1 phase-synchronized cells than for control cells (0.46 ± 0.15 × 10<sup>5</sup> vs. 225.11 ± 129.36 × 10<sup>5</sup> ; P < 0.05; **Figure 5F**). These results confirm that G0/G1-phase arrest has different effects for the two enteroviruses.

#### G2/M-Phase Synchronization Has Similar Effects on Different Strains of EV-D68 and EV-A71

G2/M synchronization with nocodazole has been shown to inhibit both the EV-D68 Fermon strain in this study (**Figures 1I–L**) and EV-A71 in our previous study (Yu et al., 2015). To determine whether similar effects of G2/M-phase synchronization are observed for the currently circulating strains of EV-D68, cells were treated with 25 ng/ml nocodazole or medium for 24 h, and were then infected at 0.8 MOI for 2 h and treated with 25 ng/ml nocodazole or fresh medium for another 24 h. Our results demonstrate that nocodazole treatment decreased the genomic RNA levels (**Figure 6A**) and the TCID50/ml value (**Figure 6B**) of Fermon, US/KY/14-18953 and US/MO/14-18947, which suggests that the virus inhibition upon G2/M synchronization may be similar for all enteroviruses.

To confirm that G2/M synchronization inhibits EV-D68 and EV-A71, we assessed the effects of an alternate agent that can exert G2/M arrest in vitro, pseudolaric acid B (PAB), which

FIGURE 6 | Synchronization in the G2/M phase inhibits the replication of EV-D68 and SV-A71. (A–C) RD cells were treated with or without 25 ng/mL nocodazole (noco) for 24 h, infected with EV-D68 Fermon (FER), US/MO/14-18947 (MO) or US/KY/14-18953 (KY) strains at an MOI of 0.8 for 2 h, and then treated again with or without 25 ng/mL nocodazole for synchronization. (A) At 24 h post-infection, intracellular EV-D68 RNA levels were detected by quantitative real-time PCR. The results were standardized using GAPDH mRNA as a control and normalized to 1.0 in mock-infected cells. (B) At 24 h post-infection, the TCID50/ml of the *(Continued)*

#### FIGURE 6 | Continued

progeny viruses was determined. (C–G) RD cells were infected with EV-D68 strains US/KY/14-18953 or EV-A71 at an MOI of 0.8 for 2 h, and then treated with 2 µM Pseudolaric acid B (PAB) for 24 h for G2/M synchronization. (C) RD cells were treated with 2 µM Pseudolaric acid B (PAB) for 24 h for G2/M synchronization. Top panel: Cell-cycle profiles were determined by flow cytometry. Bottom panel: The histograms show the percentage of cells in each phase of the cell cycle. (D,E) Intracellular EV-D68 or EV-A71 RNA levels were assessed in control medium (con+infected) or PAB containing medium (PAB+infected)-treated RD cells by real-time quantitative PCR. The results are standardized by GAPDH mRNA as a control and normalized to 1.0 in control-infected cells. (F,G) The progeny viruses of EV-D68 or EV-A71 in the supernatant were titrated using RD cells, and the TCID50/mL was shown. The results indicate the mean ± S.D of three independent experiments. \**P* < 0.05, \*\**P* < 0.01, and \*\*\**P* < 0.001.

is a diterpene acid isolated from the root and trunk bark of Pseudolarix kaempferi Grord (Pinaceae) (Yu et al., 2007, 2013). PAB was confirmed to promote G2/M arrest after 24 h (**Figure 6C**). To assess the effects of PAB on viral RNA and virulence, RD cells were infected with EV-D68 US/KY/14-18953 strain or EV-A71 at an MOI of 0.8 for 2 h and then treated with 2 µM PAB for 24 h. PAB decreased the genomic RNA levels of both EV-D68 (28.32% of control; P < 0.001; **Figure 6D**) and EV-A71 (59.87% of control; P < 0.001; **Figure 6E**). Furthermore, PAB decreased the TCID50/ml value of both EV-D68 (80.31% decrease; P < 0.01; **Figure 6F**) and EV-A71 (91.27% decrease; P < 0.01; **Figure 6G**). These results confirm that G2/M arrest inhibits both the EV-D68 and EV-A71 strains, which suggest a common approach for therapeutic intervention that might potentially target a broader range of enteroviruses.

#### DISCUSSION

Enterovirus 68 (EV-D68) usually causes mild to severe respiratory illness, including runny nose, sneezing, cough, body and muscle ache, wheezing, difficulty breathing, and in the cases of some infants, children and teens, death. Although EV-D68 was first identified in California in 1962, the number of people in one breakout in 2014 with confirmed EV-D68 infection was much greater than the number reported in previous years (http://www.cdc.gov/non-polio-enterovirus/about/ev-d68.html). It is hard to predict whether EV-D68 will emerge again in future outbreaks, but the value of resolving the pathogenic mechanism of EV-D68 is obvious. In this study, we investigated the pathogenic mechanism of EV-D68 to reveal the relationship between virus infection and the host cell cycle.

To assess the possibility that the cell cycle status affects EV-D68 viral replication, we first synchronized cells in G0/G1. Our results demonstrate that G0/G1 arrest promotes EV-D68 replication and increases viral virulence without affecting virus entry. We also assessed the effects of S phase and G2/M phase synchronization on viral production. Our results suggest that S phase synchronization does not affect viral entry, replication or production compared to the control treatment, while G2/M synchronization inhibits viral replication and decreases viral virulence, but does not affect virus entry. These results indicate that G0/G1 phase is most favorable for EV-D68 replication, that S phase can support some viral production, and that G2/M phase is inhibitory for host viral production.

Given that G0/G1 phase supports EV-D68 production, it would be advantageous for the virus to manipulate the host cell cycle to increase viral production. Indeed, the EV-D68 Fermon strain displayed significant ability to increase the percentage of cells in G0/G1 phase. The EV-D68 Fermon strain was isolated in the United States in 1962 (Schieble et al., 1967), but currently circulating strains, including EV-D68 US/MO/14-18947 and US/KY/14-18953, may be more relevant to current human health. Therefore, we examined whether these two currently circulating strains had similar ability to manipulate the cell cycle. Our results confirmed that the circulating EV-D68 strains manipulate the cell cycle in a similar manner as does the Fermon strain, though the current strains have higher virulence than Fermon. Therefore, after more than 50 years' evolution, EV-D68 still possesses the ability to arrest cells at G0/G1 phase but EV-D68 virulence has increased.

To pinpoint the cause for G0/G1 accumulation upon EV-D68 infection, we analyzed the transitions from G0/G1 into S and from G2/M into G0/G1. After cell cycle release from G0/G1, mock-infected cells entered S phase, while EV-D68-infected cells still accumulated at G0/G1 phase, thus demonstrating that EV-D68 infection prevents S phase entry from G0/G1 phase. Furthermore, after G2/M synchronization, EV-D68 infection promoted cell cycle transition from G2/M into G0/G1 phase. Therefore, EV-D68 infection regulates G0/G1 cell cycle arrest both by promoting G0/G1 phase entry and by inhibiting G0/G1 phase departure. To further analyze the mechanism of host cell cycle manipulation by EV-D68, we assessed the expression of cyclins and CDKs that are known to form complexes to regulate cell cycle progression (Oosthuysen et al., 2015). For example: cyclinD/CDK4 and cyclinD/CDK6 regulate G0/G1 progression (Massagué, 2004); cyclinE/CDK2 regulates S-phase entry from G1 (Hinds et al., 1992); cyclinA/CDK2 regulates Sphase progression by replacing cyclinE (Coverley et al., 2002; Yam et al., 2002); and cyclinB1/CDK1 prevents cell cycle transition from G2/M into G0/G1 (Yu et al., 2013; Adeyemi and Pintel, 2014). We demonstrated that cyclinD, CDK4, CDK6, cyclinE1 and CDK2 are down-regulated after EV-D68 infection, which could explain the ability of EV-D68 to inhibit the transition from G0/G1 to S phase. Furthermore, the expression of cyclinB1 and CDK1 was down-regulated after EV-D68 infection, which is consistent with the ability of EV-D68 to promoting the transition from G2/M to G0/G1. Therefore, expression of cell cycle-related proteins further supports our results suggesting that EV-D68 induces G0/G1 arrest by regulating G0/G1 phase entry and exit. We also analyzed whether the regulation of protein expression occurred at the mRNA level and found that CDK2 and CDK1 were down-regulated by EV-D68, but that the other cell cycle proteins were not. This indicates that the regulation of cellular factors related to the cell cycle by EV-D68 occurs partly at the transcriptional level, but mostly occurs at the post-translational level.

Viral non-structural proteins are often essential for viral replication, so we also evaluated whether EV-D68 might exert its host cell cycle regulatory function via its viral non-structural proteins. Our results confirmed that the non-structural protein 3D increases the percentage of cells in G0/G1 phase by decreasing the expression of the G0/G1 and S phase related-cell cycle proteins cyclinD, CDK4 and cyclinE1, while 3C decreases the percentage of cells in G2/M phase by decreasing the expression of G2/M-related proteins CDK1 and cyclinB1. These findings raise the possibility that the non-structural proteins 3D and 3C may function coordinately in the context of an EV-D68 infection to enhance the percentage of cells in G0/G1 and increase the G0/G1 to G2/M ratio. It is noted that non-structural 3D protein of EV-A71 mediates cell cycle arrest at S phase, while 3D of EV-D68 mediated cell cycle at G0/G1, although until now the detailed reason of the difference is not clear, it is speculated that 290 of same amino acid sequence might bind with a same target which is responsible for the function of cell cycle arrest, while 172 of different amino acid sequence might bind other different factors which is responsible for different ability of cell cycle regulation, and this part will be investigated in the future.

In the current study, we demonstrated that G0/G1 phase arrest was required by EV-D68 production; however, in a previous study, EV-A71 was shown to have the opposite effect (Yu et al., 2015). This difference is surprising given that both EV-A71 and EV-D68 belong to the Picornaviridae family. To further confirm the different activities between the two viruses, we analyzed them side by side, and our results confirm that G0/G1 synchronization promotes EV-D68 viral production but inhibits EV-A71 viral production. Therefore, EV-D68 an EV-A71 can both manipulate the host cell cycle, but the process and outcomes of cell cycle manipulation are entirely different, which could explain why EV-D68 and EV-A71 have varying characteristics, such as the epidemic region (Europe and Asia), the clinical symptoms, and the epidemic size. Meanwhile, these results serve as a reminder that different enteroviruses may require different therapeutic treatments targeting a different set of host factors. Unfortunately, the reason of the difference between them is not clear, but it is speculated that some a host factor(s) generated or kept high level in G0/G1 phase was required by EV-D68 production,

REFERENCES


while EV-A71 required some a host factor(s) generated or kept high level in S phase so that their non-structural 3D were obligated by two viruses to regulate host cell cycle for their own replication.

The main goal of analyzing the mechanism of EV-D68 pathogenesis was to identify new strategies for preventing and treating the disease, so we did additional experiments to further explore G2/M synchronization as an approach to inhibit the replication of different EV-D68 strains. Our results demonstrate that in addition to its effects on the EV-D68 Fermon strain, G2/M synchronization by nocodazole inhibited the replication and virulence of US/KY/14-18953 and US/MO/14- 18947. Additionally, PAB, which is another agent that can induce G2/M arrest, also significantly inhibited the production of EV-D68 and EV-A71. Because nocodazole also has been shown to be effective in inhibiting the production of EV-A71 (Yu et al., 2015), medicines that induce G2/M arrest might be considered as a common approach for inhibiting different types of antienterovirus infection, which provides a new direction for antienterovirus drug development.

#### ETHICS STATEMENT

This study has obtained ethics approval from the ethics committee at the First Hospital of Jilin University.

#### AUTHOR CONTRIBUTIONS

JY, X-fangY, and SH designed the experiments and wrote the paper, ZW and JY conducted the experiments, TZ and YW analyzed cell culture, FS, X-fengY and WZ prepared the virus, LX prepared the reagents.

#### ACKNOWLEDGMENTS

This work was supported by the National Natural Science Foundation of China (81301416) to JY, Postdoctoral Science Foundation of China (2014M561302, 2015T80299) to JY, Norman Bethune Program of Jilin University (2015202) to JY, the Jilin Provincial Science and Technology Department (20140204004YY, 20160414025GH) to JY, and the Department of Human Resources and Social Security of Jilin Province (2016014) to JY. It was also supported by funding from the Chinese Ministry of Science and Technology (2012CB911100 and 2013ZX10001005) to X-fang 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 © 2017 Wang, Zhong, Wang, Song, Yu, Xing, Zhang, Yu, Hua and Yu. 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 Multiple Sclerosis Drugs Inhibit the *In vitro* Growth of Epsilon Toxin Producing Gut Bacterium, *Clostridium perfringens*

#### Kareem R. Rumah<sup>1</sup> \*, Timothy K. Vartanian<sup>2</sup> and Vincent A. Fischetti <sup>1</sup>

*<sup>1</sup> Laboratory of Bacterial Pathogenesis and Immunology, Rockefeller University, New York, NY, USA, <sup>2</sup> The Brain and Mind Research Institute and Department of Neurology, Weill Cornell Medical College, New York, NY, USA*

There are currently three oral medications approved for the treatment of multiple sclerosis (MS). Two of these medications, Fingolimod, and Teriflunomide, are considered to be anti-inflammatory agents, while dimethyl fumarate (DMF) is thought to trigger a robust antioxidant response, protecting vulnerable cells during an MS attack. We previously proposed that epsilon toxin from the gut bacterium, *Clostridium perfringens*, may initiate newly forming MS lesions due to its tropism for blood-brain barrier (BBB) vasculature and central nervous system myelin. Because gut microbiota will be exposed to these oral therapies prior to systemic absorption, we sought to determine if these compounds affect *C. perfringens* growth *in vitro*. Here we show that Fingolimod, Teriflunomide, and DMF indeed inhibit *C. perfringens* growth. Furthermore, several compounds similar to DMF in chemical structure, namely α, β unsaturated carbonyls, also known as Michael acceptors, inhibit *C. perfringens*. Sphingosine, a Fingolimod homolog with known antibacterial properties, proved to be a potent *C. perfringens* inhibitor with a Minimal Inhibitory Concentration similar to that of Fingolimod. These findings suggest that currently approved oral MS therapies and structurally related compounds possess antibacterial properties that may alter the gut microbiota. Moreover, inhibition of *C. perfringens* growth and resulting blockade of epsilon toxin production may contribute to the clinical efficacy of these disease-modifying drugs.

#### *Edited by:*

*Venkatakrishna Rao Jala, University of Louisville, USA*

#### *Reviewed by:*

*Michael L. Vasil, University of Colorado Denver School of Medicine, USA Caitlin S. L. Parello, Biomodels, LLC, USA*

#### *\*Correspondence:*

*Kareem R. Rumah rrumah@rockefeller.edu*

*Received: 09 September 2016 Accepted: 06 January 2017 Published: 25 January 2017*

#### *Citation:*

*Rumah KR, Vartanian TK and Fischetti VA (2017) Oral Multiple Sclerosis Drugs Inhibit the In vitro Growth of Epsilon Toxin Producing Gut Bacterium, Clostridium perfringens. Front. Cell. Infect. Microbiol. 7:11. doi: 10.3389/fcimb.2017.00011* Keywords: multiple sclerosis, oral therapies, anti-bacterial agents, *Clostridium perfringens*, microbiome

## INTRODUCTION

Multiple sclerosis (MS) is the most common non-traumatic neurological disease of young adults in Western Europe and North America (Conway and Cohen, 2010). Although traditionally considered an autoimmune disease that specifically targets central nervous system myelin (Frohman et al., 2006), increasingly, investigators have been pursuing the idea that host-pathogen interactions may play a role in the MS disease process (Collins et al., 2012). Indeed, investigations into how the gut microbiota may trigger or modulate MS relapses are currently under way. With the advent of the first oral treatments for MS, Fingolimod, Teriflunomide, and dimethyl fumarate (DMF), a reasonable question arises. Do these oral medications possess antibacterial properties, and if so, could modulation of gut bacteria contribute to protection against MS relapse?

We have previously proposed that a bacterial neurotoxin, epsilon toxin, from the anaerobic gut bacterium, Clostridium perfringens, may play a pivotal role in triggering newly forming MS lesions (Rumah et al., 2013, 2015; Linden et al., 2015). Epsilon toxin (ETX) is a rational candidate MS trigger due to its tropism for the blood-brain barrier (BBB) and for the myelin sheath; both of which are specifically damaged during each MS relapse (Dorca-Arévalo et al., 2008; Rumah et al., 2013; Linden et al., 2015). Remarkably, newly forming MS lesions display evidence of BBB breakdown, oligodendrocyte cell death and early microglial activation in the absence of a peripheral inflammatory infiltrate (Barnett and Prineas, 2004). While the triggering agent of these early pathologic changes remains unknown, C. perfringens epsilon toxin serves as a provocative candidate due to its tissue specificity and resultant mechanistic plausibility (Rumah et al., 2013, 2015; Linden et al., 2015).

C. perfringens is an anaerobic, spore forming, gram-positive bacillus that is sub-classified into five distinct toxinotypes based on differential exotoxin production (**Table 1**). C. perfringens type A typically colonizes the human gut with a prevalence of 63% among healthy individuals (Carman et al., 2008), while C. perfringens types B and D, the producers of ETX, are commonly found in the intestines of ruminant animals such as sheep, goats, and cattle but not humans (Popoff, 2011). ETX toxin is a potent neurotoxin secreted as a 33 kDa inactive precursor during the logarithmic growth phase of C. perfringens in the mammalian intestine. This poorly active precursor is cleaved by gut trypsin, chymotrypsin and/or an additional clostridial exotoxin, lamda toxin. The 28.6 kDa enzymatic cleavage product permeablizes the gut epithelium, enters the blood stream and binds to receptors on the luminal surface of brain endothelial cells. Once bound to brain microvessels, ETX oligomerizes and forms a heptameric pore in the endothelial cell plasma membrane. Brain endothelial cell damage leads to breakdown of the BBB (Popoff, 2011). In addition to its known effects on BBB vasculature, ETX has been found to specifically bind to and damage myelin when incubated with mammalian brain slices (Dorca-Arévalo et al., 2008; Linden et al., 2015; Wioland et al., 2015). This unique ability to interact specifically with the tissues that are damaged in MS, the BBB, and CNS myelin, makes it a promising candidate as an environmental MS trigger.

Fingolimod was the first oral therapy to be approved for the treatment of MS. It was rationally engineered from the antifungal molecule Myriocin, which was later shown to possess immunosuppressive properties. Fingolimod and Myriocin are both structurally homologous to sphingosine, a lipid that is a necessary component of cell membrane sphingolipids (Strader et al., 2011). Similar to Myriocin, sphingosine is also known to possess antimicrobial properties. However, while Myriocin is antifungal in nature, sphingosine is antibacterial (Fischer et al., 2012). Interestingly, Fingolimod has been shown to mimic sphingosine's antibacterial properties by protecting the cystic fibrosis transmembrane conductance regulator (CFTR) knockout mouse from luminal airway infection by Pseudomonas auerginosa (Pewzner-Jung et al., 2014).

In the context of MS, Fingolimod is phosphorylated in the bloodstream and subsequently binds to the lymphocyte



sphingosine-1-phosphate receptor 1 (S1PR1), causing rapid internalization of S1PR1. In the absence of surface S1PR1, lymphocytes are unable to egress from lymphoid tissues and cannot traffic to target tissues such as the brain; thus the rationale that Fingolimod may reduce the risk of MS relapse and the severity of attacks through immune modulation (Strader et al., 2011).

Teriflunomide is the active metabolite of the immunosuppressant Lenflunomide, which is currently approved for the treatment of rheumatoid arthritis (Munier-Lehmann et al., 2013). Teriflunomide inhibits de novo pyrimidine synthesis in rapidly dividing cells such as clonally expanding lymphocytes, potentially mitigating an autoimmune attack against myelin. More specifically, Teriflunomide non-competitively inhibits dihydroorotate dehydrogenase, an enzyme involved in the first step of de novo pyrimidine synthesis. Memory B cells and T cells remain unaffected by Teriflunomide as they divide more slowly and can synthesize DNA by utilizing the pyrimidine salvage pathway (Bar-Or et al., 2014). Interestingly, dihydroorotate dehydrogenase inhibitors have been shown to arrest the growth of unicellular organisms such as plasmodium falciparum presumably by inhibiting de novo pyrimidine synthesis (Pavadai et al., 2016).

DMF is a fumaric acid ester, which was originally investigated for use as an antimicrobial preservative<sup>1</sup> . It was first used therapeutically to treat psoriasis based on a hypothesis that psoriasis is caused by a defect in fumarate mediated carbohydrate metabolism in the skin. In the early 2000s, a German neurologist noticed that MS patients taking DMF for concurrent psoriasis showed stabilization of their MS symptoms and a reduction in relapse rates (Phillips and Fox, 2013).

DMF has been shown to react with thiol-containing molecules such as the cellular antioxidant, glutathione, and the cysteine residues of proteins via a chemical reaction called the Michael

<sup>1</sup>Pest control. Patent Publication Number: US2218181 A.

addition (Brennan et al., 2015). Although DMF initially depletes mammalian cells of glutathione, its proposed protective action in MS stems from its ability to alkylate key cysteine residues in the redox sensitive protein Kelch-Like ECH-Associated Protein 1 (Keap1). Keap1 normally inhibits Nuclear factor (erythroidderived 2)-like 2 (Nrf2) from translocating to the nucleus and activating antioxidant gene expression. When the cysteine residues of Keap1 are oxidized by reactive oxygen species (ROS) or organic electrophiles such as DMF, Keap1 dissociates from Nrf-2, allowing nuclear translocation to occur. This elicits a robust antioxidant cellular response. The initial decrease in cellular glutathione after DMF treatment is followed by a sharp glutathione increase via the Nrf-2 pathway, which may protect vulnerable cells in MS (Phillips and Fox, 2013).

Although Fingolimod, Teriflunomide, and DMF have proposed mechanisms for how they protect the central nervous system from MS mediated damage, one unexplored possibility is that these orally administered agents may inhibit the growth of neurotoxin-secreting gut bacteria. Because, during log-phase growth, C. perfringens secretes ETX, a toxin that specifically targets the BBB and the myelin sheath, we chose to investigate the effect of these oral MS therapies on C. perfringens growth in vitro.

## METHODS

#### Drugs and Compounds

All drugs and compounds used in this study were purchased from Sigma Aldrich.

#### Bacterial Strains and Growth Conditions

C. perfringens ATCC 13124 (type A), ATCC 3626 (type B), ATCC 51880 (type C), ATCC 3631 (type D), ATCC 27324 (type E), and two type A clinical isolates provided by New York Presbyterian Hospital were used for the initial screen while the "type strain," ATCC 13124, was used for all subsequent experiments. Bacteria were cultured anaerobically at 37◦C overnight using the GasPak 100 system (BD). Anaerobiosis was achieved by pre-reducing the culture media using an anaerobic jar containing a GasPak EZ Anaerobe System sachet for a minimum of 6 h before inoculation. After inoculation, the GasPak sachet was replaced for the overnight culture.

#### Experimental Procedures

The compounds used for the initial growth inhibition screen were diluted to a final concentration of 500 µg/ml in Mueller Hinton broth (BD) and media was inoculated with 5 × 10<sup>6</sup> colonyforming units (CFUs) of different C. perfringens strains. Minimal Inhibitory Concentration values (≥95% growth inhibition, MIC95) were determined for inhibitory compounds using cation adjusted Mueller Hinton II broth (BD). Inhibitory compounds were diluted serially from 512 µg/ml down to 0.5 µg/ml, inoculated with 5 × 10<sup>5</sup> CFUs of C. perfringens and then anaerobically cultured at 37◦C overnight as previously described. Culture conditions for each compound were performed in triplicate and bacterial growth was determined by measuring OD600-values from 1 ml of re-suspended bacteria.

## Statistical Analysis

Results are representative of data obtained from repeated independent experiments. Each value represents the mean ± SD for three replicates. Statistical analysis was performed using the two-tailed Student t-test (GraphPad Software, San Diego, CA, USA).

## RESULTS

With renewed interest in gut bacteria and their potential involvement in the pathogenesis of MS (Bhargava and Mowry, 2014), we wished to determine if oral disease-modifying drugs (DMDs) have the ability to modulate growth of C. perfringens since type B and D strains secrete ETX during log-phase growth. Therefore, we tested if oral DMDs affected the growth of C. perfringens toxinotypes A–E. We compared the growth inhibitory effects of Fingolimod, DMF, and Teriflunomide to that of oral symptom management drugs (SMDs) Baclofen, Bupropion, and Gabapentin; drugs not thought to alter the disease course of MS. We exposed C. perfringens cultures to 500µg/ml of each compound, allowed for overnight anaerobic growth, and determined the optical density (OD600) the following day. We found that each oral DMD significantly inhibited all C. perfringens toxinotypes and strains tested, while the oral SMDs did not (**Figure 1A**). We then plotted the minimal inhibitory concentration (MIC) values for each inhibitory compound and found that Fingolimod was the most potent inhibitor at 4µg/ml (**Figure 1B**).

Because Fingolimod is a homolog of D-sphingosine and Myriocin, both of which have been shown to possess antimicrobial properties (Fischer et al., 2012), we compared the inhibitory activity of Fingolimod to these related sphingoid molecules. We exposed the type strain, C. perfringens ATCC 13124 (type A), to Fingolimod, D-sphingosine, and Myriocin and identified that, like Fingolimod, D-sphingosine also displayed inhibitory activity. Myriocin failed to inhibit C. perfringens, but instead, enhanced bacterial growth (**Figure 2A**). We then plotted and compared MICs for Fingolimod and D-sphingosine and determined that D-sphingosine displayed a similar inhibitory potency to that of Fingolimod with an MIC<sup>95</sup> of 4 µg/ml (**Figure 2B**).

Although an antioxidant mechanism has been proposed for how DMF protects cells against MS mediated damage, DMF was originally investigated for use as an antimicrobial compound<sup>1</sup> . Interestingly, DMF was also found to inhibit the growth of Clostridium botulinum (Dymicky et al., 1987), a bacterial species closely related to C. perfringens. DMF is known to be a Michael acceptor and its ability to affect the redox status of cells stems from its electrophilic nature. Michael acceptors accept electrons during the Michael reaction, while nucleophilic thiols (Michael donors) donate electrons. The Michael reaction results in covalent alkylation of the sulfhydryl group. This covalent linkage permanently inactivates thiol-containing molecules if the thiol is necessary for the molecule's function, as is the case for glutathione and its antioxidant properties (Brennan et al., 2015).

OD600-values similar to that of the DMSO vehicle control. Data are presented as means from three independent experiments. Error bars represent standard deviations, and asterisks indicate that results are statistically significant compared with the DMSO vehicle controls (Student's *t*-test, \**P* < 0.001). (B) Serial dilutions of inhibitory oral DMDs were performed and the type strain, *C. perfringens* ATCC 13124 (type A), was cultured in each condition. OD600-values for each concentration were divided by that of the corresponding dilution for the DMSO vehicle control. MIC-values were plotted for each oral DMD revealing that Fingolimod was the most potent compound with an MIC95 of 4 µg/ml, compared to 128 µg/ml for DMF and Teriflunomide.

significant compared with the DMSO vehicle control (gray); Student's *t*-test, \**P* < 0.0001. (B) Serial dilutions of inhibitory sphingoid molecules were performed and *C. perfringens* ATCC 13124 was cultured at each dilution. OD600-values for each dilution were divided by that of the corresponding DMSO vehicle control dilution. MIC-values were plotted for each of the inhibitory sphingoid compounds revealing that D-sphingosine mimics Fingolimod's antibacterial potency with an MIC95-value of 4µg/ml.

We sought to determine if DMF's antimicrobial activity extended to C. perfringens. Furthermore, we examined DMF's Michael acceptor activity as pertaining to its antimicrobial properties. We screened DMF and its metabolites, monomethyl fumarate (MMF), and fumarate and found that each compound inhibited the growth of type strain, C. perfringens ATCC 13124. However, their saturated succinate counterparts dimethyl succinate (DMS), monomethyl succinate (MMS), and succinate, molecules devoid of Michael acceptor activity due to reduction of the α, β carbon double bond, failed to inhibit C. perfringens (**Figure 3A**). We plotted MIC-values for DMF, MMF, and fumarate and found that DMF was four times more potent than either MMF or fumarate (**Figure 3B**).

Given that Michael acceptor activity was necessary for C. perfringens inhibition by DMF and its fumarate metabolites, we sought to determine if unrelated molecules that share the α, β unsaturated carbonyl structure could also inhibit C. perfringens. We screened Michael acceptors from a diverse group of chemical families and found that natural product Michael acceptors Gambogic acid (a xanthonoid), Parthenolide (a sesquiterpenoid), and Curcumin (a curcuminoid) each inhibited C. perfringens (**Figure 4A**). Interestingly, we found that Gambogic acid was particularly inhibitory with an MIC<sup>95</sup> of 1µg/ml (**Figure 4B**).

To provide additional evidence that Michael acceptor activity is indeed critical to the antibacterial properties of α, β unsaturated carbonyls, we searched the literature to find compounds for which experimental values of Michael reaction potencies have been determined. Dinkova-Kostova et al. determined the potencies of several plant derived Michael acceptors for their ability to induce cellular quinone reductase activity; a cellular marker for a compound's reactivity with sulfhydryl containing molecules (Dinkova-Kostova et al., 2001). In our study, the antibacterial potencies of Cinnamic acid, trans-Chalcone, and Curcumin mirrored the Michael acceptor potencies described by Dinkova-Kostova and colleagues (**Figure 5**). Cinnamic acid was previously shown to be inactive as a Michael acceptor, and in our hands, this compound failed to inhibit C. perfringens. Furthermore, Curcumin was found to be approximately four times more potent than trans-Chalcone (ratio = 4.13; Dinkova-Kostova et al., 2001). Likewise, we found that Curcumin was four times more potent than trans-Chalcone as a C. perfringens inhibitor (ratio = 4).

Since Michael acceptors react with thiols and deplete cellular glutathione levels (Brennan et al., 2015), we surmised that Michael acceptor inhibition of C. perfringens might be abolished by the addition of exogenous glutathione. To test this, we compared C. perfringens growth in the presence of Michael acceptors with and without an equal quantity of exogenous glutathione. We also tested the effect of glutathione on the inhibitory activity of each of the oral MS DMDs. Glutathione completely abolished growth inhibition by the known Michael acceptors in our study, but failed to abolish the inhibitory effects of Fingolimod and Teriflunomide (**Figure 6A**). Because glutathione is an antioxidant, as are vitamins C and E, we sought to determine if the glutathione's abrogation of Michael acceptor antibacterial activity is based on its nucleophilic behavior or due, more generally, to its antioxidant properties. C. perfringens was challenged with DMF in the presence of vitamin C, vitamin E, or the Michael donor, glutathione. Of the antioxidant panel, only the Michael donor, glutathione, was able to neutralize DMF's inhibitory effect (**Figure 6B**).

## DISCUSSION

In this study, we have shown that each of the oral DMDs approved for the treatment of MS, Fingolimod, Teriflunomide, and DMF, inhibits the in vitro growth of the epsilon toxinsecreting gut bacterium, C. perfringens. In contrast, oral therapies used specifically for symptomatic management fail to prevent C. perfringens growth. Of note, Fingolimod proved to be bactericidal, while Teriflunomide and DMF were bacteriostatic (**Supplemental Figure 1**). The antibacterial properties of oral DMDs raises the possibility that modulation of the intestinal microbiota may play a role in the clinical efficacy of these compounds. Preventing C. perfringens growth and toxin production may serve as a specific example of this. Furthermore, we have identified two distinct classes of molecules capable of inhibiting C. perfringens; namely sphingoid compounds such as Fingolimod and D-sphingosine, and Michael acceptors such as DMF, its fumarate metabolites, and various natural products that are α, β unsaturated carbonyls.

Important factors must be considered when attempting to extrapolate these in vitro findings to what may be occurring in the human gut. First, how do the in vitro inhibitory concentrations compare to concentrations found in the human gut? The resting volume of the human stomach is ∼0.08 L (Johnson, 1994), which yields a calculated gut concentration of 6.3 µg/ml for Fingolimod (MIC<sup>95</sup> = 4µg/ml), 87–175µg/ml for Teriflunomide (MIC<sup>95</sup> = 128µg/ml), and 1500–3000 µg/ml for DMF (MIC<sup>95</sup> = 128µg/ml). Therefore, each compound's MIC<sup>95</sup> is within the calculated range of the therapeutic concentration that will enter the small intestine. Furthermore, DMF is a delayed released capsule that dissolves in the more basic pH of the small intestine (Gold et al., 2016). Local release of DMF may increase its concentration in the small intestine where C. perfringens resides.

Second, our experimental growth conditions are likely to be more favorable to C. perfringens growth than the intestinal milieu. Anexic, in vitro growth protects C. perfringens from competition with other bacteria for nutrients. In addition, C. perfringens will not be exposed to toxic molecules secreted by competing bacteria such as bacteriocins or host derived antibacterial molecules such as defensins. Therefore, the MIC<sup>95</sup> for each of the oral DMDs may be considerably less in an environment such as the human intestine where C. perfringens must contend with a multitude of external factors.

Conversely, each of the oral DMDs possesses a significant degree of hydrophobicity, and lipid-binding molecules within the gut lumen may sequester these compounds, preventing toxic interactions with gut bacteria. Specifically considering DMF, a Michael acceptor, extracellular nucleophiles present in the gut may react with its electrophilic β carbon before it can enter the bacterial cell, possibly diminishing its antibacterial activity within the gut.

DMF's Michael reaction-dependent inhibition of C. perfringens growth may be explained by its ability to deplete this bacterium of thiol containing compounds. It is

independent experiments. Error bars represent standard deviations, and asterisks indicate that the inhibition of bacterial growth observed in the presence of unsaturated fumarates is statistically significant when compared to the bacterial growth observed in the presence of corresponding saturated succinates; Student's *t*-test, \**P* < 0.0001). (B) Serial dilutions of inhibitory fumarate compounds were performed and *C. perfringens* ATCC 13124 was cultured at each dilution. OD600-values for each dilution were divided by that of the corresponding DMSO vehicle control dilution. MIC-values were plotted for each of the inhibitory fumarates revealing that DMF is approximately four times more potent that MMF and fumarate.

*C. perfringens* ATCC 13124. Bacteria were grown anaerobically in the presence of 500 µg/ml Gambogic acid, Parthenolide, and Curcumin. Each natural product Michael acceptor successfully inhibited *C. perfringens* growth, similar to what was observed when bacteria were cultured in the presence of known antibiotic penicillin/streptomycin (pen/strep, 100 U/ml). Data are presented as means from three independent experiments. Error bars represent standard deviations, and asterisks indicate that results are statistically significant compared with the DMSO vehicle control (gray); Student's *t*-test, \**P* < 0.0001. (B) Serial dilutions of inhibitory natural product Michael acceptors were performed and *C. perfringens* ATCC 13124 was cultured at each dilution. OD600-values for each dilution were divided by that of the corresponding DMSO vehicle control dilution. MIC-values were plotted for each compound revealing Gambogic acid as the most potent with an MIC95 of 1 µg/ml when compared to Parthenolide and Curcumin, which each inhibit *C. perfringens* at 64 µg/ml.

striking that nucleophilic thiols not only play an important role in mammalian cell homeostasis, but are also necessary substrates for C. perfringens growth. This bacterium depends on an organic source of sulfur (thiols) and will not grow with strictly inorganic sources (SO2<sup>−</sup> 4 , SO2<sup>−</sup> 3 , S2O<sup>3</sup> <sup>2</sup>−, and S<sup>i</sup> ; Fuchs and Bonde, 1957). Therefore, depleting C. perfringens of thiols may contribute to Michael acceptor mediated growth inhibition.

Although Teriflunomide is an α, β unsaturated carbonyl, we have shown that glutathione has no effect on its ability to inhibit C. perfringens growth. It is tempting to speculate that Teriflunomide inhibits de novo pyrimidine synthesis in rapidly dividing bacterial cells, as it does in mammalian cells, via inhibition of dihydroorotate dehydrogenase; a gene that has been annotated for C. perfringens in the Uniprot Knowledgebase. However, we have not examined the inhibitory mechanism of Teriflunomide in the present study.

That Michael acceptors such as DMF and its fumarate metabolites inhibit C. perfringens may open the door to development of new oral MS therapies derived from the Michael acceptor functional class. Gambogic acid has been used in Eastern medicine for centuries to treat intestinal ailments and parasites (Wu et al., 2004), and in our hands, it displays an impressive antibacterial potency (MIC<sup>95</sup> = 1µg/ml).

We searched for Michael acceptors currently approved for human use that possess no known immunosuppressive properties. The naphthoquinone, Menadione (vitamin K3), is a synthetic precursor for vitamin K. It is commonly used as a dietary supplement for livestock and as a cost effective vitamin K replacement therapy in developing countries. Of note, Menadione has recently been shown to inhibit S. aureus and B. anthracis growth, and to suppress S. aureus secretion of toxic shock syndrome toxin 1 (TSST-1; Schlievert et al., 2013). Similarly, we find that Menadione inhibits C. perfringens growth, but related compounds with long aliphatic side chains, vitamin K1, vitamin K2, and ubiquinone do not (**Supplemental Figure 2A**). While Menadione's MIC95-value was found to be 64µg/ml (**Supplemental Figure 2B**), the inactivity of the Menadione related compounds, all of which are electron carriers in the electron transport chain, might be due to the fact that they are sequestered in the cell membrane by their aliphatic side chains. Membrane sequestration may protect cytosolic nucleophiles from undergoing Michael addition and subsequent depletion. Additionally, unlike Menadione but similar to Teriflunomide, these molecules possess a third σ bond at the β carbon position. This may prevent nucleophilic attack due to steric hindrance and abolish Michael acceptor activity (Schwöbel et al., 2010).

In light of the serious side effects associated with current oral DMDs, this study may be of immediate clinical importance. Some of these adverse effects are due to immunosuppression of the CNS, as evidenced by increased risk of JC virus infection and progressive multifocal leukoencephalopathy (PML, FDA Drug Safety Communication, 2014; Brooks, 2015). Perhaps new

cultured anaerobically at a Michael acceptor (DMF, Parthenolide, and trans-Chalcone), and non-Michael acceptor (Fingolimod and Teriflunomide) concentration of 500µg/ml, with (red) or without (blue) the addition of an equal quantity of exogenous Michael donor, GSH. Only Michael acceptor mediated growth inhibition could be abolished by the addition of exogenous GSH. The inhibitory activity Fingolimod and Teriflunomide remained unaffected by the presence of GSH, similar to what was observed when bacteria were cultured in the presence of pen/strep and GSH. Data are presented as means from three independent experiments. Error bars represent standard deviations, and asterisks indicate that GSH aided growth recovery is statistically significant when compared to the lack of growth recovery in the absence of GSH; Student's *t*-test, \**P* < 0.001. (B) *C. perfringens* ATCC 13124 was cultured in the presence of DMF, vitamin C, vitamin E, and GSH each at concentration of 250µg/ml. Only the Michael acceptor, DMF, inhibited bacterial growth. DMF was then paired with the antioxidants, vitamin C, vitamin E, and GSH at concentrations of 250µg/ml for each compound. The Michael donor antioxidant, GSH, abolished DMF inhibition. However, the non-Michael donor antioxidants, vitamin C and vitamin E, were unable to abolish DMF's inhibitory effect on *C. perfringens* growth. Data are presented as means from three independent experiments. Error bars represent standard deviations, and asterisks indicate that results are statistically significant compared with the DMF control (gray); Student's *t*-test, \**P* < 0.001.



antibacterial compounds based on these early oral DMDs, but lacking their immunosuppressive properties, may be of use in treating MS. For example, Fingolimod/D-sphingosine related compounds lacking hydroxyl head groups will not undergo phosphorylation and will not target lymphocyte S1PR1. Such compounds would not be immunosuppressive and may reduce the risk of JC virus infection and the development of PML. Along these lines, we have tabulated the MIC95-values for each inhibitory compound used in this study (**Table 2**).

#### AUTHOR CONTRIBUTIONS

KR conceived the study. KR, VF, and TV designed the study. KR, VF, and TV performed the literature search. KR collected the data and wrote the paper. All authors analyzed the data.

#### FUNDING

This work was generously supported by the Rockefeller University's Robertson Therapeutic Development Fund (RTDF),

#### REFERENCES


the Center for Disorders of the Digestive System (CDDS), and the Weill Cornell Tisch Family Research Fund.

#### ACKNOWLEDGMENTS

We wish to thank Mr. Nick Lewis of Downing LLP for his valuable insights.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2017.00011/full#supplementary-material

Supplemental Figure 1 | Fingolimod is bactericidal, while DMF and Teriflunomide are bacteriostatic. *C. perfringens* ATCC 13124 was anaerobically cultured to stationary phase and exposed to Fingolimod (500 µg/ml), DMF (500 µg/ml), Teriflunomide (500 µg/ml), pen/strep (100 U/ml) or DMSO vehicle control for 4 h under anaerobic conditions. Treated cultures were diluted 1000 fold in fresh, pre-reduced Mueller Hinton broth and cultured anaerobically. Fingolimod inhibited *C. perfringens* growth in a bactericidal fashion, similar to what was observed with the known bactericidal antibiotic mixture pen/strep, as post-treatment dilution and repeat culture failed to restore bacterial growth. Conversely, DMF and Teriflunomide were shown to be bacteriostatic, as post-treatment dilution and culture successfully restored bacterial growth. Data are presented as means from three independent experiments. Error bars represent standard deviations, and asterisks indicate that results are statistically significant compared with the DMSO vehicle control (gray); Student's *t*-test, <sup>∗</sup>*P* < 0.0001.

#### Supplemental Figure 2 | Synthetic vitamin K3, Menadione, inhibits

*C. perfringens* growth. (A) *C. perfringens* ATCC 13124 was anaerobically cultured in the presence of vitamin K homologs, vitamins K1, K2, K3, and ubiquinone. Only the synthetic vitamin K3 (Menadione) inhibited bacterial growth, similar to what was observed when bacteria were cultured in the presence known antibiotic penicillin/streptomycin (pen/strep, 100 U/ml). Conversely, bacteria derived vitamin K2 enhanced *C. perfringens* growth, while plant derived vitamin K1 and mammalian ubiquinone yielded OD600-values similar to that of the DMSO vehicle control. Data are presented as means from three independent experiments. Error bars represent standard deviations, and asterisks indicate that results are statistically significant compared with the DMSO vehicle control (gray); Student's *t*-test, <sup>∗</sup>*P* < 0.001. (B) Serial dilutions of Menadione were performed and *C. perfringens* ATCC 13124 was cultured at each dilution. OD600-values for each dilution were divided by that of the corresponding DMSO vehicle control dilution. MIC-values were plotted yielding an MIC95-value of 64µg/ml.


**Conflict of Interest Statement:** All authors are named as inventors on a pending patent entitled, "Methods to protect against and treat multiple sclerosis," (Publication number CA2899961 A1), which identifies Clostridium perfringens epsilon toxin as candidate trigger for multiple sclerosis.

Copyright © 2017 Rumah, Vartanian and Fischetti. 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.

# Tissue-Associated Bacterial Alterations in Rectal Carcinoma Patients Revealed by 16S rRNA Community Profiling

Andrew M. Thomas 1, 2, <sup>3</sup> , Eliane C. Jesus 1, 4, Ademar Lopes <sup>4</sup> , Samuel Aguiar Jr. <sup>4</sup> , Maria D. Begnami <sup>5</sup> , Rafael M. Rocha<sup>6</sup> , Paola Avelar Carpinetti <sup>7</sup> , Anamaria A. Camargo<sup>7</sup> , Christian Hoffmann<sup>8</sup> , Helano C. Freitas 1, 9, Israel T. Silva<sup>10</sup>, Diana N. Nunes <sup>1</sup> , João C. Setubal 2, 11 and Emmanuel Dias-Neto1, 12 \*

<sup>1</sup> Medical Genomics Laboratory, CIPE/A.C. Camargo Cancer Center, São Paulo, Brazil, <sup>2</sup> Departamento de Bioquímica, Instituto de Química, Universidade de São Paulo, São Paulo, Brazil, <sup>3</sup> Curso de Pós-Graduação em Bioinformática, Universidade de São Paulo, São Paulo, Brazil, <sup>4</sup> Department of Pelvic Surgery, A.C. Camargo Cancer Center, São Paulo, Brazil, <sup>5</sup> Department of Pathology, A.C. Camargo Cancer Center, São Paulo, Brazil, <sup>6</sup> Laboratory of Molecular Gynecology, Department of Gynecology, Medicine College, Federal University of São Paulo, São Paulo, Brazil, <sup>7</sup> Centro de Oncologia Molecular, Hospital Sirio-Libanês, São Paulo, Brazil, <sup>8</sup> Departamento de Alimentos e Nutrição Experimental, Faculdade de Ciências Farmacêuticas, Food Research Center (FoRC), Universidade de São Paulo, São Paulo, Brazil, <sup>9</sup> Department of Clinical Oncology, A.C. Camargo Cancer Center, São Paulo, Brazil, <sup>10</sup> Laboratory of Computational Biology and Bioinformatics, A.C. Camargo Cancer Center, São Paulo, Brazil, <sup>11</sup> Biocomplexity Institute, Virginia Tech, Blacksburg, VA, USA, <sup>12</sup> Laboratory of Neurosciences (LIM-27) Alzira Denise Hertzog Silva, Institute of Psychiatry, Faculdade de Medicina, Universidade de São Paulo, São Paulo, Brazil

#### *Edited by:*

Venkatakrishna Rao Jala, University of Louisville, USA

#### *Reviewed by:*

David Albert Scott, University of Louisville, USA Daniel Raimunda, CONICET–UNC, Argentina

#### *\*Correspondence:*

Emmanuel Dias-Neto emmanuel@cipe.accamargo.org.br

*Received:* 26 September 2016 *Accepted:* 24 November 2016 *Published:* 09 December 2016

#### *Citation:*

Thomas AM, Jesus EC, Lopes A, Aguiar S Jr., Begnami MD, Rocha RM, Carpinetti PA, Camargo AA, Hoffmann C, Freitas HC, Silva IT, Nunes DN, Setubal JC and Dias-Neto E (2016) Tissue-Associated Bacterial Alterations in Rectal Carcinoma Patients Revealed by 16S rRNA Community Profiling. Front. Cell. Infect. Microbiol. 6:179. doi: 10.3389/fcimb.2016.00179 Sporadic and inflammatory forms of colorectal cancer (CRC) account for more than 80% of cases. Recent publications have shown mechanistic evidence for the involvement of gut bacteria in the development of both CRC-forms. Whereas, colon and rectal cancer have been routinely studied together as CRC, increasing evidence show these to be distinct diseases. Also, the common use of fecal samples to study microbial communities may reflect disease state but possibly not the tumor microenvironment. We performed this study to evaluate differences in bacterial communities found in tissue samples of 18 rectal-cancer subjects when compared to 18 non-cancer controls. Samples were collected during exploratory colonoscopy (non-cancer group) or during surgery for tumor excision (rectal-cancer group). High throughput 16S rRNA amplicon sequencing of the V4–V5 region was conducted on the Ion PGM platform, reads were filtered using Qiime and clustered using UPARSE. We observed significant increases in species richness and diversity in rectal cancer samples, evidenced by the total number of OTUs and the Shannon and Simpson indexes. Enterotyping analysis divided our cohort into two groups, with the majority of rectal cancer samples clustering into one enterotype, characterized by a greater abundance of Bacteroides and Dorea. At the phylum level, rectal-cancer samples had increased abundance of candidate phylum OD1 (also known as Parcubacteria) whilst non-cancer samples had increased abundance of Planctomycetes. At the genera level, rectal-cancer samples had higher abundances of Bacteroides, Phascolarctobacterium, Parabacteroides, Desulfovibrio, and Odoribacter whereas non-cancer samples had higher abundances of Pseudomonas, Escherichia, Acinetobacter, Lactobacillus, and Bacillus. Two Bacteroides fragilis OTUs were more abundant among rectal-cancer patients seen through 16S rRNA amplicon sequencing, whose presence was confirmed by immunohistochemistry and enrichment verified by digital droplet PCR. Our findings point to increased bacterial richness and diversity in rectal cancer, along with several differences in microbial community composition. Our work is the first to present evidence for a possible role of bacteria such as B. fragilis and the phylum Parcubacteria in rectal cancer, emphasizing the need to study tissue-associated bacteria and specific regions of the gastrointestinal tract in order to better understand the possible links between the microbiota and rectal cancer.

Keywords: mucosa-associated microbiota, rectal cancer, 16S rRNA gene sequencing, *Bacteroides fragilis*, Bacterial diversity and community composition

## INTRODUCTION

The gut microbiota is a vast and diverse ensemble of bacteria and other microorganisms that work together to help digestion, produce vitamins, fatty acids, amino acids and other bioactive compounds, and participate in the regulation of our immune, metabolic, and neurological systems (Shapiro et al., 2014; Boulangé et al., 2016). The understanding of our microbiota, together with the determination of its composition when contrasting healthy vs. diseased states allows the identification of microorganism disturbances that are possibly related to disease development and, therefore, offers a new approach for diagnosis as well as preventive and therapeutic interventions.

Specific dietary components, tobacco and alcohol consumption, which have been linked to the development of a number of pathological states (such as obesity, allergy, diabetes, Crohn's disease, irritable colon syndrome, and cancer) are known to drive microbiome alterations and lead to dysbiosis (Turnbaugh et al., 2009; Leclercq et al., 2014; Allais et al., 2016). The direct action of these elements or of the dysbiosis they cause, appears to be instrumental in the pathogenesis of many diseases and, under certain circumstances, it is possible that dysbiosis may, per se, have a direct link with disease development (Duboc et al., 2013). In oncology, studies have been conducted in different neoplastic conditions, identifying roles for specific bacteria in carcinogenesis (Kostic et al., 2012; Riley et al., 2013; Rubinstein et al., 2013), immune evasion (Gur et al., 2015), modulation of the tumor microenvironment (Kostic et al., 2013), and interference with anti-cancer immune responses and immune-surveillance that facilitate chemotherapy activity (Zitvogel et al., 2013; Galluzzi et al., 2015; Vétizou et al., 2015). As a consequence, the emerging concept that cancer needs to be studied considering the complex tumor microenvironment, which includes components such as tumor cells, the surrounding microenviroment and the microbiome, may aid in the development and improvement of cancer treatment, including immunotherapy (Pitt et al., 2016).

Tumors of the lower digestive tract, which include colon and rectal cancer, are among the most prevalent neoplasias worldwide, as well as one of the most fatal. Colorectal cancer (CRC) is the third most commonly diagnosed cancer with 1.4 million people diagnosed annually (Torre et al., 2015). The World Health Organization estimates an increase of 77% in the number of newly diagnosed CRC cases and an increase of 80% in deaths from CRC by 2030 (Binefa et al., 2014). Whereas, colon and rectal cancers have been routinely studied together as CRC, evidences indicate these to be distinct nosological entities. Differences in embryological origin, anatomy, treatment, metastatic potential, and outcome between colon cancer and rectal cancer have led to discussions as to whether neoplastic lesions of these two anatomical sites should be considered as different diseases, with further dichotomization of colon cancers into distal and proximal (Tamas et al., 2015).

The mechanisms involved in sporadic CRC predisposition or development are still poorly understood and the long list of cancer risk factors is continuously expanding and includes age, tobacco, and alcohol consumption, lack of physical activity, increased body weight and, most importantly, diet (Moore and Moore, 1995; Bingham, 2000). Of particular importance is the fact that all these risk factors can directly or indirectly modify the microbiota, making the precise definition of their roles a very complex task. Fecal microbiota studies have contributed greatly in our understanding of the general gut microbiota composition and its dysbiosis in different scenarios (Wu et al., 2013; Sabino et al., 2016). However, possibly due to practical issues related to obtaining the required biopsy samples—from patients and controls—there are still very few available studies focused on the analysis of microbial community compositions of more specific regions of the lower digestive tract, such as the proximal and distal colon, and the rectal tissue. Furthermore, few studies contemplate the fact that fecal- and tissue-associated microbiota are significantly different (Durbán et al., 2011; Hong et al., 2011; Mira-Pascual et al., 2015; Flemer et al., 2016). This fact is particularly relevant for CRC as the intimate crosstalk between the host's epithelium layer and the gut microbial community is a key factor for cell proliferation and development, as well as the regulation of inflammation, a major driver of rectal carcinogenesis (Arthur et al., 2014). Such differences lead to a lack of representativeness with respect to the bacterial biofilm of the rectal mucosa (Durbán et al., 2011; Gevers et al., 2014) and may reflect the disease state but possibly not the tumor microenvironment, which is of great importance to study possible microbiota:disease links.

Here we have addressed such shortcomings by studying the tissue-associated microbiota of 36 subjects, 18 with and 18 without rectal adenocarcinoma. The use of 16S rDNA deep sequencing allowed us to compare non-cancer x cancer mucosa, pointing to specific OTUs and bacterial genera potentially associated with rectal adenocarcinoma.

#### MATERIALS AND METHODS

## Cohort

A total of 36 subjects were included after approval by AC Camargo Cancer Center's review board (ACCCC - 1614/11, January 30th, 2012). Tissue biopsies were collected from subjects belonging to one of the following groups:

#### Non-cancer Subjects

(Non-Cancer, NC, n = 18): All subjects had medical indication of exploratory colonoscopy due to complaints, such as bleeding, abdominal pain, constipation, and chronic diarrhea. No subjects had personal or familial history of colorectal cancer or colitis (either ulcerative, Crohn's, radiation or infectious colitis, chronic inflammatory illnesses), previous colonic or small bowel resection, nor previous colon adenomas or familial polyposis syndrome. Only individuals with complete colonoscopies that allowed the full visualization of the entire colon and showed no significant clinical alterations were included.

#### Colonoscopy and Biopsy Procedures for the NC Subjects

All patients received standard instructions for preparation for colonoscopy that included consumption of 500 ml of mannitol for bowel cleansing, luftal, and bisacodyl. Eligible subjects gave written informed consent to provide colorectal biopsies, had their anthropometric measures taken and answered questions about diet, consumption of alcohol, and tobacco. Colonoscopy was performed using a Pentax videoscope model FC38LX. During biopsy procurement, the rectum was inflated with air and care was taken not to use any suction during advancement of the scope to 7–8 cm from the anal verge. Sterile biopsy forceps were not taken out of the channel of the scope until an area that was completely clear of stool was seen with clear pink mucosa. Biopsies were taken with 2.2 mm sterile standard forceps.

#### Patients Diagnosed with Rectal Adenocarcinomas

(Rectal-Cancer, RC, n = 18): Tumor specimens, located in the higher (N = 15), mid (N = 2), and lower rectum (N = 1), were obtained from surgeries to remove the tumor mass. All subjects belonging to this group were recruited at AC Camargo Cancer Center's Pelvic Surgery Department, in São Paulo, Brazil. We included patients that were diagnosed with rectal adenocarcinoma (tumors of stage pT1 or pT2 low- or mid-straight, pT1 or pT2 or pT3 high-straight), that had not undergone any neoadjuvant therapy and had their tumors surgically resected at the Pelvic Surgery Department, AC Camargo Cancer Center, with diagnosis confirmed by the Pathology Department of the same institution. After the histopathologic confirmation of rectal adenocarcinoma diagnosis, surplus samples were macrodissected by an experienced pathologist and used for DNA extraction and bacterial community profiling. Exclusion criteria were: Patients subjected to neoadjuvant therapy prior to tissue collection; patients reporting inflammatory bowel diseases or with hereditary cancer syndromes. We also excluded all subjects (cases and controls) who reported the use of antibiotics for at least 4 weeks prior to samplecollection.

## DNA Extraction

DNA extraction started after incubating the samples for 18 h in 600 µl of a lysis buffer (Qiagen) and 15µl of proteinase K (20µg/µl) at 55◦C. After this period, DNA samples were extracted using a standard phenol chloroform protocol, followed by ethanol precipitation, quantification using a spectrophotometer (Nanodrop—Thermo Scientific), and visualized on 2% agarose gels to inspect DNA integrity.

## PCR Amplification and Sequencing of the V4–V5 Region of 16S rRNA Gene

#### Oligonucleotide Primer Selection and Coverage Analysis

The V4–V5 region was amplified using a primer set designed to generate amplicons compatible with the chemistry available for the Ion Torrent PGM platform, that allowed ∼400 nt of high quality sequences (Ion PGM Sequencing 400 Kit). Coverage of the primer set was evaluated using the Ribosomal Database Project's (RDP—Release 11.2) ProbeMatch (Cole et al., 2014) and the ARB Silva's (Release 115) TestPrime (Klindworth et al., 2013). The forward primer (5′ -AYTGGGYDTAAAGNG-3′ ) and reverse primer (5′ -CCGTCAATTCNTTTRAGTTT-3′ ) corresponded to positions 562 and 906, respectively, of the Escherichia coli 16S rRNA gene.

#### PCR Amplification and Amplicon Sequencing

Three 50 µl amplification replicate reactions were performed per sample, each containing: 2.5 µM of each primer; 25 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; followed by a last extension step at 72◦C for 5 min. Amplicons of the three reactions from each subject were pooled and purified using a MinElute PCR Purification Kit (Qiagen). The purified products were run on 1.5% agarose gels and gel 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. Equimolar amounts of amplicons from each sample were pooled, using the Ion Torrent qPCR quantitation kit (Thermo Scientific, Carlsbad, USA), and used for emulsion PCR. All samples were sequenced on the Ion torrent PGM platform (Thermo Scientific, Carlsbad, USA) using two 318 v2 chips. Samples from both groups were processed simultaneously, to avoid possible batch effects.

## Sequence Analysis

#### Sequence Filtering

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., 2010a). We first removed sequences with an average quality score <20 using a 50 nt sliding window. Then, we identified barcodes used for subject-assignment, allowing a maximum of 2 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.

#### Sequence Clustering and OTU Filtering

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. The RDP classifier (Wang et al., 2007), as implemented within the Qiime interface (default parameters), was used to assign taxonomic ranks using a minimum confidence value of 80% and, subsequently, to each operational taxonomic unit (OTU). Unless otherwise stated, OTUs that occurred in less than 25% of all samples and with less than 3 reads were not considered.

#### Alpha and Beta Diversity Analysis

We rarefied the OTU table to 17,414 sequences per sample in order to calculate species diversity, using the Shannon-Weaver index (Shannon, 1948) and the Simpson index (Simpson, 1949), and richness (by using the observed species) implemented in the R Phyloseq package (McMurdie and Holmes, 2013).

For beta diversity analysis, OTU-representative sequences were aligned using PyNAST (Caporaso et al., 2010b) against the aligned green genes core set (DeSantis et al., 2006) with Qiime default parameters, and the alignments were lanemask-filtered (Lane, 1991). A phylogenetic tree was built using FastTree (Price et al., 2009), weighted and unweighted UniFrac (Lozupone and Knight, 2005) distances were calculated and a distance matrix was generated. Using the R phyloseq package, distance matrices were used to calculate coordinates for principal coordinate analysis (PCoA).

#### Enterotypes

Community types of each sample were analyzed by the Dirichlet multinomial mixture model-based method (Holmes et al., 2012) using rarefied genera level counts of 16S rRNA sequencing reads. Partioning around medoids (PAM) enterotyping was performed in R using genera level relative abundances and the "cluster" package (Maechler et al., 2015). We applied 4 distance metrics: Weighted UniFrac, Unweighted UniFrac, root Jensen-Shannon divergence, and Bray-Curtis and assessed the quality of the clusters using prediction strength (Tibshirani and Walther, 2005), silhouette index (Rousseeuw, 1987), and the Calinski- ´ Harabasz statistic (Calinski and Harabasz, 1974) using the "fpc" R package (Hennig, 2015).

## Differential Abundance Analysis

To investigate differences in OTU, phyla and genera abundances between both groups, raw counts were normalized then log transformed using the normalization method below, as performed by a previous study (Sanapareddy et al., 2012):

$$\begin{aligned} \text{Normalized count} &= \log\_{10}((\frac{\text{raw count}}{\text{number of sequences in that sample}})) \\ &\times \text{average number of sequences per sample} + 1) \end{aligned}$$

We also evaluated high-level phenotypical differences in microbial composition between both groups. Quality control passed sequences were closed-reference picked at 97% identity using UCLUST\_Ref (Edgar, 2010) and the green genes core set (Version 13.5). The resulting OTU table was rarefied to 13,944 sequences and submitted to BugBase (http://github.com/ danknights/bugbase) in order to calculate differences between both groups in terms of microbial phenotypes.

### Data Validation

#### Digital Droplet PCR of Bacteroides Fragilis 16S rRNA

We detected and quantified the absolute number of 16S rRNA B. fragilis copies in our samples using the QX200TM Droplet DigitalTM PCR System (Bio-Rad). The primers used to amplify the B. fragilis 16S rRNA gene were: BF-fwd 5′ -TCRGGAAGAAAGCTTGCT-3′ and BF-rev 5′ - CATCCTTTACCGGAATCCT-3′ (Tong et al., 2011) and to ensure further specificity, a labeled probe BF-p 5′ (FAM)- ACACGTATCCAACCTGCCCTTTACTCG-3′ (BHQ1) (Tong et al., 2011) was included in the reaction. We used a commercial RNAseP Copy Number Reference Assay (Thermo-Fisher) to detect and quantify human DNA. Microdroplets (≈20.000/reaction) were generated on the Bio-Rad QX-100 following the manufacturer's instructions. RNAse P and B. fragilis ddPCR were performed in 96 well-plates, in a final volume of 20 µl, containing: 15 ng of total DNA, 10 ul of ddPCR supermix for probes (Bio-Rad), 8 pmol of each PCR BF-primer and 2 pmol of the BF-probe, or 1 µl of RNAse P assay. PCR conditions were: 50◦C- 2 min; 95◦C- 10 min; 95◦C- 15 s and 60◦C- 1 min for 40 cycles. After cycling, the 96-well plate was immediately transferred on a QX200 Droplet Reader (Bio-Rad), where flow cytometric analysis determined the fraction of PCR-positive droplets vs. the number of PCR-negative droplets in the original sample. Data acquisition and quantification was carried out using QuantaSoft Software (Bio-Rad). To ensure the accuracy of the results, a minimum of 10,000 acceptable droplets per reaction were required for quantification using the QuantaSoft software. Samples yielding a minimum of 3 positive droplets from 10–15,000 droplets analyzed were scored as positive.

#### Immunohistochemistry

Immunohistochemistry was performed in an automated Benchmark platform (Ventana Medical Systems, USA) for Anti-B. fragilis LPS antibody (mouse monoclonal—Abcam 1265/30) in whole slide tissues. Alkaline phosphatase conjugated to secondary polymeric system was used for IHC visualization. The selection of positive and negative samples was guided by the high-throughput sequencing (HTS) data and used to confirm the presence of B. fragilis in the sample set. The primary antibody was omitted to evaluate background staining.

#### Statistical Analysis

Wilcoxon tests were used to compare mean differences between tumor and biopsy samples for phyla, genera and OTU logabundances. Considering t = total number of taxa tested, p = raw p-value and R = sorted rank of the taxon, P-values were corrected for multiple testing (Sanapareddy et al., 2012) using:

$$Adjusted \, p\,\, value \, = \, \frac{t \times p}{R}$$

Fold changes for each genera/OTU were calculated using:

$$\text{Log2FC} = \log\_2(\text{RC average } + 1) - \log\_2(\text{NC average } + 1)$$

Chi-Square tests were performed on subject's categorical data such as gender, alcohol and tobacco use and vital status. Student t-tests were performed to compare differences in the means between both groups for age, height, weight, BMI, and alpha diversity. We used ANOSIM and ADONIS (Oksanen et al., 2016) to compare differences in beta-diversity between groups using 3 distance metrics weighted UniFrac, unweighted UniFrac and Bray-Curtis for categorical, and numerical variables, respectively. Linear models were built using normalized counts at the genera and OTU level to investigate associations with clinicalpathological characteristics of rectal-cancer samples, such as lymph node and perineural neoplastic invasion status. Unless otherwise stated, values were reported as mean ± SD (standard deviation) and P-values <0.05 were considered statistically significant. All calculations were performed within the R statistical computing environment (R Foundation, 2011) unless otherwise stated.

#### RESULTS

### Subjects and Tissue Sample Characteristics

We analyzed tissue-associated bacteria from mucosal biopsies of 18 non-cancer controls and 18 rectal adenocarcinoma tumors using 16S rRNA high throughput amplicon sequencing. We found no significant differences between rectal-cancer and noncancer subjects regarding age and gender distribution, tobacco, and alcohol use and other risk factors (**Table 1**). All samples consisted of rectal-biopsies. The biopsies of individuals with no tumor lesions derived from the mid rectum and were distributed along the ∼12 cm-long human rectum, with most samples deriving from the higher-mid rectum (94%).

#### TABLE 1 | Subject and sample data.


N.A., Not applicable.

#### Primer Coverage

Our analyses indicate that the PCR primers used here (V4–V5 region of the 16S rRNA gene) cover 84.4 and 52.1% of all eubacterial sequences present in the ARB SILVA database and the Ribosomal Database Project, respectively (Supplementary Table 1). Coverage rates were evenly distributed among most bacterial phyla, except for Verrucomicrobia, where coverage rates were 21 and 10.9%, dropping below the 75 and 48% averages of taxa present in the SILVA and RDP databases, respectively.

#### Sequence Analysis

#### Sequence Generation and Filtering

A total of 12,078,140 sequence reads were generated, with a mean sequence length of 304.5 ± 97.34 nt (standard deviation—std). After quality filtering and primer trimming, 5,593,020 (46.3%) sequences remained, with an average of 155,361 sequences/sample and a mean sequence length of 315 ± 30 nt.

#### Sequence Clustering and OTU Filtering

When all individuals were considered, a total of 3222 OTUs were obtained. Thirty-one (0.7%) OTUs were identified as chimeras by UCHIME and 209 (4.7%) could not be assigned to a taxonomic rank. After filtering OTUs with less than three sequences and not present in at least 25% of all samples (NC and RC combined), 1492 OTUs remained.

#### Alpha and Beta Diversity Species Richness and Diversity

We observed significantly higher species richness and species diversity in rectal cancer samples compared to controls. This was observed for the number of OTUs, the Shannon index and the Simpson Index (P-values = 0.002, <0.001, and <0.001, respectively) (**Figures 1A,B**). When we stratified rectal-cancer samples into smaller (pT2) and larger tumors (pT3), we observed an increase in species richness, with an average of 280 and 366 OTUs, respectively, compared to 236 OTUs in NC; however this effect reached no statistical significance between pT2 and pT3, maybe because of the reduced number of pT2 samples (N = 5, compared to N = 13 for pT3) (**Figure 1B** and data not shown).

#### Beta Diversity

Using three distance metrics we observed consistent and statistically significant differences between the sample groups when considering cancer status (Bray-Curtis, Unweighted and Weighted UniFrac; p-value: 0.001; ANOSIM using 999 permutations), but not for any other categorical or numerical variable, which included amplicon library construction, age, gender, BMI, alcohol, and tobacco use (**Figure 1D**; Supplementary Table 2).

#### Enterotypes

Enterotyping analysis using a Dirichlet multinomial mixture model divided our cohort in two clusters (**Figures 2A–C**). Enterotype I was significantly enriched for rectal-cancer samples, whilst enterotype II was composed mostly of non-cancer samples (p-value: 0.0001, Fisher's exact test). Enterotype I had higher abundances of Bacteroides, Clostridiales, Dorea, and other genera, whilst enterotype II was characterized by elevated amounts of Pseudomonas and Brevundimonas (**Figure 2D**). When using the PAM based enterotyping method and criterion adopted by a meta-analysis of human enterotypes (Koren et al., 2013), we found two enterotypes with prediction strength above 0.9 (meaning that 90% of the data points fall within the cluster and 10% are outliers) using the Weighted UniFrac distance (Supplementary Figure 1).

## Global Signatures of the Microbial Community

#### Phyla Log Abundances

We observed a significant difference in the log abundances of 6 out of 12 detected phyla between both groups (Supplementary Figure 2). The most abundant phyla identified were (in decreasing order) Proteobacteria, Firmicutes, Bacteroidetes, Fusobacteria, Actinobacteria, and Verrucomicrobia. In non-cancer samples, we observed higher log abundances of Actinobacteria, Cyanobacteria, Proteobacteria, and Planctomycetes, whose presence was detected in 9/18 NC samples, with an average log abundance of 0.54 and was absent from all RC individuals (p-value < 0.001). In rectal-cancer we found greater log abundances of Bacteroidetes and of the much less known candidate phylum OD1 (also known as Parcubacteria), whose presence was detected in 14/18 RC samples with an average log abundance of 0.71 vs. 1/19 NC samples and an average log abundance of 0.02 (p-value < 0.001).

#### Genera Log Abundances

At the genus level, 86 out of 260 genera (33%) showed significant differential log abundances between both groups (**Figure 3A** and Supplementary Table 3). The top five genera with differential log abundances between the groups were Bacteroides, Phascolarctobacterium, Odoribacter, Parabacteroides, Desulfobrio (more abundant in the cancer group), and Lactobacillus, Pseudomonas, Bacillus, Escherichia, Acinetobacter (more abundant in the non-cancer set) (**Figure 3B**).

#### OTU Log Abundances

Of the 1492 OTUs identified, 163 (10.9%) were found to have significant differential log abundances between both groups (**Figure 3C**). Three OTUs assigned to the genus Bacteroides, two belonging to B. fragilis and one to B. uniformis, as well as OTUs assigned to Bilophila sp. and Fusobacterium sp., were significantly more abundant in rectalcancer samples (**Figure 3D**). In non-cancer samples, OTUs assigned to Alcaligenes faecalis, Bacillus cereus, Lactobacillus delbruecki, Prevotella melaninogenica and Pseudomonas ssp had higher log abundances compared to rectal-cancer samples. Four OTUs belonging to the Bacilli class were more abundant among non-cancer samples, including L. delbrueckii (**Figure 3D**).

When analyzing high-level phenotypical differences, the most striking differences included a higher abundance of anaerobic bacteria and a deficit in biofilm-forming bacteria in rectalcancer samples (Supplementary Figure 3). In our searches for associations between rectal-cancer samples' clinical data and genera/OTU log abundances using linear regression, we found significant associations between genera/OTUs with regards to the presence of lymph node disease (Supplementary Table 4) and perineural invasion (data not shown). We found a significant increase of Coprococcus, Dorea, Roseburia, and Mogibacterium in lymph node positive rectal-cancer (Supplementary Figure 4).

### ddPCR Confirms the Higher Counts of *B. fragilis* in Tumor Samples

As two OTUs classified as B. fragilis were among the smallest p-values found and with the highest fold change between the groups, we designed a specific ddPCR assay for B. fragilis in order to verify the validity of the results using an alternative approach. As can be seen in **Figures 4A,B**, we observed the expected correlation (R <sup>2</sup> = 0.78) between both methods and confirmed the higher ratio of B. fragilis/human DNA in rectal cancer samples, validating the results of our sequencing approach (P-value = 0.04, Wilcoxon Rank-Sum Test). To further evidence the presence of B. fragilis in tumor specimens, we performed an immunohistochemistry assay on 3 rectal-cancer samples using an anti-B. fragilis LPS antibody and found that this bacterium was present in rectal-cancer tissue (**Figures 4C,D**).

## DISCUSSION

In face of the microbiota gradient found in the human digestive tract (Zhang et al., 2014; Gao et al., 2015; Flemer et al., 2016) and the possibility that tissue-associated microorganisms could play a more direct role in immunomodulation and cancer

development, we investigated bacterial populations present in tissue biopsies, which may be relevant to pathological processes. Instead of studying colon and rectum samples together, our work is more specific as it is focused and contains only rectal tumors. Whereas, we achieved high 16S rRNA coverage from a large spectrum of bacteria from cancer samples, before any therapeutic intervention, we also see limitations, such as our relatively small sample size of 36 individuals. However, effect size analysis (Kelly et al., 2015) between both groups revealed an ω 2 ranging from 0.13 to 0.26, depending on the metric of pairwise distance, with PERMANOVA p-values <0.001 and power of 1 (data not shown), a finding that indicates

that this sample size allows the observation of significant microbial differences between our two sample groups. We also need to point out that, whereas the primer pair used here gives a good coverage of most phyla, it has a poor coverage of the two closely related bacteria phyla Lentisphaerae and Verrucomicrobia.

In our study, we observed increased species-diversity and -richness among rectal-cancer samples. Higher speciesdiversity and -richness were seen in rectal tissue samples from adenomas compared to normal samples (Sanapareddy et al., 2012) and CRCs vs. adenomas (Nakatsu et al., 2015) and increased richness was found in CRCs compared to both adenomas and controls (Mira-Pascual et al., 2015). However, when looking at fecal samples, studies have had conflicting results. One study found increased diversity of both genes and genera along the adenoma-carcinoma transition (Feng et al., 2015), whereas another found a decrease in diversity when comparing carcinoma samples and normal controls (Ahn et al., 2013) and a third found no differences between controls, adenomas and carcinomas (Zeller et al., 2014). It is noteworthy to state that these fecal studies grouped proximal and distal colon cancers together with rectal cancers, which could have led to differences in their results. We should note that the five cases of early-stage lesions (pT2) showed, on average, intermediate microbial richness, when compared to non-cancer biopsies and a more advanced neoplastic stage (pT3). This suggests that increased species richness of cancer lesions could have an early role in rectal carcinogenesis.

Inter-individual microbial community heterogeneity of the human gut is influenced by spatial distribution, micro-heterogeneity, host genetics, dietary preferences,

and mucin content (Eckburg et al., 2005; Hong et al., 2011; Zhang et al., 2014), and has posed a longstanding challenge when investigating microbial signatures implicated in CRC tumorigenesis. However, our results show that despite the high inter-individual differences, a common microbial community pattern appears to emerge, as shown in the PCoA analysis that clustered noncancer and rectal-cancer groups separately (**Figure 1D**), suggesting a common dysbiotic setting related to this neoplasia.

with arrows) using magnification of 1000X.

We performed a global analysis of high-level phenotypical differences for bacteria identified in both groups. We highlight the higher abundance of anaerobic bacteria in the RC group in agreement with a previous study (Warren et al., 2013) and the reduction of biofilm-forming bacteria. The latter is a finding that may point to barrier breakage that would contribute to rectal colonization by relevant bacteria (Reid et al., 2001) (Supplementary Figure 3).

The alterations we found at the phylum level include higher levels of Cyanobacteria (possibly Melainabacteria) (Soo et al., 2014), Actinobacteria, Bacteroidetes, OD1, Proteobacteria, and Planctomycetes in the RC-group. We should note an important abundance difference for bacteria of the candidate phylum OD1 (Parcubacteria). These highly adapted organisms have not been isolated in vitro yet; they have small genomes (<1 Mb) and reduced metabolic properties identified in a range of anoxic environments. The absence of biosynthetic capabilities and DNA repair enzymes, derived from the genomic analyses of some OD1 bacteria, suggests a role as ectosymbionts (Nelson and Stegen, 2015). However, the putative role of these microbes in rectal cancer remains to be determined. A second phylum, Planctomycetes, which are atypical bacteria (Fuerst and Sagulenko, 2011) relatively close to Verrucomicrobia (Hou et al., 2008) and more frequently observed in aquatic environments (such as saltwater, fresh water, and acidic mud), also showed potential as a biomarker for RC, with striking differences between the groups.

Interestingly, our study also indicated the differential abundance of more specific microbes after comparing NC and RC groups. B. fragilis, a symbiotic organism common to the human intestinal tract, was found to be more abundant in rectalcancer samples seen by 16S rRNA HTS and confirmed by ddPCR. Other studies that investigated tissue-associated bacteria also found increased abundance of B. fragilis in tumor samples (Wang et al., 2012; Zeller et al., 2014; Nakatsu et al., 2015). B. fragilis has been identified as an important human intestinal symbiont and has been suggested to act as a "keystone pathogen" in the development of CRC (Hajishengallis et al., 2012). B. fragilis is an obligate anaerobe and is a minority member of the normal colonic microbiota with a propensity for mucosal adherence (Sears et al., 2014). Previous reports have linked enterotoxigenic B. fragilis (ETBF) to human diarrheal illnesses and increased tumorigenesis in an IL-23-dependent and STAT3-dependent manner (Wick et al., 2014). The toxin fragylisin, produced by ETBF, is a zinc-dependent metalloprotease that triggers NF-kB signaling and cleaves E-cadherin, and has been suggested to be oncogenic (Wu et al., 2009). Bacterial genera known for their role in butyrate production, such as Ruminococcus, Roseburia, and Butyricimonas were more abundant among rectal-cancers, differing from results reported so far. An explanation for this difference could involve the fact that most data has been derived from fecal samples and/or grouping different anatomical tumor sites (such as proximal, distal, and rectal). An OTU assigned to Bilophila, a bile-resistant, strictly anaerobic bacterial genus, was also more abundant among rectal-cancer samples, and evidence suggests that products of bacterial bile acid conjugation, secondary bile acids, are carcinogenic (McGarr et al., 2005; Ridlon et al., 2014). Desulfovibrio, a commensal sulfate-reducing bacterium, may contribute to mucosal inflammation through hydrogen sulfide production, a resulting by-product of sulfated mucin metabolism (Earley et al., 2015). Phascolarctobacterium, known to produce propionate via succinate fermentation, was also increased among cancer samples. On the other hand, we found that L. delbrueckii was more abundant in non-cancer samples. Probiotic Lactobacilli can modify the enteric flora and are thought to have a beneficial effect on enterocolitis. Treatment of IL-10-deficient mice with the probiotic Lactobacillus salivarius ssp. reduced the intensity of mucosal inflammation and the incidence of colon cancer from 50 to 10%. These effects were accompanied by significant reductions in fecal coliform, enterococci, and Clostridium perfringens levels (O'Mahony et al., 2001). This study exemplifies the effect of changes at the flora level on the development of inflammation, and supports the hypothesis that there are "protective" species and "harmful" species in the normal bacterial flora.

After identifying relevant cancer-related microorganisms, the next steps of microbiome studies will certainly involve microbial manipulations to reduce disease-associated agents, or increase the frequency of protective and health-associated microbes. This can be achieved through diet, exemplified by a previous study using animal models that showed taurine consumption lead to a reduction of Proteobacteria (especially Helicobacter), as well as an elevation in short-chain fatty acids (SCFA) and a reduction in fecal lipopolysaccharides (LPS) (Yu et al., 2016). Duque et al., recently demonstrated, using SHIME <sup>R</sup> (Simulator of the Human Microbial Ecosystem), that the consumption of non-pasteurized fresh orange juice was able to significantly increase levels of Lactobacillus spp., Enterococcus spp., Bifidobacterium spp., and Clostridium spp. and to reduce enterobacteria (Duque et al., 2016).

Long before associations between cancer and the microbial flora started to be uncovered, diet recommendations—including low consumption of red meat and fat, and high ingestion of fibers and vegetables—have been recognized as protective against the development of colorectal cancer. Current evidences suggest that diet recommendations may be effective, together with tissue environment and host-related factors, because they also help shape the gut microbiota (Sonnenburg and Bäckhed, 2016). Further research may show that treatment of rectal dysbiosis may contribute to the prevention of inflammation-induced rectal carcinoma development and aid in chemotherapy and overall treatment response (Yang and Pei, 2006).

## AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: AT, EJ, AL, SA, DN, ED; Performed the experiments: AT, EJ, RR, PC; Analyzed the data: AT, CH, DN, JS, ED; Contributed reagents/samples/analysis tools: MB, AL, SA, AC, HF, IS; Wrote or edited the manuscript: AT, HF, CH, JS, DN, ED. All authors read and approved the final manuscript.

## FUNDING

AT was supported by a fellowship from FAPESP (2015/01507-7). CH was supported by CAPES grant (88887.062078/2014-00) and FAPESP grant (2013/07914-8). This project was supported by PRONON (SIPAR 25000.055.167/2015-23), by Associação Beneficiente Alzira Denise Hertzog Silva (ABADHS) and by CAPES grant 3385/2013.

## ACKNOWLEDGMENTS

The authors are grateful to the institutional tumor bank of the AC Camargo Cancer Center. ED and JS are research fellows of the Conselho Nacional de Desenvolvimento Cientifico e Tecnologico (CNPq).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcimb. 2016.00179/full#supplementary-material

#### Availability of Supporting Data

Nucleotide sequences used for this study have been deposited in the SRA under accession SRP077097.

#### REFERENCES


cancer: not just a different anatomic site. Cancer Treat. Rev. 41, 671–679. doi: 10.1016/j.ctrv.2015.06.007


**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 reviewer DAS and handling Editor declared their shared affiliation and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2016 Thomas, Jesus, Lopes, Aguiar, Begnami, Rocha, Carpinetti, Camargo, Hoffmann, Freitas, Silva, Nunes, Setubal, and Dias-Neto. 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.

digital media

of impactful research

article's readership