# INTEGRATIVE APPROACHES TO THE MOLECULAR PHYSIOLOGY OF INFLAMMATION

EDITED BY : Enrique Hernández-Lemus, María Elena Soto and Carlos Rosales PUBLISHED IN : Frontiers in Physiology and Frontiers in Pharmacology

#### Frontiers Copyright Statement

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

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

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

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

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

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

The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use. ISSN 1664-8714 ISBN 978-2-88945-800-4 DOI 10.3389/978-2-88945-800-4

### About Frontiers

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

### Frontiers Journal Series

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

### Dedication to Quality

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

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

### What are Frontiers Research Topics?

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

# INTEGRATIVE APPROACHES TO THE MOLECULAR PHYSIOLOGY OF INFLAMMATION

Topic Editors:

Enrique Hernández-Lemus, Instituto Nacional de Medicina Genómica, Mexico María Elena Soto, Instituto Nacional de Cardiología Ignacio Chávez, Mexico Carlos Rosales, Instituto de Investigaciones Biomédicas, UNAM, Mexico

Cover: Sebastian Kaulitzki

"Integrative Approaches to the Molecular Physiology of Inflammation" presents contributions from the many different fields and approaches to the physiology and the molecular origins of inflammation; particularly those that may be involved in the development and evolution of diseased phenotypes.

We selected among the wide scope and multiple views used to probe into the molecular origins of complex inflammatory phenotypes. This book consists of an Introductory Editorial and 6 thematic chapters encompassing 24 articles: 17 original

research contributions and 7 review articles (5 reviews, 1 systematic review, and 1 minireview). Both, the research papers and the reviews provide varied and insightful approaches to different facets of inflammation with approaches ranging from general inflammation and signaling depictions deeply rooted on functional biology and physiology, to computational systems biology analyses, translational medicine, and pharmacological explorations. Model systems are also quite diverse: human subjects, mice and other mammal models, cell cultures and *in silico*, complex networks and database studies.

Citation: Hernández-Lemus, E., Soto, M. E., Rosales, C., eds. (2019). Integrative Approaches to the Molecular Physiology of Inflammation. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-800-4

# Table of Contents

### *07 Editorial: Integrative Approaches to the Molecular Physiology of Inflammation*

Enrique Hernández-Lemus, Maria Elena Soto and Carlos Rosales

### CHAPTER 1

### DISSECTING THE INFLAMMATORY COMPONENTS OF COMPLEX DISEASE

*10 Remodeling of Retinal Architecture in Diabetic Retinopathy: Disruption of Ocular Physiology and Visual Functions by Inflammatory Gene Products and Pyroptosis*

Rubens P. Homme, Mahavir Singh, Avisek Majumder, Akash K. George, Kavya Nair, Harpal S. Sandhu, Neetu Tyagi, David Lominadze and Suresh C Tyagi

*28 Osteoarthritis-Like Changes in Bardet–Biedl Syndrome Mutant Ciliopathy Mice (*Bbs1M390R/M390R*): Evidence for a Role of Primary Cilia in Cartilage Homeostasis and Regulation of Inflammation*

Isaac D. Sheffield, Mercedes A. McGee, Steven J. Glenn, Da Young Baek, Joshua M. Coleman, Bradley K. Dorius, Channing Williams, Brandon J. Rose, Anthony E. Sanchez, Michael A. Goodman, John M. Daines, Dennis L. Eggett, Val C. Sheffield, Arminda Suli and David L. Kooyman

### CHAPTER 2

### INFLAMMATION AND METABOLISM

*38 Participation of Arachidonic Acid Metabolism in the Aortic Aneurysm Formation in Patients With Marfan Syndrome* María E. Soto, Verónica Guarner-Lans, Karla Y. Herrera-Morales and

Israel Pérez-Torres


### CHAPTER 3

### INFLAMMATION AND SIGNALING

*100 "Thinking" vs. "Talking": Differential Autocrine Inflammatory Networks in Isolated Primary Hepatic Stellate Cells and Hepatocytes Under Hypoxic Stress*

Yoram Vodovotz, Richard L. Simmons, Chandrashekhar R. Gandhi, Derek Barclay, Bahiyyah S. Jefferson, Chao Huang, Rami Namas, Fayten el-Dehaibi, Qi Mi, Timothy R. Billiar and Ruben Zamora

*112 Coactivation of TLR2 and TLR8 in Primary Human Monocytes Triggers a Distinct Inflammatory Signaling Response*

Korbinian Bösl, Miriam Giambelluca, Markus Haug, Marit Bugge, Terje Espevik, Richard K. Kandasamy and Bjarte Bergstrøm

### CHAPTER 4

### SYSTEMS PHYSIOLOGY OF INFLAMMATION


Eun Jung Kim, So Youn Park, Seung Eun Baek, Min A. Jang, Won Suk Lee, Sun Sik Bae, Koanhoi Kim and Chi Dae Kim


Wan-Yu Lo, Ching-Tien Peng and Huang-Joe Wang

*187 Prognostic Values of Long Noncoding RNA GAS5 in Various Carcinomas: An Updated Systematic Review and Meta-Analysis* Qunjun Gao, Haibiao Xie, Hengji Zhan, Jianfa Li, Yuchen Liu and Weiren Huang

### CHAPTER 5

### INTEGRATIVE AND COMPUTATIONAL APPROACHES TO INFLAMMATION


### CHAPTER 6

### THERAPEUTIC STRATEGIES TO MODULATE INFLAMMATION

*224 Inhibition of Toll-Like Receptor Signaling as a Promising Therapy for Inflammatory Diseases: A Journey From Molecular to Nano Therapeutics* Wei Gao, Ye Xiong, Qiang Li and Hong Yang

### *244 LASSBio-897 Reduces Lung Injury Induced by Silica Particles in Mice: Potential Interaction With the A2A Receptor*

Vinicius F. Carvalho, Tatiana P. T. Ferreira, Ana C. S. de Arantes, François Noël, Roberta Tesch, Carlos M. R. Sant'Anna, Eliezer J. L. Barreiro, Carlos A. M. Fraga, Patrícia M. Rodrigues e Silva and Marco A. Martins

*260 Volatile Oil From Amomi Fructus Attenuates 5-Fluorouracil-Induced Intestinal Mucositis*

Ting Zhang, Shan H. Lu, Qian Bi, Li Liang, Yan F. Wang, Xing X. Yang, Wen Gu and Jie Yu

*273 Carnosic Acid Alleviates BDL-Induced Liver Fibrosis Through miR-29b-3p-Mediated Inhibition of the High-Mobility Group Box 1/Toll-Like Receptor 4 Signaling Pathway in Rats*

Shuai Zhang, Zhecheng Wang, Jie Zhu, Ting Xu, Yan Zhao, Huanyu Zhao, Fan Tang, Zhenlu Li, Junjun Zhou, Dongyan Gao, Xiaofeng Tian and Jihong Yao

*284 Age-, Gender-, and* in Vivo *Different Doses of Isoproterenol Modify* in Vitro *Aortic Vasoreactivity and Circulating VCAM-1*

Betzabé Nieto-Lima, Agustina Cano-Martínez, María E. Rubio-Ruiz, Israel Pérez-Torres and Verónica Guarner-Lans

*295 The Modulatory Properties of Chronic Antidepressant Drugs Treatment on the Brain Chemokine – Chemokine Receptor Network: A Molecular Study in an Animal Model of Depression*

Ewa Trojan, Joanna Ślusarczyk, Katarzyna Chamera, Katarzyna Kotarska, Katarzyna Głombik, Marta Kubera and Agnieszka Basta-Kaim

*311 Much More Than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain*

Luiz H. A. Cavalcante-Silva, Éssia de Almeida Lima, Deyse C. M. Carvalho, José M. de Sales-Neto, Anne K. de Abreu Alves, José G. F. M. Galvão, Juliane S. de França da Silva and Sandra Rodrigues-Mascarenhas

*319 Corrigendum: Much More Than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain*

Luiz H. A. Cavalcante-Silva, Éssia de Almeida Lima, Deyse C. M. Carvalho, José M. de Sales-Neto, Anne K. de Abreu Alves, José G. F. M. Galvão, Juliane S. de França da Silva and Sandra Rodrigues-Mascarenhas

*320 Key Inflammatory Processes in Human NASH are Reflected in Ldlr−/−.Leiden Mice: A Translational Gene Profiling Study* Martine C. Morrison, Robert Kleemann, Arianne van Koppen, Roeland Hanemaaijer and Lars Verschuren

# Editorial: Integrative Approaches to the Molecular Physiology of Inflammation

#### Enrique Hernández-Lemus <sup>1</sup> \*, Maria Elena Soto<sup>2</sup> and Carlos Rosales <sup>3</sup>

<sup>1</sup> Computational Genomics, Instituto Nacional de Medicina Genómica, Mexico City, Mexico, <sup>2</sup> Departamento de Inmunología, Instituto Nacional de Cardiología Ignacio Chávez, Mexico City, Mexico, <sup>3</sup> Departamento de Inmunología, Instituto de Investigaciones Biomédicas, Universidad Nacional Autónoma de México, Mexico City, Mexico

Keywords: inflammation, molecular physiology, integrative biology, immunity, pathophyisology

#### **Editorial on the Research Topic**

### **Integrative Approaches to the Molecular Physiology of Inflammation**

Inflammation is the generic name given to a number of complex biological processes related to the organismal response to a disparate set of stimuli (most of them harmful or pathogenic), either intrinsic (DNA damage, metabolic deregulation, etc.) or extrinsic (pathogens, irritants, etc.) in nature. Such processes are commonly related to a protective reaction to disease related events that involve immune response, vascularization, and cellular signaling among many other features. Recent years have witnessed an increased interest in the study of inflammation, since it was discovered that inflammatory processes are associated with a growing number of pathologies, many of which had not been previously classified as "inflammatory."

Complex chronic diseases such as cancer, diabetes, or even Alzheimer's or Parkinson's have recently been discovered to be strongly associated with inflammatory responses. Other maladies such as cardiovascular, rheumatic, and autoimmune diseases have been traditionally studied from the standpoint of inflammation. Processes such as the ones leading to aging and fragility or even hormone dysfunctions are also starting to be related to inflammatory responses.

The various stages and processes of inflammation can be related to genetic changes, environmental and molecular patterns associated with damage (DAMPs) or molecular patterns associated with microbes (MAMPs) or molecular patterns associated with pathogens (PAMPs). These molecular patterns, when recognized by different cells, can induce multiple vascular and cellular responses that lead to the clinical diversity of inflammatory conditions related to pathogenesis.

For the reasons mentioned above, an important goal in contemporary biomedical science is the integrated study of the physiology of inflammation and the molecular pathways associated with it. The aim of this Research Topic is hence to gather contributions from the many different fields and approaches to the physiology and the molecular origins of inflammation; particularly those that may be involved in the development and evolution of diseased phenotypes. By presenting them together we want to cooperate to unveil the commonalities and differences that so many of these phenomena have, particularly in relation to their molecular origins as well as to any issues that may enlighten prognostics, diagnostics, and therapeutic decisions.

We believe that this Research Topic on Integrative Approaches to the Molecular Physiology of Inflammation is a good selection of the wide scope and multiple (seemingly disparate but often convergent) views to approach the molecular origins of complex inflammatory phenotypes. It consists on 24 articles: 17 original research contributions and 7 review articles (5 reviews, 1 systematic review, and 1 minireview). Both, the research papers and the reviews provide varied and

#### Edited by:

Geoffrey A. Head, Baker Heart and Diabetes Institute, Australia

#### Reviewed by:

Kulmira Nurgali, Victoria University, Australia, Australia

#### \*Correspondence: Enrique Hernández-Lemus

ehernandez@inmegen.gob.mx

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 25 October 2018 Accepted: 06 December 2018 Published: 18 December 2018

#### Citation:

Hernández-Lemus E, Soto ME and Rosales C (2018) Editorial: Integrative Approaches to the Molecular Physiology of Inflammation. Front. Physiol. 9:1825. doi: 10.3389/fphys.2018.01825

**7**

insightful approaches to the different facets of inflammation with approaches ranging from general inflammation and signaling depictions deeply rooted on functional biology and physiology, to computational systems biology analyses, translational medicine, and pharmacological explorations. Model systems are also quite diverse: human subjects, mice and other mammal models, cell cultures and in silico, complex networks and database studies.

These systems allowed the analysis of questions that include: dissecting the inflammatory components of complex disease phenotypes, such is the case of the review by Homme et al. on diabetic retinopathy and its relationship with pyroptosis as well as downstream products of inflammation; and of the research paper by Sheffield et al. where they present their findings on how deregulation of inflammation leads to changes in cartilage homeostasis with strong consequences for Bardet-Biedl cilliopathy.

Inflammatory processes are often intertwined with abnormal metabolic regulation, such is the case of inflammation induced changes in the arachidonic acid metabolism leading to disruption of the aortic function of patients with Marfan syndrome. In the aortic tissue of these patients, it has been shown that the inflammatory process through an imbalance in the synthesis of prostaglandins is a consequence of a differential protein expression of the isoforms of cyclooxygenases (COXs) COX1 and COX2. This imbalance ultimately contributes to distorted vascular smooth muscle contraction and relaxation. Also, the overexpression of COX2 induces metalloproteinases whose activation directs the degradation of the extracellular matrix. Together, these effects lead to a significant increase in the aortic aneurysm of patients with Marfan syndrome as presented in the article by Soto et al.

Related changes in metabolism, driven by inflammasome activity and oxidative stress may even challenge therapeutic interventions in diabetic patients with cardiovascular complications as is thoroughly reviewed by the Sharma et al. The multiple roles of neutrophils in inflammation are thoroughly explored in the review paper by Rosales. Systemic chronic low-grade inflammatory processes are also associated with protein metabolism and anabolic sensitivity in age-related sarcopenia as it is discussed by Dalle et al.

A functional approach to study differential signaling networks related to autocrine processes in hepatic stellate cells and hepatocytes under stress is presented by Vodovotz et al. The authors were able to differentiate between networks associated with intracellular information processing ("thinking") and networks devoted to extracellular information transfer ("talking"), whose interplay results determinant for the differences in autocrine response in both cell types. Also in connection with systemic signaling regulation, the research paper by Bösl et al. provide a nice example of cooperative signaling by two TLRs, enhancing the regulatory processes in a way no single receptor cascades may actually achieve. The way in which system-wide molecular processes affect tissues and physiological behaviors is well exemplified by the research article contributed by Jarkovska et al. The authors identified cellular and molecular mechanisms behind the loss of cardiac tone by myocardial depression, a condition associated with septic shock. The associated molecular mechanisms constitute a novel tool to unveil potential therapeutic targets useful for the prevention and treatment of sepsis-induced myocardial dysfunction.

The relation between metabolic processes in the muscular tissue and inflammation is further discussed in the research paper by Kim et al. on IL-1β production in vascular smooth muscle cells and in the work by Miao et al. on the attenuation of cancer cachexia by crosstalk induction by inflammation products.

The molecular origins of inflammatory processes affecting the architecture and function of the endothelium have also been covered in this special issue: On the one hand, the work of Wang and Lo identifies the action of the basic fibroblast growth factor on protecting the laminar shear flow medium of the arteries from the action of TNF-alpha induced endothelial dysfunction. On the other hand, the same group identified microRNAmediated changes in glucose metabolism leading to endothelial inflammation, as it can be found in the manuscript by Lo et al. Also discussing the role of non-coding RNAs in inflammation is the systematic review article by Gao et al., there they present an up to date review as well as a meta-analysis of the prognostic applications of lncRNA GAS5 in several carcinomas.

Neurological disorders have also been linked to nonresolved inflammation scenarios. In this regard, an integrative computational approach by the group of Ravichandran et al. shows how the analysis of molecular networks involved in inflammation led to the discovery of specific sites linked to Alzheimer's disease. Mathematical and computational models are increasingly providing insight, not only in the pathophysiology of inflammation and its influence on disease but are also helpful in the development of systematic therapeutics. Particularly useful are approaches that allow drug repurposing, since these provide a significantly faster and easier transition from research findings to patients' treatment. The work by de Anda-Jáuregui et al. provides a powerful example of high-throughput computational analysis of massive experimental databases combined with a network approach to the repositioning of anti-inflammatory drugs.

The development of novel therapeutic strategies and pharmacological approaches for the modulation of inflammatory processes is deeply covered in this research topic. Gao et al. contributed with a review that covers some of the therapeutic uses of the TLR signaling pathway in the development of antiinflammatory therapeutics based on molecular approaches going up to nanomedicine. There are also a number of research papers dealing with pharmacological and therapeutical developments to treat inflammation. This is the case of the study of the effects of the small molecule LASSBio-897 to treat silicosis as discussed by Carvalho et al.

The use of plant-derived pharmaceuticals to treat inflammation is covered by the study on the effects of the extract of the Amomum villosum ginger (rich in bornyl acetate) to treat intestinal mucositis as presented in the contribution by Zhang et al. Another natural product with important antiinflammatory effects is carnosic acid, a benezenediol compound with anti-oxidant properties that can be extracted from both rosemary and sage. In their contribution, Zhang et al. showed that by an epigenomic mechanism of miR-29b-3p-mediated inhibition of HMGB1/TLR4 carnosic acid is able to alleviate liver fibrosis in an animal model. The use of isoproterenol to modify aortic vasoreactivity and VCAM-1 modulation was investigated in an animal model and reported by Nieto-Lima et al. They found important age and gender associations between drug dose and response to the treatment. The differences found are attributed to differential beta-adrenergic stimulation. Authors discuss how then it is relevant to consider age and gender variability in the design of animal models, as it is customarily made in the clinical settings. In an additional animal model investigation, Trojan et al. developed an in-depth network study of chemokine interactions of anti-depressants that unveil the role of such drugs on the modulation of chemokine-chemokine interactions. The mini review by Cavalcante-Silva et al. on the modulatory properties of ouabain highlights a broader applicability of this cardiotonic steroid that includes regulatory actions of inflammatory events such as cell migration, vascular permeability, and cytokine production. Therapeutics is highly benefitted from translational research studies. In the case of nonalcoholic steatohepatitis, the work by Morrison et al. on gene profiling to unveil key modulators of inflammation in a mouse model of this disease results enlightening by showing that the effects of master regulator molecules and specific inflammation

pathways are similar between the human-based and animal models.

Together, these articles provide a sample of the multiple and complex roles of inflammation and its involvement in infections, immunity, and multiple pathological conditions. They also provide guidance for novel therapeutic approaches and future research on the physiology of inflammation.

### AUTHOR CONTRIBUTIONS

EH-L, MS, and CR planned and edited this special topic. EH-L, MS, and CR wrote this editorial.

**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 Hernández-Lemus, Soto and Rosales. 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.

# Remodeling of Retinal Architecture in Diabetic Retinopathy: Disruption of Ocular Physiology and Visual Functions by Inflammatory Gene Products and Pyroptosis

Rubens P. Homme1,2† , Mahavir Singh1,2 \* † , Avisek Majumder1,3, Akash K. George1,2 , Kavya Nair1,2, Harpal S. Sandhu4,5, Neetu Tyagi<sup>2</sup> , David Lominadze<sup>2</sup> and Suresh C Tyagi<sup>2</sup>

<sup>1</sup> Eye and Vision Science Laboratory, Department of Physiology, University of Louisville School of Medicine, Louisville, KY, United States, <sup>2</sup> Department of Physiology, University of Louisville School of Medicine, Louisville, KY, United States, <sup>3</sup> Department of Biochemistry and Molecular Genetics, University of Louisville School of Medicine, Louisville, KY, United States, <sup>4</sup> Department of Ophthalmology and Visual Sciences, University of Louisville School of Medicine, Louisville, KY, United States, <sup>5</sup> Kentucky Lions Eye Center, University of Louisville School of Medicine, Louisville, KY, United States

#### Edited by:

Geoffrey A. Head, Baker Heart and Diabetes Institute, Australia

#### Reviewed by:

Guillermo De Anda Jáuregui, Instituto Nacional de Medicina Genómica (INMEGEN), Mexico Enrique Hernandez-Lemus, Instituto Nacional de Medicina Genómica (INMEGEN), Mexico

#### \*Correspondence:

Mahavir Singh mahavir.singh@louisville.edu †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 16 October 2017 Accepted: 21 August 2018 Published: 05 September 2018

#### Citation:

Homme RP, Singh M, Majumder A, George AK, Nair K, Sandhu HS, Tyagi N, Lominadze D and Tyagi SC (2018) Remodeling of Retinal Architecture in Diabetic Retinopathy: Disruption of Ocular Physiology and Visual Functions by Inflammatory Gene Products and Pyroptosis. Front. Physiol. 9:1268. doi: 10.3389/fphys.2018.01268 Diabetic patients suffer from a host of physiological abnormalities beyond just those of glucose metabolism. These abnormalities often lead to systemic inflammation via modulation of several inflammation-related genes, their respective gene products, homocysteine metabolism, and pyroptosis. The very nature of this homeostatic disruption re-sets the overall physiology of diabetics via upregulation of immune responses, enhanced retinal neovascularization, upregulation of epigenetic events, and disturbances in cells' redox regulatory system. This altered pathophysiological milieu can lead to the development of diabetic retinopathy (DR), a debilitating vision-threatening eye condition with microvascular complications. DR is the most prevalent cause of irreversible blindness in the working-age adults throughout the world as it can lead to severe structural and functional remodeling of the retina, decreasing vision and thus diminishing the quality of life. In this manuscript, we attempt to summarize recent developments and new insights to explore the very nature of this intertwined crosstalk between components of the immune system and their metabolic orchestrations to elucidate the pathophysiology of DR. Understanding the multifaceted nature of the cellular and molecular factors that are involved in DR could reveal new targets for effective diagnostics, therapeutics, prognostics, preventive tools, and finally strategies to combat the development and progression of DR in susceptible subjects.

Keywords: chemokines, cytokines, diabetic retinopathy, epigenomics, homocysteine, inflammation, pyroptosis, signaling pathways

**Abbreviations:** AR, aldose reductase; BRB, blood-retinal barrier; CDC, center for disease control; CVDs, cardiovascular diseases; DM, diabetes mellitus; DME, diabetic macular edema; DR, diabetic retinopathy; ECM, extracellular matrix; HbA1C, hemoglobin A1C; Hcy, homocysteine; HNE, 4-hydroxynonenal; MI, myocardial infarction; MMPs, matrix metalloproteinases; NV, neovascularization; PDR, proliferative diabetic retinopathy; PRP, pan-retinal photocoagulation; PUFAs, poly unsaturated fatty acids; RPE, retinal pigment epithelium; T1D, type I diabetes; T2D, type II diabetes; TGF-β, transforming growth factor-β; TRD, tractional retinal detachment; VEGF, vascular endothelial growth factor; WHO, World Health Organization.

## INTRODUCTION

fphys-09-01268 September 4, 2018 Time: 9:44 # 2

Despite advances in medical health and technologies, the incidence of diabetes has reached epidemic proportions globally and thus diabetic complications are increasing throughout the world, e.g., DR. According to WHO, the diabetic population has risen by 314 million between 1980 and 2014 (Mathers and Loncar, 2006) and the CDC projects that Americans with diabetes will at least double or even triple in 2050, meaning one in every three adults in the United States would be affected if the current trend continues unabated. It is hard to imagine what this steep increase might have on the overall health care system in United States and the world since the prevalence of diabetes-related complications will keep rising proportionally in the affected patient populations (Mattei et al., 2015). In short, WHO believes that diabetes is an epidemic and therefore it may lead to a major health crisis in future. DM is divided into two main categories: T1D and T2D. T1D is characterized by a total or near total loss of insulin production, and thus patients are insulin-dependent. In contrast, T2D is characterized by insensitivity to insulin, and the patients can be treated with other drugs along with insulin during the early phases of the disease (Ferris et al., 1999; Nathan, 2015). Diabetes complications generally group into microvascular or macrovascular complications. Microvascular disorders include retinopathy, nephropathy, and neuropathy while macrovascular include coronary artery disease, stroke, and other peripheral vascular disease (Barrett et al., 2017; Huang et al., 2017). If these pathologies are not managed appropriately, then significant vascular and thus end-organ damage can occur, including the loss of retinal architecture. Eventually, new blood vessels (NV) will grow profusely superficial to and inside the retina leading to devastating outcomes in the untreated patients.

Diabetes can affect almost every part of our visual system, but vision loss and blindness occur predominantly because of retinal complications if timely and effective interventions are not sought by the patients (Gori et al., 2005). DR has already become a leading cause of acquired blindness in the United States and other developed nations. Unfortunately, ∼2.6% of global blindness cases can be attributed directly to diabetes (Bourne et al., 2013). Approximately 40% of diabetics over 40 years of age have some form of retinopathy; of which 8.2% exhibit visual symptoms and vision related complications (Ciulla et al., 2003; Klein et al., 2008). Landmark clinical trials established that hyperglycemia is a critical factor in the development of DR (Nathan, 2014). Thus, identifying early pathogenetic events in the disease process may provide useful insights that could lead to the identification and deployment of appropriate diagnostics and therapeutic targets for preventing the serious micro- or macrovascular complications that could arise later, particularly retinopathy (van Wijngaarden et al., 2005; Gross and Glassman, 2016).

Hyperglycemia's harmful effects are directly related to the structural damage inflicted on small retinal blood vessels and are governed by the systemic and local ocular factors. Although there are genetic factors also at play, but individuals with elevated blood glucose levels have severe retinopathy than those with strict blood glucose control (Kuo et al., 2014; Pradhan et al., 2016). HbA1c measures overall glucose level in blood over a period of 3 months. HbA1c levels less than 7.0% are generally recommended to minimize the risk of vascular complications, including DR. On the other hand, increased levels of blood lipids may also impart some role in the progression of DR as lipid-lowering compounds such as fenofibrate offers benefits in preventing the progression of DR. The complete details of the mechanism(s) underlying this protection are not well understood (Knickelbein et al., 2016; Ju et al., 2017; Liu et al., 2017). Moreover, the genetics and epigenetics of an individual can influence susceptibility to DR and response to the treatment. However, the exact number and nature of the genes involved or the epigenomics elements that are involved remain elusive. Other demographic risk factors for DR include Hispanic or African American ethnicity, systemic hypertension, duration of disease, and pregnancy (Gupta and Misra, 2016; Penman et al., 2016; Varma et al., 2016; Davidson, 2017; Jin et al., 2017; Kowluru, 2017; Mastropasqua et al., 2017; Zhang et al., 2017). DME is one of the most common complications and the cause of vision loss in diabetics. The early stage of DR is called nonproliferative diabetic retinopathy (NPDR) and is characterized by the presence of intra-retinal hemorrhages, microaneurysms, intra-retinal microvascular abnormalities, and cotton wool spots (fluffy white patches in the retina caused by focal swelling in the retinal nerve fiber layer). Development of retinal NV is called PDR, an advanced form of DR that carries a risk of other structural complications. Retinal changes occur after approximately a decade of living with T1D. One study reported nearly a 50% chance of developing PDR after 25 years of diabetes. Recently, progression to PDR and severe visual loss has become less, reflecting a strict control of hyperglycemia, serum lipids, blood pressure, and earlier diagnosis (Scanlon et al., 2013).

There are three principal ways by which DR can cause vision loss: (1) vitreous hemorrhage or TRD due to PDR (Crawford et al., 2009), (2) DME (Wilkinson et al., 2003; Lee et al., 2015), and (3) macular ischemia (Sim et al., 2013). In PDR, the new vessels are incompetent and can easily hemorrhage into the pre-retinal space or vitreous cavity, causing severe vision loss, often instantaneously. The vessels ultimately become fibrotic if left untreated and can exert traction on the retina, causing a TRD phenotype. DME involves fluid leakage out of retinal capillaries into the central retina (macula), the visually significant portion of the retina. Finally, macular ischemia, as the name implies, involves a critical lack of perfusion to the macula due to microvascular complications, causing dysfunction and even death of retinal cells (Hardy et al., 2005; Simo et al., 2006; Wong et al., 2016). Without therapeutic intervention, DR will result into the vision loss or blindness.

Management of DR starts with screening patients for the signs of retinopathy and then treating them if vision-threatening lesions are identified. Typically, DME is treated with intravitreal administration of anti-VEGF based biologics or small organic molecules such as steroids (Hodzic-Hadzibegovic et al., 2017; Koyanagi et al., 2018). Anti-VEGF agents have primarily replaced focal or grid laser photocoagulation of the macula as the firstline treatment of DME, although this remains an important adjunct therapy in refractory cases. In practice, the management of DR aims at preventing or delaying the onset of retinopathy by

controlling high blood pressure, blood sugar, and lipids levels. Introduced approximately 60 years back, photocoagulation is still an effective treatment for PDR (Evans et al., 2014). Recent clinical trials have established that PDR can also be treated with repeated anti-VEGF injections, though it is still unclear how long a patient be treated if one opts for the injection strategy. Other drugs that are currently being tried are the potential agonists for peroxisome proliferator activated receptors (PPARs), plant extracts such as forskolin (it binds to glucose receptor, specifically GLUT1), minocycline (it serves as an anti-inflammatory agent), celecoxib, another COX-2, and angiopoietin 2 antagonists. Anti-VEGF agents have been shown to have a disease-modifying effect (Ip et al., 2015). However, such an approach carries a significant treatment burden (Blanco-Garavito et al., 2017; Reich et al., 2017; Wong et al., 2017). Thus, there remains a considerable unmet medical need for the development of effective strategy (ies) to prevent DR or at least slow its progression. Critical aspects of this endeavor are to ensure that the anticipated intervention is safe, well tolerated, and less burdensome than the currently available options. Some investigations have implicated inflammation and pyroptosis as a possible cell-molecular mechanism in DR biology (Feenstra et al., 2013; Volpe et al., 2016).

### INFLAMMATION AND PYROPTOSIS DISRUPT PHYSIOLOGICAL HOMEOSTASIS IN THE EYE

During the pathogenesis of DR, inflammatory mediators are elevated before any anatomical or histopathological alterations are seen because hyperglycemia induces metabolic dysregulation that causes cells to malfunction and give rise to a chronic state of low-level inflammation in the body. A concurrent infection and the subsequent mounting of an immune along with existing inflammation in diabetics can further activate additional proinflammatory pathways. These can potentially further disturb the homeostatic balance in the eyes (Joshi et al., 1999; Casqueiro et al., 2012; Kitagaki et al., 2016; Heller et al., 2017). This heightened immune reaction, in the long run, plays havoc with the host's cellular functions. New evidence now indicates that retinal inflammation plays a dominant role in the pathogenesis of DR (Zhang et al., 2011; Whitcup et al., 2013; Ascaso et al., 2014). Data from experimental findings and clinical studies revealed a set of inflammatory cytokines, chemokines, and their receptors that were acutely expressed in the RPE, blood, vitreous, and aqueous humor of the patients. In other words, upregulation of the inflammatory mediators and cross-talk between them cause microvascular changes and breakdown of the BRB (**Figure 1**).

Homocysteine; a homolog of the amino acid cysteine has been implicated in a variety of thrombotic and vascular occlusive diseases via an array of inflammatory signatures. In a clinical study of plasma homocysteine from diabetics and controls, the mean Hcy level recorded was 11.75 ± 0.24 in the control group, 13.46 ± 0.74 in people with diabetes who had no DR, 14.56 ± 0.64 in subjects with NPDR, and 15.86 ± 1.34 in subjects with PDR. Thus, Hcy levels seem correlative with diabetes, and the severity of DR. In general, hyperhomocysteinemia (HHcy) is defined as a condition when Hcy levels are higher than the 15 µM/L. The data indicated that HHcy was significantly higher in the NPDR and PDR groups relative to the control group. These observations and other findings suggest that HHcy is associated with DR, and it can explain the increased risk of microvascular angiopathy in diabetics (Goldstein et al., 2004). Interestingly, Hcy is metabolized to homo-thiolactone (HCTL) which is an intra-molecular thioester of Hcy and is regarded as more toxic than Hcy. Paraoxonase (PON) exhibits esterase activity and lactonase activity (PON-HCTLase) that help detoxify HCTL. PDR patients show a significant increase in HCTL and PON-HCTLase activities and that have been confirmed in in vitro studies using the bovine retinal endothelial cells (RECs) depicting a direct proportional relationship in PON-HCTLase activity and mRNA expression levels of PON2 in a dose- and timedependent manner when treated with Hcy and HCTL. It was concluded that elevations were probably a protective effect of eliminating HCTL, which mediated endothelial cell dysfunction. Furthermore, a bioinformatics analysis revealed that HHcy could modulate the structural and functional aspects of PON in PDR and could also affect the dual enzyme activity of PON. These findings indicate the possibility of PON activity and changes in HCTL expression as potential biomarkers for screening and detection of DR patients (Barathi et al., 2010). In our early study, we also demonstrated that Hcy could potentially mediate the expression of inflammatory markers (e.g., chemokines, cytokines, and interleukins receptor molecules) in human retinal cells without interfering with their cellular morphologies or their genomic integrities (Singh and Tyagi, 2017a).

Inflammation plays an essential role in the early pathogenesis of DR (Kern, 2007; Tang and Kern, 2011; Rangasamy et al., 2012; Kastelan et al., 2013). As the disease progresses, a corresponding increase in inflammatory factors along with leukocyte adhesion to inflamed endothelium affect the integrity of the BRB in diabetic eyes. As we know, maintenance and regulation of the BRB are required for normal retinal function. Loss of this barrier can result in the panoply of retinovascular diseases, including DR. In fact, loss of BRB integrity is associated with DR which can lead to vascular leakage and development of macular edema. In this context, DME can result in decreased vision and even blindness if patients fail to seek timely medical intervention due to leakage and blockage of micro-vessels (Bhagat et al., 2009). In diabetic patients and diabetic animal model studies, studies have shown an increase in expression of many proinflammatory mediators including TNF-α, IL-6, and IL-1β in both the vitreous and the retina. These inflammatory markers were associated with an increase in permeability in retinal vessels, particularly in diabetic eyes. Studies have also shown that hyperglycemia injures retinal vascular morphology through activation of a pro-inflammatory phenotype in endothelial cells that is characterized by the upregulation of cell surface adhesion molecules (e.g., ICAM-1, VCAM-1) (Kim et al., 1994; Morigi et al., 1998). These molecules, in turn, facilitate the adhesion and subsequently the transmigration of leukocytes across endothelial lines, leading to a retinovascular inflammatory hotbed, resulting in capillary occlusion, and apoptosis of the endothelial cell. These injuries eventually induce the breakdown of the BRB

(Miyamoto et al., 1998, 1999; Miyahara et al., 2004). Therefore, an anti-inflammatory treatment could be very useful in preventing further damage from DR (**Figure 1**) (Joussen et al., 2001, 2003; Miyahara et al., 2004).

Sequence of events during the immune response that leads to BRB alterations and subsequent breakdown appears to be as follows: (a) upregulation of endothelial adhesion molecules [e.g., ICAM-1, VCAM-1, platelet and endothelial cell adhesion molecule 1 (PECAM-1)], (b) leukocyte adhesion to the endothelium, (c) simultaneous release of chemokines, cytokines, and vascular permeability factors, (d) decrease in tight-junction associated proteins, and finally (e) inability of white blood cells to penetrate the neuro-retinal parenchymal structures. Important pro-inflammatory chemokines and adhesion molecules contributing to the initiation and progression of DR phenotypes, include endothelin-1, Interleukins [interleukin 1β (IL-1β), interleukin 8 (IL-8), interleukin 1α (IL-1α), interleukin 6 (IL-6)], monocyte chemotactic protein 1 (MCP-1), p-selectin, and CXCL10/IFN induced protein 10 (IP-10). Additional cytokines that are also associated with DR are angiopoietin 2 (Ang 2), tumor necrosis factor alpha (TNF-α), platelet-derived growth factor AA (PDGF-AA), the members of VEGF family, and TGF-β. While a variety of chemokines are elevated in diabetic eyes, chemokine ligand 2 (CCL2) is more prevalent in both the vitreous and the serum (Elner et al., 1995; Dominguez et al., 2016). CCL2/MCP-1 influences vascular inflammation via activating and recruiting leukocytes during hyperglycemia in various retinal cells (e.g., pigmented epithelial cells, Muller's glial cells, retinal vascular endothelial cells) (Bian et al., 2001). Furthermore, CCL2 gene polymorphisms were found to be associated with the gradual development of DR in susceptible hosts (Maier et al., 2008; Dong et al., 2014; Urias et al., 2017). BRB breakdown was inhibited in a diabetic CCL2 knock-out mice model, hence directly implicating this molecule. Thus, inhibition of CCL2 may selectively prevent disruption of BRB in DR related pathologies (Rangasamy et al., 2014). By-products of glucose metabolism are transformed non-enzymatically into advanced glycation end-products (AGEs). AGEs can further react to make reactive species like dicarboxylic compounds. Furthermore, it was reported that AGEs tend to accumulate in the retina and cornea in diabetics indicating the onset of pathological alterations inside eyes. In vivo studies confirmed that exposure to AGEs increases VEGF expression along with inflammatory mediators such as ICAM-1 via nuclear factor-kappa beta (NF-κB)

(Stitt, 2003; Alghadyan, 2011). As DR progresses, it causes significant thickening of the capillaries' basement membrane in different compartments of the retina [e.g., the inner plexiform layer (IPL) relative to the outer plexiform layer (OPL)] (Anderson et al., 1995). DR can also lead to the induction of apoptosis in glial, ganglion, and Müller cells as well as in the capillary network of the retina (Kowluru et al., 2012).

It is becoming increasingly evident that MMPs are primary regulators of the innate as well as acquired immune networks in our body (Kessenbrock et al., 2010; Singh and Tyagi, 2017c), shedding light on the development of DR pathology (Parks et al., 2004). MMPs are primarily produced as zymogens; inactive enzymes which contain a pro-peptide domain that needs to be cleaved. The pro-peptide domain contains a cysteine residue that interacts with zinc in the enzyme-binding site, thus preventing enzyme activity. Some of the MMPs do have a prohormone convertase cleavage site, a furin-like entity, which when cleaved activates the enzyme. Some MMPs also include a transmembrane segment in their domains (Pei et al., 2000). Immune-related molecules such as CCL/monocyte chemoattractant family protein members are activated by MMP cleavage. MMPs and their regulators are upregulated in the diabetic vitreous (Tuuminen and Loukovaara, 2014). This observation was consistent in animal models of diabetes which showed the retinal upregulation of MMP-2 and MMP-9 (Giebel et al., 2005). Accordingly, MMP inhibitors have been shown to significantly decrease retinal vascular permeability and loss of junctional proteins in diabetic animals (Navaratna et al., 2007; Rangasamy et al., 2012). These observations suggest a plausible mechanism for BRB breakdown via the proteolytic degradation of VE-cadherin (Navaratna et al., 2007). In fact, hyperglycemia can activate many other types of soluble mediators such as AGEs and a range of potent inflammatory cytokines that can increase MMPs' expression levels. Findings demonstrated that retinal inflammation could attract an increased number of leukocytes to the retina, as mentioned earlier, which in turn can bind to the vascular endothelium to activate cellular proteinases like elastase which in turn cleaves VE-cadherin and its associated proteins from the cell surface repertoire. Based on these observations, it appears that these proteinases may serve as potential therapeutic targets for DME (Allport et al., 2000).

Interestingly, remodeling of the retina and its altered architecture in DR can be explained by the active role played by a host of MMPs and other events like pyroptosis. As we know that MMPs are implicated in numerous cellular functions such as degradation of ECM, cell proliferation, apoptosis, corneal collagen formation, organ development as well as in the causation of internal inflammation. Some of these proteins like MMP-2 are most prevalent in cells and tissues while MMP-9 is quite complicated in its functionality. Notably, MMP-9 levels are different among various ethnic groups while there is no significant difference between genders (Tayebjee et al., 2005). MMPs are involved in the pathogenesis of DR since MMP-9 is activated through Ras/Raf/mitogen-activated protein MAPK/ERK pathway in the retinal and retinal capillary cells of diabetics. MMP-9 plays a role in the death of retinal capillary cells and that increases vascular leakage through disruption of the tight junction's complex, and degradation of the tight junction protein occludin. They also participate in the process of DR NV (Kowluru et al., 2012). Studies show that MMP-9 participates in the development of DR via two routes: by inducing apoptosis in retinal capillaries cells during the early disease stage and NV in the later stage (Salzmann et al., 2000; Giebel et al., 2005; Kowluru and Zhong, 2011; Kowluru et al., 2011, 2012). MMPs also participate in a host of organ-specific (e.g., retinopathy, nephropathy, and cardiomyopathy) diabetic complications due to a high glucose environment that stimulates secretion of these zinc-dependent proteinases during disease progression. In fact, animal models that have been specifically developed and studied simulating DR phenotypes and when their retinal and vitreous samples were examined along with diabetic retinopathic patients they all showed a significant increase in the levels of MMP-9 and MMP-2 (Kowluru et al., 2012). Thus, hyperglycemia does affect the activity of MMPs in the eye, and they can certainly compromise the integrity of the BRB. Interestingly, physical activity such as exercise can ameliorate hyperglycemia as well as can lower the levels of MMPs, and therefore can potentially help prevent or at least slow down the pathogenesis of DR in susceptible individuals.

Evolutionarily, the eukaryotic cell is fully capable of starting several cell death programs when confronted with that option that may fall into two broader categories: non-inflammatory or pro-inflammatory. Cell death programs include necrosis, apoptosis, and pyroptosis (Sarks et al., 1988; Kaneko et al., 2011; Tarallo et al., 2012). Apoptosis is non-inflammatory while others induce inflammation and various cell death pathways can be identified by certain markers. Apoptosis is identified via the caspase while pyroptosis is identified by the inflammasome activity (Bergsbaken et al., 2009). Pyroptosis is inherently an inflammatory event in its origin and can be triggered by a host of stimuli including the CVDs such as stroke and heart attack, cancer, neurological disorders, etc. Pyroptosis is also crucial for controlling microbial infections, and it is known that the competition between the host and pathogen can modulate the process of pyroptosis. Pyroptosis signaling pathways that have been studied extensively employ the NLR family pyrin domain containing 3 (NLRP3) inflammasome. This pathway includes apoptosis-associated speck-like protein containing a carboxy-terminal CARD, caspase-1, bridging adaptor, and of course the NLRP3 and is activated when NF-κB is activated and that increases IL-1β and NLRP3 expression. It also triggers inflammasome assembly to generate IL-18 and IL-1β. Recent studies have identified pyroptosis in aged-related macular degeneration leading to the RPE cell death (Gao et al., 2015). As discussed above for DR, all these factors are present such as the activation of NF-κB, upregulation of IL-1β and IL-18. Other related studies have shown that diabetic animal models deficient in NLRP3 have attenuated DR signatures (Schroder et al., 2010). Additionally, some studies showed the presence of NLRP3 in the vitreous of PDR patients (Loukovaara et al., 2017). It is worth to point out that we are just beginning to understand the finer details and related molecular mechanism(s) that govern pyroptosis and other related processes which are downstream of the caspase-1 activation pathway in the eyes.

In future, identification of potential markers may likely provide further insights that can allow us to comprehend this process in more details. However, it is important to remember that composition of the inflammasome may too play pathogenic role(s) and thus can determine the fate of the inflamed cells in the eyes that have been exposed to varied responses because of chronic stimuli environment. Thus, pyroptosis and inflammation are relevant to our understanding of their effects during serious medical conditions for which inflammation is central to the pathophysiology of diseases such as DR. In fact, DR shares many similarities with other inflammatory diseases and as a result it is being recognized now as the central player in DR via various mechanisms including pyroptosis (Joussen et al., 2001; Singh and Tyagi, 2017b).

### RETINAL NEOVASCULARIZATION LEADS TO BLINDNESS DURING DIABETIC RETINOPATHY

Similar to the pathology of the neovascular glaucoma (NVG); the slow growth of new blood vessels that occurs in the iris as well as the trabecular meshwork (the proverbial "drain" of the eye) inhibits outflow of the aqueous humor fluid. Consequently, intraocular pressure rises precipitously which can cause irreversible damage to the optic nerve leading to blindness if left untreated. Same way, the corneal epithelium in people with diabetes is prone to sloughing and epithelial erosions (e.g., breakdown of the epithelial surface or corneal abrasions) and that can develop into a chronically non-healing epithelial defect which is painful and that can decrease the vision (Bikbova et al., 2012; Vieira-Potter et al., 2016). Likewise, in early DR the patients may be asymptomatic which is clinically called as NPDR. The only cause of vision impairment in NPDR is DME, which is present only in a minority of cases with mild NPDR. However, as stated above, during the late stage of the disease, small, new blood vessels (NV) grow on top of the retina and into the preretinal space or the vitreous (**Figure 2**). As we know that the retinal vascular unit is mainly composed of endothelial cells, pericytes, and astrocytes. Diabetes over time changes the integrity and structure of this retinal neurovascular unit (Shin et al., 2014). The production of various growth factors by the inflamed retina may also trigger an inflammatory cascade that further drives NV. Unlike NPDR, PDR causes hemorrhage into the vitreous cavity, which frequently results in a sudden and profound loss of vision. On the other hand, the principal cause of retinal NV in PDR is hypoxia wherein VEGF is released from retinal cells in response to hypoxia. VEGF being an anti-apoptotic factor can promote endothelial cell division, vasodilation, and increase in vascular permeability (**Figure 2**). VEGF is also capable of activating PKCβ which phosphorylates the tight junction protein, occludin. The phosphorylated occludin is then ubiquitinated and targeted for degradation leading to an increased vascular permeability (Lai et al., 2005; Murakami et al., 2012; Manolov et al., 2014). VEGF is made by many types of retinal cells, including RPE, Muller, ganglion, glial, and endothelial cells (Huang et al., 2011; Sun et al., 2012) and it has been studied extensively. It is considered a critical factor related to the BRB breakdown and its levels have been found to be significantly higher in DME patients than non-diabetics ones (Funatsu et al., 2002; Caldwell et al., 2003). The biology of VEGF is unique in the sense that by serving as a vasoactive cytokine, it can also increase vascular permeability, thereby causing extravasation of fluid into the retinal space. VEGF can also efficiently phosphorylate important proteins, e.g., VE-cadherin, occludin, and ZO-1, leading to BRB impairments (Caldwell et al., 2003). Additionally, VEGF can stimulate leukostasis in small retinal vessels. These leukocytes further release more cytokines or migrate via the trans-endothelial routes, thus exacerbating BRB dysfunction in the eyes (**Figure 1**) (Aiello et al., 1994, 1997).

There are several other pathways beyond VEGF that are important in retinal NV and are plausible therapeutic targets. The one under active investigation is the angiopoietin-Tie2 signaling pathway. Angiopoietin 1 (Ang1) is a protein that acts as an agonist of Tie2, a receptive tyrosine kinase mostly expressed on endothelial cells. This ultimately results in the stabilization of the endothelium and inhibits retinal NV when VEGF levels are high. Conversely, angiopoietin 2 (Ang2) dephosphorylates Tie2 which destabilizes endothelial cells and thus promotes NV (Campochiaro, 2013). Vitreous levels of Ang-2 are significantly upregulated in patients with severe macular edema, thus indicating that Ang-2 may be associated with impairment of the BRB. Furthermore, in a study when Ang-2 was injected in the vitreous of non-diabetic rats, it was concluded that it significantly increased vascular permeability in the retina. Other studies found that Ang-2 contributed to VE-cadherin loss via phosphorylation (Patel et al., 2005; Rangasamy et al., 2011). Ang-2 can further increase inflammation via inducing sensitivity of endothelial cells to TNF-α, which then upregulates the expression of ICAM-1 thus promoting adhesion of monocytes. Moreover, VEGF and Ang2 act synergistically to cause NV, but interestingly Ang2 in the setting of low VEGF levels can actually inhibit the process. The synergy displayed by Ang2 and VEGF has prompted interest in combined VEGF and Ang2 inhibition. A phase II trial of a single monoclonal antibody that inhibits both targets was completed in early 2018 and showed additional efficacy over anti-VEGF therapy alone for DME (Dugel et al., 2018). This represents a promising new target both for achieving more rapid visual gains in DME and possibly in treating the subset of patients whose DME is non-responsive to anti-VEGF therapy. Since anti-VEGF therapies remain the treatment of choice for DME and are also a reasonable alternative treatment to PRP for the treatment of PDR. Despite some concerns about its heavy usage, clinical trials involving patients followed for 5 years have not shown any such toxicity. The injection procedure does carry risks of vitreous hemorrhage, retinal tear, retinal detachment, cataract formation, and endophthalmitis, but thankfully these complications are infrequent and are generic ones for an injectable agent. In a trial of repeated bevacizumab, ranibizumab, or aflibercept for the treatment of DME, conducted by Diabetic Retinopathy Clinical Research Network, 15–16 injections over 2 years were required on average to control edema (DRCR Protocol T, 2-year results). Drugs which are currently in clinical use for treating DME include VEGF inhibitors such as ranibizumab (Lucentis) and

bevacizumab (Avastin) but there are few other options which include the soluble receptor for VEGF commonly known as the VEGF-Trap or aflibercept (Eylea) and intravitreal triamcinolone.

As discussed, because of the complexity and multifactorial nature of DR, there appears to be potential for discovering numerous biomarkers based on inflammatory molecules (like cytokines, chemokines and their respective receptors) or metabolites (such as homocysteine; Hcy) or various species of RNAs molecules (for example; miRNAs, circRNAs, ncRNAs, etc.) all stemming from inflamed retinal cells during DR disease pathogenesis. Anti-VEGF therapy in PDR is effective but requires indefinite treatment and for DME it is the mainstay of treatment, but approximately 40% patients in the DRCR Protocol I study, still had persistent edema at 6 months despite repeated injections. Furthermore, current treatments of anti-VEGF are inefficient as proven with the re-occurring of the edema within weeks or months of administration and thus repeated doses are invariably needed. Therefore, VEGF inhibition alone may not be enough and able to achieve neutralization of the inflammatory cascade and pyroptosis process that are responsible for BRB breakdown in the eyes. Thus, inhibition of inflammatory mediators could prove fruitful in DME therapies since cytokines and chemokines are upregulated in blood, aqueous and vitreous humor and these factors tend to accelerate DR pathogenesis. Other interventions or approaches are urgently needed to reverse the growth of retinal NV and treat DME (Cheung et al., 2010). A few newer versions of therapeutic strategies that can successfully target vascular complications in diabetes have also been developed such as the selective inhibition of sodium-glucose co-transporter 2 (SGLT2). These drugs have been recommended for the treatment of patients suffering from diabetes because of low risk of hypoglycemia and no weight gain. The principal mode of actions of the drugs rely upon the observations that up to 90% of the glucose is filtered by the kidney is absorbed back by a low-affinity/high-capacity SGLT2 that is expressed in the S1 and S2 segments of the proximal tubule in the kidneys. Thus, the blockade of SGLT2 promotes the urinary glucose excretion leading to the improvement in hyperglycemia in an insulinindependent manner. In fact, clinical studies proved that SGLT2 inhibitors reduce blood pressure, body weight, and serum uric acid levels and thus could also ameliorate cardiovascular risks in diabetics (Yamagishi and Matsui, 2016). It was proposed that patients with steroid-resistant DME, and those with refractory to the dipeptidyl peptidase-4 inhibitors (DDP4) the visual symptoms and macular edema might be treated by these SGLT2 inhibitors (Yoshizumi et al., 2018).

Also, DR treatment varies depending on the extent of the damage done by the disease itself. Some diabetics may require laser surgery to seal their leaking blood vessels or to even prevent blood vessels from leaking further while other might need immediate intravitreal injections to decrease inflammation or to reverse the course of NV. PDR patients may also need

surgery to remove and replace a part of the vitreous or to repair TRDs. In the worst-case scenario, over time scar tissue in or on top of the retina can form which can cause retinal detachment and eventual cell death. While many of these detachments can be anatomically repaired with surgery, visual outcomes are mixed and frequently remain poor (**Figure 2**). DR can be diagnosed through a comprehensive eye examination. It is essential to know the patient's history to determine vision difficulties because other co-morbidities may affect vision too. Therefore, visual acuity measurements are needed to determine how much central vision has been altered or lost. The doctor can also check refraction to determine whether a new eyeglass prescription would help the affected patient. Measurements of the intraocular pressure can also assist in deciding the appropriate treatment modalities for the diabetics. Supplemental testing may include retinal photography or optical coherence tomography (OCT) to analyze retinal layers' integrity as well as to screen for macular edema (Bhaduri et al., 2017). Fluorescein angiography can readily identify NV, microaneurysms, and intra-retinal or intravitreal leakage (Takase et al., 2015; Hwang et al., 2016; Nesper et al., 2017). As detailed above, the treatment for DR depends on the stage of the disease, and the main aim of any treatment is to diminish or prevent disease progression. During the early stage of NPDR, strict monitoring is advised. Following physicians' recommendation concerning diet and exercise coupled with blood sugar monitoring can help in following the disease progression. Lifestyle changes that promote physical activities such as regular exercise along with a healthy diet can potentially reduce blood glucose levels in patients suffering from T2D and reduce the risk of developing CVD and related pathologies. Physical activity increases insulin sensitivity and increases expression of the GLUT-4 transporter, which increases glucose uptake. Also, exercise tends to decrease blood MMP-9 levels in the diabetic patients (Kadoglou et al., 2010; Dehghan et al., 2016). Furthermore, because DR is typically diagnosed via a formal examination of the eyes by an expert in the clinic, this limits the number of patients that can be screened at a given time in a day. Therefore, an easily accessible and a reliable screening biomarker centered approach for timely DR detection (and prognosis) would also have tremendous practical value for disease management. This discovery could allow clinicians to identify susceptible patients early on that are in an urgent need of further assessment and/or treatment.

### EPIGENETIC MODIFICATION OF THE GENOME IN DIABETIC RETINOPATHY

Emerging evidence suggests that apart from one's genetic makeup, the environment does affect DR phenotypic outcomes. Previous clinical trials have underscored the advantages of glucose control in preventing and diminishing DR-related complications, but studies have also demonstrated that strict control of glucose in diabetics does not suddenly stop DR progression (Mishra and Kowluru, 2016). In other words, the benefits arising from good control usually persist beyond the period of glycemic control. These specific observations hint at the importance and the existence of epigenetic regulatory mechanism(s) that have been revealed during the last few years. Recently, many types of epigenetic regulations, such as histone post-translational modifications in chromatin or DNA methylation, and the involvement of specific signature(s) governed by various RNA species have been associated with DR. Such discoveries point to the fact that the new field of "epigenomics" and its implications in DR and related pathologies are continually evolving such as the expression of pathological genes was found to be altered in endothelial and vascular smooth muscle cells without inducing any changes in the underlying DNA sequences of these cells (**Figure 3**). Further, besides the phenomenon of "metabolic memory" other events at play may determine the long-term epigenetics outcome(s) in cells and organs of the patients even after the underlying causes have been remedied (Zhang et al., 2012; Kowluru, 2017). Although some progress regarding the role of epigenetics in diabetes complications such as CVD has been made, the significance and importance of epigenetic mechanism(s) in DR remain mostly unknown (Pasquier et al., 2015). Researchers strongly believe that epigenetic modifications seem to play essential roles in the progression of DR (Zhong and Kowluru, 2011). Due to incredibly rich mechanistic pathway(s) that involves epigenetics (e.g., miRNAs' expression, DNA methylation patterns, histone post-translational modifications, etc.), there exists a true potential for further research in DR to discover new details with potential therapeutic implications. We know that hyperglycemia has been shown to influence changes in the structure of chromatin through activating signaling pathways and hitherto unknown components (Qiu, 2006). In fact, many essential enzymes are specifically activated depends on the state (i.e., active or repressed) of chromatin. Equally, inflammatory genes in retinal cells that are regulated by epigenetics have also been demonstrated. Histone modifications by enzymes such as histone methyltransferases, histone demethylases, histone acetyltransferases, and histone deacetylases can reversibly change gene expression patterns (Pradhan et al., 2016; Zhang et al., 2017). Therefore, elucidating epigenetic regulation could certainly lead to the discovery and development of novel therapeutic agents for DR. Popularly known as 'epi-drugs,' such as histone demethylases, HATs, DNA methylation inhibitors, and HDACs are presently being analyzed for other medical conditions like cancer (Shi, 2007). It is hoped that available medicines could be used for their potential ability to alter epigenetic markers and thus also become an essential part of the therapeutic regimen for DR in the coming future (Zeng and Chen, 2014). As such, well-defined animal and cell based disease models with and without related interventions could lead us to understand the finer details of epigenetic regulations and how to employ them to prevent DR (Reddy et al., 2015). It is anticipated that the ongoing human epigenome research project will improve our understanding further of epigenetic states in normal health and the disease states (**Figure 3**) (Slomko et al., 2012).

Again, poor glycemic control and epigenomics have close relationship with DR as demonstrated in preclinical studies showing upregulation of HDAC1, HDAC2, and HDAC8 in the retinas of streptozotocin (STZ)-treated rats. Additionally, impairments in histone H3-specific acetyltransferase activity

suggested that epigenetic metabolic memory might be a reason for continued disease progression even when blood glucose level has returned to normal (Zhong and Kowluru, 2010). Furthermore, antioxidants such as Sod2 (manganese superoxide dismutase parent gene) can prevent DR development in animals suggesting its role in DR. It is appropriate to mention that Sod2 is heavily regulated by epigenetic controls. Studies have demonstrated that epigenetic changes of retinal Sod2 play an essential role in the development of DR and the creation of traceable metabolic memory (Zhong and Kowluru, 2011, Kowluru, 2013). Also, hyperglycemia increases acetyl H3K9, H4K20me3 via NF-κB controlled genomic sequences encoding retinal Sod2, SUV420h2, and concurrently enhances the interactions between NF-κβ subunit of p65 to H4K20me3 and acetyl H3K9. Surprisingly, hyperglycemia reversal did not prevent these observations. SUV420h2; the methylation enzyme, is upregulated in retinal capillary cells and the retina during high glucose concentration. Furthermore, silencing of SUV420h2 prevents activation of H4K20me3 in Sod2 indicating the role of SUV420h2 in creating the metabolic memory in DR (Zhong and Kowluru, 2013). More specifically, hyperglycemia downregulated the signature methylation patterns (such as H3K4me1 and H3K4me2) and that can precisely be regulated by lysine-specific histone demethylase 1 (LSD-1) along with Sp1 interactions at the Sod2 promoter. It is known that LSD-1 demethylases (H3K4 and H3K9), and the subsequent methyl signature removal in H3K4 is associated with repression of the transcriptional profile. Interestingly, decreased H3K4me2 and increased LSD-1 at Sod2 also lead to the development of DR (**Figure 3**) (Forneris et al., 2008).

Matrix metalloproteinases play important roles not only in diabetic nephropathy and cardiac diseases but in retinopathy also as discussed earlier. Retinal MMP-9 is activated during diabetic damage to the eye initiating apoptosis of the capillary cells. During diabetes the promoter region of retinal MMP-9 exhibits an upregulation of acetyl H3K9, p65, LSD-1, and a downregulation of H3K9me2. Inhibition of LSD-1 via siRNA improves those changes and the impaired MMP-9 detrimental effects in DR (e.g., apoptosis) (Kowluru and Shan, 2017). From these observations and others, it appears that regulation of MMP-9 activities by epigenetic modifications as well as by genetic polymorphisms have important roles in the pathogenesis of DR as reviewed by Singh and Tyagi (Singh and Tyagi, 2017c). Also, the aberrant DNA methylation patterns of CpG dinucleotides in certain genetic elements are fundamental mechanisms in the regulation of gene expression status, thus disrupting the standard transcriptional regulatory machinery of a cell irrespective of a disease state (Micevic et al., 2017). Overexpression of methylation of the CpG dinucleotides in the polymerase-γ region affects mtDNA binding compromising the activity of transcription because replication enzyme remains downregulated along with mtDNA 'D' loop region damage. These changes were not reversed in the absence of high blood glucose level (Tewari et al., 2012).

### THE MICRORNAs AND THEIR ROLE(S) IN THE DEVELOPMENT OF DIABETIC RETINOPATHY

The expression levels of miRNAs are also affected during DR development. For instance, miRNA-200b is significantly upregulated in T1D model of DR. Moreover, a mimic of miR-200b decreased oxidase resistance 1-impaired oxidative stress progression during transfection. STZ-treated rats exhibited changes in the expression of 37 miRNAs in DR: upregulation of miRNA-96, 124, 182, 183, 204, and 211 while miRNA-10a, 10b, 199a-3, 219-2-3p, 144, and 338 were downregulated (Wu et al., 2012; Murray et al., 2013). Other studies have also shown an increase in the expression of miRNAs (miRNA-21, 132, 146, 155) that are responsive to NF-κB in RECs. The micro-RNAs (miRNA-17-5p, 18a, 20a, 21, 31,133) which are regulated by VEGF are also upregulated in RECs and retinas. As widely known that NF-κB is a crucial regulatory molecule in our immune system and is associated with DR development wherein it induces apoptosis in retinal pericytes. The miRNAs (miRNA-21, 146, and 155) that may be regulated via NF-κB were also upregulated in diabetic RECs indicating that NF-κB can activate miRNA-146 expression. The higher expression of miRNA-146 inhibits activation of interleukin-1β (activation-induced NF-κB) in RECs. Therefore, enhanced expression of miRNA-146 could be influential in controlling DR progression (Kowluru and Koppolu, 2002; Brooks et al., 2004; Taganov et al., 2006; Gatto et al., 2008; Sheedy et al., 2010; Kovacs et al., 2011). Increased expression of VEGF, and the miRNAs that respond to p-53 indicate the association of miRNAs in the pro-angiogenic or pro-apoptotic role(s) of VEGF in DR. Downregulation of miRNA-200b and upregulation of VEGF has been observed in human umbilical vein endothelial cells (HUVECs) and RECs. The miRNA mimic treatments in vitro in endothelial cells and in vivo (intravitreal injections) might counter the VEGF upregulation in DR. In animal models, miR-200b mimics intravitreal injections that were shown to be downregulated by VEGF-A (Long et al., 2010; McArthur et al., 2011; Li et al., 2017).

### PERTURBATION OF THE REDOX SYSTEM IN DR

Diabetes creates an excess of oxidants at the cellular level, i.e., oxidative stress. Many investigations have indicated a direct association between inflammation, oxidative stress, and diabetes during DR development. This relationship may help explain the harmful effects of inflammation and pyroptosis in both the retinal vasculature and retina and the consequent abnormalities in visual function. Oxygen consumption by mammalian cells is essential for sustaining aerobic life and the standard homeostatic mechanisms. This process is predicated upon a strict control in the level of reactive oxygen species (ROS) production and their removal from almost every tissue and cell type in our body. Although the initiators of oxidative stress in DR are not well understood, it is agreed that oxidative stress is influential in DR (**Figure 4**). Antioxidants have been shown to alter the course of retinopathy development in animal models (Tanito et al., 2005; Rosales et al., 2010; Kumar et al., 2012; Kiang et al., 2014; Arellano-Buendia et al., 2016), but the results obtained from limited clinical trials have not demonstrated the same effects (Berman and Gombos, 1989; Chiarelli et al., 2004; Garcia-Medina et al., 2011). However, antioxidants are routinely prescribed for a range of other chronic disease conditions. Thus, more evidence-based clinical trials are needed to justify therapeutic effects of antioxidants in the development and progression of DR (Kowluru and Chan, 2007).

Reactive oxygen species such as superoxides are potent secretagogues for cytokines secretion. They can promote inflammation through various pathways that can serve as a key pathogenetic event that can damage vascular cells and thus increasing the vascular permeability. Cytokines and chemokines released into the retina become potent recruiters of neutrophils and a host of other cells (De Groef et al., 2015; Liu et al., 2016). ROS induce NF-κB activation which drives an inflammatory response via increasing the levels of nitric oxide (NO), prostaglandins, and other cytokines (Schreck et al., 1992). The level of interleukins significantly increased in the vitreous of PDR patients, and the retina of experimental animal models such as diabetic rats and mice, particularly IL-1β, IL-6, and IL-8 are (Yuuki et al., 2001). This study showed that IL-1β increased significantly in retinal capillary cells under hyperglycemic conditions (Carmo et al., 2000; Kowluru and Odenbach, 2004b). Interestingly, IL-1 activation can lead to release of more ROS and NF-κB translocation to the nucleus, thus creating a continuous inflammatory response (**Figure 1**) (Chang and LoCicero, 2004; Koenig et al., 2016). Further, IL-1β is associated with apoptosis of retinal capillary cells, probably via NF-κB and caspase-3 activation (Kowluru and Odenbach, 2004a; Gupta et al., 2017; Mendiola and Cardona, 2017). IL-1 could upregulate prostaglandin E<sup>2</sup> (PGE2) via the catalytic enzyme cyclooxygenase-2 (COX-2). Both are associated with DR development by modulating vascular permeability and angiogenesis through VEGF (Ozawa et al., 2011).

Oxidative stress has also been linked to chronic inflammation leading to the changes in messages that are encoded by stress response genes (Reuter et al., 2010). These signals, in turn, induce diabetic complications such as retinopathy. It is worth noting here that retina contains relatively higher levels of polyunsaturated fatty acids (PUFA) and uses more oxygen by mass than any other part in the human body. It also exhibits higher glucose oxidation than any other tissue in mammals. Various species of ROS (e.g., superoxide, hydrogen peroxide, etc.) are elevated in retinal tissues of diabetic rats while the activities of the enzymes that specifically quench these ROS such as superoxide dismutase, catalase, glutathione reductase and peroxidase are decreased in the diabetic retinal cells. Prominent antioxidants such as beta-carotene, vitamin C, and E, and GSH are decreased likewise in the diabetic patients. Biochemically, glucose phosphorylation results in fructose 3-phosphate which further breaks down into three deoxyglucoses. Both are glycation agents which can then enter the advanced glycation pathway as oxidative stress activates signals that can lead to the generation

of AGE products, the induction of the hexosamine biosynthesis, enhanced generation of insulin-like growth factor 1 (IGF-1), and, most importantly, activation of the polyol pathway (**Figure 4**) (Singh et al., 2015). Glucose is reduced to sorbitol via AR which is then converted to fructose by sorbitol dehydrogenase (SDH) in the polyol pathway. AR and SDH are two primary enzymes that are heavily used in the polyol pathway. The conversion of glucose to sorbitol requires abundant NADPH. This competes with the reaction that forms the antioxidant GSH, which also requires NADPH. The use of NAD<sup>+</sup> may also create an imbalance in the overall ratio of NADH/NAD+, the excess NADH could become a substrate for NADH oxidase, which in turn, can generate more oxidative species in the cells. Many retinal cell types contain AR, including ganglion cells, Müller cells, endothelial, pericytes, as well as retinal pigment epithelial (RPE) cells. They all are subjected to polyol pathway metabolism in diabetic eyes. Also, retinal cells in diabetic animals show an increase in the levels of sorbitol, fructose, and oxidative stress, and these could be prevented through compounds that can specifically inhibit the biological activity of the AR enzyme (Sato et al., 1993; Lorenzi, 2007). AR can potentially deplete the levels of NADPH in the polyol pathway thereby lowering its availability for glutathione reductase, an important antioxidant. In other words, the overall ability of a cell to respond to oxidative stress decreases significantly under hyperglycemic conditions (Lorenzi, 2007).

### ROLE OF SIGNALING PATHWAYS DURING DIABETIC RETINOPATHY

Obesity and T2D are associated with galectin-3 (Gal-3) upregulation which is known to mediate inflammation and clearance of glucose adducts. Gal-3 knock-out (KO) mice were shown to upregulate adipose and inflammation in systemic tissues. These animals also exhibited impaired fasting glucose levels, increased response to glucose tolerance test (GTT) along with decreased expressions of adiponectin, GAL-12, ATGL, and PPARγ. Impaired glucose metabolism can precede the development of excess fat deposition thereby leading to the systemic inflammation. These observations indicate the essential role of Gal-3 in adiposity, inflammation, and glucose metabolism (Pang et al., 2013). Further, Gal-3 has been suggested as an obvious therapeutic target for altering degeneration of the retina due to its role in causing hypoperfusion. Additionally, the H-Rasmediated signaling cascade has been shown to activate MMP-9 in RECs in humans and in animal models. This suggests that Ras-ERK pathway plays a vital role in MMP-9 activation in the retina. In fact, the sustained activation of this pathway can result in capillary cells apoptosis in retina. Thus, MMP-9 may present a plausible therapeutic target (Mohammad and Kowluru, 2012). The biology of visual processing has been linked to NO signaling pathway and changes in this pathway have been noticed during diabetes. Interestingly, upregulation of neuronal nitric

oxide synthase (nNOS) occurs via adrenomedullin (ADM) in this pathway. Diabetic individuals exhibit an upregulation of ADM in their eyes. Inhibiting ADM/NO signaling pathway in DR might downregulate NO in the retinal cultures treated with high glucose. Ruboxistaurin (PKC β inhibitor) downregulates ADM expression and activity, and electroretinography (ERG) showed that this change helped preserve the retinal functions suggesting that ADM/NO pathway could also serve as a potential therapeutic target for DR (Midena and Pilotto, 2017).

### DISCUSSION

Blindness constitutes a significant global health concern and has a powerful impact not only on afflicted patients but also on their families. It carries an enormous socio-economic burden on the society as well. Blindness is further augmented by an increasing incidence of diabetes cases and DR in particular because DM patients are living longer than ever before (Tabish, 2007; Zheng et al., 2012; Abuyassin and Laher, 2016; Leasher et al., 2016). The most threatening complications of DR are DME and PDR. To address this public health menace, effective strategies must be devised to control severe visual disability by implementing appropriate screening tools for early detection by devising standard operating procedures (SOPs) that should essentially include: (1) DR prevention, (2) reinforcing DR treatments, (3) using appropriate technologies to diagnose and treat early, and (4) increasing awareness about DR prevention methods (Romero-Aroca et al., 2010; Khandekar, 2012; Mi et al., 2014).

By presenting numerous pieces of evidence both at the cellular and molecular levels, we have tried to combine together an explanation that may enlighten future paths toward understanding and developing appropriate DR prognostics, diagnostics and therapeutics tool-kits that might help researchers and clinicians to make sound decisions regarding diagnosis, treatment, and future research. Taking cues from other chronic medical conditions such as cancer, Alzheimer, Parkinson or CVD, it has recently been discovered that inflammation is strongly associated with DR. Even the processes of aging and hormonal abnormalities are also being considered from an inflammatory perspective or at least related to inflammatory responses in our body (Joussen et al., 2004; Wellen and

Hotamisligil, 2005; Sjoholm and Nystrom, 2006; Sears and Ricordi, 2011). Early detection is the key and can undoubtedly help institute appropriate management practices for patients with available options that may still prevent, delay, or mitigate vision loss from DR. The primary mechanisms that may explain DR pathology include formation of AGEs, the polyol pathway (**Figure 5**), oxidative stress, epigenetic modifications, inflammation, and pyroptosis (Safi et al., 2014). However, the main underlying factors that are linked directly with diabetes remain hyperglycemia, oxidative stress, increased production and synthesis of growth factors, and the activation and secretion of a host of inflammatory molecules that in principle can deregulate retinal physiology. AGE formation can lead to an imbalance in growth factors and inflammation imbalances in the ocular compartment of diabetic patients (Obrosova and Kador, 2011). Further, upregulation of polyol via the hexosamine pathway is also associated with DR (Sharma et al., 2005; Kowluru and Chan, 2007; Safi et al., 2014; Singh et al., 2017).

Vision problems arising from DR become evident in later stages of the disease but molecular changes that are operate and regulate the disease process in general and in the retina in particular start quite early when patients are still at the prediabetic stage. Thus, efforts should be focused on seeking new and robust ways to detect DR at an earlier time point before a patient presents with an advanced disease. To do this, work should also be directed to harness the power of the latest 'omics' technologies including the refinement of currently available imaging tools for detecting subtle early types of changes in the retina and its elegant vascular architectural network. New evidence reflects that epigenomics affects gene expression in diabetes, confirming the metabolic memory concept, although much remains to be discovered in this evolving space. Interestingly, epigenomics can also aid in determining the interactive role, if any, played by genetic variants that stand to protect or promote diabetes susceptibility and disease progression in various ethnicities and races around the globe (Singh and Tyagi, 2017c). Also, it will be worthwhile to assess whether the combination of one's lifestyle such as exercise coupled with a healthy diet can help reduce complications by altering the epigenetics profile of the ocular cells. In that context, numerous investigations have demonstrated the benefits of using marker(s) associated with diabetes (Ronn and Ling, 2013). Current understanding espouses that epigenetic changes are reversible, and some therapies are under development to harness this feature with drugs that are being touted as "epi-drugs" and antagomirs (inhibitors of miRNA molecules) which might someday be used to complement existing treatment options (Baylin and Jones, 2011; Kato and Natarajan, 2014). As outlined earlier, inflammation appears to be a driver of complications due to diabetes. Therefore, variations in epigenetics could be analyzed in circulating inflammatory cells (e.g., monocytes) (Reddy et al., 2015).

Finally, the standard of clinical care for DR has evolved dramatically in the last 10 years. For decades, laser photocoagulation has been the first-line therapy for both PDR and DME (Evans et al., 2014). However, at present, intravitreal anti-VEGF injections are the standard for DME and their effectiveness has been demonstrated in clinical trials. They are also easy to administer and provide rapid improvement in vision. Anti-VEGF therapy also provides an alternative to panretinal photocoagulation for the treatment of PDR. DRCR.net protocol S trial revealed that intravitreal ranibizumab (Lucentis) was non-inferior to pan-retinal laser photocoagulation in treating PDR. All of this has given confidence to ophthalmologists to prefer anti-VEGF injections over the laser intervention for DME, which has essentially become a second-line therapy for this indication. However, most agree that anti-VEGF agents are far from perfect as they are short-acting, relatively costly, and carry the risks of endophthalmitis and stroke, which are not concerns with retinal laser photocoagulation. A clinical trial performed by Diabetic Retinopathy Clinical Research Network (DRCR.net) protocol T found that almost half of patients with DME required additional laser treatment after 6 months of anti-VEGF therapy.

### CONCLUSION

Globally, diabetes incidence is rising. Currently, therapeutic methods are unable to restore vision in advanced cases, and thus there is an unmet need for innovating newer approaches to tackle important cellular processes like oxidative stress, inhibition of signal transduction pathways, inflammation and pyroptosis (e.g., the protein kinase C cascade and AR pathways). Future treatments should be able to target different molecules that are vital to DR development (e.g., hepatocyte growth factors, MMP-9). Finally, novel mechanisms of therapeutic delivery should also occupy a front seat to treat and manage people suffering from this seemingly unstoppable health crisis (Dedania and Bakri, 2015). It appears that the mechanisms, events and pathways that we covered in this manuscript seem to act in conjunction in several ways. Since epigenetics is still a nascent discipline, we should be cautious not to draw hasty conclusions. Impact of the high prevalence of diabetes on people's health and the economic burden that it brings with it deserve a fresh look. We see such an intervention as being crucial to reduce the socio-economic burden of this incurable malady on global economy.

## AUTHOR CONTRIBUTIONS

All authors participated and contributed intellectually toward collecting literature pertaining to inflammation, pyroptosis and diabetes along with writing, reviewing and editing the manuscript before submission for its publication.

### FUNDING

This work was supported by the following grants: HL-74815, HL-107640, HL-139047, AR071789, and NS-084823.

### ACKNOWLEDGMENTS

Members in the laboratory are gratefully acknowledged for their continuous support and encouragement.

### REFERENCES

fphys-09-01268 September 4, 2018 Time: 9:44 # 14


macular edema before and after intraocular injection of triamcinolone. Arch. Ophthalmol. 122, 1801–1807. doi: 10.1001/archopht.122.12.1801






**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 Homme, Singh, Majumder, George, Nair, Sandhu, Tyagi, Lominadze and Tyagi. 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.

# Osteoarthritis-Like Changes in Bardet–Biedl Syndrome Mutant Ciliopathy Mice (Bbs1M390R/M390R): Evidence for a Role of Primary Cilia in Cartilage Homeostasis and Regulation of Inflammation

Isaac D. Sheffield<sup>1</sup> , Mercedes A. McGee<sup>1</sup> , Steven J. Glenn<sup>1</sup> , Da Young Baek<sup>1</sup> , Joshua M. Coleman<sup>1</sup> , Bradley K. Dorius<sup>1</sup> , Channing Williams<sup>1</sup> , Brandon J. Rose<sup>1</sup> , Anthony E. Sanchez<sup>1</sup> , Michael A. Goodman<sup>1</sup> , John M. Daines<sup>1</sup> , Dennis L. Eggett<sup>1</sup> , Val C. Sheffield<sup>2</sup> , Arminda Suli<sup>1</sup> and David L. Kooyman<sup>1</sup> \*

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

#### Reviewed by:

Rajeev Rohatgi, Northport VA Medical Center, United States David Meyre, McMaster University, Canada

#### \*Correspondence:

David L. Kooyman david\_kooyman@byu.edu

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 18 September 2017 Accepted: 22 May 2018 Published: 19 June 2018

#### Citation:

Sheffield ID, McGee MA, Glenn SJ, Baek DY, Coleman JM, Dorius BK, Williams C, Rose BJ, Sanchez AE, Goodman MA, Daines JM, Eggett DL, Sheffield VC, Suli A and Kooyman DL (2018) Osteoarthritis-Like Changes in Bardet–Biedl Syndrome Mutant Ciliopathy Mice (Bbs1M390R/M390R): Evidence for a Role of Primary Cilia in Cartilage Homeostasis and Regulation of Inflammation. Front. Physiol. 9:708. doi: 10.3389/fphys.2018.00708 <sup>1</sup> Department of Physiology and Developmental Biology, Brigham Young University, Provo, UT, United States, <sup>2</sup> Departments of Pediatrics and Ophthalmology, University of Iowa, Iowa City, IA, United States

Osteoarthritis (OA) is a debilitating inflammation related disease characterized by joint pain and effusion, loss of mobility, and deformity that may result in functional joint failure and significant impact on quality of life. Once thought of as a simple "wear and tear" disease, it is now widely recognized that OA has a considerable metabolic component and is related to chronic inflammation. Defects associated with primary cilia have been shown to be cause OA-like changes in Bardet–Biedl mice. We examined the role of dysfunctional primary cilia in OA in mice through the regulation of the previously identified degradative and pro-inflammatory molecular pathways common to OA. We observed an increase in the presence of pro-inflammatory markers TGFβ-1 and HTRA1 as well as cartilage destructive protease MMP-13 but a decrease in DDR-2. We observed a morphological difference in cartilage thickness in Bbs1M390R/M390R mice compared to wild type (WT). We did not observe any difference in OARSI or Mankin scores between WT and Bbs1M390R/M390R mice. Primary cilia appear to be involved in the upregulation of biomarkers, including pro-inflammatory markers common to OA.

Keywords: Bardet–Biedl syndrome, Bbs, osteoarthritis, primary cilia, inflammation

### INTRODUCTION

Osteoarthritis (OA) is a debilitating disease characterized by joint pain and effusion, loss of mobility, and deformity that may result in functional joint failure and significant impact on quality of life (1–3). It is also one of the most common chronic diseases in the United States, with recent estimates suggesting that more than 27 million adults suffer from clinical OA in this country. Vulnerable populations include the elderly, where 80% of people above 65 years of age are symptomatic for OA (Felson et al., 1988; Guccione, 1994; Lawrence et al., 2008), and the obese. As upward obesity trends continue it is probable that an increase in clinical OA will also be observed (1–3). There is currently no cure, and treatment options are minimal, with pain management and surgical joint replacement procedures being the only reprieve.

Once thought of as a passive "wear and tear" disease, it is now widely recognized that OA has a considerable metabolic component. It has been shown that whether the onset of the disease is due to aging, obesity, joint misalignment, acute injury, or genetic predisposition, all result in the activation of a common molecular pathway (Murwantoko et al., 2004; Lawrence et al., 2008; Polur et al., 2010; Holt et al., 2012; Larkin et al., 2013). Previous work has successfully identified key mediators in this pathway, including transforming growth factor beta 1 (TGF-β1), high temperature receptor A-1 (HTRA1), discoidin domain-containing receptor 2 (DDR-2), matrix metalloprotease 13 (MMP-13), and other matrix-degrading enzymes that are secreted by chondrocytes and contribute to the degradation of articular cartilage. Stress to the chondrocyte leads to an increase in TGF-β1-mediated chondroplasia but also has the potential to stimulate osteogenesis. Perhaps to offset this phenomena, HTRA1 is secreted to cleave TGF-β1. Being a non-specific serine protease, HTRA-1 also cleaves proteins of the pericellular matrix and brings the collagenous matrix in contact with the surface of chondrocytes, binding to DDR2 and subsequently increasing production of MMP-13, which specifically cleaves the type II collagen, among other proteins, making up the majority of articular cartilage.

Activation of MMP-13 gene expression appears to occur through a number of molecular pathways that work through either inflammation or primary cilia. That is not to say there are not some common themes. Stress-inducible nuclear protein 1 (NUPR1) has been shown to regulate MMP-13 expression in vitro (Yammani and Loeser, 2014). Yammani and Loeser (2014) showed that NUPR1, expressed in cartilage, is required for expression of MMP-13 via IL-1β. This might be a pathway for the catabolic effects of OA to be mediated through inflammation. This is especially interesting in light of the study done by Xu et al. (2015) in which they analyzed differential expression of genes in cartilage involved in OA and rheumatoid arthritis (RA). While these researchers identified multiple genes associated with the regulation of MMPs, the predominant ones were associated with inflammation. This might give greater credence for the role of early inflammatory signals (i.e., AGEs, IL-1) in the initiation and progression of OA. While more obvious, a similar role for inflammation appears to be present in RA. Araki et al. (2016) reported that histone methylation and the binding of signal transducer activator of transcription 3 (STAT3) was associated with RA and OA. They report that histone H3 methylation is associated with elevated expression of MMP-1, -3, -9, and -13. However, STAT3 was shown to increase expression, either spontaneous or IL-6 activated, of MMPs 1, 3, and 13 but not 9. As previously indicated, primary cilia appear to also be involved in OA. Sugita et al. (2015) reported that transcription factor hairy and enhancer of split-1 (Hes1) is involved in the upregulation of expression of MMP-13. Normally Hes1 acts as a transcriptional repressor but under the influence of calcium/calmodulin-dependent protein kinase 2 (CaMK2) it becomes a transcriptional activator, thus upregulating MMP-13 expression (Ju et al., 2004). Thus Hes1 acts to increase expression of MMP-13. It is of particular interest to note that HES1 acts through Notch signaling pathway (Kageyama et al., 2007). Notch has previously been shown to modulate sonic hedgehog signaling and work through primary cilia (Ezratty et al., 2011; Kong et al., 2015). In an apparent unrelated mechanism, Niebler et al. (2015) showed that the transcription factor AP-2e is intimately involved in the upregulation of MMP-13 as OA progresses.

Bardet–Biedl syndrome (BBS, MIM 209900) is a pleiotropic genetically heterogeneous disorder characterized by obesity, retinopathy, polydactyly, renal anomalies including polycystic kidney disease, intellectual disabilities and hypo-genitalism (Chiang et al., 2004; Kaushik et al., 2009; Marion et al., 2011). To date, 21 BBS genes have been identified (Katsanis et al., 2000; Slavotinek and Biesecker, 2000; Mykytyn et al., 2001; Nishimura et al., 2001, 2005; Ansley et al., 2003; Badano et al., 2003; Chiang et al., 2004, 2006; Li et al., 2004; Stoetzel et al., 2006a,b, 2007). Although the cellular functions of BBS proteins are not yet fully understood, evidence from a variety of organisms, including BBS mouse models, demonstrate that BBS proteins play a role in cilia assembly and/or function, as well as intracellular vesicle trafficking. BBS mouse models lack spermatozoa flagella (Fan et al., 2004; Badano et al., 2005). Knock down of multiple BBS genes in zebrafish have been shown to interfere with function of nodal cilia, as well as to result in delay of intracellular vesicle transport. C. elegans homologs of Bbs1, Bbs2, Bbs7 and Bbs8 have been shown to be expressed in ciliated cells and Bbs8 was found in the ciliary basal body but not in the microtubulebased ciliary axoneme (Ansley et al., 2003). BBS1, BBS4, BBS5, BBS7, and BBS8 have also been reported to localize to the centrosome and/or basal body of widely used mammalian cell lines (Badano et al., 2005). The depletion of BBS4 through RNA interference has been shown to lead to the disruption of cytoplasmic microtubule arrays, the mis-localization of some pericentriolar proteins including pericentriolar material 1 protein (PCM1), and cell division arrest and apoptosis in cultured cells (Kim et al., 2004). Recent studies indicate that the seven most conserved BBS proteins (BBS1, BBS2, BBS4, BBS5, BBS7, BBS8, and BBS9), along with BBS18 form a stable complex known as the BBSome. The BBSome associates with centriolar satellites in the cytoplasm and transitions to a membrane-associated form at the base of the cilium. Depletion of some components of the BBSome affects ciliogenesis by interfering with membrane trafficking to the primary cilium (Nachury et al., 2007). Studies involving renal cells have shown a predisposition of primary cilia to act as surface receptors (Marion et al., 2011) and function as both mechanoand chemoreceptors for specific environments (Poole et al., 1985, 1997, 2001; Kouri et al., 1996, 1998; Jensen et al., 2004; McGlashan et al., 2006; Kaushik et al., 2009). For a recent reviews of BBS genes and associated diseases see the following papers (Heon et al., 2016; Kaur et al., 2017; Weihbrecht et al., 2017). We chose to work with mice with the Bbs1M390R/M390R since it accounts for 25% of all human BBS cases and is directly involved with the BBSome.

Previous work has visualized the presence of primary cilia on chondrocytes, with each chondrocyte containing a single primary cilium (Scherft and Daems, 1967; Hart, 1968; Wilsman, 1978; Meier-Vismara et al., 1979; Wilsman et al., 1980; Kaushik et al., 2009). One study demonstrated cartilage abnormalities associated

with defective primary cilia (Kaushik et al., 2009). Studies employing Oak Ridge polycystic disease mouse chondrocytes show that the primary cilia in chondrocytes are not involved in the initial mechanosensation, but are associated with signal processing in response to increased loads (Mykytyn and Sheffield, 2004). We hypothesized that dysfunctional primary cilia result in OA in Bbs1M390R/M390R mice as they age through intra-cellular pathways resulting in up-regulation of the previously identified HTRA1-DDR2-MMP13 degradative pathway common to OA.

### MATERIALS AND METHODS

### Tissue Samples

Knees from wild type (WT) n = 44 (n = 8 10–14 weeks old; n = 24 15–18 weeks old; n = 12 21+ weeks old) and Bbs1M390R/M390R n = 61 (n = 18 10–14 weeks old; n = 39 15–18 weeks old; n = 4 21+ weeks old) mutant mice were obtained from the laboratory of Val Sheffield at the University of Iowa. Knees were harvested and fixed in 4% paraformaldehyde prior to decalcification and embedding according to a previously reported procedure (Spencer et al., 2015). Knees were serially sectioned and sections analyzed histologically after staining with Safranin O and fast green as previously reported (Spencer et al., 2015). Using a light microscope equipped with a digital camera, photographs of each knee joint were taken at 100× and 200× magnifications. The articular cartilage in two representative sections from each stained slide were analyzed using the OARSI scoring system (Glasson et al., 2010) to quantify the pathological state of each joint, with a score of zero representing unaltered articular cartilage and six representing severe OA.

### Immunohistochemical Analysis

Immunohistochemistry (IHC) was performed on slides representing serial sections of mouse knee joints from all animals. Separate slides were stained with antibodies against HTRA1, DDR-2, MMP-13, and TGF-β1. Each slide was deparaffinized and then blocked with 5% bovine serum albumin for 1 h. Primary antibodies against HTRA-1 (ab38611) (Abcam, Cambridge, MA, United States), DDR-2 (SC-8989) (Santa Cruz Biotechnology, Santa Cruz, CA, United States), MMP-13 (AB8120) (Chemicon, Temecula, CA, United States) and TGF-β1 (Abcam- ab92486, Rabbit Polyclonal IgG) were used. All antibodies were diluted 1:200, applied to specimens, and incubated overnight at 4◦C. On the second day, samples were rinsed with PBS and then incubated with an avidin/biotin ABC mix (Vectastain elite ABC Kit). Slides were rinsed again with PBS and incubated with a species appropriate biotinylated secondary antibody. After a third rinse, a color reaction was initiated using a peroxidase substrate (Vector Labs, NovaRED). Negative controls were prepared by staining without the addition of primary antibody. Differences in staining intensity were compared qualitatively with WT controls. Blind counting of stained cells was performed using ImageJ (NIH, Bethesda, MD, United States). IHC and histological analysis focused on different parts of the joint to demonstrate the universal nature of the OA in the Bbs1M390R/M390R model. Animals were grouped into three general age categories (10–14 weeks, 15–18 weeks, and 21+ weeks) for comparison purposes. The age categories selected indicate young, middle age, and old mice. For the purposes of IHC and histological analysis, i.e., cell counting (see section "ImageJ Analysis" below), the number of mice used in each category was 10–14 weeks: WT n = 8, Bbs1M390R/M390R mutant n = 18, 15–18 weeks: WT n = 24, Bbs1M390R/M390R mutant n = 39, 21+ weeks: WT n = 12, Bbs1M390R/M390R mutant n = 4.

### Zebrafish Immunohistochemistry

Zebrafish IHC techniques were employed as previously described (Suli et al., 2014). Antibodies used for visualization: Primary cilia: anti-mouse acetylated tubulin (1:500, Sigma, T7451) with anti-mouse IgG Alexa488 (1:400, Invitrogen, A-28175); Primary DDR2: anti-mouse DDR2 (1:200, Santa Cruz SC-8989) with anti-rabbit IgG Alexa405 (1:400, Invitrogen, A31556).

Larvae were fixed with 4% paraformaldehyde either for 2 h at room temperature, washed three times, 20 min each, with PBST (0.1% Tween in PBS) and incubated 1 h in distilled water. They were placed in block solution [1% bovine serum albumin, 1% dimethyl sulfoxide (DMSO), and 0.02% sodium azide in PBST, 10% normal goat serum] for 1 h and then incubated with primary antibody overnight at 4◦C. After four 20 min washes with PBST, they were incubated with secondary antibody for 3 h at room temperature, washed four times, 10 min each, in PBST and cleared in 50% glycerol/PBS. Embryos were imaged using an Olympus FV1000 confocal microscope.

### Osteoarthritis Scoring

The articular cartilage in two representative sections from each stained slide was analyzed using the OARSI scoring system to quantify the pathological state of each joint, with a score of zero representing unaltered articular cartilage and six representing severe OA. Overall OARSI scoring was based on an osteoarthritic damage 0–6 subjective scoring system applied to all four quadrants of the knee (Glasson et al., 2010). The articular cartilage in two representative sections from each stained slide was also analyzed using a modified Mankin score to quantify the pathological state of each joint, with a score of zero representing unaltered articular cartilage and 14 representing severe OA. Overall Mankin scoring was based on a subset of scores including cartilage erosion score (0–6), chondrocyte periphery staining (0– 2), spatial arrangement of chondrocytes (0–3), and background staining intensity (0–3) (Mankin et al., 1971; Xu et al., 2003; Bomsta et al., 2006). Statistical significance of the combined Mankin scores for the 28-day Bbs1M390R/M390R mutant and wildtype surgery groups using a two-way ANOVA test conducted by the Department of Statistics at Brigham Young University. As in the case of IHC, animals were grouped into three general age categories (10–14 weeks, 15–18 weeks, and 21+ weeks) for comparison purposes. The age categories selected indicate young, middle age, and old mice. For the purposes of OARSI scoring analysis the number of mice used in each category was 10–14 weeks: WT n = 8, Bbs1M390R/M390R mutant n = 18, 15–18 weeks: WT n = 24, Bbs1M390R/M390R mutant n = 39, 21+ weeks: WT n = 12, Bbs1M390R/M390R mutant n = 4.

### ImageJ Analysis

fphys-09-00708 June 15, 2018 Time: 13:44 # 4

The expression levels of HTRA1, DDR-2, MMP-13, and TGFβ1 were analyzed quantitatively by calculating the percentage of cells staining positive for the respective biomarkers and the total number of chondrocytes in a defined 200×900 pixel area of articular cartilage immediately distal to the tibial plateau (Larkin et al., 2013). All quantitative analysis was performed using ImageJ (National Institutes of Health, Bethesda, MD, United States). Cell counting was done by two independent investigators who were blinded to the strain of mouse (Bbs1M390R/M390R mutant vs. WT). The quantitative results were subsequently analyzed statistically using an ANOVA test to detect differences in the mean percentages of positive staining for key OA biomarkers and mean chondrocyte counts between the Bbs1M390R/M390R mutant and WT samples. The numbers of animals analyzed for each age group were previously indicated in the IHC methods section.

### Adobe Photoshop Analysis

To compare the thickness of cartilage between Bbs1M390R/M390R mutant vs. WT, cartilage measurements above the subchondral bone for each mouse specimen was captured using the Lasso Tool in Adobe Photoshop. This was done to measure the area between the apical surface of the cartilage and where the articular cartilage interfaces with subchondral bone. The technique was perfected in our laboratory and previously reported (Black et al., 2015). To help standardize the human measurement error, two different lab members independently measured the tissue samples.

### Statistics

Statistical analysis was performed using a mixed models analysis of variance (ANOVA) on the OARSI or Mankin score as well as. The dependent variables were the OARSI, Mankin scores or biomarker staining for the knee. The independent variables were age and genotype along with their interaction. Blocking was done on each animal to account for their multiple measures. Post hoc t-tests were performed to determine differences in genotype biomarker expression level at each age, p-values of <0.05 were considered significant. The percent of chondrocyte biomarker expression level staining was compared using t-tests. A Bonferroni adjust was made for these four tests. Thus p-values of <0.0125 were considered significant.

### RESULTS

### Co-expression of DDR-2 and Primary Cilia

We observed primary cilia (acetylated α-tubulin) in zebrafish chondrocytes residing in the mandible (**Figure 1A**) indicated in green staining. DDR-2, indicated by purple staining (**Figure 1B**), did not co-localize with primary cilia.

### Cartilage Morphology

In a histological comparison of 18-week-old WT and Bbs1M390R/M390R mutant knee joints stained with Safranin-O, Fast Green we observed thicker cartilage (p < 0.01) in

FIGURE 1 | Normal primary cilia in zebrafish chondrocytes residing in the mandible, as indicated in green staining for acetylated α-tubulin (A). Ddr-2 is indicated by purple staining (B), demonstrating a lack of co-localization between primary cilia and Ddr-2.

Bbs1M390R/M390R mutant knees with noted surface erosion in some cases. See **Figure 2** for representative pictures.

### OARSI and Mankin Scores

Although Bbs1M390R/M390R mutants tended to have higher OARSI scores than WT mice (**Figure 3**), the mixed models analysis showed no statistically significant difference was observed in OARSI or Mankin (data not shown) scoring between the two groups studied. The alteration from normal biomarker concentrations in Bbs1M390R/M390R mutant mice, suggests that the metabolic changes occur prior to cartilage degradation.

### MmMP-13, HTRA1, and TGF-β1 Expression in Bbs1M390R/M390R Mutant Mice

T-tests of immunohistochemical staining revealed significant differences in the percent of chondrocytes staining positive for MMP-13 (p < 0.001) (**Figure 4A** vs. **Figure 4E**), HTRA1 (p < 0.001) (**Figure 4B** vs. **Figure 4F**), and TGF-β1 (p < 0.001) (**Figure 4C** vs. **Figure 4G**), but not DDR-2 (p > 0.05) (**Figure 4D** vs. **Figure 4H**) was not significant with the Bonferroni adjustment, in Bbs1M390R/M390R mutant mice when compared to age matched WT mice. These findings remained consistent across all ages examined.

When comparing all age groups, WT mice exhibited significantly less staining for HTRA1 and MMP-13 compared to Bbs1M390R/M390R mutants. WT mice exhibited significantly more staining for TGF-β1 compared to Bbs1M390R/M390R mutants, indicative of earlier stages of OA. This is consistent with previous observations of an inverse relationship between HRTA1 and TGF-β1 (Larkin et al., 2013). We did not observe a decrease in DDR2 expression in Bbs1M390R/M390R mutant mice compared to WT controls (**Figure 4D** vs. **Figure 4H**), contrary to general

FIGURE 2 | Histological comparison of WT and Bbs1M390R/M390R mutant knee joints. (A) Knee joint stained with Safranin-O Fast Green from a 17-week-old male WT mouse. (B) Knee joint stained with Safranin-O, Fast Green from an 18-week-old male Bbs1M390R/M390R mutant mouse. Note the surface erosion and exaggerated thickness of cartilage. Bbs1M390R/M390R mutant knees exhibited significantly thicker cartilage compared to WT controls (p < 0.01). However, Bbs1M390R/M390R mutant mice do not exhibit higher than normal OARSI or Mankin scores.

observations of upregulation in OA cartilage compared to controls. Notwithstanding the verified catabolic role of DDR2, no protection was afforded Bbs1M390R/M390R mutant cartilage by its decreased expression. Visualization of the primary cilia (acetylated α-tubulin) and DDR2 in embryonic mandibular zebrafish chondrocytes (**Figure 1A**) indicated that DDR2 did not co-localize with primary cilia. The data are summarized in **Table 1**.

### DISCUSSION

It is imperative to fully understand the metabolic pathway involved in the age-related onset and progression of OA if an effective treatment is to be developed. Common to the pathogenesis of human OA is the activation of the HTRA1- DDR2-MMP13 axis. HTRA1 is strongly expressed in the presence of stressors in murine OA models, and its activation augments the expression of DDR-2 (Polur et al., 2010). Upregulation of DDR2 results in excess binding of the receptor to its ligand, type II collagen (Leitinger et al., 2004), which in turn stimulates high levels of expression of the MMP-13 (Su et al., 2009). This ultimately leads to the destruction of articular cartilage. Aside from the wide array of evidence showing HTRA1-DDR2- MMP13 activity in OA of multiple modalities, the convergence of diverse noxious stimuli upon this axis is further supported in the protective effects exerted in DDR-2 hypomorphic strains of mice. When subjected to the destabilization of the medial meniscus (DMM) surgical procedure they demonstrate a significant decrease in the progression of OA compared to WT littermates

FIGURE 4 | Representative images showing the results of the immunohistochemical and histological staining performed to analyze the presence of OA biomarkers. Although, Bbs1M390R/M390R mutant and WT mouse knees did not exhibit a statistical difference in cartilage erosion scoring, we analyzed separately cell staining within the three major age (week) categories examined – 10–14 (WT n = 8; Bbs1M390R/M390R mutant n = 18), 15–18 (WT n = 24; Bbs1M390R/M390R mutant n = 39), 21+ (WT n = 12; Bbs1M390R/M390R mutant n = 4). All tissue samples represented in this figure were age matched at 14 weeks. The patterns of cell staining for biomarkers examined did not differ significantly with age. (A) WT MMP-13; (B) WT HTRA-1; (C) WT TGF-β1; (D) WT DDR-2; (E) Bbs1M390R/M390R mutant MMP13; (F) Bbs1M390R/M390R mutant HTRA-1; (G) Bbs1M390R/M390R mutant TGF-β1; (H) Bbs1M390R/M390R mutant DDR-2.

TABLE 1 | Percent Bbs1M390R/M390R mutant cells stained with biomarkers compared with WT.


<sup>∗</sup>P < 0.001; ∗∗P < 0.05.

(Xu et al., 2010). We aimed to demonstrate the significance of primary cilia as it pertains to the onset and progression of OA. It is noteworthy that ciliopathy related genes do not just function in cilia. For example, the BBSome plays a role in leptin receptor trafficking and insulin receptor trafficking to the plasma membrane in neurons in the hypothalamus. Thus, BBS genes may play a role in OA pathways even if not directly involving the primary cilia organelle. Curiously, WT mice consistently exhibited significantly more staining for DDR-2 than Bbs1M390R/M390R mutants at all age observed age groups, inconsistent with previous observations correlating high levels of DDR-2 expression with more advanced OA. This may suggest a role for primary cilia in processing signals associated with DDR-2 activation and/or expression of the receptor.

In order to identify primary cilia as a component in the chondrocyte destruction pathway associated with OA, we used Bbs1M390R/M390R mutant mice, which are known to have defective primary cilia (**Figure 2B**) (Mykytyn et al., 2004) and have been previously shown to exhibit OA-like phenotype (Kaushik et al., 2009). We compared Bbs1M390R/M390R mutants to WT at different ages and in each instance found significant differences in the concentrations of key biomarkers previously shown to be associated with OA. Among these differences were increased concentrations of chondrocytes expressing HTRA-1 and a decreased concentration of chondrocytes expressing TGF-b1 and increased levels of MMP-13. All three observations are consistent with previous studies on biomarkers associated with OA (Murwantoko et al., 2004; Polur et al., 2010; Holt et al., 2012; Larkin et al., 2013). The increased presence of HTRA1 and MMP13 may help explain why the Bbs1M390R/M390R mutants respond more vigorously to DMM surgery (Peterson et al., 2011).

TGFβ-1 functions largely in the SMAD signaling pathway; it is a cytokine involved in a number of cellular processes including cell growth, proliferation, differentiation, apoptosis, and the elaboration of extracellular matrix (van Beuningen, 2000). HTRA-1 has a degrading effect on the peri- and extracellular matrix, as well as an inhibitory effect on TGFβ1 (Oka et al., 2004; Tocharus et al., 2004). Thus HTRA-1 may contribute to OA not only through direct matrix destruction, but also by inhibiting possible protection to the matrix offered by TGF-β1. Expression of MMP-13, a protein with a degrading effect on extracellular matrix similar to HTRA-1, appears to be upregulated. This is consistent with alterations in MMP-13 expression as in OA cases resulting from other causes, such as aging, obesity, joint misalignment, and acute injury.

Our results indicate that age-related OA-associated expression of DDR-2 is altered in Bbs1M390R/M390R mice. Curiously, DDR-2 expression appeared to be lower in Bbs1M390R/M390R mutant mice than in WT mice at all ages. Whereas in cases of OA resulting from the aforementioned etiologies, DDR-2 expression is upregulated in mice presenting with OA. We believe a possible explanation for this could be that primary cilia might have a role in processing signals associated with DDR-2 activation and/or expression of the receptor. However, we have demonstrated that DDR2 does not co-localize to primary cilia (**Figure 2A**). Therefore, any interaction(s) between primary cilia and DDR2 would presumably be indirect.

We did not observe a statistical difference between Bbs1M390R/M390R mutant and WT mice when using knee histological scoring methods (OARSI and Modified Mankin). We did observe a difference in the thickness of cartilage in Bbs1M390R/M390R mutant mice compared to WT as others have previously noted (Kaushik et al., 2009).

Finally, much has been published regarding the potential role of inflammation in OA. More recently, the interaction of adipokines and inflammation in initiating OA has been demonstrated (Azamar-Llamas et al., 2017). Leptin, a peptide hormone involved in maintaining insulin sensitivity and contributing to the sensation of satiety, is expressed at very high levels in obese individuals. It appears to be correlated with OA as well, with intervention at the level of MMP-13 expression occurring. Downregulation of leptin mRNA translation via small interference RNA molecules inhibits MMP-13 expression in cultured osteoarthritic chondrocytes (Iliopoulos et al., 2007; Pallu et al., 2010). The situation appears to be most exacerbated in cases of extreme obesity where a strong positive correlation exists between the responsiveness of the MMP-13 gene to leptin and the BMI of osteoarthritic individuals. This is intriguing in light of the observation that primary cilia are intimately involved in leptin homeostasis (Seo et al., 2009; Guo et al., 2016).

### CONCLUSION

Our findings support our initial hypothesis that BBS mutations are involved in the identified molecular pathway common to OA, with the exception of DDR-2. Since the BBSome plays a role in leptin receptor trafficking and insulin receptor trafficking to the plasma membrane in neurons in the hypothalamus, BBS genes may play a role in OA pathways even if not directly involving the primary cilia organelle. We have shown that MMP-13, HTRA-1, and TGF-β1 expression mimic their expression during all other investigated cases of OA. This suggests that primary cilia act as a surface receptor on chondrocytes and work to relay the condition of the cartilaginous matrix and its integrity to the associated chondrocyte. Primary cilia may also be an important link between inflammatory signals and OA. Future research should work to identify the specific role primary cilia play in the onset and progression of OA.

### STRENGTHS AND LIMITATIONS

This paper adds comprehensive and important data to a paucity of information regarding the involvement of primary cilia in the pathogenesis of OA. Some significant and important findings in this work are that primary cilia are a link between proinflammatory bio-markers and OA and that the BBSome may play a role in this disease as well. While BBS is a pleiotropic disease with 21 known genes, the work associated with this paper employed the BBS mutant (Bbs1M390R/M390R) which is associated with 25% of all human cases. Some weaknesses of this work include the lack of elucidating the exact molecular pathway(s) or link(s) between primary cilia and OA as well as no data from human BBS patients.

### INSTITUTIONAL SAFETY

All U.S. National Institutes of Health guidelines for research involving Recombinant or Synthetic Nucleic Acid Molecules were followed. All chemical use was according to the U.S. Occupational Safety and Health Administration guidelines as overseen by the Brigham Young University Office of Risk Management.

### ETHICS STATEMENT

This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved Institutional Animal Care and Use Committee (IACUC) protocols 5061426 and 6081818, of the University of Iowa and the Brigham Young University Institutional Animal Care and Use Committee protocol 13-0601. Animals were housed according to IACUC recommendations. Methods of euthanasia used were carbon dioxide inhalation followed by cervical dislocation, or anesthesia induced by ketamine/xylazine followed by transcardial perfusion. Humane endpoints were strictly observed, and every effort was made to minimize suffering.

## AUTHOR CONTRIBUTIONS

IS provided direct oversight for the project. MM, SG, DB, JC, BD, CW, BR, AS, MG, and JD performed all of the histological work including analyses associated therewith. DE was responsible for all statistical work assisted by MG. DE also assisted with experimental design. VS provided the BBS mice and assisted with experimental design. AS provided the zebrafish and oversaw all of that work, assisted by SG and JD. DK conceived the project and was the overall coordinator for it.

## FUNDING

This work was supported, in part, by Brigham Young University Office of Research and Creative Activities Undergraduate and

Mentoring Environment Grants. This work was supported in part by US National Institutes of Health grants R01 EY-011298 and RO1 EY-017168 (to VS) and the Roy J. Carver Charitable Trust.

### REFERENCES


Hart, J. A. L. (1968). Cilia in articular cartilage. J. Anat. 103:222.


### ACKNOWLEDGMENTS

The authors wish to thank Matthew Sterling for help in preparing this manuscript.

sedc mouse associated with the HtrA1-Ddr2-Mmp-13 degradative pathway: a new model of osteoarthritis. Osteoarthritis Cartilage 20, 430–439. doi: 10.1016/ j.joca.2011.11.008




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

Copyright © 2018 Sheffield, McGee, Glenn, Baek, Coleman, Dorius, Williams, Rose, Sanchez, Goodman, Daines, Eggett, Sheffield, Suli and Kooyman. 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 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.

# Participation of Arachidonic Acid Metabolism in the Aortic Aneurysm Formation in Patients with Marfan Syndrome

#### María E. Soto<sup>1</sup> , Verónica Guarner-Lans <sup>2</sup> , Karla Y. Herrera-Morales <sup>3</sup> and Israel Pérez-Torres <sup>4</sup> \*

<sup>1</sup> Department of Immunology, Instituto Nacional de Cardiología "Ignacio Chávez", Mexico City, Mexico, <sup>2</sup> Department of Physiology, Instituto Nacional de Cardiología "Ignacio Chávez", Mexico City, Mexico, <sup>3</sup> Cardiothoracic Surgery, Instituto Nacional de Cardiología "Ignacio Chávez", Mexico City, Mexico, <sup>4</sup> Department of Pathology, Instituto Nacional de Cardiología "Ignacio Chávez", Mexico City, Mexico

#### Edited by:

John D. Imig, Medical College of Wisconsin, United States

#### Reviewed by:

Marius Catalin Staiculescu, Washington University in St. Louis, United States Bysani Chandrasekar, University of Missouri, United States Luis A. Martinez-Lemus, University of Missouri, United States

> \*Correspondence: Israel Pérez-Torres pertorisr@yahoo.com.mx

#### Specialty section:

This article was submitted to Vascular Physiology, a section of the journal Frontiers in Physiology

Received: 14 August 2017 Accepted: 22 January 2018 Published: 12 February 2018

#### Citation:

Soto ME, Guarner-Lans V, Herrera-Morales KY and Pérez-Torres I (2018) Participation of Arachidonic Acid Metabolism in the Aortic Aneurysm Formation in Patients with Marfan Syndrome. Front. Physiol. 9:77. doi: 10.3389/fphys.2018.00077 Marfan syndrome (MFS) is a pleiotropic genetic disease involving the cardiovascular system where a fibrillin-1 mutation is present. This mutation is associated with accelerated activation of transforming growth factor β (TGFβ1) which contributes to the formation of aneurysms in the root of the aorta. There is an imbalance in the synthesis of thromboxane A<sup>2</sup> (TXA2) and prostacyclin, that is a consequence of a differential protein expression of the isoforms of cyclooxygenases (COXs), suggesting an alteration of arachidonic acid (AA) metabolism. The aim of this study was to analyze the participation of AA metabolism associated with inflammatory factors in the dilation and dissection of the aortic aneurysm in patients with MFS. A decrease in AA (p = 0.02), an increase in oleic acid (OA), TGFβ1, tumor necrosis factor alpha (TNFα), prostaglandin E<sup>2</sup> (PGE2) (p < 0.05), and COXs activity (p = 0.002) was found. The expressions of phospholipase A<sup>2</sup> (PLA2), cytochrome P450 (CYP450 4A), 5-lipoxygenase (5-LOX), COX2 and TXA2R (p < 0.05) showed a significant increase in the aortic aneurysm of patients with MFS compared to control subjects. COX1, 6-keto-prostaglandin 1 alpha (6-keto-PG1α) and 8-isoprostane did not show significant changes. Histological examination of the aortas showed an increase of cystic necrosis, elastic fibers and collagen in MFS. The results suggest that there are inflammatory factors coupled to genetic factors that predispose to aortic endothelial dysfunction in the aortic tissue of patients with MFS. There is a decrease in the percentage of AA, associated with an increase of PLA2, COX2/TXA2R, CYP450 4A, and 5-LOX which leads to a greater synthesis of PGE<sup>2</sup> than of 6-keto-PGF1α, thus contributing to the formation of the aortic aneurysm. The evident loss of the homeostasis in these mechanisms confirms that there is a participation of the AA pathway in the aneurysm progression in MFS.

Keywords: aortic aneurysm, marfan syndrome, arachidonic acid, inflammation, cyclooxygenase

## INTRODUCTION

The Marfan syndrome (MFS) is a pleiotropic genetic disease with involvement of the cardiovascular, ocular and skeletal system with a very wide clinical variability. It is a connective tissue disorder with autosomal dominant associations in which the gene encoding for the fibrillin-1 (FBN-1) protein is affected (Wheeler et al., 2014).

FBN-1 is an essential protein in the formation and maturation of elastic fibers in the arteries and it is responsible for the assembly of the networks of microfibrils. The microfibrils store growth factors which are released at specific times to control growth and reparation of the tissues and organs of the body. Microfibrils and elastin form part of the extracellular matrix of tissues. FBN-1 is considered as a constitutive protein of connective tissue (Milewicz et al., 2008). The alteration in the expression of this protein can lead to destruction of the assembly of normal microfibrils and the production of abnormal elastic fibers and, as a consequence it leads to changes in elasticity in the aortic tissue causing growth and instability (Granata et al., 2016). This damage results in structural variations within the same arterial vessel with inherent heterogeneity in its content, thickness and cellular composition (Soto et al., 2014). In addition, defects or deficiencies in FBN-1 are associated with the regulation, bioavailability and accelerated activation of transforming growth factor β1 (TGFβ1) in the extracellular matrix. TGFβ1 contributes to the formation of aortic root aneurysms (Neptune et al., 2003). Also, the blockage of TGFβ1 release through anti-TGFβ1 antibodies reduces aortic root dilation in studies in mice (Cohn et al., 2007).

In addition, several non-genetic diseases are associated with aortic damage, and the etiology can therefore be multifactorial (Hendel et al., 2015). Oxidative stress(OS),the degree of inflammation and edema may contribute to the progression of aortic tissue damage associated with endothelial dysfunction (Soto et al., 2016a). In a preliminary study of our group, we evaluated the involvement of OS in different aortopathies in humans and correlated it with sub-endothelial basement membrane proteins and found the existence of dysfunction and progression of aortic damage (Soto et al., 2014). In another study, we showed that OS in MFS is associated with alterations in enzymes that employ glutathione, leading to increased chronic inflammation (Zúñiga-Muñoz et al., 2017).Furthermore, the vasomotor function in Marfan thoracic aorta is associated with an imbalance in the synthesis of thromboxane A<sup>2</sup> (TXA2) and prostacyclin derived from the differential protein expression of cyclooxygenase (COXs) isoforms (Chung et al., 2007b). Levels of prostaglandin E<sup>2</sup> (PGE2), TXA2, and IL-6 were significant higher in aortic aneurysms and were associate with overexpression in the aneurysmal wall of the inflammatory cyclooxygenase 2 (COX2) (Cheuk and Cheng, 2007).

On another hand, aortic damage and endothelial dysfunction with chronic inflammatory involvement have been associated to changes in arachidonic acid (AA) metabolism in several animal models and in cell lines both in vivo and in vitro (Aggarwal et al., 2012). Various physical and hormonal stimuli may stimulate the entrance of Ca2<sup>+</sup> ions to the endothelial cell, increasing its intracellular concentration and activating the ion-dependent phospholipase A<sup>2</sup> (PLA2) that releases the AA from the phospholipids of the cell membrane. Phospholipids can then be metabolized by the COXs isoforms, thromboxane synthase (TXs), lipoxygenase (LOXs) and cytochrome (CYP450) (Jamieson et al., 2017). The end products of these pathways are eicosanoids that can act in an autocrine and paracrine manner modulating vascular tone, platelet activation, signaling, proliferation, cell migration, fever and inflammatory processes. (Gauthier et al., 2011) Furthermore, AA can be non-enzymatically oxidized by free radicals, resulting in 8-isoprostanes which are oxidizing agents associated with OS and inflammation (Yang et al., 2010). Therefore, the evaluation of the participation of the AA metabolism is important since it may be involved in the development of aortic damage through activation of chronic inflammatory processes that lead to dilatation and dissection of the aortic aneurysm in MFS. The aim of this study was to analyze the participation of AA metabolism associated with inflammatory factors in the dilation and dissection of the aortic aneurysm in patients with MFS.

### MATERIALS AND METHODS

## Study Design: The Study Was Done on an Observational Cohort, and It Is Descriptive and Prospective

### Population in Study

Patients with a Bentall and/or De Bono procedure (Bentall and De Bono, 1968) with aortic arch replacement, stent placement in the thoracic aorta, mitral valve replacement, mitral valvular change, coronary revascularization, Tirone David procedure and the procedure for the replacement of the thoracic-abdominal aorta were included. The following inclusion criteria were taken into account: aortic tissue was obtained from patient with a MFS diagnosis, based on the Ghent criteria. They had aortic dilatation and their cases had been presented in the surgical medical session to standardize the requirements of the type of aortic surgery to be considered according to the pathology of each patient. They were treated at the National Institute of Cardiology "Ignacio Chavez." Exclusion criteria taken into account were a doubtful diagnosis and/or the lack of agreement to sign the informed consent form for the research study. Elimination Criteria were the insufficient tissue sample taken at the moment of surgery even when the patients met the inclusion criteria or the inappropriate conditions for the sample obtainment when considering the requirements for the research process.

### Ethical Considerations

The study was carried out according to the international ethical standards and the General Health Law, as well as according to the Helsinki declaration, modified at the Congress of Tokyo, Japan.

In all of the MFS patients and control subjects, studies of echocardiography, computerized tomography or magnetic resonance were performed. In the MFS patients, these studies were performed in order to determine the aortic lesion and valvular disease extension. In the control group, they were used to discard aortic damage additional to valvular damage. Therefore, only subjects who had isolated aortic stenosis without sclerosis and had no disease in the aorta were selected. Also, all patients selected had tricuspid aortic valve, patients with bicuspid valve were excluded. In the patients with MFS, dilatation of the aortic diameter (>50 mm), which is a criterion for the indication of surgery, was present and the risks and benefits of a surgical intervention were assessed in a clinical and pathological meeting with cardiovascular specialists. When patients were accepted for surgery, the preoperative protocol included coagulation tests, X-rays, electrocardiogram, anesthesia evaluation, and individualized medical intervention. Cases were dealt with caution, to avoid including patients undertaking treatment with antioxidants, allopurinol, or potential inhibitors of pathways involved in ROS production. Aspirin, warfarin, clopidogrel, and other antiplatelet or anticoagulant medications were suspended. In control subjects, routine laboratory tests were performed to determine acute phase reactants, triglycerides, and HDL cholesterol. None of the control subjects were taking antiinflammatory drugs or statins. Control tissue was selected from patients that had an indication for surgery and in which aortic tissue could be obtained during the procedure that they required. The control subjects had trivalva aorta and underwent surgery for aortic stenosis. The surgery performed implied substitution of aortic valves, and the need to perform plastia or resection of aortic tissue surrounding the valvular area. During surgery, sample tissues of root thoracic aorta were obtained. Once the surgery was performed, the tissue was placed in liquid nitrogen and was kept at −30◦C until used. The research protocol was approved by the Research and Ethics Committee of our institution (Institutional protocol number: 09654).

### Thoracic Aorta Homogenization

A sample from thoracic aorta was taken and placed in liquid N<sup>2</sup> for homogenization. The homogenization process was previously described by Soto et al. (2014). Protein concentration in the thoracic aortic homogenate was determined by the method of Bradford (1976).

### Extraction and Derivatization of the Arachidonic and Oleic Acid

For AA and oleic acid (OA) extraction, 100µg of protein from the aortic homogenates were used in the presence of 100µg of nonadecanoic acid as internal standard and 2 ml of chloroformmethanol (2:1, vol/vol) with 0.002% BHT, as described by Folch (Folch et al., 1957) and previously report by López (López et al., 2016).

### Immunoblotting

One hundred microgram of protein from eachof the aortic homogenate were separated by SDS-PAGE (8% polyacrylamide gel) and transferred to a nitrocellulose Hybond-P membrane 45µm (Millipore), according to methods previously described by Soto (Soto et al., 2014). The membranes were incubated overnight at 4◦C with rabbit primary polyclonal anti-bodies against COX1(H-62, sc-7950), COX2 (H-62, sc-7951),PLA2(N-216, sc-438), TXA2R (H-120, sc-30036), 5-LOX (H-120, sc-20785) and mouse primary monoclonal antibody against CYP450 4A1/A2/A3 (clo4, sc-53247) (Santa Cruz Biotechnology, Santa Cruz, CA, USA).

### Cyclooxygenases Activity Assays

COXs activity was performed by monitoring the rate of O<sup>2</sup> uptake using an oximeter (YSI oximeter model 5300A-1) which was coupled to an O2Clark type electrode. To 3 ml of a buffer of 0.1 M tris-HCl buffer, 1 mM phenol, 85µg bovine hemoglobin, pH 8 and 100µM AA at 37◦C, 100µg of aortic homogenate were added to initiate the reaction. Inhibition and discrimination of the catalytic activity of COXs was performed by the addition of 10µM NS398 (COX2 specific inhibitor) and 1, 5, and 10µM SC-560 (COX1 specific inhibitor). A unit of cyclooxygenase activity is defined as the ability of the enzyme to catalyze oxygenation of 1 nmol AA per minute at 37◦C. The calibration curve was made with human COX2 (C0858-1000UN) provided by Sigma-Aldrich.

### Histology

The histological sections were processed according to conventional histological procedures and stained by methods previously described by Soto et al. (2014).

### Interleukins

6-keto-PGF1α, PGE<sup>2</sup> and 8-isoprostane were provided by Cayman Chemical Company. TGFβ1 and TNFα were quantified by ELISA using commercial kits obtained by Elab science Biotechnology Co., Ltd. (Cat No. E-EL-ROO19) and Enzo Life Sciences (Cat No. ADI-900-155), respectively and read in a visible light microplate reader of 492/630 nm.

### Statistical Analysis

For the analysis of continuous quantitative variables of normal distribution, t student test was used. For nonparametric data such as in general demographic characteristics Mann-Whitney U-test was employed. The program Sigma Plot version 11, Jandel Corporation was used to obtain the graphs. The data are presented as mean ± standard error. The differences were considered as statistically significant when p ≤ 0.05.

### RESULTS

A total of 14 subjects with MFS and 6 controls were studied; the overall mean age was 40 ± 16 years. Age in patients with MFS had a median of 35 with a minimum value of 16 and a maximum of 59 and in controls it had a median of 63 with a minimum value 49 and maximum of 72 (p = 0.001), age and gender showed no differences. Demographic characteristics are shown in **Tables 1**, **2**.

**Table 3** shows the concentrations in pg/mg protein of the different markers measured. The 8-isoprostane AA oxidation marker was measured by a non-enzymatic assay. In patients with MFS it tended to increase but did not reach a statistically significant difference in comparison with the control subjects.




F, Female; M, male; BMI, Body mass index; C, control; MFS, Marfan syndrome; n, number; HDL, cholesterol-high density lipoprotein; LDL, cholesterol low density lipoprotein. Global n = 20, MFS n = 14, and Control n = 6.

The same tendency was observed in the 6-keto-PG1<sup>α</sup> stable metabolite of COX1. However, PGE2, TGFβ1 and TNFα showed a significant increase in patients with MFS when compared to controls (p = 0.02, p = 0.04, p = 0.03, respectively).

**Figure 1A** shows that the AA had a statistically significant percentage reduction in the aorta homogenate of MFS patients when compared with control subjects (p = 0.02). **Figure 1B** shows an increase of OAin MFS patients (p = 0.04) in comparison with control subjects.

**Figure 2A** shows that the activity of the cyclooxygenase isoforms was statistically higher in patients with MFS compared to control subjects (p = 0.002). **Figure 2B** shows that inhibition of COX2 by NS398 was statically significant in patients with MFS compared to control subjects (p = 0.01). However, the inhibition of COX1 by SC560 in patients with MFS showed a tendency to decrease at concentrations of 5 and 10µM (p = 0.08 and p = 0.06 respectively), but did not reach a statistically significant difference in comparison with the control subjects.

In patients with MFS, the expression of PLA2, CYP450 4A, and 5-LOX showed a statistically significant elevation from that in control subjects (p = 0.03, p = 0.001, and p = 0.004, **Figures 3A–C**, respectively).

The expression of COX1 showed a tendency to increase in patients with MFS without reaching a statistically significant difference when compared to control subjects (p = 0.08, **Figure 4A**). However, expression of COX2 and TXA2R presented a significant increase in patients with MFS compared to control subjects (p = 0.05, **Figures 4B,C**, respectively).

In **Figures 5A–C** representative photomicrographs of the hematoxylin-stained aorta, Masson's trichrome, and Weigers method for elastic fibers respectively of control subjects are shown. Elastic fibers shown in black can be seen alternating with limited collagen fibers in reddish brown. **Figures 5D–F** represent photomicrographs of the aortic tissue of patients with MFS respectively. There is an increase in collagen between broken elastic fibers that separate and form cavities resulting from the breakage of the elastic fibers. These characteristics correspond to the presence of cystic necrosis and suggest a lack of elasticity, thickening and high disorganization of the elastic lamellar structure and fibrosis due to excess collagen in patients with MFS.

**Figure 6** shows the same representative photomicrographs reported of control subjects in **Figures 6A–C** and of MFS patients in **Figures 6D–F** using the same histological techniques mentioned in **Figure 5**, but with a higher amplification of 40x. The histopathological changes are more evident than in **Figure 5**, where the magnifications were to 16x. These last pictures are panoramic and allow for the appreciation of part of the aortic vessel.

### DISCUSSION

The aim of the study was to determine the participation of AA metabolism associated with inflammation factors in the aneurysm aortic dilatation and dissection in patients with MFS. AA is a major component of the membrane phospholipids necessary for the reparation and growth of skeletal muscle tissues. It is also a precursor of numerous eicosanoids (Hadley et al., 2016). The initiation of AA metabolism begins by the activation and translocation of the PLA<sup>2</sup> and this enzyme releases AA from membrane phospholipids (Balsinde and Balboa, 2005). Different mechanisms can stimulate PLA<sup>2</sup> such as ionophore A23187, increase in Ca2<sup>+</sup> concentration through voltage dependent Ca2<sup>+</sup> channels, activation of K+-ATP channels, stimulation of PKC (Linkous and Yazlovitskaya, 2010) and a pro-inflammatory state. The latter, in MFS may cause alterations in membrane lipid packaging and asymmetry which might increase PLA2. Our results show that PLA<sup>2</sup> expression was increased in the aortic aneurysm of MFS patients. This increase suggests a high percentage of AA to be metabolized. However, the results show that AA decreases in MFS patients. This reduction can be, in part, attributed to the increase in OA since this acid can reduce the percentage of AA as has been described in serum and breast muscle (Haug et al., 2010). A recent study by our group described that OA is increased and involved in the development of aortic aneurysm in MFS patients. This study further shows a possible increase of the AA percentage without a reaching a significantly significant difference. However, in this new patient series, a statistically significant decrease of AA percentage was observed (Soto et al., 2016a). Although our results seem paradoxical, the


MFS, Marfan syndrome; F, female; M, male; FH, family history.

TABLE 3 | Inflammation, oxidative stress markers and vasodilator and vasoconstrictor prostaglandins products of the COX1 and COX2 respectively in control subjects and MFS.


MFS, Marfan syndrome. Control vs. MFS † p = 0.02, \*p = 0.04, \*\*p = 0.03.

AA reduction could depend on the degree of inflammation, aneurysm progression and the type of mutation present (Radonic et al., 2012).

The inverse relation between the percentage OA and AA could be caused by a positive feedback regulation, since a reduced percentage of AA could probably be expected if OA inhibits 16desaturase, elongase-5, and/or 15desaturase, the enzymes governing the formation of AA by the desaturation and elongation process from linoleic acid (Obukowicz et al., 1998; Høstmark and Haug, 2013). Furthermore, OA may induce the expression of COX2 (Fang et al., 2009), since an injection of OA increased prostanoids derived from COX2 in the pulmonary artery and was associated tothe presence of edema (Selig et al., 1987). Another possible explanation for the reduction of the percentage of AA may be its use as a substrate for the enzymes which metabolize it. This would lead to an increased expression and/or activity as is shown in our results. In addition, mice deficient in COX1, but not in COX2 exhibit a reduction in AA which contributes to edema and inflammation (Langenbach et al., 1995). This suggests that the AA percentage has to have the appropriate threshold since its broken equilibrium can cause adverse effects on the enzymatic pathways that depend on it.

On the other hand, our results show that the activity of COX isoforms is increased in MFS patients; however, this experiment does not allow for the discrimination of which of the two isoforms displays a greater participation. The immunoblot and the inhibition of the activity with the specific inhibitors show that COX2 expression and the inhibition of its activity were increased in MFS patients. However, the COX1 expression and inhibition also showed a tendency. This indicates that the enzyme that participates more is COX2, and it is therefore more important in the development of aortic aneurysm in MFS. Also, in ruptured abdominal aortic aneurysm, COX2 expression is significantly elevated (Chung et al., 2007b). Another study showed an increase in the COX2 expression in the thoracic aorta of the Fbn1C1039G/+MFS mouse model, while COX1 was down-regulated (Chung et al., 2007a).

PGE<sup>2</sup> biosynthesis by COX2 is important for many biological functions and is generally very low in un-inflamed tissues, but increases immediately in acute inflammation prior to the recruitment of leukocytes and infiltration of immune cells (Ricciotti and FitzGerald, 2011). Under these physiological conditions, PGE<sup>2</sup> can modulate various steps of inflammation through receptors such as those for IL-1β, IL-6, and MCP-1 (Babaev et al., 2008). In wild-type mice, the deficiency of COX2 reduces the level of PGE<sup>2</sup> production by approximately 75%, while the deficiency of COX1 reduces the PGE<sup>2</sup> level by 25%. This indicates that both COX isoforms contribute to PG production during inflammation and also that COX2 derived PGs appear to be more important in both the acute inflammatory process and in the resolution phase (Langenbach et al., 1999). Our results show an increase in PGE<sup>2</sup> in comparison to 6-keto-PG1<sup>α</sup> in aortic aneurysm tissue from MFS patients that were associated with increase and decrease of COX2 and COX1 respectively. This indicates a high contribution of PGE<sup>2</sup> in the aortic aneurism

formation in MFS. Furthermore, PGE<sup>2</sup> biosynthesis by COX2 is increased in human abdominal aortic aneurysm and the infiltration of leukocytes in the aortic wall and may potentially contribute to the PGE<sup>2</sup> increase (Solà-Villà et al., 2015). In addition, PGI<sup>2</sup> is a potent vasodilator and inhibitor of platelet aggregation, leukocyte adhesion and it increases the production of the anti-inflammatory cytokine IL-10 (Francois et al., 2005), and is rapidly converted by non-enzymatic processes to an inactive hydrolysis product, 6-keto-PGF1<sup>α</sup> (Wu and Liou, 2005). Additionally, PGI<sup>2</sup> is more abundant in the aorta than PGE<sup>2</sup> (Qi et al., 2006). However, when COX1 is inhibited PGI<sup>2</sup> is significantly decreased while PGE<sup>2</sup> is increased in the aorta by COX2 overexpression, and the inflammatory chronic process in MFS can contribute to this (Qi et al., 2006).

Additionally, the limiting step in the synthesis of TXA<sup>2</sup> is COX2/TXs enzyme on AA, and the increase of this enzyme has been associated with a TXA<sup>2</sup> increase (Jiang et al., 2007). In the thoracic aortic dissection from Marfan mice at 3, 6, and 9

months of age that were heterozygous for the Fbn1C1039g/<sup>+</sup> allele, COX1 expression was down-regulated, COX2 was increased and there was an imbalance in the synthesis of TXA<sup>2</sup> which was associated with vascular hyperplasia, thrombosis events and vascular remodeling (Chung et al., 2007a). Our results show that the TXA2R expression was increased in aneurysm tissue from MFS patients. This indicates that TXA<sup>2</sup> synthesized by COX2/TXs enzyme can increase TXA2R and contribute to the progression of aortic aneurism in MFS patients (Chung et al., 2007b). In addition, TXA<sup>2</sup> produced by COX2 promotes inflammatory mediator production that participates in vascular injury, hypertrophy, platelet aggregation and extracellular matrix formation. Thus, it may be a factor influencing the expansion rate and eventual rupture of the aneurysm (Cheuk and Cheng, 2007). Therefore, our results suggest that, in the aorta the MFS patients, there exists an imbalance in the synthesis of the prostaglandins given by an increase of PGE2, TXA<sup>2</sup> and decrease of 6-keto-PG1<sup>α</sup>

which result from the altered metabolism of AA through the rate limiting enzymes COX1 and COX2. This imbalance contributes in part to the compromised aortic vasomotor function in MFS, impairing the release of endothelial relaxant molecules (Di Marzo, 1995; Soto et al., 2016a).

On the other hand, PGE<sup>2</sup> and TXA<sup>2</sup> increase transcription of type IV collagen, laminin and fibronectin. These proteins are involved in the thickening of the aortic sub-endothelial layer of the aortic aneurysm (Pricci et al., 1996). The increases in PGE<sup>2</sup> and TXA2, participate in the alteration of the contraction and relaxation of vascular smooth muscle, together with the inflammatory chronic processes, which influence metalloproteinase (MMPs) expression (Dorn et al., 1992). Furthermore COX2 over-expression induces MMPs whose activation can result in the extracellular matrix degradation that is essential for vascular remodeling and inflammatory

tissue; CN, cystic necrosis. 16x magnifications.

cell infiltration that contribute to the formation of the aortic aneurysm (Gitlin et al., 2007). In addition, the chronic inflammatory disease present in the aortic aneurysm of MFS is produced by the FBN-1 mutation (Holm et al., 2011). This FBN-1 mutation could easily affect the organization of the collagen fibers in the aortic adventitia through an increase in TGFβ1 signaling that occurs in the extracellular matrix and contributes to increased collagen production (Holm et al., 2011). These changes lead to aortic damage due to impaired resistance to pressure, and by creating a positive feedback that may cause collagen damage. The degradation products resulting from this process may induce chronic inflammation in MFS patients (Radonic et al., 2012). Furthermore, TGFβ1 can also regulate the over expression of elastase and many MMPs such as -2 and -9 (Neptune et al., 2003). Also, an increased level of elastase might increase elastin degradation mediated by the MMPs, and can be responsible for the disintegration of elastic fibers that reduce connective tissue elasticity and lead to weakness of the

FIGURE 6 | Representative photomicrographs of the aortic root at a 40x magnification. Aortic root aneurysms of control subjects (A–C) and of MFS patients (D–F). Sections were stained with same histological techniques as in Figure 5. All MFS patients, had cystic necrosis, bundles of elastic fibers, thickening and rupture of elastic fibers, and increase in collagen between the elastic fibers in the aorta in comparison to C subjects. E, Endothelium; EF, elastic fibers; Col, collagen; CT, connective tissue; CN, cystic necrosis.

aortic wall (Benke et al., 2013). Besides, the soluble peptide fragments derived from the degradation of extracellular matrix components, including elastin, laminin and fibronectin may also serve as chemotactic agents for infiltration by macrophages which are responsible for the enhanced release of inflammatory mediators (Cheuk and Cheng, 2008). However, hemodynamic forces, trans-mural inflammation, imbalance of MMPs, increase of TGFβ1, inflammatory cell infiltrates, apoptosis, ROS, fatty acid alteration and interleukins such as TNFα also participate (Soto et al., 2016a).

Our results show that TGFβ1 and TNFα concentrations in the homogenized tissue from the aortic aneurysm the MFS patients were increased, and this increase was associated with accumulation of collagen, cystic necrosis and degradation of elastic fibers as observed in the photomicrographs. A study shows that indomethacin, an unspecific inhibitor of COXs isoforms, significantly improves elastin integrity and reduces the numbers of macrophages in the adventitia of heterozygous mgR/mgR FBN1 in a Marfan mice model. These changes coincide with decreased MMP-2, -9, and -12 expressions, and are blocked by a decrease in the TGFβ1 activity in the aorta, and with improved elastic lamellae architecture. These changes were associated with a decrease in COX2 expression and PGE<sup>2</sup> concentration (Guo et al., 2013). In addition, an alteration in the AA percentage has been associated with increased production and secretion of TNFα (Cubero and Nieto, 2012). TNFα can induce a switch from the PGD<sup>2</sup> to the PGE<sup>2</sup> synthesis pathway by regulating PGE<sup>2</sup> synthase expression and/or activity. It can also act on activators of PKA that potentiate markedly the TNFαinduced increase in PGE<sup>2</sup> through up-regulation of COX2 gene expression (Fournier et al., 1997). In a study using pre- to post-operative arterial blood samples for 13 days in patients with ruptured aneurysms and intact aneurysms, an increase of TNFα, IL-1β, and IL-6, was found (Swartbol et al., 2001). Also, TNFα derived from intrinsic vessel wall components or the cells of the tymphomonocytic infiltrate, is part of an accelerated proteolytic cascade that is responsible for progressive destruction of structural matrix proteins, particularly collagen and elastin. This destruction leads to a thin degraded media layer with significant loss of the elastic component and a fibrotic and or inflammatory adventitia (Fernandez-Moure et al., 2011). TNFα is also involved in the early phase of the cytokine cascade in a proinflammatory state that promotes endothelial dysfunction and induction of pro-inflammatory genes including those of iNOS and COX2 (Lu et al., 2015). TNFα may stimulate endothelial cells to express vascular cell adhesion molecule-1, intracellularadhesion molecule-1 which facilitates macrophages infiltration with the secretion of collagen, elastin and proteoglycans to form a fibrous matrix resulting in the changes in the aortic wall architecture (Grötzinger, 2002). Similarly, the excess release of TGFβ1 from the connective tissue of the matrix can cause an increased activation of the smooth muscle cells and a haphazard and inappropriate remodeling response, which is characterized by excess deposition of matrix elements such as collagen, proteoglycans, MMPs-2, -9 and infiltration of macrophages (Zhang et al., 2013).

The results of TNFα and TGFβ1 determinations obtained in this study are associated with the thickening and high disorganization of the structure of elastic lamella in the aortic aneurysm of MFS patients. This alteration of the aortic medial layer can generate cystic necrosis and fibrosis by collagen excess in MFS (Guo et al., 2006). This contributes to the reduced elasticity and compliance in the aortic aneurysm, the loss of elastic fiber integrity and the increase in stiffness which result in a significantly higher breaking stress in the aorta the MFS patients (Yuan and Jing, 2011). Besides, a recent paper where a histopathological analysis in MFS patients was performed, showed an increase MMP-9, angiotensin II and TGF-β1 productions which was associated with cytolytic necrosis and elastic fiber degeneration (Grewal et al., 2016). In a similar manner, high concentrations of TGFβ1 and TNFα, favor apoptosis in endothelial cells and activate MMPs-1 and -2 in vascular smooth muscle cells (Ramachandra et al., 2015). It therefore promotes collagen and elastin degradation in the extracellular matrix, which contributes to physically inhibit the normal dilatation of the aorta. MMP-2 has also been associated with the development of aneurysms in the thoracic aorta (Ramachandra et al., 2015). Our results showed that all MFS patients had cystic necrosis and an increase of collagen and elastic fibers rupture.

On another hand, AA can also be metabolized through other pathways, such as 5-, 12-, 15-LOX, and CYP450 pathway (Funk, 2001; Ferrucci et al., 2006). Our results show that CYP450 4A and 5-LOX expressions were increased in the aortic aneurysm of MFS patients. These results suggest that other enzymes involved in AA metabolism also contribute to the aortic aneurysm progression in MFS, and provide another possible explanation to the contribution of the AA reduction. This could be related to an increase in use AA by CYP450 4A and 5-LOX enzymes, which can contribute to aneurysm progression in MFS. Hence, AA has an anti-thrombogenic effect via CYP450 generated EETs; this eicosanoids possess vasodilator properties, and may inhibit the COX2 (Krötz et al., 2004).This suggests that the CYP450 4A overexpression can be a compensatory mechanism to counteract the other pathways of AA metabolism that are altered. However, this pathway is unable to completely counteract the other effects of the AA metabolism. In contrast, various components of the 5- LOX pathway are involved in human vascular disease (Cao et al., 2007). In Apoe−/<sup>−</sup> mice with an atherosclerotic producing diet, 5-LOX in the lamina adventitia rather than in the lamina intima contributed to the formation of aortic aneurysms associated with production of systemic inflammatory leukotrienes that indirectly affected extracellular matrix components including MMP-2 (Zhao et al., 2004). A study in advanced lesions in human coronary arteries showed that there is abundant 5-LOX in the vascular adventitia (Spanbroek et al., 2003). Another study showed that leukotriene via 5-LOX inhibited the production of PGI<sup>2</sup> by COX1 in the vascular endothelium and indirectly contributed to overall vascular constriction (Chawengsub et al., 2009).

Furthermore, OS is associated with endothelial dysfunction in a mouse model and in MFS patients. It contributes to increases the aneurysm formation (Zúñiga-Muñoz et al., 2017). OS is associated with decreased nitric oxide and PGI<sup>2</sup> mediated endothelium-dependent relaxation via eNOS and COX1. Furthermore, OS is associated with the aortic structural changes that lead in part to the formation of aneurysms (Soto et al., 2016b; Zúñiga-Muñoz et al., 2017). Significantly increased levels of isoprostane 8-epi-PGF2α have been recognized as oxidative markers associated with vascular diseases such as vascular reperfusion, and have also been found in plasma and aortic homogenate of MFS mice. However, 8-isoprostane is also a product of the AA metabolism which is oxidized by free radicals and by the non-enzymatic pathway (Łuczaj et al., 2015). The results in the homogenized tissue from the aortic aneurism from MFS patients showed a tendency of 8-isoprostane to increase without reaching a significant value (Yang et al., 2010). This indicates that high concentrations of this product of oxidation,

could favor endothelial cell apoptosis and activate MMPs-1 and -2 in vascular smooth muscle cells. It therefore promotes the degradation of various components of the extracellular matrix such as collagen and elastin (Fujiwara et al., 2008). This could, in part, contribute to progression in the aortic aneurysm in MFS patients.

### CONCLUSION

In the aortic tissue of MFS patients there are inflammatory factors coupled with the genetic factors that predispose to aortic endothelial dysfunction. There is a decrease in the percentage of AA, which is associated with an increase of PLA2, COX2/TXs, CYP450 4A and 5-LOX which leads to a greater synthesis of PGE<sup>2</sup> than 6-keto-PGF1<sup>α</sup> thus contributing to the formation of the aortic aneurysm. The evident loss of homeostasis in these mechanisms confirms that there is a participation of the AA pathway associated to the degree of inflammation and the aneurysm progression in MFS.

### Study Limitations

The use of aortas from MFS patients and C subjects constitutes an important limitation of this study. The obtainment of tissue from aortic samples is very difficult despite the informed consent of MFS patients and control subjects and the aortic sample size is very small. The incidence MFS is of 2-3 per 10,000 individuals, being an autosomal dominant disorder of the connective tissue caused by mutations. The monitoring of each

### REFERENCES


patient prospectively for a long time is also practically impossible. Retrospective studies only allow for the evaluation of some aspects but do not give the opportunity to correct some biases. The results from this study suggest the importance of studying the general inflammatory profile in this disease to correlate it to the local inflammation in the aortic tissue in future studies. Another limitation to this study is the improbability of having matched controls for age and gender, since it is not possible to obtain tissue from healthy people. The only way to obtain the tissue is from subjects having a surgical indication where there is a possibility to ethically withdraw a small sample. This depends on the surgical technique of the treatment applied to the subjects and the informed consent of the patients. However, even though there are age differences in controls and patients, there is certainty that there were no co-morbidities or aortic damage as shown by the imaging studies reported.

### AUTHOR CONTRIBUTIONS

MS: designed the study, diagnosed the patients and designed the tables. VG-L: wrote, restructured and reviewed the manuscript. KH-M: obtaining the sample, IP-T: designed the study, wrote the manuscript, performed the assays and interpreted the results.

### ACKNOWLEDGMENTS

We thank Benito Chávez Rentería for histology technical support.

aortic aneurysm? World. J. Surg. 32, 55–61. doi: 10.1007/s00268-007- 9279-9


aortopathies. Oxid. Med. Cell. Longev. 2014:760694. doi: 10.1155/2014/ 760694


**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 Soto, Guarner-Lans, Herrera-Morales and Pérez-Torres. 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.

# Oxidative Stress and NLRP3-Inflammasome Activity as Significant Drivers of Diabetic Cardiovascular Complications: Therapeutic Implications

#### Edited by:

*Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico*

#### Reviewed by:

*James Todd Pearson, National Cerebral and Cardiovascular Center, Japan Zhihong Yang, University of Fribourg, Switzerland*

#### \*Correspondence:

*Judy B. de Haan judy.dehaan@baker.edu.au † Rebecca H. Ritchie orcid.org/0000-0002-8610-0058 ‡ These authors have contributed equally to this work.*

§*Joint senior authors.*

#### Specialty section:

*This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology*

Received: *17 November 2017* Accepted: *05 February 2018* Published: *20 February 2018*

#### Citation:

*Sharma A, Tate M, Mathew G, Vince JE, Ritchie RH and de Haan JB (2018) Oxidative Stress and NLRP3-Inflammasome Activity as Significant Drivers of Diabetic Cardiovascular Complications: Therapeutic Implications. Front. Physiol. 9:114. doi: 10.3389/fphys.2018.00114* Arpeeta Sharma1‡, Mitchel Tate2‡, Geetha Mathew<sup>3</sup> , James E. Vince4,5 , Rebecca H. Ritchie2†§ and Judy B. de Haan<sup>1</sup> \* §

*<sup>1</sup> Oxidative Stress Laboratory, Basic Science Domain, Baker Heart and Diabetes Institute, Melbourne, VIC, Australia, <sup>2</sup> Heart Failure Pharmacology Laboratory, Basic Science Domain, Baker Heart and Diabetes Institute, Melbourne, VIC, Australia, <sup>3</sup> Cellular Therapies Laboratory, Westmead Hospital, Sydney, NSW, Australia, <sup>4</sup> Inflammation Division, Walter and Eliza Hall Institute, Melbourne, VIC, Australia, <sup>5</sup> Department of Medical Biology, University of Melbourne, Melbourne, VIC, Australia*

It is now increasingly appreciated that inflammation is not limited to the control of pathogens by the host, but rather that sterile inflammation which occurs in the absence of viral or bacterial pathogens, accompanies numerous disease states, none more so than the complications that arise as a result of hyperglycaemia. Individuals with type 1 or type 2 diabetes mellitus (T1D, T2D) are at increased risk of developing cardiac and vascular complications. Glucose and blood pressure lowering therapies have not stopped the advance of these morbidities that often lead to fatal heart attacks and/or stroke. A unifying mechanism of hyperglycemia-induced cellular damage was initially proposed to link elevated blood glucose levels with oxidative stress and the dysregulation of metabolic pathways. Pre-clinical evidence has, in most cases, supported this notion. However, therapeutic strategies to lessen oxidative stress in clinical trials has not proved efficacious, most likely due to indiscriminate targeting by antioxidants such as vitamins. Recent evidence now suggests that oxidative stress is a major driver of inflammation and vice versa, with the latest findings suggesting not only a key role for inflammatory pathways underpinning metabolic and haemodynamic dysfunction in diabetes, but furthermore that these perturbations are driven by activation of the NOD-like receptor family pyrin domain containing 3 (NLRP3) inflammasome. This review will address these latest findings with an aim of highlighting the interconnectivity between oxidative stress, NLRP3 activation and inflammation as it pertains to cardiac and vascular injury sustained by diabetes. Current therapeutic strategies to lessen both oxidative stress and inflammation will be emphasized. This will be placed in the context of improving the burden of these diabetic complications.

Keywords: NLRP3 inflammasome, diabetic nephropathy, diabetic atherosclerosis, diabetic cardiomyopathy, inflammation, inflammatory cytokines, oxidative stress, diabetic complications

### INTRODUCTION

Therapeutic strategies to limit diabetic micro and macrovascular complications have focussed first and foremost on eliminating known risk factors. However, despite the availability of numerous drugs such as statins to lower lipids, angiotensin converting enzyme (ACE) inhibitors and angiotensin receptor blockers to control blood pressure, orlistat to reduce weight gain (Wilding, 2018), and a range of insulin-sensitizing drugs such as metformin, sulfonylureas, glitazones, gliptins, α-glucosidase inhibitors, and sodium-glucose transporter-2 inhibitors, for many diabetic patients the inevitable slide toward heart disease, renal failure, neuropathy and diabetic blindness continues unabated, leading to increased morbidity and mortality. In looking further afield at significant drivers of these diabetic complications, it is now increasingly appreciated that inflammatory processes per se play a key role, underpinning all of the complications associated with diabetes. Thus, inflammation is now considered a driving force in the progression of diabetic complications and is no longer simply viewed as an epiphenomenon. Indeed, it is now recognized that inflammation is not limited to the control of pathogens by the host, but rather that sterile inflammation which occurs in the absence of viral or bacterial pathogens accompanies the complications that arise as a result of hyperglycaemia.

In the years since a unifying mechanism of hyperglycemiainduced cellular damage (Brownlee, 2001) was proposed linking elevated blood glucose levels with oxidative stress and dysregulation of metabolic pathways, pre-clinical evidence has, in most cases, supported this notion. However, therapeutic strategies to lessen oxidative stress in clinical trials have not proved efficacious, most likely due to the indiscriminate targeting by antioxidants such as vitamins. With evidence now suggesting that oxidative stress begets inflammation and vice versa, the latest findings suggest not only a key role for inflammatory pathways underpinning metabolic and haemodynamic dysfunction in diabetes, but furthermore that these perturbations are driven by activation of the NOD-like receptor family pyrindomain-containing-3 (NLRP3) inflammasome. This review addresses these latest findings with an aim of highlighting the interconnectivity between oxidative stress, NLRP3 activation and inflammation as it pertains to cardiac, vascular and renal injury sustained by diabetes (**Figure 1**). Current therapeutic strategies to lessen both oxidative stress and inflammation are emphasized, and placed in the context of improving the burden of these diabetic complications.

### OXIDATIVE STRESS, INFLAMMATION AND DIABETES

Prior to the proposal of the Brownlee unifying mechanism (Brownlee, 2001), research into the pathophysiology of diabetic complications suggested that several important metabolic pathways, namely the polyol, hexosamine biosynthesis, protein kinase C, and the production of advanced glycosylation endproducts, were compromised as a result of elevated blood glucose.

diabetes. Metabolic changes in diabetes including insulin resistance, hyperlipidaemia and hyperglycaemia (and the subsequent production of DAMPs), lead to an increase in inflammation, oxidative stress and NLRP3 inflammasome activation and subsequent development of diabetic complications including diabetic nephropathy, atherosclerosis, diabetic cardiomyopathy, diabetic neuropathy and diabetic retinopathy. DAMPs, damage associated molecular patterns; RAGE, receptor for advanced glycation end-products; IAPP, islet amyloid polypeptide protein.

In understanding how these pathways were affected, it was discovered that the reactive oxygen species (ROS) superoxide (O<sup>2</sup> .−), derived from the mitochondria or elsewhere, were capable of damaging a key glycolytic enzyme glyceraldehyde dehydrogenase (GAPDH), which in turn, diverts upstream glycolytic metabolites into the four pathways of glucose overutilization. Subsequent studies have also implicated the NADPH oxidases (NOX) family of enzymes as major cytosolic sources of superoxide, and it is now appreciated that several sources exist within the cell that contribute to the increased oxidative stress accompanying diabetes (de Haan and Cooper, 2011). In addition to the metabolic perturbations, ROS are also known to cause alteration at the molecular level, with numerous reports of enhanced lipid peroxidation, protein modifications, and nucleic acid damage as a consequence of elevated glucose. Thus, it was postulated that newer more targeted antioxidants (such as SOD mimetics or catalase) might offer better protection against ROS damage (de Haan and Cooper, 2011), but in the years since those initial findings, clinical trials with more basic antioxidants (Vitamins C and E) such as HOPE and GISSI have shown no cardiovascular health benefits (Marchioli et al., 2001). More recent antioxidant approaches of note are based on positive pre-clinical data showing a role for specific isoforms of the NOX enzymes in driving diabetic complications (Gray and Jandeleit-Dahm, 2015; Di Marco et al., 2016). Indeed, current trials using specific Nox1/Nox4 inhibitors to limit diabetic nephropathy are ongoing.

One aspect not covered by Brownlee's unifying mechanism (Brownlee, 2001) is the role that ROS play in modulating the signaling cascade of immune factors. It has become apparent that increased ROS production leads to the increased production of inflammatory cytokines, and reciprocally, an increase in inflammatory cytokines can stimulate ROS production. This cyclical process, where oxidative stress begets inflammation and vice versa, drives a highly pro-inflammatory state. Evidence for this stems from the plethora of investigations across numerous diabetic complications, in which the emphasis has shifted away from a purely metabolic state to one that additionally encompasses an inflammatory state (Hotamisligil, 2006; Engelbertsen et al., 2012; Ruiz et al., 2013; Paterniti et al., 2015), thereby modifying the therapeutic approach to now also include drug regimens that target the heightened inflammation.

### INFLAMMASOME ACTIVATION AND REGULATION

### Inflammasomes Activate Capase-1 to Process IL-1β, IL-18, and Trigger Cell Death

Microbial, host-derived and environmental molecules are all capable of activating large cytosolic protein complexes known as inflammasomes. The ability of inflammasomes to specifically detect cellular stressors, metabolic changes and danger molecules makes them central governors of the innate immune response. Although protective against infections and important for an appropriate immune response following tissue damage, excessive inflammasome activity has now been linked to numerous diseases, including cancer, neurodegenerative disorders, and metabolic dysfunction (Menu and Vince, 2011). NLRP3 activation and subsequent IL-1β secretion is wellcharacterized in cells of the innate immune system, such as monocytes, macrophages and neutrophils, however, other cell types such as endothelial cells are able to produce IL-1β via the NLRP3 inflammasome (Lopez-Castejon and Brough, 2011; Xiao et al., 2013). Activation of the NLRP3 inflammasome is a two-step process (Signal 1 and Signal 2). In the first step (Signal 1), damage associated molecular pattern (DAMPs) or pathogen associated molecular patterns (PAMPs) activate toll-like receptors (TLRs) to induce the expression of inactive NLRP3, IL-β and IL-18. Subsequently, a second signal, such as bacterial toxins or host metabolites (e.g., ATP), are detected by inflammasome sensor proteins, resulting in their recruitment of the adaptor protein ASC [apoptosis-associated speck-like protein containing a caspase activation and recruitment domain (CARD)] through homotypic pyrin-pyrin domain interactions. A single perinuclear oligomerized ASC "speck" subsequently forms in each cell, which is reported to adopt a prion-like structure (Cai X. et al., 2014; Franklin et al., 2014; Lu et al., 2014). Oligomerized ASC binds pro-caspase-1 through homotypic CARD-CARD interactions resulting in proximity-induced caspase-1 processing and activation. Caspase-1 cleaves the inactive precursor proteins of IL-1β and IL-18 into their bioactive fragments that are subsequently released from macrophages into the extracellular milieu. In parallel with caspase-1 cleavage of IL-1β, caspase-1 or the related inflammatory caspase, caspase-11, can cleave gasdermin D (GSDMD)(Kayagaki et al., 2015; Shi J. et al., 2015) thereby triggering an oligomerized N-terminal GSDMD fragment to form pores in the plasma membrane and induce the lytic pro-inflammatory cell death known as pyroptosis (Aglietti et al., 2016; Ding et al., 2016; Liu et al., 2016; Sborgi et al., 2016) (**Figure 2**). While GSDMD-induced membrane pore formation, and presumably cell death, is important for bioactive IL-1β release from macrophages, other cell types such as neutrophils have been reported to secrete active IL-1β without any compromise in membrane permeability (Chen et al., 2014). This suggests that cell-specific differences, and possibly distinct mechanisms, in IL-1β release exist and that the cellular exit of IL-1β can occur due to an active, yet ill-defined, secretory process (Conos et al., 2016). Alternatively, it has become clear that other cell death modalities that result in a compromised macrophage plasma membrane, such as Mixed Lineage Kinase Domain Like (MLKL) mediated necroptosis, may also cause activated IL-1β release even when GSDMD is genetically deleted (Conos et al., 2017; Gutierrez et al., 2017).

### Excessive Inflammasome Activation Results in Autoinflammatory Disease

A variety of inflammasome sensor proteins have been identified. These include several members of the NOD-like Receptor family, such as NLRP1, NAIP/NLRC4, NLRP3, NLRP6, NLRP7, and NLRP9 in addition to the HIN-200 family member, AIM2, and the tripartite motif-containing (TRIM) family member Pyrin (also known as TRIM20). Evidence from the literature suggests that several inflammasome sensors have well-defined activating ligands (such as the DNA-binding AIM2) (Latz et al., 2013) and/or mechanisms (such as toxin-induced Rho-GTPase inactivation, which then permits Pyrin to be released from 14- 3-3 protein inhibition) (Latz et al., 2013). However, a consensus on precisely how the inflammasome sensor proteins implicated in the metabolic syndrome and diabetic conditions, NLRP1 and NLRP3, come to be activated in these states, has yet to be reached. Regardless, the ability of inflammasomes to trigger damaging autoinflammation is undeniable, as evidenced by the rare hereditary inflammatory diseases resulting from activating

cleavage and further processes pro IL-1β and pro IL-18 into their mature forms for secretion. In addition, caspase-1 also cleaves GSDMD which results in plasma membrane pore formation and pyroptototic cell death. PAMPs, pathogen associated molecular patterns; DAMPs, damage associated molecular patterns; LPS, lipopolysaccharides; ASC, apoptosis-associated speck-like protein containing a caspase recruitment domain (CARD); GSDMD, gasdermin D; CARD, caspase recruitment domain; PYD, PYRIN domain; NACHT, a domain containing NAIP, CIITA, HET-E and TEP1 domains.

mutations in NLRP1, NLRP3, Pyrin, and NLRC4 (Manthiram et al., 2017).

The caspase-1 substrate IL-1β has typically been implicated in inflammasome-driven autoinflammation, and IL-1 blockade has shown remarkable success in the clinic in the treatment of inflammasomopathies. More recently, NLRC4 activating mutations have been documented to result in macrophage activation syndrome (Canna et al., 2014), a condition similar to hemophagocytic lymphohistiocytosis, which are both associated with exacerbated IL-18 levels (Kaplanski, 2018). A pathologically pro-inflammatory role for increased IL-18 is also suggested by findings implicating IL-18 signaling in vascular pathologies (Bhat et al., 2015). Despite these observations, other studies have implicated a protective function for inflammasome-activated IL-18 in obesity and the metabolic syndrome (Netea et al., 2006; Zorrilla et al., 2007; Murphy et al., 2016). In this context, Murphy et al. (Murphy et al., 2016) recently identified that IL-18 production is driven by activation of the inflammasome sensor NLRP1, and that NLRP1-induced IL-18 maturation protected against an obese phenotype in mice by stimulating lipolysis. The cell types in which NLRP1 functions to cleave IL-18 remains to be determined although production of IL-18 appeared to be local rather than systemic and most likely occurred within the adipose tissue (Murphy et al., 2016). The extent to which inflammasome and caspase-1 driven GSDMD activity participates in autoinflammatory disease, to either promote IL-1β or IL-18 release or trigger pyroptosis, remains to be clarified.

### Mechanism of NLRP1 and NLRP3 Activation NLRP1

Studies suggest that NLRP1 and NLRP3 activity levels alter metabolic homeostasis and can thereby contribute to glucose intolerance and insulin resistance (Stienstra et al., 2010, 2011; Vandanmagsar et al., 2011; Wen et al., 2011; Youm et al., 2011; Murphy et al., 2016). As discussed above, murine studies have documented how the deletion of NLRP1 results in an obese phenotype with increased lipid accumulation and glucose intolerance, and that the NLRP1 deficient phenotype is exacerbated further when animals are placed on a high fat diet (Murphy et al., 2016). Conversely, the same study reported reduced fat mass and metabolic dysfunction in an autoactivating NLRP1 mutant mouse, which was linked to excessive IL-18 activation. Overall, these findings suggest that the NLRP1- Caspase-1-IL-18 axis represents an important metabolic rheostat, and that the elevated IL-18 levels observed in the metabolic syndrome, obese people or type 2 diabetes (Fischer et al., 2005; Hung et al., 2005; Zirlik et al., 2007) may result from an attempt to counteract metabolic dysfunction and insulin resistance. How NLRP1 is activated in the context of excess energy intake remains unknown. However, recent findings from the study of autoactivating NLRP1 mutations suggest that autolytic cleavage of NLRP1 is an important step, removing the inhibitory pyrin domain to allow interactions of its C-terminal CARD domain with either ASC or Caspase-1(Yu et al., 2018).

### NLRP3

Pathological NLRP3 inflammasome signaling and IL-1β processing has been implicated in widespread conditions beyond diabetes and heart disease, such as gout, cancer and Alzheimer's disease (Guo et al., 2015). It therefore comes as no surprise that NLRP3 activity is normally kept tightly incheck by a number of molecular mechanisms, including both transcriptional and post-translational regulatory events, such as phosphorylation, ubiquitylation, S-nitrosylation, and cleavage (Baker et al., 2017). Defining the precise roles and timing of many of these modifications, and how they impact and co-ordinate with reported modifications of ASC or IL-1β itself, remains a substantial challenge.

A diverse number of molecules implicated in inflammatory conditions, such uric acid crystals associated with gout (Martinon et al., 2006), increased lipids [palmitic acid (Abderrazak et al., 2015), ceramide (Vandanmagsar et al., 2011), cholesterol crystals (Duewell et al., 2010)] and glucose or islet amyloid polypeptide (IAPP) (Masters et al., 2010) associated with the metabolic syndrome, atherosclerosis and type 2 diabetes, have been reported to activate NLRP3. Consistent with this, NLRP3 deficiency or IL-1 inhibition is beneficial in mouse models of these diseases and type 2 diabetic patients treated with anti-IL-1 therapy display improvements in glucose control and markers of systemic inflammation (Esser et al., 2014).

Although numerous pathways leading to NLRP3 signaling by diverse chemical and structural stimuli have been proposed, such as ROS production and lysosomal rupture, the unifying molecular switch that triggers NLRP3 inflammasome assembly remains to be clarified (Lawlor and Vince, 2014). However, despite significant controversy in this area, in recent years there has been a growing consensus that the common determinant dictating NLRP3 activation in nearly all circumstances are levels of intracellular potassium. This stance is consistent with the fact that almost all known NLRP3 activators result in decreased levels of intracellular potassium, and if potassium release is prevented, then NLRP3 activation is inhibited (Pétrilli et al., 2007; Muñoz-Planillo et al., 2013). Similarly, media depleted of potassium to promote potassium efflux suffices to trigger spontaneous NLRP3 activation. NEK7 was recently identified as binding to NLRP3 upon potassium efflux to induce inflammasome formation (He et al., 2016; Schmid-Burgk et al., 2016; Shi et al., 2016), although how decreases in cellular potassium trigger this event is not known. Regardless, potassium efflux invariably occurs following membrane damage. This is likely to explain why NLRP3 is activated by genetically-distinct programmed cell death pathways that all cause membrane damage (Vince and Silke, 2016), and importantly, may also account for why the diverse number of cellular stressors linked to atherosclerosis and diabetes all cause pathological NLRP3 activity.

In summary, it is likely that the two inflammasome sensors, NLRP1 and NLRP3, exert opposite effects in obesity; NLRP1 cleavage of IL-18 acts to limit metabolic dysfunction, while NLRP3 activation of IL-1β is detrimental and promotes glucose intolerance. Given that both the NLRP3 and NLRP1 inflammasome produce mature caspase-1 to cleave IL-18 and IL-1β, how these opposing effects occur, and in which cellular compartments, to drive or limit obesity remain unanswered questions that will be important to resolve.

### INFLAMMASOMES IN DIABETIC COMPLICATIONS

### Atherosclerosis

Atherosclerosis is a multi-factorial disease of the large arteries characterized by the deposition and accumulation of lipids and inflammatory cells. It is the underlying cause of life-threatening cardiovascular complications including myocardial infarction and stroke. Diabetes is a prominent risk factor for atherosclerosis with diabetic patients demonstrating a 2 to 4-fold increased incidence in the development of atherosclerosis than the nondiabetic population (Khaleeli et al., 2001).

A plethora of information, resulting from the analysis of rodent and human atherosclerotic plaques, has demonstrated that IL-1β and IL-18, both of which are products of the NLRP3 inflammasome activation, play a key role in the initiation and progression of atherosclerosis. Deficiency in IL-1β or IL-18 as well as the delivery of antagonist to the IL-1 receptor, has demonstrated marked reduction in atherosclerotic lesion size (Elhage et al., 2003; Kirii et al., 2003; Rader, 2012). Mechanistically, in early and advanced atherosclerotic lesions, the presence of cholesterol crystals was shown to act as the endogenous "danger signal" for the release of cytokines from the inflammasome pathway (**Figure 3**). This has now been confirmed in vitro, in lipopolysaccharide (LPS)-primed peripheral blood mononuclear cells and macrophages exposed to increasing concentrations of cholesterol crystals (Duewell et al., 2010). The release of IL-1β from LPS-primed macrophages subjected to exposure to cholesterol crystals was diminished in macrophages isolated from NLRP3-deficient and ASC-deficient mice, clearly suggesting that cholesterol crystals can act at least in part via the inflammasome pathway to stimulate the inflammatory response (Duewell et al., 2010). Injection of cholesterol crystals results in a state of acute inflammation (assessed by the intraperitoneal accumulation of neutrophils in wild-type mice); genetic deletion of components proposed to activate NLRP3, such as cathepsin B and cathepsin L, or loss of NLRP3 itself, protects mice against this acute inflammation (Duewell et al., 2010). Futhermore, lethallyirradiated low density lipoprotein receptor (LDLR) –/– mice reconstituted with bone marrow from either NLRP3-, ASC- , IL-1α/β or caspase 1/11-deficient mice are protected from the development of diet-induced atherosclerosis (Duewell et al., 2010; Hendrikx et al., 2015), further strengthening the role of myeloid-cell derived NLRP3, ASC, caspase-1 and IL-1β in the pathogenesis of atherosclerosis. Interestingly, examination of the

FIGURE 3 | Graphic summary of NLRP3 activation to promote atherogenesis. In endothelial cells, atheroprone or disturbed flow activates sterol regulatory element binding protein 2 (SREBP2) which induces expression of the ROS producing enzyme Nox2 and NLRP3. In macrophages, NLRP3 activation is mediated by the increased presence of glucose-mediated ROS, ox-LDL and cholesterol crystals in diabetic and atherogenic settings. The resultant secretion of IL-1β and IL-18 by both cell types in turn contribute to endothelial dysfunction, macrophage recruitment, migration and activation, and the upregulation of inflammatory mediators including VCAM-1, MCP-1, and NF-κB, leading to progression of atherosclerosis.

role of NLRP3 inflammasomes in the ApoE KO mouse model of atherosclerosis has however yielded contradictory results. The initial study by Menu et al, demonstrated no differences in atherosclerotic progression, macrophage infiltration in plaques and plaque stability in NLRP3/ApoE, ASC/ApoE and caspase-1/ApoE double KO mice, with the authors concluding that NLRP3 inflammasome activation is not a critical factor in atherogenesis (Menu et al., 2011). Following this, two other studies subsequently reported that caspase-1 deficiency in the high-fat-diet-fed ApoE KO mouse conferred atheroprotection (Gage et al., 2012; Usui et al., 2015). The underlying causes for these discrepancies between the three studies remain unclear, however differences in experimental design, in microbiota environment, and in formulation of the diet (at the level of cholesterol content) might have contributed to the differences in atherosclerosis observed.

Endothelial cells and macrophages are considered the primary cell types that participate in development and progression of atherosclerosis. Recently, several other molecular mechanisms by which NLRP3 inflammasomes contribute to atherogenesis in both cell types have been elucidated. Firstly, the oxidized form of low-density lipoprotein (ox-LDL), a prominent molecule implicated in atherosclerotic plaque development, was shown to play a direct role in NLRP3 activation (**Figure 3**) and NLRP3-induced macrophage pyroptosis. It was proposed that ox-LDL promoted robust ROS formation and direct ROS-driven activation of NLRP3 and caspase-1, leading to pyroptosis of macrophages and atherosclerotic lesion instability (Lin et al., 2013). Secondly, sterol regulatory element binding protein-2 (SREBP 2), a master regulator of cholesterol biosynthesis, plays a critical role in athero-prone flow-mediated endothelial inflammation via the activation of the NLRP3 inflammasome in endothelial cells. (Li et al., 2013; Xiao et al., 2013). Atheroprone flow activated the endothelium and resulted in the upregulation of SREBP 2, which then induced the transcription of NADPH oxidase 2 (Nox2) and NLRP3 expression, thereby leading to IL-1β expression and endothelial inflammation (Xiao et al., 2013). SREBP-inflammasome-endothelial innate immune responses provide a novel link between endothelial activation and monocyte recruitment, which forms the key process that causes the imbalance in vascular homeostasis, thus leading to atherosclerosis (**Figure 3**). In addition, this pivotal data provides evidence that endothelial cells play an important role in inflammatory responses through activation of NLRP3 inflammasomes.

In the context of diabetes, the role of inflammasomes in diabetes-associated atherosclerosis has attracted only modest attention. The evidence that has emerged however favors a role for inflammasome-derived cytokine production in this setting. In a short-term model of diabetes-induced endothelial dysfunction, treatment with the IL-1 receptor antagonist, anakinra, restored endothelial-dependent relaxation, with concomitant reduction of vascular NADPH oxidase and NF-κB activation (Vallejo et al., 2014). Moreover, several inflammasome components NLRP3, ASC and IL-1β are elevated at the gene expression level, in a diabetic porcine model of atherosclerosis (Li et al., 2013). In streptozotocin (STZ)-induced diabetic ApoE KO mice, aortic protein expression of NLRP3, ASC, caspase-1, IL-1β, and IL-18 was upregulated in comparison to non-diabetic controls, and was associated with enhanced lesion development and an elevation of ROS levels (Leng et al., 2016). This was confirmed in vitro in isolated bone marrow-derived macrophages, which demonstrated increased protein expression of NLRP3 and secretion of IL-1β following stimulation with high glucose and in the presence of LPS (Leng et al., 2016). Thioredoxin-interacting protein (TXNIP), a redox signaling regulator, is upregulated significantly in response to hyperglycemia and is reported to be a direct ligand of the NLRP3 inflammasome, at least in some contexts, although conflicting observations have been reported (Masters et al., 2010). In type 2 diabetes, islet amyloid polypeptide protein (IAPP), a protein responsible for the deposition of amyloid in the pancreas, has also been identified as a trigger for the NLRP3 inflammasome with resultant mature IL-1β secretion from pancreatic islets, thereby further contributing to the inflammatory response in diabetes (Masters et al., 2010). Saturated free fatty acids, for example palmitate, elevated in obese diabetic patients, induces the activation of hematopoietic NLRP3 inflammasomes via a ROS-mediated pathway, ultimately leading to the impairment of insulin signaling in target tissues to reduce glucose tolerance and insulin sensitivity (Lee et al., 2013). However, whilst there has been identification of endogenous "danger signals" that can trigger the inflammasome machinery, their direct link to diabetes-induced macrovascular complications needs further attention.

In the clinical setting, the protein and gene expression of NLRP3, ASC, caspase-1, IL-1β, and IL-18 was significantly increased in unstable carotid atherosclerotic plaques as compared to stable plaques and control patients that had no evidence of coronary artery stenosis (Shi X. et al., 2015; Paramel Varghese et al., 2016). This correlated with increased serum IL-1β and IL-18 from the same patient cohort. Moreover, immunohistochemical localization of NLRP3 has been observed in the cytoplasm of macrophages and foam cells, and associated with cholesterol crystal clefts inside and outside of the cell (Shi X. et al., 2015). Furthermore, polymorphisms in the NLRP3 gene are strongly correlated with an increased risk of macrovascular complications, in particular myocardial infarction, in Type 2 diabetic patients (Klen et al., 2015). Clinical verification of the role of NLRP3 and products of NLRP3 activation in the development of macrovascular complications may serve as a basis to incorporate this inflammasome family, as useful clinical predictors of cardiovascular events, in diabetic patients.

### Diabetic Cardiomyopathy

It is becoming increasingly apparent that, in addition to an increase in ROS, inflammasome activation is important in the pathogenesis of type 2 diabetes-mediated cardiomyopathy. Diabetic patients develop a distinct form of cardiomyopathy (Rubler et al., 1972), termed diabetic cardiomyopathy, characterized by structural and functional alterations of the heart (Tate et al., 2017). NLRP3-inflammasome activation and its role in the progression of heart failure in the absence of diabetes is well-described (Butts et al., 2015; Turner, 2016). Indeed, in the setting of acute experimental myocardial infarction, NLRP3 deletion or pharmacological inhibition reduced infarct size (Mezzaroma et al., 2011; Marchetti et al., 2014), whilst activation of NLRP3 led to inflammasome hyperactivation and amplified cardiac injury (Mezzaroma et al., 2011).

Recent reports reveal that inflammasome activation is also crucial in the setting of diabetes-induced cardiomyopathy, with archetypic metabolic disturbances exacerbating diabetes-induced cardiac dysfunction. Both type 1 and type 2 diabetes mellitus are associated with oxidative and nitrosative stress, insulin resistance and deficiency, leading to low-grade inflammation in several tissues (Singh, 2014). The healthy heart consumes a huge amount of energy that is derived from mitochondrial oxidative phosphorylation of fatty acids (∼70%) and glucose oxidation (∼30%) (Bayeva et al., 2013; Jia et al., 2018). However, in diabetes there is shift toward fatty acid metabolism that is cardiotoxic and contributes toward cardiomyopathy. Recently, the activation of the NLRP3 inflammasome was reported to be associated with this process, whereby NLRP3 activation accompanies increased CD36 expression. CD36 is responsible for the uptake of fatty acids in the heart, whereas glucose uptake is primarily mediated via insulindependent glucose transporter, GLUT4. In the healthy heart, nutrients increase plasma insulin levels triggering myocardial insulin signaling and the presence of both GLUT4 and CD36 at the myocyte sarcolemma, supporting metabolic flexibility (Jia et al., 2016). However, when insulin resistance is present, as seen in type 2 diabetes patients, GLUT4 is preferentially internalized, favoring CD36-mediated fatty acid oxidation and metabolic instability, thereby ultimately leading to a state of decreased cardiac efficiency (Sheedy et al., 2013; Jia et al., 2016; Shah and Brownlee, 2016).

NLRP3 is expressed in the key effector cell types in the diabetic heart, namely cardiomyocytes, fibroblasts and leukocytes (Lee et al., 2013; Bracey et al., 2014; Ruscitti et al., 2015; Monnerat et al., 2016). In untreated type 2 diabetic patients, monocytes exhibited increased levels of NLRP3, ASC and proinflammatory cytokines (Lee et al., 2013). Monocytes isolated from these patients were more susceptible to damage associated molecular patterns (DAMPs) (Turner, 2016) and caspase-1 cleavage (Lee et al., 2013). DAMPs are molecular signals from the damaged myocardium that include, but are not limited to intracellular molecules not normally accessible to the immune system, cytokines released from damaged cells and ECM degradation products (Turner, 2016). These effects were shown to be mitochondrial ROS- and AMPK-dependent, and interestingly, 2 months of metformin treatment, a known AMPK activator, blunted this pro-inflammatory response (Li et al., 2013). Furthermore, macrophage-dependent IL-1β increased the propensity to develop cardiac arrhythmias in diabetic mice, an effect attenuated by NLRP3 inflammasome blockade and IL-1 receptor inhibition (Monnerat et al., 2016).

As mentioned, the diabetic heart is particularly susceptible to extracellular matrix remodeling. It is now becoming apparent that excessive NLRP3 inflammasome activation is partly responsible for driving the structural and functional changes in the diabetic heart, promoting cardiac inflammation, apoptosis, and fibrosis (Cai J. et al., 2014; Luo et al., 2017). Luo et al. demonstrated that the NLRP3 inflammasome plays a direct role in the progression of diabetic cardiomyopathy (Luo et al., 2014a). Moreover, subsequent activation of IL-1β and caspase-1 initiates cell death via pyroptosis, an effect attenuated by NLRP3 gene silencing (Luo et al., 2014a). Furthermore, Rosuvastatin, a 3-hydroxy-3- methylglutaryl coenzyme A reductase inhibitor shown to have anti-inflammatory and anti-oxidant properties in several disease pathologies (Sharma et al., 2011; Zhang et al., 2012), conferred improvements in the setting of diabetic cardiomyopathy, via an NLRP3 inflammasome-dependent mechanism (Luo et al., 2014b). These beneficial effects were associated with suppressed MAPK signaling. Indeed, the suppression of MAPK signaling is of significance as these pathways are triggered by hyperglycaemia and known to accelerate the development of cardiac fibrosis (Van Linthout et al., 2007; Rajesh et al., 2012).

Cardiac fibroblasts differentiate into myofibroblasts in order to aid wound healing following stimulation primarily by TGFβ and angiotensin II (van den Borne et al., 2010). However, in certain pathologies, including diabetic cardiomyopathy, chronic activation leads to excessive collagen deposition, tissue fibrosis and detrimental cardiac remodeling (Petrov et al., 2002; Tate et al., 2016; Turner, 2016). NLRP3 expression is increased in cardiac fibroblasts following TGF-β stimulation, with NLRP3-deficient cells exhibiting impaired differentiation and Smad signaling (Bracey et al., 2014). Furthermore, NLRP3 inflammasome activation suppressed cAMP release in cardiac fibroblasts and inhibited cardiac contraction (Zhang et al., 2012). The response to tissue damage seems to be a highlyorchestrated response that is temporal in nature and includes all of the key effector cell types, which express NLRP3 and associated proteins. Although no studies to date have investigated fibroblast-to-myofibroblast differentiation in the setting of diabetic cardiomyopathy, it remains possible that the NLRP3 axis may represent a useful therapeutic target to limit this process.

### Diabetic Nephropathy

Inflammation is now recognized as an important causal factor of diabetic nephropathy (Wada and Makino, 2013). Understanding which inflammatory pathways are of significance has identified molecules that contribute to disease progression including various transcription factors, pro-inflammatory cytokines, chemokines, adhesion molecules, Toll-like receptors, adipokines and nuclear receptors. Of significance, inflammatory mediators produced by the NLRP3 inflammasome are implicated in diabetic nephropathy (DN). For example, in a recent study by Fu et al., temporal increases in the expression of NLRP3-related proteins (IL-1β, NLRP3, caspase-1) in rats with DN, and in human glomerular mesangial cells under high glucose conditions, were reported (Fu et al., 2017).

The exact mechanism by which the inflammasome is activated in DN is still unclear. Inflammasome activation through the ROS/TXNIP pathway has been reported in glomerular mesangial cells exposed to high-glucose and lipopolysaccharide (Feng et al., 2016). Long coding RNAs (lncRNAs) are also known to play important roles in several diseases. In a recent study by Yi et al. to evaluate the role of lncRNAs in DN, long intergenic noncoding RNA (lincRNA)-Gm4419 was found to be involved in inflammation and fibrosis in mesangial cells exposed to high-glucose through the NFκB/NLRP3 inflammasome signaling pathway (Yi et al., 2017). Glomerular apoptosis mediated by caspase-1-dependent inflammasome activation can also lead to DN (Shahzad et al., 2016a). Mitochondrial ROS is yet another potential mediator which activates the inflammasome in the diabetic milleu leading to DN (Shahzad et al., 2015). In a study to investigate the role of hyperuricemia in DN, Kim et al. reported that uric acid can stimulate the NLRP3 inflammasome in murine macrophages, leading to pro-inflammatory signaling in proximal tubular cells and thereby contributing to DN (Kim et al., 2015).

In DN, treatment with the DPP-4 inhibitor saxagliptin attenuated diabetes-induced activation of the inflammasome and delayed the progression of diabetic nephropathy in experimental models (Birnbaum et al., 2016). In fact, saxagliptin attenuated protein levels of ASC, NLRP3, TNFα, and caspase-1 in renal and adipose tissue (Birnbaum et al., 2016). Interestingly, the effects of saxagliptin on the NLRP3 inflammasome components and subsequent kidney remodeling were comparable in both type 1 and type 2 murine models of diabetes. This occurred despite saxagliptin having no effect on glycaemic control in type 1 diabetes, thereby inferring that the beneficial actions of saxagliptin may, at least in part, be independent of its established glucose lowering actions (Birnbaum et al., 2016).

Recently, Wang S. et al. (2017) reported that IL-22 gene therapy ameliorated renal damage, mesangial expansion and improved renal function in a mouse model of STZinduced experimental DN. Interestingly, IL-22 therapy inhibited NLRP3 activation, caspase-1 cleavage and IL-1β maturation suggesting an anti-inflammatory action mediated through NLRP3 suppression in the kidney (Wang S. et al., 2017). Furthermore, IL-22 dose-dependently reduced glucose-induced activation of NLRP3 inflammasome in renal mesangial cells suggesting an action independent of improved glycemic control (Wang S. et al., 2017). The glucose-lowering thiazolinedione, pioglitazone, was recently shown to ameliorate glomerular NLRP3 inflammasome activation in diabetic ApoE KO mice. Pioglitazone reduced the expression of AGEs, RAGE, and NFκB, leading to reduced NLRP3 and downstream pro-inflammatory mediators (Wang Y. et al., 2017).

Recent data now suggests a role for the redox-sensitive transcription factor nuclear erythroid 2-related factor 2 (Nrf2) in the protection against DN via inhibition of the NLRP3 inflammasome. The tetracycline antibiotic minocycline is now thought to afford anti-inflammatory effects in diabetic patients, and renoprotection in animal models of DN, via the Nrf2/NLRP3 axis (Shahzad et al., 2016b). Minocycline enhances Nrf2 levels by inhibiting Nrf2 ubiquitination (Shahzad et al., 2016b). Furthermore, the minocycline-mediated NLRP3 inflammasome inhibition and its subsequent therapeutic effect in DN was absent in diabetic Nrf2 KO mice (Shahzad et al., 2016b), thereby implicating Nrf2 protection at the level of the inflammasome. In addition, curcumin, a known Nrf2 activator, ameliorated endpoints of DN and improved kidney function by suppressing NLPR3 signaling in the db/db type 2 diabetic mouse model (Lu et al., 2017). Hence, given the emerging importance of the NLRP3 inflammasome in DN, as evidenced by the above studies, targeting the NLRP3 inflammasome may be a potential therapeutic strategy to effectively treat DN (Sakai and Wada, 2015).

### NEW THERAPIES THAT TARGET THE INFLAMMASOME

In the recent years there has been a paradigm shift in our understanding of inflammatory diseases. Dysregulation of inflammasome activity has been heavily implicated in autoimmune disorders, such as rheumatoid arthritis and cryopyrin-associated autoinflammatory syndrome (CAPS), which led to the development of several IL-1 and IL-18 decoy binding proteins and neutralizing antibodies, such as anakinra and canakinumab (Guo et al., 2015). However, it is becoming more apparent that the inflammasome-driven inflammation is also implicated in sterile inflammatory diseases, such as diabetes, metabolic syndrome, Alzheimer's disease and atherosclerosis, and that there is a need for therapies specifically targeting the mechanistic pathway of inflammasome activation to enhance clinical success. Current therapeutics that target the inflammasome include compounds such as β-hydroxybutyrate (BHB), MCC950, as well as caspase-1 and ASC inhibitors (Guo et al., 2015; Ozaki et al., 2015). We will review drugs/compounds that were originally developed for different indications, but have shown promise at modulating the inflammasome pathway in vivo and in vitro, as well as the latest results from the Canakinumab Antiinflammatory Thrombosis Outcome Study (CANTOS) trial.

### SGLT2 Inhibitors

In type 2 diabetic patients at high risk of experiencing cardiovascular events, treatment with sodium–glucose cotransporter 2 (SGLT2) inhibitors, a new class of glucoselowering agent, led to a lower rate of death from all cardiovascular causes in the EMPA-REG OUTCOMES trial (Zinman et al., 2015). SGLT2 inhibitors are considered to primarily act by decreasing renal glucose reabsorption via the early proximal tubules of the kidney, subsequently increasing urinary glucose excretion. Interestingly, they also have anti-inflammatory actions and repress the advancement of diabetic nephropathy (De Nicola et al., 2014). In experimental models of both type 1 and type 2 diabetes, treatment with ipragliflozin reduced levels of oxidative stress biomarkers and inflammatory markers in liver and plasma (Tahara et al., 2013), further supporting the premise that SGLT2 inhibition not only improves hyperglycaemia but also diabetes-associated metabolic and inflammatory abnormalities. Ye et al. (2017) were the first to report in vivo that SGLT-2 inhibition with a known inhibitor, dapagliflozin, supressed the structural and functional changes that occur in diabetic cardiomyopathy. The findings of this study provided partial mechanistic insight into the encouraging results of the EMPA-REG OUTCOMES trial (Zinman et al., 2015). Superficially, the clinical and experimental findings are somewhat surprising as SGLT2 is not expressed in the heart (Vrhovac et al., 2015), however the in vivo dapagloflozin findings were replicated in LPS-induced inflammasome activation studies in isolated cardiomyocytes (Ye et al., 2017), suggesting the beneficial effects are SGLT2-independent. Further interrogation of off-target mechanisms of the SGLT2 inhibitors may yield further mechanistic insights into the net protective actions of this new class of glucose-lowering therapy. For example, the inflammatory effects of dapagliflozin were shown to be at least in part AMPK-dependent (Ye et al., 2017), correlating with findings in other organs that AMPK activation suppresses the upregulation of the inflammasome (Lee et al., 2013; Bae et al., 2016).

Treatment of high glucose-stimulated human proximal tubular cells with empagliflozin supressed the expression of inflammatory and fibrotic markers, seemingly by blocking glucose entry into the cell (Panchapakesan et al., 2013). Furthermore, empaglifozin treatment in high fat fed mice reduced renal tubular damage and attenuated cardiac lipid accumulation, and this was associated with a decrease in NLRP3 inflammasome activation in the kidney (Benetti et al., 2016). Notably, in this study NLRP3 inflammasome activation in the heart was not increased in high fat fed mice after treatment (Benetti et al., 2016). In the vasculature, the SGLT2 inhibitor, dapagliflozin, lessened atherosclerotic lesions in the diabetic ApoE KO mouse, specifically reduced macrophage infiltration in the plaque and enhanced the stability of lesion (Leng et al., 2016). Mechanistically, this was correlated with lowered serum levels of IL-1β and IL-18, NLRP3 protein, and mitochondrial ROS in aortic tissue. Interestingly, inhibition of the NLRP3 pathway was not observed in isolated macrophages treated with dapagliflozin (Leng et al., 2016). Therefore, the exact signaling mechanism that elicits the inhibition of components of the NLRP3 inflammasome by SGLT2 inhibitors may be tissuespecific and requires further evaluation. Nevertheless, these findings clearly provide a rationale to further study SGLT2 inhibition on NLRP3 inflammasome activation in diabetes patients, and represents a target with considerable therapeutic promise.

### Glyburide

Another drug that is routinely used for the treatment of type 2 diabetes is glyburide, belonging to the sulfonylurea drug class. Glyburide acts as an inhibitor of ATP-sensitive K+ channels in pancreatic β cells (Ashcroft, 2005; Lamkanfi et al., 2009). However, in addition, Glyburide has demonstrated anti-inflammatory activity by inhibiting IL-1β, IL-18, caspase-1 activation and macrophage cell death (Lamkanfi et al., 2009). ATP signals through the P2X7 K+ channel to facilitate K+ efflux which is known to promote "signal 2" in inflammasome activation. Glyburide not only inhibited IL-1β production and caspase-1 activation downstream of the P2X7 receptor, but was also effective against ATP-, nigericin- and IAPP-induced inflammasome activation (Masters et al., 2010). Moreover, glyburide's anti-inflammatory action has been shown to be NLRP3-specific since other members of the inflammasome family, such as NLRC4- and NLRP1- were not inhibited in their ability to produce activated IL-1β (Lamkanfi et al., 2009). This is seen as advantageous as host defense against pathogens is not compromised. Additionally, glyburide treatment delayed LPS-induced endotoxic lethality in mice (Lamkanfi et al., 2009) and improved survival in diabetic patients that presented with gram-negative sepsis (Koh et al., 2011). Indeed, in the later study, the cohort that was taking oral glyburide had lower blood infection and better survival than patients that were not taking any drugs, collectively demonstrating glyburide's efficacy against inflammation of the immune system (Koh et al., 2011).

### Nrf2 Activators

As mentioned above, Nrf2 is the master regulator of endogenous anti-oxidant enzymes, however, more recently the role of this transcription factor in preventing inflammation has gained attention. In an elegant study by Liu et al. (2017) an exogenous Nrf2 activator, tert-butylhydroquinone (tBHQ), was shown to limit ROS production by upregulating NADPH quinone dehydrogenase 1 (NQO1), one of the downstream antioxidant enzymes that is modulated by Nrf2. This subsequently led to reduced NLRP3 activation, caspase-1 cleavage, IL-1β production and alum-induced peritonitis, an effect that was dependent on Nrf2-regulated ROS production (Liu et al., 2017). As mentioned above, the antibiotic minocycline is an Nrf2 activator that reduces renal NLRP3 inflammasome activation in pre-clinical models of diabetic nephropathy in a ROS-dependent manner (Shahzad et al., 2016b). Sulforaphane, a natural isothiocyanate present in cruciferous vegetables such as broccoli, is another known Nrf2 activator and was shown to reduce IL-1β production, however, it was not limited to NLRP3-induced IL-1β and could inhibit multiple inflammasome complexes including NLRP1, NLRC4, and AIM2(Greaney et al., 2016). It is important to note that there is a caveat to modulating Nrf2, as endogenous Nrf2 in certain situations has been shown to be a positive regulator of the NLRP3 inflammasome (Zhao et al., 2014; Garstkiewicz et al., 2017) and Nrf2 deficient mice on an ApoE-deficient background are protected against atherogenesis (Freigang et al., 2011). Equally intriguing, recent data by Kobayashi show an opposite effect where Nrf2 binds to a proximal region of the IL-1β and IL-18 gene to inhibit transcription (Kobayashi et al., 2016), whilst Nrf2 knockout mice on an LDL-R deficient background show increased atherogenesis (Barajas et al., 2011). These opposing effects of Nrf2, and compounds that activate Nrf2, reinforce the need for careful consideration of dosage, type of Nrf2 activator and genetic environment, and their impact on the NLRP3 inflammasome pathway.

### IL-1β Antibodies to Lessen Inflammation: The Cantos Trial

The recent reporting of primary outcomes in the CANTOS (Ridker et al., 2017a) trial has buoyed the field with respect to the link between inflammation and cardiovascular disease. The much anticipated first-of-its-kind randomized, doubleblind placebo-controlled trial to use a monoclonal antibody directed against IL-1β (known as canakinumab), involved 10,061 patients with previous myocardial infarction and indications of elevated inflammation (a high-sensitivity C-reactive protein level of ≥2 mg/L). Canakinumab was administered at three doses (50, 150, and 300 mg) subcutaneously every 3 months. All doses lowered inflammatory burden with a 26, 37, and 41% reduction in hsCRP reported respectively. A dosage of 150 mg was found to be optimal and led to significantly lower rates of recurrent cardiovascular events than placebo, and was independent of lipid lowering. However, canakinumab was associated with a higher incidence of fatal infections or sepsis in the pooled group of participants assigned to any active dose of canakinumab, suggesting that despite its positive effects on cardiovascular disease, long term inhibition of this important cytokine may compromise host immune defenses (Ridker et al., 2017a). Interestingly, in an additional analysis of the data, the authors report separately that IL-1β inhibition also led to significant reductions in the incidence of lung cancer and lung cancer mortality (Ridker et al., 2017b). Although not specifically reported on, 40% of patients in all 3 treatment arms of the CANTOS trial were diabetic, leading to speculation that an appropriate anti-inflammatory strategy could be clinically beneficial for diabetic patients (Tenenbaum and Fisman, 2017), although patients who died from infection were more likely to have diabetes. Importantly, the results of this trial have made significant progress in answering whether inflammation is a significant contributor to coronary artery disease, and have opened up a new frontier for novel therapies, besides the targeting of cholesterol, to reduce cardiovascular disease. Recent evidence from a prospective randomized trial of the dipeptidyl peptidase-4 (DPP4) inhibitor vildagliptin added to metformin treatment in T2D patients led to significant reductions in IL-1β levels, in addition to the lowering of HbA1c and hsCRP levels (Younis et al., 2017). This additional anti-inflammatory benefit of a drug designed to lessen glucagon to ultimately lower blood glucose, may additionally lessen cardiovascular outcomes in diabetic patients who are at high risk for these cardiovascular events.

### REFERENCES


### CONCLUSION AND PERSPECTIVES

This review highlights the interconnectivity between oxidative stress, NLRP3 activation and inflammation as it pertains to cardiac, vascular and renal injury sustained by diabetes. It is no longer appropriate to restrict our view of diabetic complications to one simply resulting from an abnormal metabolic state. The ability of the activated inflammasome to trigger cell death (incorporating pyroptosis, necroptosis or apoptosis) and autoinflammatory disease, as well as contributions from other regulating factors (TXNIP, IAPP, etc.) together with changes in substrate utilization, exemplify the complexity of the milieu from which these diabetic complications emerge. The development of glucose-optimizing agents, namely the DPP4 inhibitors that additionally lower inflammation and the sodium/glucose cotransporter (SGLT)2 inhibitors that furthermore confer benefits on cardiovascular outcomes, together with novel experimental approaches, highlight a new era in diabetes research, which is likely improve clinical outcomes for patients living with diabetes.

### AUTHOR CONTRIBUTIONS

AS and MT: Partially wrote manuscript, edited manuscript, designed, and constructed figures. GM, JV, and RR: Partially wrote manuscript, edited manuscript: JdH: Partially wrote manuscript, edited manuscript, edited figures.

### ACKNOWLEDGMENTS

This work was supported in part by both the National Health and Medical Research Council (NHMRC) of Australia, including APP1045140 (to RR), APP1101405 (to JV), and GNT1091003 (to AS), and the Victorian Government's Operational Infrastructure Support Program. RR is an NHMRC Senior Research Fellow (APP1059960). JdH acknowledges support from Reata Pharmaceuticals and a Baker Fellowship.

effects on plasma lipoproteins and cholesterol transport that overshadow antioxidant protection. Arterioscler. Thromb. Vasc. Biol. 31, 58–66. doi: 10.1161/ATVBAHA.110.210906


blocking proinflammatory cytokine transcription. Nat. Commun. 7:11624. doi: 10.1038/ncomms11624


saxagliptin, a DPP4 inhibitor. Cardiovasc. Drugs Ther. 31, 119–132. doi: 10.1007/s10557-017-6725-2


**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 Sharma, Tate, Mathew, Vince, Ritchie and de Haan. 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.

# Neutrophil: A Cell with Many Roles in Inflammation or Several Cell Types?

### Carlos Rosales\*

Departamento de Inmunología, Instituto de Investigaciones Biomédicas, Universidad Nacional Autónoma de México, Ciudad de México, Mexico

Neutrophils are the most abundant leukocytes in the circulation, and have been regarded as first line of defense in the innate arm of the immune system. They capture and destroy invading microorganisms, through phagocytosis and intracellular degradation, release of granules, and formation of neutrophil extracellular traps after detecting pathogens. Neutrophils also participate as mediators of inflammation. The classical view for these leukocytes is that neutrophils constitute a homogenous population of terminally differentiated cells with a unique function. However, evidence accumulated in recent years, has revealed that neutrophils present a large phenotypic heterogeneity and functional versatility, which place neutrophils as important modulators of both inflammation and immune responses. Indeed, the roles played by neutrophils in homeostatic conditions as well as in pathological inflammation and immune processes are the focus of a renovated interest in neutrophil biology. In this review, I present the concept of neutrophil phenotypic and functional heterogeneity and describe several neutrophil subpopulations reported to date. I also discuss the role these subpopulations seem to play in homeostasis and disease.

### Edited by:

Giovanni Li Volti, Università degli Studi di Catania, Italy

### Reviewed by:

Abhishek D. Garg, KU Leuven, Belgium David Sharkey, University of Adelaide, Australia

> \*Correspondence: Carlos Rosales carosal@unam.mx

### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 02 November 2017 Accepted: 05 February 2018 Published: 20 February 2018

### Citation:

Rosales C (2018) Neutrophil: A Cell with Many Roles in Inflammation or Several Cell Types? Front. Physiol. 9:113. doi: 10.3389/fphys.2018.00113 Keywords: neutrophil, bacteria, infection, inflammation, cancer

## INTRODUCTION

Neutrophils, also known as polymorphonuclear (PMN) leukocytes, are the most abundant cell type in human blood. They are produced in the bone marrow in large numbers, ∼10<sup>11</sup> cell per day. Under homeostatic conditions, neutrophils enter the circulation, migrate to tissues, where they complete their functions, and finally are eliminated by macrophages, all in the lapse of a day. Neutrophils are important effector cells in the innate arm of the immune system (Mayadas et al., 2014). They constantly patrol the organism for signs of microbial infections, and when found, these cells quickly respond to trap and kill the invading pathogens. Three main antimicrobial functions are recognized for neutrophils: phagocytosis, degranulation, and the release of nuclear material in the form of neutrophil extracellular traps (NETs) (**Figure 1**). These functions were considered, until recently, the only purpose of neutrophils. However, current research by investigators in several fields of neutrophil cell biology has revealed that neutrophils possess a much diverse repertoire of functional responses that go beyond the simple killing of microorganisms. Neutrophils respond to multiple signals and respond by producing several cytokines and other inflammatory factors that influence and regulate inflammation and also the immune system (Nauseef and Borregaard, 2014; Scapini and Cassatella, 2014). Nowadays it is recognized that neutrophils are transcriptionally active complex cells (Ericson et al., 2014) that produce cytokines (Tecchio and Cassatella, 2016), modulate the activities of neighboring cells and contribute to the resolution

of inflammation (Greenlee-Wacker, 2016), regulate macrophages for long-term immune responses (Chen et al., 2014), actively participate in several diseases including cancer (Uribe-Querol and Rosales, 2015; Mishalian et al., 2017), and even have a role in innate immune memory (Netea et al., 2016).

The multitude of neutrophil functional responses is induced by transcriptional activation and by changes in expression of surface molecules or activity. These phenotypic changes are usually detected in only a subset of neutrophils, suggesting that great neutrophil heterogeneity exists (Beyrau et al., 2012). Neutrophils display different phenotypes from the time they leave the bone marrow and enter the circulation (fresh neutrophils) to the time they disappear from the circulation (aged neutrophils). This shift in phenotype is known as aging, since it takes place within a single day, and results in various neutrophils with distinct properties (Adrover et al., 2016). In addition, the microenvironment in different tissues can induce neutrophils to acquire specialized functions. Thus, the fact that neutrophils can display many functional phenotypes further supports the existence of several neutrophil subsets.

Cancer is a particular condition, in which the number of neutrophils in circulation increases, and the phenotype of these cells changes along tumor progression. In advanced cancer, several subpopulations of circulating neutrophils with different characteristics of maturity, tumor cytotoxicity, and immune suppression have been described (Sagiv et al., 2015), including the granulocytic myeloid derived suppressor cells (G-MDSC). However, these different cell types are not clearly defined and their existence as bona fide neutrophil subsets or even a complete different cell type is today a controversial topic. For most researchers, it is clear that the differences in plasma membrane proteins and functional responses of neutrophils under diverse settings are strong evidence for the remarkable plasticity of neutrophils. Yet, this neutrophil heterogeneity is not supported by consensus criteria by which to define populations of neutrophils in blood and in tissues. Additionally, the stability of observed phenotypes in these subpopulations is not clear. Are these bona fide neutrophil subsets or just various activating stages displayed in response to local factors?

It is the purpose of this review to highlight the differences in functional responses of neutrophils and other cell types, such as G-MDSC, that may or may not be subpopulations of circulating neutrophils. We present the characteristics of the different neutrophil types, describe their functions, and discuss the possible relations among them.

### NEUTROPHIL LIFE CYCLE

Neutrophils represent about 70% of all leukocytes and more than 10<sup>11</sup> cells are produced every day in the bone marrow (Dancey et al., 1976). From there, neutrophils enter the blood where they circulate until they leave into tissues. Once neutrophils reach the end of their lifespan within tissues, they are cleared mostly by macrophages through the process of phagocytosis (Bratton and Henson, 2011). Despite this impressive turnover, the number of neutrophils in circulation remains relatively constant thanks to a fine balance between production and elimination (neutrophil homeostasis; von Vietinghoff and Ley, 2008). In addition, neutrophils actively change to be able to perform special functions at different times or places.

### Granulopoiesis

Neutrophils are produced in large numbers in the bone marrow from hematopoietic stem cells (Görgens et al., 2013). These cells differentiate into multipotent progenitor (MPP) cells that cannot self-renew themselves. MPPs then transform into lymphoidprimed multipotent progenitors (LMPPs), which differentiate into granulocyte–monocyte progenitors (GMPs). These GMPs, under control of the granulocyte colony-stimulating factor (G-CSF) commit to neutrophil generation by turning into myeloblasts (**Figure 2**). These cells then follow a maturation process that includes the stages of promyelocyte, myelocyte, metamyelocyte, band cell, and finally a mature neutrophil (von Vietinghoff and Ley, 2008) (**Figure 2**). During differentiation, the developing neutrophil changes its nucleus from a round shape into a banded and then a lobulated morphology, and also the expression of various receptors. The integrin α4β1 (VLA4) and the CXC chemokine receptor 4 (CXCR4) are downregulated, while CXCR2 and Toll-like receptor 4 (TLR4) are upregulated. The bone marrow stroma cells express vascular cell adhesion molecule 1 (VCAM1), a ligand for VLA4, and the chemokine stromal-derived factor-1/SDF-1 (CXCL12), a ligand for CXCR4, in order to retain the progenitor cells in the bone marrow. Mature neutrophils also contain granules and secretory vesicles that store specific proteins relevant to their functions (Häger et al., 2010). These granules are formed at particular differentiation stages. Primary (azurophil) granules are found at the myeloblast to promyelocyte stage. Secondary (specific) granules are detected at myelocyte and metamyelocyte stages. Tertiary (gelatinase) granules are found at the band cell stage. Finally, secretory vesicles are detected only in mature neutrophils (**Figure 2**). These granules store an arsenal of antimicrobial enzymes, including elastase, myeloperoxidase, cathelicidins, defensins, and matrix metalloproteinases, which are used to destroy invading pathogens.

### Neutrophil Exit from the Bone Marrow

Once neutrophils mature they can leave the bone marrow into circulation. The release of neutrophils is tightly controlled since only 1 or 2% of all neutrophils in the body are found in the blood under normal homeostatic conditions. Mature neutrophils are kept in the bone marrow through the action of two chemokine receptors, CXCR2 and CXCR4. Osteoblasts and other bone marrow stromal cells produce CXCL12 and keep CXCR4-expressing neutrophils in the bone marrow. G-CSF induces neutrophil exit from the bone marrow by interfering with the CXCR4-CXCL12 interaction (Summers et al., 2010). In addition, ligands for CXCR2, such as CXCL1, CXCL2, CXCL5, and CXCL8 (in humans) are expressed by endothelial cells outside the bone marrow when neutrophils need to be mobilized into the blood (Eash et al., 2010; Köhler et al., 2011). G-CSF prompts the release of neutrophils by inducing upregulation of CXCR2 ligands on megakaryocytes (Köhler et al., 2011), reduced expression of CXCL12 by bone marrow stroma cells (Petit et al., 2002; Semerad et al., 2005), and also reduced expression of CXCR4 on neutrophils themselves (Kim et al., 2006).

Outside the bone marrow, neutrophil production is also regulated by a cytokine network that involves interleukin (IL)-23 produced by phagocytes and IL-17 produced by T lymphocytes. In this mechanism, macrophages and also dendritic cells phagocytose apoptotic neutrophils (Gordy et al., 2011; Jiao et al., 2014) leading to a reduction on IL-23 (Stark et al., 2005), which in turn controls expression of IL-17 by T lymphocytes (Gaffen et al., 2014). Because, IL-17 promotes granulopoiesis and neutrophil release by up-regulation of G-CSF (von Vietinghoff and Ley, 2008), the lower levels of IL-17 then result in reduced expression of G-CSF, and a steady-state release of neutrophils. During

inflammation, IL-1 can also stimulate neutrophil production through the IL-17-G-CSF axis (Ueda et al., 2009), and neutrophils themselves create a positive loop for neutrophil recruitment. Neutrophils can produce IL-17 (Eskan et al., 2012), and attract IL-17-producing T lymphocytes (Th17 cells) (Weaver et al., 2013). In turn, Th17 cells recruit more neutrophils (Pelletier et al., 2010; Zenobia and Hajishengallis, 2015). Recently, it was also found that microbiota can induce neutrophil production by increasing IL-17 production (Deshmukh et al., 2014).

### Neutrophil Trafficking and Clearance

Neutrophils from the blood can be mobilized to sites of infection or inflammation through the process known as the leukocyte adhesion cascade (Ley et al., 2007; Chavakis et al., 2009). Endothelial cells of blood vessels close to the affected site get activated and express adhesion receptors such as E-, and P-selectins. These receptors bind glycoprotein ligands on neutrophils, causing them to roll on the endothelium. Next, the neutrophil is activated by chemokines, which induce a high affinity state in β2 integrins. Binding of integrins to their ligands such as intercellular adhesion molecule-1 (ICAM-1) and ICAM-2 on endothelial cells causes firm adhesion of the neutrophil. Next the neutrophil transmigrates into peripheral tissues (Hajishengallis and Chavakis, 2013). Once neutrophils are in the peripheral tissues, they follow gradients of chemoattractans such as formyl-methionyl-leucyl-phenylalanine (fMLF), and the anaphylatoxin C5a to complete their functions (Kolaczkowska and Kubes, 2013).

The constant numbers of neutrophils in the circulation are also controlled by central signals delivered by the sympathetic nervous system. In this case, adrenergic nerves induce the temporal expression of adhesion molecules on endothelial cells, allowing neutrophils to bind the endothelium and leave the circulation (Scheiermann et al., 2012). This regulation of neutrophil migration into tissues follows a circadian pattern (Scheiermann et al., 2013; see below).

Once in tissues, neutrophils undergo apoptosis and are finally cleared through phagocytosis by resident macrophages and dendritic cells. Senescent neutrophils in blood upregulate expression of CXCR4, which allows them to return to the bone marrow for final clearance (Martin et al., 2003). The clearance of apoptotic neutrophils is also important for controlling neutrophil production in the bone marrow (Stark et al., 2005). Phagocytosis of apoptotic neutrophil triggers an anti-inflammatory response characterized by a reduction in IL-23 by macrophages. As described above, less IL-23 leads to reduced IL-17 levels and to less G-CSF production, and finally, in consequence to reduced granulopoiesis (Stark et al., 2005).

### NEUTROPHIL SUBPOPULATIONS IN HEALTH

The previous description of the neutrophil life cycle gives the idea that these cells are produced in the bone marrow, go to the circulation, migrate to sites of infection or inflammation, execute their antimicrobial functions, and then quietly are cleared by tissue-resident macrophages. This simple view for neutrophils being only pathogen-killing cells, is far from the complex behavior neutrophils actually display. It was already mentioned that in fact neutrophils are transcriptionally active cells with the potential to change the expression of several membrane molecules, and to produce cytokines (Tecchio et al., 2014; Tecchio and Cassatella, 2016), and consequently neutrophils are capable of performing different cell functions depending on the tissues where they are found (Borregaard, 2010; Mayadas et al., 2014; Nauseef and Borregaard, 2014). Thus, neutrophils do not seem to be a homogeneous population that always behaves the same. In fact, the existence of several subpopulations of neutrophils has been suggested in various conditions of health and disease.

### Neutrophil Subpopulations in the Circulation

In normal conditions, neutrophils remain in circulation for just few hours (the half-life is estimated at 6–12 h) before they leave into tissues (Summers et al., 2010). During this time, neutrophils appear to change their phenotype. Looking at neutrophils in circulation of healthy mice every 4 h over the course of a day, it was found that these cells change their morphology and phenotype (Casanova-Acebes et al., 2013). Freshly released neutrophils from the bone marrow undergo several changes that accumulate until the cells begin to move from the circulation into the tissues. These changes in neutrophil phenotype from the time they are released from the bone marrow (fresh neutrophils) until they leave the circulation (aged neutrophils) in the absence of inflammation are referred to as aging (Adrover et al., 2016). The number of total neutrophils oscillates in a circadian way. Fresh neutrophils are released into the blood when the mice begin their activity phase, and aged neutrophils are cleared at the end of their resting stage (Casanova-Acebes et al., 2013).

Previous in vitro studies on neutrophil aging indicated that there is a spontaneous upregulation of the receptor CXCR4 in cells that are kept in culture. Freshly isolated blood neutrophils present this increase in CXCR4 expression after only 4 h in culture (Nagase et al., 2002). As mentioned before, the chemokine CXCL12 produced by bone marrow stroma cells functions as a retention signal for neutrophils. Thus, re-expression of CXCR4 on neutrophils is thought to encourage "senescent" neutrophils to return to the bone marrow (Martin et al., 2003). However, in mice with CXCR4-deficient myeloid cells, no changes in neutrophil clearance were observed (Eash et al., 2009), suggesting that other organs besides the bone marrow also contribute to neutrophil clearance (Adrover et al., 2016). In addition, neutrophils in culture also downregulate the expression of CXCR2, which has opposite effects to CXCR4 and promotes the release of cells from the bone marrow (Eash et al., 2010). Therefore, it seems that these phenotypic changes prepare neutrophils to leave the circulation into tissues.

In vivo studies have identified other phenotypic changes in neutrophils. In mice, neutrophils that were forced experimentally to stay longer in the circulation presented lower expression of Lselectin (CD62L), together with a higher expression of CXCR4 (Casanova-Acebes et al., 2013). These aged neutrophils appeared in the circulation following circadian oscillations over time. They increased in numbers during the day (when mice are at rest), and disappeared in the evening, when mice begin their active phase (Casanova-Acebes et al., 2013). Other molecules are also expressed al higher levels on the cell membrane of these aged neutrophils, including CD11b (αM) and CD49d (α4), the alpha subunits for integrins Mac-1 and VLA4, respectively. These integrins are involved in adhesion to activated endothelial cells at sites of inflammation. In contrast, the expression of the molecule CD47, a "don't eat me" signal for apoptotic cells (Jaiswal et al., 2009), was reduced on the membrane of these neutrophils. This would suggest that aged neutrophils are more easily phagocytosed by macrophages. However, in a more recent study, the expression of CD47 was not reduced (Zhang et al., 2015). More recently, the surface expression of other molecules has also been found increased in aged neutrophils. Some of these molecules include TLR4, ICAM-1, CD11c, CD24, and CD45 (Zhang et al., 2015) (**Figure 3**). Opposite to this, the mouse neutrophil marker Ly6G was also reduced in aged cells (Zhang et al., 2015) (**Figure 3**). Together with the changes in surface expression of these molecules, the cells presented morphological changes. Aged neutrophils are smaller, contain fewer granules, and display a granular multilobullar nucleus (Casanova-Acebes et al., 2013; **Figure 3**).

In addition, transcriptomic analysis of aging neutrophils revealed that several signaling pathways are different from those active pathways in fresh neutrophils. Signaling related to cell activation, microbial detection, adhesion, migration, and cell death is altered (Zhang et al., 2015). These changes in aged neutrophils are similar to changes in neutrophils at inflammation sites. Thus, it might be that aged neutrophils are in an activated state (Adrover et al., 2016). Together these reports suggest that neutrophils spontaneously change their phenotype in circulation, such that different surface molecules may facilitate migration into tissues. At present, it is not known how these differences in receptor expression may control neutrophil clearance into particular tissues, but a similar mechanism to the one controlling neutrophil exit from and return to the bone marrow is likely to be at work. Identifying particular neutrophil phenotypes will certainly help us understand how these cells are directed to various parts of the body.

### Neutrophil Subpopulations in Tissues

In normal homeostasis, neutrophils are found in many tissues (von Vietinghoff and Ley, 2008, 2009), where they perform specialized functions. For the most part, there is very little knowledge on how neutrophils are directed to the different organs, and the particular functions they perform in each tissue. Evidence suggests that indeed neutrophils display different phenotypes in various organs. For example, in the lung a large number of neutrophils accumulate adhered to the vascular lumen and in the interstitial space. These neutrophils are kept in the tissue by a CXCR4-dependent mechanism (Devi et al., 2013). Because aged neutrophils express more CXCR4 it is possible that they preferentially migrate to the lungs (Adrover et al., 2016). Similarly, in the spleen many neutrophils are found in the marginal zone, producing cytokines that induce somatic hypermutation and antibody production by marginal B lymphocytes (Puga et al., 2011). Also, neutrophils in spleen display the phenotype CD62Llow CD11bhi ICAM-1hi, and have a tendency to produce NETs (Cerutti et al., 2013), similar to aged neutrophils in the circulation (Summers et al., 2010; Casanova-Acebes et al., 2013).

Another subpopulation of neutrophils has been found to preferentially move into lymph nodes to interact with T lymphocytes, carrying antigens to the lymph nodes and mediate T cell activation (Duffy et al., 2012; Hampton and Chtanova, 2016). These neutrophils can selectively migrate to the lymph node because they express the CCR7 receptor (Beauvillain et al., 2011), as well as the integrin LFA-1 and the chemokine receptor CXCR4. These receptors are also involved in neutrophil trafficking through afferent lymphatics (Gorlino et al., 2014).

Yet another subtype of neutrophils in tissues is capable of inducing angiogenesis. These neutrophils display the phenotype CD49dhi CXCR4hi VEGFR1. The latter is the receptor for vascular endothelial growth factor-A (VEGF-A). These neutrophils are efficiently recruited to non-vascularized tissues under hypoxia conditions, and promote angiogenesis (Massena et al., 2015). Interestingly, this proangiogenic subset of neutrophils is similar to neutrophils that support tumor vascularization (Jablonska et al., 2010; see later).

The examples described above indicate that indeed neutrophils can display great phenotypic and functional heterogeneity, and support the idea that at least some of the neutrophil subtypes may derive from aging neutrophils migrating into tissues.

### Migrating Neutrophil Subpopulations

Some neutrophils have been observed to perform reverse transendothelial migration (rTEM) or reverse interstitial migration (rIM), depending on their initial location (Nourshargh et al., 2016). Neutrophils doing rTEM have been observed in mice (Woodfin et al., 2011), whereas neutrophils doing rIM have been seen only in the transparent zebra fish model (Mathias et al., 2012). Neutrophils performing rTEM present the phenotype ICAM-1hi CXCR1low, which is different from the phenotype ICAM-1low CXCR1hi of circulatory neutrophils and the phenotype ICAM-1low CXCR1low of neutrophils in tissues (Buckley et al., 2006).

### Neutrophil Subpopulations Induced by the Microbiota

Neutrophils can modify their functional responses after being exposed to multiple factors, through the process named neutrophil priming (Downey et al., 1995; El-Benna et al., 2016). Recent studies in mice suggest that microbial products derived from the microbiota can induce neutrophil diversity. For example, diaminopimelic acid-bearing peptides, which are recognized by the nucleotide—binding oligomerization domain containing 1 (NOD1) receptor, modify the lifespan of neutrophils (Hergott et al., 2016) and prime neutrophils for improved antimicrobial function (Clarke et al., 2010). In addition, endotoxins from the gut microbiota can enter the blood

circulation and influence neutrophil aging (Zhang et al., 2015) and also the development of B cell-helper neutrophils in the spleen (Puga et al., 2011). Thus, neutrophils can display different phenotypes after priming by the microbiota.

aged neutrophils are then cleared from the blood by migration into tissues or by returning to the bone marrow.

### NEUTROPHIL SUBPOPULATIONS IN DISEASE

As discussed earlier, it is becoming increasingly apparent that neutrophils are much more than just microbe-killing cells. They can display several phenotypes and perform a wide array of cellular functions. Several subsets of neutrophils are found in tissues under homeostatic conditions. We still have much to learn on how these different neutrophil subtypes are generated and recruited to tissues. In addition, various subsets of neutrophils with distinct properties are also detected in pathological conditions particularly in inflammation and in cancer (Silvestre-Roig et al., 2016; Yang et al., 2017).

### Neutrophil Subpopulations in Inflammation

Neutrophils are the first cell type recruited to sites of inflammation. From there, they can switch phenotypes and generate various subpopulations with different cell functions. Neutrophils can also interact, directly, or via cytokines and chemokines, with other immune cells to modulate both innate and adaptive immune responses. There is not a complete understanding of these subpopulations of neutrophils, but some clear examples showing that bona fide inflammatory subsets occur are mentioned next.

Upon infection with an antibiotic-resistant Staphylococcus aureus, two clear subsets of murine neutrophils can be observed. They differ in cytokine production, macrophage activation potential, expression of TLR, and expression of surface molecules. These subsets were named PMN-1 and PMN-2 (Tsuda et al., 2004). PMN-1 cells produce IL-12; classically activate macrophages, express TLR2/TLR4/TLR5/TLR8, and CD49dhi CD11blow. In contrast, PMN-2 cells produce IL-10; alternatively activate macrophages, express TLR2/TLR4/TLR7/TLR9, and CD49dlow CD11bhi (Tsuda et al., 2004). In systemic inflammation condition, another subset of neutrophils is generated with low doses of endotoxin. These cells have a hypersegmented nucleus and display the phenotype CD62low CD11bhi CD11chi, which is similar to the one described for murine aged neutrophils (Casanova-Acebes et al., 2013; Zhang et al., 2015). Also, they are capable of inhibiting T lymphocytes by direct cell contact involving the integrin Mac1, and by local delivery of reactive oxygen species (ROS) (Pillay et al., 2012).

In certain organs such as liver and adipose tissue, few neutrophils are detected in normal homeostatic conditions. However, upon an inflammatory state induced by experimental obesity, neutrophil numbers increase rapidly and a metabolic imbalance is slowly generated (Talukdar et al., 2012). First, neutrophils release elastase from azurophilic granules. This enzyme can destroy insulin receptor substrate 1 (IRS1) in adipocytes and hepatocytes, and in consequence induce insulin resistance and lipogenesis (Talukdar et al., 2012). Supporting a direct role for this neutrophil subtype in metabolic disorders was the observation that altered levels of elastase or its inhibitor (1-antitrypsin) are associated with metabolic syndrome and the onset of diabetes (Mansuy-Aubert et al., 2013).

### Neutrophil Subpopulations Induced by Metabolic Deregulation

As mentioned before, neutrophil effector functions are markedly enhanced after priming. When certain metabolic functions are altered, neutrophils can be primed to present stronger proinflammatory responses.

In hyperglycemia human and mouse neutrophils are primed to undergo NETosis (Wong et al., 2015). First, neutrophils in circulation respond to high glucose levels by releasing S100 calcium-binding proteins A8 (S100A8) and S100A9, which interact with the receptor for advanced glycation end products (RAGE), and induce macrophages and GMPs in the bone marrow to secrete G-CSF (Kraakman et al., 2017). In consequence, production of neutrophils is enhanced (Xiang et al., 2012). These new released neutrophils are primed for ROS production and formation of NETs (Wong et al., 2015). Similarly, during hypercholesterolemia, neutrophils showed a primed state characterized by elevated ROS production, increased release of myeloperoxidase (MPO) and increased expression of CD11b (Mazor et al., 2008).Together, these reports show that hyperglycemia and hyperlipidemia generate primed, proinflammatory neutrophils that may contribute to diabetes, adipose tissue inflammation, and cardiovascular inflammation.

### Neutrophil Subpopulations and NET Formation

NETosis, the process for producing NETs can be activated by multiple types of microorganisms (Fuchs et al., 2007; Yipp et al., 2012). Yet, the capacity of neutrophils to undergo NETosis can vary with physiological states, suggesting a neutrophil diversity that could be clinically relevant. In fact, several reports indicate that NETs can influence thrombosis (Fuchs et al., 2010) and vascular inflammation (Kessenbrock et al., 2009; Chistiakov et al., 2015), cancer (Berger-Achituv et al., 2013; Garley et al., 2016) and autoimmunity (Stephenson et al., 2016). As mentioned before some metabolic conditions associated with states of chronic inflammation, can increase neutrophil predisposition to form NETs. Hence, neutrophils from diabetic patients (Wong et al., 2015) and from systemic lupus erythematosus (SLE) patients (Garcia-Romo et al., 2011; Villanueva et al., 2011) have been shown to be more prone to NET formation.

Nowadays NETs have been described as a player of several pathophysiological processes, including vascular diseases, such as atherosclerosis and venous thrombosis (Qi et al., 2017; Bonaventura et al., 2018), and inflammatory pathologies, such as gout and pancreatitis (Hahn et al., 2016).

Atherosclerosis is a cardiovascular disease accompanied by chronic vascular wall inflammation and endothelial cell dysfunction (Gisterå and Hansson, 2017). Hyperlipidemia can damage endothelial cells, promoting lipid deposition and plaque formation. This usually characterizes the onset of atherosclerosis. Hyperlipidemia and also hypercholesterolemia induce neutrophilia, which is positively associated with atherosclerotic plaque burden (Drechsler et al., 2010). Neutrophils attach themselves to atherosclerotic plaques, primarily through NETs formation, where cholesterol crystals function as danger signals, inducing IL-1β-mediated NETs release from neutrophils. Then, components of NETs, such as cathepsin G and cathelicidin-related antimicrobial peptide (CRAMP), can attract monocytes and macrophages to plaques (Döring et al., 2012; Wang et al., 2014). NETs can also regulate cytokine production from macrophages in atherosclerosis (Warnatsch et al., 2015), and induce endothelial dysfunction directly by activation and damage of endothelial cells (Knight et al., 2014). Furthermore, proteinases from NETs contribute to plaque instability (Hansson et al., 2015). These reports indicate that NETs directly participate in rupture of atherosclerotic plaques (Döring et al., 2017), which triggers platelet aggregation and fibrin deposition at the initial site of atherothrombosis. After plaque rupture, thrombin-activated platelets interact with neutrophils at the injured site inducing more formation of NETs (Stakos et al., 2015). Thus, neutrophils and NETs are major contributors to atherothrombosis. Different from arteries, thrombosis in veins is usually initiated by endothelial injury (Di Nisio et al., 2016) triggered by alteration in the blood flow or endothelial dysfunction (Xu et al., 2017). Subsequently, damaged endothelial cells secrete massive amounts of von Willebrand factor and P-selectin, which adhere to platelets and recruit leukocytes (Etulain et al., 2015; Michels et al., 2016). Platelets then interact directly with neutrophils and promote the production of NETs (Clark et al., 2007). NETs can also stimulate the activation of coagulation cascades (Brill et al., 2012) and not only platelet adhesion (Massberg et al., 2010), but also erythrocyte adhesion (Fuchs et al., 2010). Reciprocally, NETs induce endothelial cell activation through NET-derived proteases, histones, and defensins (Saffarzadeh et al., 2012), creating a positive feedback loop for thrombosis. These findings and studies in mice showing an association between the risk of venous thrombosis and high neutrophil counts (Ramacciotti et al., 2009), confirm that NETs make a substantial contribution to maintenance of venous thrombi.

Another inflammation process in which NETs play an important role is pancreatitis. Acute pancreatitis is an inflammatory disorder of pancreas for which no specific treatment is available. Important risk factors of acute pancreatitis are formation of gallstones and alcohol abuse (Spanier et al., 2008). The most severe cases of the disease are associated with mortality, with acute respiratory distress syndrome being the most frequent cause of death in the early phase of the disease (Pandol et al., 2007). Obstruction of the pancreatic duct causes blockage of pancreatic secretion, which is accompanied by disorders in organelle function of pancreatic acinar cells (Gukovskaya et al., 2016). These disorders promote colocalization of zymogens-containing vesicles and lysosomes leading to formation of co-localization organelles. In co-localized organelles cathepsin B activates trypsinogen to trypsin (Halangk et al., 2000; Van Acker et al., 2002), which mediates acinar cell death and thus causing severe inflammation of the pancreas (Lankisch et al., 2015; Manohar et al., 2017). The destroyed tissue will eventually be replaced by fatty tissue, typical of chronic pancreatitis (Braganza et al., 2011), if the original acute pancreatitis is not resolved (Braganza et al., 2011). Neutrophils infiltrate the pancreatic parenchyma during this acute inflammatory response (Lankisch et al., 2015), and can augment trypsinogen activation via ROS. Both trypsin activation and pancreatic injury were reduced in NADPH oxidasedeficient mice(Gukovskaya et al., 2002). In addition, infiltrating neutrophils aggravate inflammation by releasing NETs in the pancreas and at the sites of systemic injury, namely, the lungs (Merza et al., 2015). NETs incubated in vitro with pancreatic acinar cells led to trypsin activation in these cells, degradation of the NETs by treatment with deoxyribonuclease (DNAse) abolished the trypsin activation, reduced local acinar damage, and systemic inflammation (Merza et al., 2015). Furthermore, neutrophils enter the pancreatic ducts and there they form large deposits of NETs, also known as aggregated NETs (aggNETs). AggNETs in turn obstruct secretory flow and thereby perpetuate inflammation (Leppkes et al., 2016). Therefore, NETs formed after an initial inflammatory stimulus become the activators of further inflammation in the pancreas.

Contrary to the situations mentioned above, Gout is a disease where NETs appear to have a positive effect. Gout is an acute inflammatory reaction originated from precipitation of uric acid in the form of needle-shaped monosodium urate (MSU) crystals (So and Martinon, 2017). Aggregates of MSU crystals known as tophi, induce inflammation in the joints and tissues. Local immune cells such as macrophages and dendritic cells take up the crystals via phagocytosis. The MSU-containing phagosomes then fuse with lysosomes. The low pH in phagolysosomes causes a massive release of sodium and consequently raises intracellular osmolarity, which is balanced by passive water influx through aquaporins. This process dilutes intracellular sodium and potassium concentrations. The low potassium is a trigger for activation of NLRP3 inflammasomes (Schorn et al., 2011). Inflammasome activation by MSU crystals has been considered a response to "danger," since damaged cells release urate and ATP into the environment (Busso and So, 2012). As a consequence of inflammasome activation, proIL-1β is cleaved to release active IL-1β and other pro-inflammatory cytokines (Kingsbury et al., 2011; So and Busso, 2014). Cytokine release then leads to a rapid and dramatic recruitment of neutrophils. Recruitment of neutrophils is further mediated by CXCR2, CXCL-8, CXCL-1, CXCL-2, and CXCL-3 (Terkeltaub et al., 1998). This neutrophil influx is accompanied by the infamously intense clinical symptoms of inflammation during an acute gout attack (So and Martinon, 2017). Due to the intense local inflammation, cytokines produced in large quantities can also enter the circulation, resulting in an acute phase response that can trigger fever and leukocytosis (Maueröder et al., 2015). Continuous recruitment of neutrophils to the site of inflammation results in very high neutrophil densities (Shah et al., 2007). After the neutrophil concentration in the tissue exceeds a certain threshold, NETs begin to aggregate and build aggNETs in which the MSU crystals are embedded in a mesh of DNA and proteins from neutrophil granules. As such, these aggNETs block MSU crystals and also trap and degrade pro-inflammatory mediators by serine proteases attached to the DNA fibers (Schauer et al., 2014). MSU crystal-induced aggNET formation is augmented by release of ATP and lactoferrin from activated neutrophils. The release of ATP during NETs formation is of high importance since extracellular nucleotides initiate antiinflammatory clearance of dead cells by mononuclear phagocytes (Elliott et al., 2009). In addition, lactoferrin on NETs abrogates further recruitment of neutrophils and thus contributes to the anti-inflammatory action of NETs in highly infiltrated tissues (Bournazou et al., 2009). Clearly, in this case NETs have a positive anti-inflammatory effect.

The inflammation processes described before show that NETs are much more than a simple anti-microbial tool and they can be produced in very different conditions of neutrophil activation. However, NETs are a double-edged sword. On the one hand, "bad" NETs are involved in stimulating inflammation, as shown in the obstruction of blood vessels and pancreatic ducts. On the other hand "good" NETs are able to contribute to the resolution of inflammation as shown in gout (Hahn et al., 2016).

In addition, neutrophils primed by microbiota-derived products can form NETs more easily than neutrophils newly released from the bone marrow (Zhang et al., 2015). In contrast, the formation of NETs can be blocked by phagocytosis of small microorganisms via the C-type lectin receptor Dectin-1, which acts as a sensor of microorganism size (Branzk et al., 2014). Dectin-1 downregulates the translocation of neutrophil elastase (NE) to the nucleus. This protease promotes NETosis by degrading histones in the nucleus (Branzk et al., 2014). Also, neutrophils that phagocytose apoptotic cells, lose their capacity to up-regulate β2 integrins and to respond to activating stimuli that induce NETs formation (Manfredi et al., 2015). This could explain in part why in conditions in which phagocytosis of apoptotic cells is compromised, such as SLE and anti-neutrophil cytoplasmic antibody (ANCA)-associated vasculitis, a state of persistent inflammation is observed. These findings imply that heterogeneity in neutrophil function (for example for NETs formation) is also regulated by physiological signals.

It is evident that infection and inflammation can regulate the appearance of neutrophil phenotypes with unique properties. Unfortunately, nowadays we can only (incompletely) describe the function of these neutrophil subtypes. It would become important to characterize these cells at the level of surface markers, functional responses, and transcriptional profiles in order to understand their role in multiple diseases.

### Neutrophil Subpopulations in Cancer

The changes in neutrophil phenotype during cancer are perhaps the most impressive and best studied so far. Neutrophils play important and contradictory roles in cancer development, as reflected by several recent reviews (Sionov et al., 2015; Swierczak et al., 2015; Uribe-Querol and Rosales, 2015; Coffelt et al., 2016; Mishalian et al., 2017). In tumor-bearing mice, the number of circulating neutrophils increases along with tumor progression. Similarly, in patients with advanced cancer counts of neutrophils in blood are also increased. It is not clear how tumors can induce neutrophilia, but a common mechanism seems to be the production by tumors of cytokines that influence granulopoiesis, including G-CSF (McGary et al., 1995), IL-1, and IL-6 (Lechner et al., 2010). The presence of elevated numbers of neutrophils in the circulation is associated with poor outcome in several types of cancers (Schmidt et al., 2005). In addition, the presence of neutrophils in tumors also seems to be an indicator of poor outcome (Sionov et al., 2015). For this reason, the counts of neutrophils in blood in relation to other leukocytes have been suggested as a prognostic value in cancer. Therefore, the neutrophil to lymphocyte ratio (NLR) was introduced as a simple and inexpensive biomarker for many types of cancer (Peng et al., 2015; Faria et al., 2016). In general, the blood NLR is high in patients with more advanced or aggressive cancers (Guthrie et al., 2013), and correlates with poor survival of patients with many solid tumors (Paramanathan et al., 2014; Templeton et al., 2014). Despite the simplicity for using the NLR, it has not been accepted in many clinical settings. One reason for this is that neutrophilia can be the result of elevated granulopoiesis and as a consequence, it is not always a bad sign for cancer progression. Another reason is that neutrophilia does not correlate with poor clinical outcome in all types of cancer. In gastric cancer, for example a high NLR is indicative of positive prognosis (Caruso et al., 2002). This is indicative of the great plasticity neutrophils have. They can directly kill tumor cells and control cancer (Yan et al., 2014), but they can also acquire a pro-tumor phenotype and favor cancer (Fridlender and Albelda, 2012). Therefore, the exact role of neutrophils within the tumor is a controversial matter (Sionov et al., 2015; Uribe-Querol and Rosales, 2015)**.**

### Myeloid-Derived Suppressor Cells (MDSC)

In several types of cancer, not only an increase in the number of neutrophils in blood is observed, but also an increase in immature myeloid cells (Brandau et al., 2013). These immature cells are at various stages of differentiation, accumulate in the spleen of tumor-bearing animals, and present an immunosuppressive phenotype that supports tumor progression (Nagaraj et al., 2010; Raber et al., 2014; Keskinov and Shurin, 2015). For this reason, they were named myeloidderived suppressor cells (MDSC) (Peranzoni et al., 2010). These MDSC are a heterogeneous mixture of cells that can at least be divided into two subgroups: the granulocytic (G-MDSC) and the monocytic (Mo-MDSC) subgroups (Raber et al., 2014). The G-MDSC group resembles neutrophils. Hence, some researchers considerer them to be a bona fide phenotype of neutrophils (Pillay et al., 2013). However, the relationship among these cells is not clear since immature neutrophils do not have immunosuppressive properties (Solito et al., 2017), and neutrophils in the circulation are differentiated cells characterized by a lobulated nucleus (Pillay et al., 2013); while MDSCs are cells with clear immature morphology, including band or myelocyte-like nuclei (Pillay et al., 2013) (**Figures 2**, **4**).

Murine neutrophils are defined as CD11b<sup>+</sup> Ly6G<sup>+</sup> cells (Daley et al., 2008). Murine G-MDSC are CD11b<sup>+</sup> and Ly6G+, thus they are considered neutrophils. In contrast Mo-MDSC express CD11b and Ly6C (Brandau et al., 2013; Keskinov and Shurin, 2015), making them more monocytic- than neutrophillike cells. In humans the problem is more complex, since the Ly6G antigen does not exist. Human mature neutrophils are defined by the phenotype CD14<sup>−</sup> CD15<sup>+</sup> CD16<sup>+</sup> CD66b<sup>+</sup> (Dumitru et al., 2012). MDSC share all these markers, making it impossible to differentiate these cells from mature neutrophils. An extended panel of six markers (adding CD11b, CD33, and HLA-DR) was used to evaluate human MDSC (Damuzzo et al., 2015). Mo-MDSC were described as CD11b<sup>+</sup> CD14<sup>+</sup> CD15<sup>−</sup> CD33<sup>+</sup> CD66b<sup>+</sup> HLA-DR−/low; while G-MDSC were described as CD11b<sup>+</sup> CD14<sup>−</sup> CD15<sup>+</sup> CD33<sup>+</sup> CD66b<sup>+</sup> HLA-DR<sup>−</sup> (Brandau et al., 2013; Favaloro et al., 2014; Keskinov and Shurin, 2015). Although, these markers show differences between Mo-MDSC and G-MDSC, they do not allow for clear separation of these cells from neutrophils. At present, it is not possible to distinguish whether MDSC are indeed subpopulations of neutrophils or a separate cell type (Solito et al., 2017),

but an international effort continues to find better ways for identification of these cells by flow cytometry (Mandruzzato et al., 2016).

### Low-Density Neutrophils (LDNs)

An interesting subpopulation of neutrophils is the so-called low-density neutrophils (LDNs). Traditionally, neutrophils are purified by a Ficoll density gradient (Böyum, 1968; García-García et al., 2013) where they appear at the bottom of the tube (high-density fraction), separated from mononuclear cells, which are found at the interphase of plasma and Ficoll (low-density fraction) (**Figure 4**). In contrast, LDNs are found in the lowdensity fraction (Sagiv et al., 2015, 2016) (**Figure 4**). Interestingly, the proportion of LDNs in the low-density fraction increases with tumor growth and progression (Mishalian et al., 2017), and includes cells with mature and immature neutrophil morphology (Sagiv et al., 2015) (**Figure 4**). These LDNs are a subpopulation of neutrophils with characteristics and functions not well described.

Although, LDNs were first reported in the blood of patients with SLE, rheumatoid arthritis, or rheumatic fever (Hacbarth and Kajdacsy-Balla, 1986), they attracted attention only recently because they seem to be associated with cancer (Brandau et al., 2011; Sagiv et al., 2015). These LDNs have been found in many other pathological conditions including sepsis (Morisaki et al., 1992), psoriasis (Lin et al., 2011), HIV infection (Cloke et al., 2012), asthma (Fu et al., 2014), ANCA-associated vasculitis (Grayson et al., 2015), and malaria (Rocha et al., 2015). In addition, LDNs have been reported in natural pregnancy (Ssemaganda et al., 2014). During pregnancy, downregulation of T cell functions is required to ensure materno-fetal tolerance. One way to inhibit T cell function is through the enzyme arginase, which depletes <sup>L</sup>-arginine, an essential amino acid required for proper expression of the T cell receptor CD3 ζ chain and for T cell proliferation (Rodriguez et al., 2004; Raber et al., 2012). Arginase activity is significantly increased in the peripheral blood of pregnant women and also in term placentae (Kropf et al., 2007). The source for arginase in these tissues was identified as LDNs with the phenotype CD15<sup>+</sup> CD33<sup>+</sup> CD66b<sup>+</sup> CD16low (Ssemaganda et al., 2014). This phenotype is suggestive of a activated neutrophil (Fortunati et al., 2009), and is similar to the phenotype of G-MDSC (Favaloro et al., 2014; Keskinov and Shurin, 2015). Thus, situations of chronic inflammation and immunosuppression appear to induce neutrophil diversity. Very little is known about the function of these subpopulations of neutrophils. LDNs from SLE patients were also reported to readily form NETs (Villanueva et al., 2011), and because these NETs presented autoantigens, it has been suggested that LDNs in these patients are responsible for sustaining chronic inflammation leading to autoimmunity (Garcia-Romo et al., 2011; Khandpur et al., 2013). Similarly, it was recently reported that the number of CD66b<sup>+</sup> LDNs was markedly elevated in peritoneal cavity after abdominal surgery of gastric cancer (Kanamaru et al., 2018). These LDNs readily formed NETs that selectively attached cancer cells (Kanamaru et al., 2018). These NETs could then assist the clustering and growth of free tumor cells disseminated in the abdomen.

The origin of LDNs remains unclear. Since LDNs are a mixture of cells with segmented or banded nuclei and myelocytelike cells, one thought is that LDNs are immature neutrophils that are released from the bone marrow during chronic inflammation or immunosuppression (Denny et al., 2010; Carmona-Rivera and Kaplan, 2013). Another possibility is that these LDNs are activated neutrophils that have undergone degranulation and therefore they have a reduced density (Rocha et al., 2015; Deng et al., 2016). In mouse models of cancer, these LDNs seem to derive either from immature cells released by the bone marrow (Youn et al., 2008) or from normal-density neutrophils (Sagiv et al., 2015). It is important to notice that these LDNs are still not properly characterized. The immunosuppressive function of these cells has not been directly determined and the transition of mature (normal density) to LDNs appears to involve an increase in volume rather than degranulation (Sagiv et al., 2015). Thus, most likely these LDNs are not activated normal neutrophils. In addition, since SLE patients have chronic inflammation, it is unlikely that their LDNs present immunosuppressive activity. All these possibilities need to be further studied in the future. However, one serious limitation for the characterization of these cells is that there are not specific molecular markers that could distinguish among these possible neutrophils subpopulations.

### Tumor-Associated Neutrophils (TANs)

The phenotypic changes of circulating neutrophils during tumor progression are also related to infiltration of neutrophils into tumors. Unfortunately, the relationship among immunosuppressive cells (MDSC), LDNs, high-density (normal) neutrophils, and tumor-associated neutrophils (TANs) is just beginning to be elucidated (Mishalian et al., 2013; Uribe-Querol and Rosales, 2015; **Figure 5**).

Depending on the phenotype displayed by TANs in tumorbearing mice, they have been classified as N1 or N2 (Fridlender et al., 2009). This classification is analogous to antitumor tumorinfiltrating macrophages (M1) or protumor macrophages (M2) (Galdiero et al., 2013). Murine N1 TANs are proinflammatory and antitumorigenic. In contrast, N2 TANs are protumorigenic (Fridlender et al., 2009). When tumor-bearing mice were treated to inhibit transforming growth factor-beta (TGF-β functions, the CD11b<sup>+</sup> Ly6G<sup>+</sup> neutrophils recruited to tumors were hypersegmented, more cytotoxic and more proinflammatory (N1). In contrast, in presence of TGF-β, had a protumor (N2) phenotype. Therefore, it seems that TGF-β within the tumor microenvironment induces a population of TANs with a protumor activity (Fridlender et al., 2009). This means that TANs can display an antitumor (N1) phenotype or a pro-tumor (N2) phenotype depending on the tumor microenvironment (Sionov et al., 2015). In addition, in tumor-bearing animals the LDNs increased progressively in circulation, were not cytotoxic, and had reduced expression of cytokines (Sagiv et al., 2015). The authors proposed that these immunosuppressive LDNs (G-MDSC) are the source of N2 TANs (Sagiv et al., 2015) (**Figure 5**). This is consistent with the idea that some of the LDNs are indeed immature neutrophils (Sagiv et al., 2015). Also, in the same murine model mature neutrophils were capable of becoming LDNs upon treatment with TGF-β (Sagiv et al., 2015). This

FIGURE 5 | Neutrophils in the circulation display different phenotypes. Mature (normal) neutrophils (PMN) leave the bone marrow and display the classical pro-inflammatory and anti-tumor properties of these cells. It is thought that these PMN can migrate into tumors and display an anti-tumor (N1) phenotype. In tumor-bearing mice, immature neutrophils, such as band cells, also leave the bone marrow into the circulation. These "low-density" neutrophils include granulocytic myeloid derived suppressor cells (G-MDSC) and neutrophils with immunosuppressive properties. These cells can infiltrate tumors and display pro-tumor (N2) phenotype. Under the influence of transforming growth factor-beta (TGF-β, normal PMN can change into "low-density" neutrophils. The exact origin of recruited neutrophils is not known. Also, it is not clear if N1 cells can change into N2 cells and vice versa under the influence of the tumor microenvironment.

has been interpreted as normal mature neutrophils having the capacity to infiltrate tumors and being cytotoxic, but under the influence of TGF-β, being able to change into N2 cells (Mishalian et al., 2013) (**Figure 5**). Although attractive, this hypothesis requires further evidence to be validated, since TGF-β was able to induce this change in tumor-bearing mice, but it had no effect on tumor-free mice (Sagiv et al., 2015).

It is important to emphasize here that the paradigm of N1 and N2 TANs has been described only in murine models of cancer, and that the nature and function of TANs in the tumor microenvironment remains largely unknown, particularly with human tumors. There are only two reports on isolation of human TANs, but they describe controversial data on their immunosuppression capacity. In one report, TANs were isolated from digested human lung tumors (Eruslanov et al., 2014). These TANs had an activated phenotype (CD62Llow CD54hi) and produced l proinflammatory cytokines. In consequence, these TANs stimulated T cell proliferation and interferon gamma (IFN-γ) production (Eruslanov et al., 2014). In another report, TANs were isolated from a colorectal tumor and found to present a typical neutrophil morphology. These TANs had the phenotype CD45<sup>+</sup> Lin<sup>−</sup> HLADR<sup>−</sup> CD11b<sup>+</sup> CD33<sup>+</sup> CD66b+, and were classified as G-MDSC (Wu et al., 2014). These TANs secreted arginase 1 (ARG1) and ROS, and inhibited proliferation of activated autologous T cells and IFN-γ production (Wu et al., 2014). Together these reports suggest that neutrophils are present in three subpopulations in cancer: normal high-density neutrophils, immature LDNs (G-MDSC), and large mature LDNs (**Figure 5**). Early TANs (N1) are not immunosuppressive, but rather stimulate T cell responses (Eruslanov et al., 2014). Latter, the cells acquire an N2 phenotype and become immunosuppressive (Wu et al., 2014) (**Figure 5**).

### Immunogenic Cell Death

Recently, another important role of neutrophils in anticancer therapy has been described. Immunogenic cell death can stimulate neutrophils to utilize cytotoxicity against residual live cancer cells after therapy. The concept of immunosurveillance explains how only the most immunoevasive or highly mutagenic neoplastic cells are able to generate clinically relevant tumors (Dunn et al., 2002; Schreiber et al., 2011). Yet, the immune system is capable of recognizing altered-self molecules in damaged cells through endogenously derived danger signals or alarmins (Bianchi, 2007; Garg et al., 2014). As a consequence of cell death, alarmins, collectively referred to as "damage-associated molecular patterns" (DAMPs), enhance sensing of dying cells by innate immune cells (Zitvogel et al., 2010; Garg et al., 2015b). Release of DAMPs can either be achieved in an unregulated fashion by necrosis (Aaes et al., 2016) or in a regulated fashion by immunogenic cell death, also called immunogenic apoptosis (Kepp et al., 2014; Garg et al., 2015a). Anti-cancer therapyinduced cancer cell death can be subdivided into three distinct types i.e., tolerogenic cell death, inflammatory cell death, and immunogenic cell death. The details of these types of cell death are beyond the scope of the present publication, but the reader is directed to some excellent recent reviews (Green et al., 2009; Garg et al., 2016; Garg and Agostinis, 2017). Briefly, immunogenic cell death is characterized by a defined discharge of DAMPs, type I interferon response, and the production of pathogen responselike chemokines (CXCL1, CCL2, and CXCL10) (Garg et al., 2017b) that together raise the immunogenic potential of dying cancer cells.

Through immunogenic cell death, after some types of anticancer therapy (Galluzzi et al., 2012; Bezu et al., 2015; Garg et al., 2017a), a damaged cancer cell produces specific inflammatory chemokines to recruit neutrophils as first innate immune cells (Kolaczkowska and Kubes, 2013; Garg et al., 2017b). The damaged cancer cells also express two important "eat me" signals on their membranes, namely, phosphatidylserine, and calreticulin. These signals trigger neutrophil phagocytosis of cancer cells and pro-inflammatory stimulation, leading to a change in neutrophil phenotype (Garg et al., 2017b). Neutrophils expressed a mature phenotype characterized by expression of CD86hi MHC-IIhi, and an activated phenotype characterized by IL-6hi IL-1β hi IL-10low expression (Garg et al., 2017b). As a result, neutrophils stimulated by immunogenic cell death showed cytotoxicity against residual live cancer cells (Garg et al., 2017b). Thus, neutrophils interacting with immunogenic apoptotic cells gain a pro-inflammatory profile, culminating into neutrophil dependent cytotoxicity against residual cancer cells.

### CONCLUSION

There is a lot of excitement around the many "new" neutrophil functions just discovered recently. The realization that neutrophils do in fact perform many more functions than just antimicrobial responses, and the fact that neutrophils with different phenotypes have been reported in various tissues and pathological conditions, suggest that indeed different neutrophils exist. However, in most reports the evidence is circumstantial and we are in need for solid experimental proof that the cells described are in fact novel neutrophil subsets. What we have learned for sure so far is that in various pathological conditions, particularly cancer, distinct populations of mature and immature neutrophils are found in circulation. After gradient density centrifugation of blood, the mature high-density (or more properly normal-density) neutrophils mostly represent cells with a pro-inflammatory phenotype, while the LDNs are comprised of immature neutrophils and "activated" mature neutrophils. These cells in turn may migrate to tumors and become, at least in mice, N1 and N2 TANs (**Figure 5**). However, we have to highlight that the actual cell type responsible for the immunosuppressive properties of MDSC remains a mystery.

Many questions remain open, but at least two topics seem to be relevant at the moment. One important topic that needs to be addressed is whether mature neutrophils in circulation can be reprogrammed by external stimuli, or whether defined phenotypes are programmed in the bone marrow and neutrophils exit with a particular phenotypic signature. Evidence suggests that neutrophils are very plastic cells and in consequence the various subtypes described seem to be acquired in the tissues. However, these possibilities need to be formally tested. Another relevant topic is that the many functions described have not been assigned to particular phenotypes of neutrophils. This remains a complex issue, as there are currently no appropriate molecular markers to readily identify these different neutrophil subpopulations. This confusing scenario is the fuel for new and even more exciting research. We expect to learn new tricks from our favorite cell type in the near future.

### AUTHOR CONTRIBUTIONS

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

### REFERENCES


### FUNDING

Research in the author's laboratory was supported by grant 254434 from Consejo Nacional de Ciencia y Tecnología (CONACyT), Mexico.

### ACKNOWLEDGMENTS

The author thanks Francis Valeria Eliosa García for preparing the list of references, and Dr. Eileen Uribe-Querol for preparing the figures.


stem cells and leukemia cells to avoid phagocytosis. Cell 138, 271–285. doi: 10.1016/j.cell.2009.05.046


from an interim study. Cancer. Immunol. Immunother. 65, 161–169. doi: 10.1007/s00262-015-1782-5


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

# The Role of Inflammation in Age-Related Sarcopenia

#### Sebastiaan Dalle<sup>1</sup> , Lenka Rossmeislova<sup>2</sup> and Katrien Koppo<sup>1</sup> \*

<sup>1</sup> Exercise Physiology Research Group, Department of Kinesiology, KU Leuven, Leuven, Belgium, <sup>2</sup> Department for the Study of Obesity and Diabetes, Third Faculty of Medicine, Charles University, Prague, Czechia

Many physiological changes occur with aging. These changes often, directly or indirectly, result in a deterioration of the quality of life and even in a shortening of life expectancy. Besides increased levels of reactive oxygen species, DNA damage and cell apoptosis, another important factor affecting the aging process involves a systemic chronic low-grade inflammation. This condition has already been shown to be interrelated with several (sub)clinical conditions, such as insulin resistance, atherosclerosis and Alzheimer's disease. Recent evidence, however, shows that chronic low-grade inflammation also contributes to the loss of muscle mass, strength and functionality, referred to as sarcopenia, as it affects both muscle protein breakdown and synthesis through several signaling pathways. Classic interventions to counteract age-related muscle wasting mainly focus on resistance training and/or protein supplementation to overcome the anabolic inflexibility from which elderly suffer. Although the elderly benefit from these classic interventions, the therapeutic potential of anti-inflammatory strategies is of great interest, as these might add up to/support the anabolic effect of resistance exercise and/or protein supplementation. In this review, the molecular interaction between inflammation, anabolic sensitivity and muscle protein metabolism in sarcopenic elderly will be addressed.

#### Edited by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

#### Reviewed by:

Gautham Yepuri, University of Fribourg, Switzerland Rui Curi, University of São Paulo, Brazil

#### \*Correspondence:

Katrien Koppo katrien.koppo@kuleuven.be

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 15 September 2017 Accepted: 29 November 2017 Published: 12 December 2017

#### Citation:

Dalle S, Rossmeislova L and Koppo K (2017) The Role of Inflammation in Age-Related Sarcopenia. Front. Physiol. 8:1045. doi: 10.3389/fphys.2017.01045 Keywords: muscle wasting, muscle protein metabolism, NSAID, sarcopenia, protein supplementation, resistance training, inflammation

## INTRODUCTION

Aging is generally associated with numerous changes that may, directly or indirectly, affect health and/or life span. One of the major problems in the aging population is a progressive loss in skeletal muscle mass, muscle strength, and/or functionality, described as age-related sarcopenia. The elderly suffering from this multifactorial pathological condition are at risk of adverse outcomes such as physical disability, injuries, frailty, social exclusion, hospitalization and eventually an increased mortality (Cruz-Jentoft et al., 2010; Visser and Schaap, 2011). Several strategies to attenuate the loss of muscle mass and other muscle impairments that comes with aging have been developed (Sakuma and Yamaguchi, 2012; Denison et al., 2015). However, none of these have been proven successful to fully reverse the muscle wasting condition. Given the high prevalence of sarcopenia in the aging population and the associated high health care costs, it is of importance to reveal and elucidate the working mechanisms which underlie muscle protein metabolism in the elderly, in order to optimize the classic interventions and/or to develop new ones.

Muscle protein metabolism is carefully regulated by counterbalanced fluctuations in muscle protein breakdown (MPB) and muscle protein synthesis (MPS) (Churchward-Venne et al., 2014).

**83**

In elderly, the balance between MPB and MPS seems to be disturbed, which progressively increases the loss of skeletal muscle mass (Churchward-Venne et al., 2014). Many underlying factors such as hormonal changes, decreased activity, diminished nutrient intake, and neuronal changes were reported in the literature (Dirks et al., 2006; Budui et al., 2015), but lately, the role of inflammation on the regulation of muscle protein metabolism has gained more and more interest among gerontologists. Generally, aging is associated with a chronic state of slightly increased plasma levels of pro-inflammatory mediators, such as tumor necrosis factor α (TNFα), interleukin 6 (IL-6) and C-reactive protein (CRP) (for an overview see **Table 1**). This state is often referred to as a low-grade inflammation (LGI) and is, at least partly, the manifestation of increased numbers of cells leaving the cell cycle and entering the state of cellular senescence. Indeed, senescent cells acquire a Senescence-Associated Secretory Phenotype, which induces the production of pro-inflammatory cytokines (TNFα, IL-6 and an overactivation of NF-κB) (Tchkonia et al., 2013). Moreover, there is a growing interest in the association between the telomere/telomerase system and LGI, as cellular senescence can be triggered by critically short telomeres, representing irreparable DNA damage (Kordinas et al., 2016). Also, there are indications that LGI can directly cause telomere/telomerase dysfunction, enforcing the vicious LGI circle and stimulating an accelerated aging phenotype (Jurk et al., 2014).

Although it has been suggested that inflammatory mediators affect muscle protein metabolism (Frasca and Blomberg, 2016), it is not fully understood to what extent and through which signaling pathways they induce muscle wasting. Populationbased data suggest that circulating concentrations of IL-6 and TNFα are significantly elevated in sarcopenic elderly (Bian et al., 2017) and it was reported that higher IL-6 and CRP levels increase the risk of muscle strength loss (Schaap et al., 2006). In a 10-year longitudinal study in community-dwelling elderly, plasma concentrations of TNFα, IL-6, and IL-1 were shown to be strong predictors of morbidity and mortality in older subjects (Baylis et al., 2013). Furthermore, systemic inflammation was also reported as one of the primary mediators of skeletal muscle wasting in diseases such AIDS, COPD, chronic heart failure and cancer (Sakuma et al., 2015) and it was shown to accelerate aging in general (Jurk et al., 2014). Without pronouncing on causality, these findings suggest that there is a link between inflammatory mediators and muscle mass and function.

This review is particularly concerned with the recent progress that has been made in understanding the role of inflammation in age-related sarcopenia and in muscle protein metabolism in general. In the first sections, pathways of both MPB and MPS will be described and related to inflammation. The following sections will give an overview of the classic approaches (i.e., aerobic/resistance exercise, protein supplementation) that attempt to attenuate age-related sarcopenia. Finally, interventions that target the administration of compounds with anti-inflammatory properties (vitamin D, poly-unsaturated fatty acids, non-steroidal anti-inflammatory drugs) will be discussed as an alternative approach.

### MUSCLE PROTEIN BREAKDOWN AND ITS REGULATION BY INFLAMMATION

As mentioned above, elevated levels of pro-inflammatory plasma cytokines and acute-phase proteins are often observed in sarcopenic elderly. Despite strong correlations between inflammatory markers and the risk for functional decline and mortality in elderly, conclusive evidence on a causative link between these markers and age-related sarcopenia is hard to establish (Reuben et al., 2003; Schaap et al., 2006). Recent evidence, however, suggests that chronic LGI is, at least partially, involved in the onset and/or progression of age-related muscle wasting (Beyer et al., 2012; Argilés et al., 2015). Higher levels of inflammatory markers were reported to negatively affect skeletal muscle metabolism through direct catabolic or indirect mechanisms, such as decreases in growth hormones (Sakuma et al., 2015). Though, the challenge remains to further uncover the specific molecular mechanisms through which inflammation interacts with muscle protein metabolism.

Acute inflammation models are widely applied to study the molecular mechanisms which link inflammation and muscle protein metabolism. In rodents, the effect of lipopolysaccharide injection has been shown to induce muscle catabolism through inflammatory signaling (Schakman et al., 2012). Accordingly, blunting an acute infection-induced inflammation improved muscle performance in elderly (Beyer et al., 2011). However, it should be noted that acute high-grade inflammation has a limited translational potential compared to the LGI present in the aged population.

In general, proteolysis is regulated through four main pathways (Combaret et al., 2009; Fanzani et al., 2012; Bowen et al., 2015). The contribution of each pathway is context- (e.g., stress vs. basal) or tissue- (e.g., brain vs. muscle) dependent (de Oliviera Nunes Teixeira et al., 2012). Their mechanisms are also inter-dependent and might co-regulate proteolysis on a different hierarchical level (Dirks et al., 2006). In the following section, the four main proteolytic pathways (ubiquitin proteasome, calpains, macro autophagy, apoptosis) are briefly discussed, as well as their interaction(s) with inflammatory signaling.

## ATP-Dependent Ubiquitin-Proteasome Pathway

The ATP-dependent ubiquitin-proteasome pathway is responsible for degrading the majority of proteins (80–90%) (Lilienbaum, 2013). The proteolytic core of the 26S complex, a 20S proteasome, degrades ubiquitin-conjugated proteins after binding with a regulatory 19S particle. This conjugation to ubiquitin marks abnormal (misfolded, denaturated) or short-lived proteins for rapid hydrolysis (Rock et al., 1994). Ubiquitination of specific substrates is closely regulated by ubiquitin protein (E3) ligases, such as MuRF1 and MAFbx, specifically in muscle cells (Bodine and Baehr, 2014). Evidence indicates that the expression of both E3 ligases is upregulated in several models of muscle atrophy, such as immobilization, denervation, hind limb unloading, dexamethasone treatment, and IL-1-induced cachexia (Bodine and Baehr, 2014). The role of



TNFα, tumor necrosis factor α; IL-6, interleukin 6; hs-CRP, high-sensitive C-reactive protein.

the ubiquitin-proteasome pathway in sarcopenia, however, is less clear (Bowen et al., 2015; Fan et al., 2016; Wilson et al., 2017). Some studies reported elevated levels of MAFbx and MuRF1 in aged rat muscle (Clavel et al., 2006; Altun et al., 2010), while others observed decreased levels of atrogenes with aging (Erik et al., 2017). Likewise, in human studies, some authors reported no differences in skeletal muscle protease activity between different age groups (Bossola et al., 2008), while others noticed higher mRNA levels of proteins encoding protease components, ubiquitin-like proteins and proteins related to ubiquitin function in elderly compared to younger adults (Welle et al., 2003). Furthermore, elevated levels of the ubiquitin protein in old human muscles mainly occurred in fast-twitch muscles, such as extensor digitorum longus, while no differences were detected in the slow-twitch soleus muscle (Cai et al., 2004). Dissimilarities between studies might therefore be attributed to the studied muscle type, as well as the specific study population, as it was suggested that age-related muscle wasting is mechanistically different from muscle wasting-related confounders such as disuse, bedrest or disease (Erik et al., 2017). A recent review suggested that elevated levels of pro-inflammatory mediators (due to LGI), such as TNFα and IL-6, might upregulate this proteolytic pathway through activation of FOXO3a, which regulates the ubiquitin-proteasome system (Xia et al., 2017). It remains to be elucidated, however, whether low levels of these pro-inflammatory mediators are (on the long-term) sufficient to induce this proteolytic pathway. To conclude, the ubiquitinproteasome pathway is likely to be involved in the regulation of muscle wasting with aging, partially due to its upregulation with inactivity. On the other hand, findings on the association between LGI and the ubiquitin-proteasome pathway are scarce and need to be further elucidated.

### Calpains

Calpains are a family of cysteine proteases with a wide range of cellular calcium-regulated functions. They are responsible for the proteolysis of several substrates, including cytoskeletal and membrane proteins, enzymes and transcription factors (Dargelos et al., 2008). Furthermore, calpains are also involved in apoptosis through two major pathways, triggered by cell death signals (TNF) or endoplasmatic reticulum (ER) stress, respectively (Dargelos et al., 2008). Reactive oxygen species (ROS) accumulation with advanced age is able to upregulate calpain activity directly (Nakashima et al., 2004) or indirectly by inducing Ca2<sup>+</sup> release from the ER or oxidation of MAPK-dependent pathways (Kefaloyianni et al., 2006). Although the evidence is scarce, the contribution of calpains to age-related muscle wasting is not excluded. Higher calpain availability and activity, and lower availability of its endogenous inhibitor (calpastatin), were observed in muscles of old (23 months) compared to young rats (3 months) and treatment of dystrophic muscle with calpain inhibitors attenuated muscle degeneration (Dargelos et al., 2007). It was reported that intracellular [Ca2+] were elevated in aged rodent muscle cells, potentially contributing to an increased calpain activation with aging (Andersson et al., 2011). To our knowledge, there are no data which compare calpain acitivity/availability in old vs. young human muscles. Calpains can also promote inflammation through several mechanisms leading to NFκB (pro-inflammatory transcription factor) activation and the production of pro-inflammatory cytokines (Ji et al., 2016). However, it is unclear whether calpains are involved in muscle proteolysis through regulation of inflammation in aged muscles. Therefore, more studies are required to further elaborate the mechanistic association between calpains and age-related muscle wasting in humans.

### (Macro)Autophagy Pathway

The (macro)autophagy pathway is as a housekeeping mechanism responsible for the clearance of dysfunctional organelles and damaged macromolecules (e.g., protein aggregates). The autophagosome, a double-membrane vesicle, encloses macromolecules or organelles for delivery to the lysosome, in which hydrolases stand in for the degradation into recycled "building blocks," e.g., amino acids. The lysosomal protease cathepsin L was shown to be involved in muscle proteolysis upon acute infection (Deval et al., 2001). Both in vitro and in vivo studies reported that elevated levels of the pro-inflammatory cytokine IL-6 is a first mechanism which links inflammation to cathepsin-induced muscle atrophy (Ebisui et al., 1995; Tsujinaka et al., 1995). Whether cathepsin L plays a crucial role in agerelated muscle wasting is less straightforward, as higher levels of cathepsin L were observed in the soleus muscle of old (30 months) vs. young (3 months) rats (Pattison et al., 2003), while no different expression patterns were found in the gastrocnemius muscle of old (30 months) and young (3 months) rats (O'Connell et al., 2007).

Autophagic activity in the fast-twitch plantaris muscle decreases with aging, resulting in insufficient clearance and accumulation of intracellular waste products, such as lipofuscin, protein aggregates and damaged mitochondria (Wohlgemuth et al., 2010). In general, autophagy is accepted to be a promising target to counteract sarcopenia (Fan et al., 2016). Until now, physical exercise and caloric restriction (CR) were proven successful in preventing/attenuating the age-related decrease in autophagic activity. White et al. (2016) reported that long-term voluntary resistance wheel exercise in elderly mice increased basal autophagy (increased LC3III/I ratio, marker of autophagy) compared to sedentary age-matched controls, while MuRF1 and MAFbx remained unaffected. Another study found that a mild CR (8%) positively affected muscle composition, oxidative stress, cell death and autophagy in old rats, suggesting that mild CR might be an applicable intervention to combat age-related sarcopenia, given a sufficiently qualitative energy intake (Wohlgemuth et al., 2010). In human studies, markers of autophagy were downregulated in Duchenne muscular dystrophy patients when compared to healthy controls, suggesting a decreased autophagic activity in these patients (Palma et al., 2012). Although evidence is scarce, Jiao and Demontis (2017) suggested that comparable mechanisms may apply to the development of age-related muscle wasting in elderly.

Autophagy is also involved in the regulation of the production of inflammatory mediators upon acute inflammatory stimuli (e.g., LPS) (Cadwell, 2016). However, to our knowledge, a crosstalk between autophagy and LGI was not yet established in the literature. It is conceivable that inflammation contributes to a dysregulation of the autophagic/mitophagic activity by stimulation of oxidative stress (Li et al., 1999; Suematsu et al., 2003). Nevertheless, it remains to be clarified whether LGI is a sufficiently strong signal for inducing oxidative stress in muscle cells. Furthermore, upregulation of the classic NF-κB signaling is capable of inhibiting autophagy by suppression of autophagyrelated genes (atg5 and beclin 1), but this mechanism was not confirmed yet in old skeletal muscle cells.

To summarize, skeletal muscle autophagy is downregulated with advanced aged and indirectly contributes to muscle wasting due to insufficient clearance of intracellular waste products or damaged organelles, such as mitochondria. An accumulation of damaged mitochondria and cellular waste generally induces oxidative stress, which enforces a pro-inflammatory environment in favor of catabolic processes.

### Apoptosis

Apoptosis is another main player contributing to the onset and progression of sarcopenia. Apoptosis, the process of programmed cell death, generally occurs during development and aging as a homeostatic mechanism to maintain cell populations in tissues (Elmore, 2007). Dirks and Leeuwenburgh (2002) reported higher levels of apoptosis (+50%) (measured by nucleosome fragmentation) in 24 months old rats compared to 6 months young rats. The same research group also looked at the mechanisms through which age-related apoptosis affected muscle wasting in skeletal muscles of old rats (Marzetti et al., 2008). They concluded that muscle weight declined progressively with advancing age, concomitant with increased apoptotic DNA fragmentation. A higher susceptibility of aged type II muscle fibers to TNFα-stimulated apoptotic signaling partially explains the greater loss of fast twitch muscle fibers with aging (Phillips and Leeuwenburgh, 2005).

In skeletal muscle, two main routes are described to induce apoptosis, i.e., an internal and an external pathway. In the (mitochondrial-dependent) internal pathway, nonreceptor-mediated stimuli in the cytosol, such as elevated ROS and Ca2<sup>+</sup> levels, can directly act upon mitochondrial homeostasis, increasing their permeability (Nitahara et al., 1998). Consequently, pro-apoptotic proteins are released from the intermembrane space into the cytosol. In the external pathway, ligands (such as TNFα) bind to their respective receptor and initiate the recruitment of adaptor proteins, eventually resulting in a death-induced caspase-cascade. As TNFα protein levels are generally elevated in both circulation and skeletal muscle in elderly (Greiwe et al., 2001), this pathway might also play a role in age-related muscle wasting. On the other hand, no apoptosis of muscle cells was observed in inflammatory myopathies, which questions the role of this pathway in age-related LGI (Migheli et al., 1997). Furthermore, it is not fully clear which apoptotic pathway is the main player in age-related muscle wasting. Future research should further reveal the contribution of each apoptotic pathway in order to develop targeted anti-apoptotic strategies. Similar to autophagy, exercise and mild CR were found to be effective in attenuating apoptosis with aging (Dirks and Leeuwenburgh, 2004; Phillips and Leeuwenburgh, 2005; Wohlgemuth et al., 2010). Despite the extensively studied association between age-related muscle wasting and apoptosis in rodents, evidence emerging from human samples are currently lacking to our knowledge. Therefore, human data studying the role of apoptosis in sarcopenia are required in order to see whether the mechanisms/observations are comparable to rodent models. As concluded by Marzetti and Bernabei (2012), targeting apoptosis might be an effective intervention to counteract age-related muscle wasting.

To conclude, several pathways are involved in the regulation of MPB in sarcopenia. However, it should be noted that possible other pathways might be active and remain to be confirmed in sarcopenic human muscles. These pathways contribute to muscle wasting to a certain extent, which depends on the unique context of each muscle fiber, i.e., fiber type, (im)mobility, oxidative damage, etc. Until now, no studies focused on the relative contribution of each proteolytic pathway. Therefore, future research should elucidate the contribution of each proteolytic pathway, taking into account the specific research sample/population/condition, in order to allow the development of effective and specific strategies, which target one or more proteolytic mechanisms.

### MUSCLE PROTEIN SYNTHESIS AND ITS REGULATION BY INFLAMMATION

Besides its putative role in MPB, inflammation negatively affects MPS. Already in 1984, Klasing and Austic (1984) established that high-grade inflammatory challenges, such as E. coli injection, decreased the in vivo MPS in broilers. Also, a more prolonged exposure to a high-grade inflammatory milieu, e.g., sepsis, reduced MPS, through decreased activation of the mTORC1 signaling pathway, and hence muscle mass, as reviewed by Frost and Lang (2011). The mechanistic link between low-grade inflammation and the downregulation of aged-related MPS, however, is less understood.

Muscle protein metabolism is the result of a balance between an increased post-prandial whole body protein synthesis and a decreased post-absorptive protein synthesis (Dardevet et al., 2012). This regulation is mainly orchestrated by the mTORC1 signaling pathway that integrates environmental stimuli to control cellular growth (Deldicque et al., 2005). In the muscle, many mTORC1 stimuli are described, including amino acids (leucine in particular), insulin, hormones released during mechanical stimuli as well as contractions per se, while energy deprivation and stress/hypoxia inhibit mTORC1 activity. Therefore, combined resistance exercise and protein supplementation has been proven effective in stimulating muscle growth in adults (Churchward-venne et al., 2012). Under basal conditions, MPS does not appear to be compromised in elderly, but its response to physiologic stimuli, such as amino acids, exercise or insulin (rather as a permissive mediator) is blunted (Dardevet et al., 2012; Drummond et al., 2012; Haran et al., 2012). This phenomenon is called "anabolic resistance" and implicates the need for higher intakes of amino acids and/or higher loads of resistance exercise, which is difficult to meet in a frail population that generally suffers from a loss of appetite (Rennie, 2009; Wysokinski et al., 2015).

### Protein Intake As Anabolic Signal Animal Studies

Inflammation is an important underlying factor that contributes to the insensitivity to anabolic signals. The research group of Dardevet confirmed the interference of inflammatory background (NF-κB, TNFα, IL-6 etc.) with anabolic signaling in several animal studies. They observed that rats which developed LGI at 25 months were unresponsive, in their MPS, to food intake, while in rats without LGI MPS was significantly increased (Balage et al., 2010). Another study revealed that rats, whose LGI was attenuated by ibuprofen (NSAID) treatment, exhibited a restored protein anabolism at post-prandial state (Rieu et al., 2009). Similarly, the supplementation of an antioxidant mixture during 7 week, which reduced LGI, was effective in improving the anabolic response to leucine in skeletal muscles of aged rats (Marzani et al., 2008). In contrast, Mayot et al. (2008) observed no differences in MPS between aged-matched 24 months old LGI and no-LGI rats. However, the authors did not specifically mention when MPS was measured. Previous studies showed differences in post-prandial MPS with similar post-absorbative MPS between LGI and no-LGI aged rats (Rieu et al., 2009; Balage et al., 2010).

### Human Studies

To our knowledge, only two human studies looked at the association between LGI and meal-induced stimulation of MPS. A study executed by the research group of Dardevet found that elderly with LGI (assessed by increased CRP levels) had no different post-absorptive or post-prandial MPS, compared to elderly without LGI (Buffière et al., 2015). The authors suggested that other pro-inflammatory cytokines (TNFα, IL-1/6) might be more suitable to establish the association between LGI and anabolic resistance. Furthermore, the post-protein bolus time interval during which MPS is measured was 5 h in this study, while other studies recommend a shorter period of 1.5–2 h (Bohé et al., 2001; Norton et al., 2009; Atherton et al., 2010; Wilson et al., 2011). Similarly, it was found that ibuprofen (1,800 mg.d−<sup>1</sup> ) administration during 1 week in elderly with LGI did not affect MPS in response to a whey protein bolus when compared to placebo (Dideriksen et al., 2016). However, this might be due to the lack of efficacy of a short NSAID administration period of 1 week, demonstrated by the lack of decrease in CRP levels in this group. Furthermore, it is possible that the anabolic effect of anti-inflammatory strategies (e.g., NSAIDs) might partially occur via a decreased MPB rather than an increased MPS. Future human trials are warranted in order to reproduce the findings of earlier animal studies in which the association between LGI and anabolic resistance was established (Marzani et al., 2008; Rieu et al., 2009; Balage et al., 2010). The studies should take into account the exposure time to LGI, the treatment duration, applied LGI biomarker(s), sufficiently large sample sizes, and the post-prandial time interval to assess MPS.

### Exercise As Anabolic Signal

Besides the effect of inflammation on the anabolic response to food/protein intake, its effect on the anabolic response to exercise is broadly studied. These findings might also contribute to our understanding of why some elderly gain less muscle mass with relatively identical exercise stimuli compared to younger adults. Interesting to note is that NSAID administration in younger adults vs. elderly results into different adaptations to exercise training. In younger adults, studies which looked at the acute effects of NSAID intake on the muscle response to a single bout of resistance exercise observed a decrease in MPS compared to placebo (Trappe et al., 2002) and an attenuated satellite cell proliferation up to 8 days after resistance exercise (Mikkelsen et al., 2009) or endurance exercise (Mackey et al., 2007). Also, when long-term (8 week) resistance training was combined with a daily high-dose (1,200 mg.d−<sup>1</sup> ) of ibuprofen intake, muscle strength and muscle hypertrophic adaptations to resistance training were impaired compared to a control condition (Lilja et al., 2017). In contrast, a moderate dose of ibuprofen (400 mg.d−<sup>1</sup> ) was not sufficient to negatively affect muscle strength and hypertrophy following 6 week resistance training (Krentz et al., 2008). Similar to the results observed in younger adults, a study performed in young rats reported that ibuprofen administration (∼20 mg.kg−<sup>1</sup> .d−<sup>1</sup> ) during chronic overload by synergist ablation for 14 days inhibited muscle hypertrophy with ∼50% (Soltow et al., 2006). The authors, however, were not sure whether the inhibition of hypertrophy by the NSAID was caused by its interference with the regeneration process (decreased satellite cell activity), an attenuation of MPS, or both. A downregulated MPS response to resistance exercise with NSAID intake could be ascribed to a blunted prostaglandin F2<sup>α</sup> (PGF2α) increase, which normally stimulates skeletal MPS (Rodemann and Goldberg, 1982).

Contrary to the younger adults, long-term resistance training combined with NSAID intake, unexpectedly induced additional gains in muscle mass and/or muscle strength in elderly (Trappe et al., 2011) and osteoarthritis patients (Petersen et al., 2011). Trappe et al. (2011) speculated that COX inhibition, an important mechanism of action of NSAIDs, might have a relatively stronger inhibitory effect on MPB compared to MPS in elderly. Besides blunting PGF2α, it was also shown that COX inhibition might suppress protein degradation by reducing intramuscular production of prostaglandin E<sup>2</sup> (PGE2), eventually resulting in a higher net protein balance (Rodemann and Goldberg, 1982). It cannot be excluded that the effect of NSAID administration on training adaptations is partly related to its pain-relieving feature, as it was reported that osteoarthritis patients were able to produce higher maximal strengths after 12 week of ibuprofen (1,200 mg.d−<sup>1</sup> ) administration (Petersen et al., 2011).

In general, it is important to interpret findings concerning the role of inflammation in muscle protein metabolism upon exercise with caution, taking into account the study population (adults vs. elderly with LGI/patient populations), type of exercise/muscle adaptation (endurance vs. resistance exercise vs. muscle damage following severe resistance exercise), and the time span of evaluation of muscle effects (immediately following exercise vs. long-term effects). Furthermore, studies should focus on mechanisms through which MPS is affected by LGI. Currently, many studies do not or only superficially report findings concerning the COX or prostaglandin pathways.

### SATELLITE CELLS AND THEIR REGULATION BY INFLAMMATION

Satellite cells are adults muscle stem cells which play an important role in muscle growth and repair (Yin et al., 2013). Under basal conditions, satellite cells remain sublaminal in a quiescent state. Upon muscle damage, satellite cells exit their quiescent state, start to proliferate through the sarcolemma and fuse with existing muscle fibers. These processes are accompanied by specific expression patterns of myogenic regulator factor (MRF) genes and protein levels. Adult quiescent satellite cells express Pax7, while Myf5 and/or MyoD expression is rapidly upregulated following satellite cell activation, both regulated by Pax 7 (Cornelison and Wold, 1997; Rudnicki et al., 2008).

The lower regenerative potential of aged muscles can be explained by a deterioration in satellite cell differentiation and a reduced Pax7 pool of myogenic stem cells (Collins et al., 2007; Bernet et al., 2014). Furthermore, recent evidence suggests that many aged satellite cells switch from the quiescent state to an irreversible senescence state, and fail to activate and expand upon injury (Sousa-Victor et al., 2014). The reduced satellite cell function with aging might be due to altered systemic factors which affect satellite cell activity and differentiation (Conboy et al., 2005), such as altered Notch signaling in muscles, altered circulating levels of protein growth differentiation factor 11, reduced levels of IGF-1, increased inflammation and proinflammatory cytokines (Harridge, 2003; Degens, 2010; Jang et al., 2011; Sinha et al., 2014). It was shown that TNFα, in particular, reduces the expression of MyoD and myogenin in myoblasts and destabilizes MyoD in regenerating mice muscles (Szalay et al., 1997; Langen et al., 2004). Therefore, it can be hypothesized that some effects of LGI on muscle protein metabolism are mediated through changes in satellite cell function, as TNFα levels affect MRF expression.

### CLASSIC APPROACHES TARGETING AGE-RELATED SARCOPENIA

Many studies currently focus on developing strategies to combat age-related sarcopenia. In this regard, life style interventions are of great interest due to their relatively straightforward applicability. These interventions can be subdivided in two main categories, i.e., exercise and nutrition. Both aerobic and resistance exercise, as well as the supplementation of amino acids/proteins, vitamin D (vit D) and polyunsaturated fatty acids (PUFAs), have ergogenic implications on the regulation of muscle mass in elderly. In addition, aerobic exercise, and the supplementation of vit D or PUFAs showed interesting interactions with the modulation of inflammation, and might therefore decrease LGI and subsequently reduce muscle wasting in elderly.

### Aerobic Exercise

Aerobic exercise has been suggested as a meaningful strategy to combat age-related sarcopenia. Besides its effects on cardiovascular fitness and endurance, aerobic exercise is a potent inducer of muscle size and strength (Konopka and Harber, 2014). Recent evidence indicates that endurance exercise-induced increases in muscle size and strength are due to an increased MPS, similarly in elderly and young adults (Harber et al., 2012; Ozaki et al., 2013; Konopka and Harber, 2014). Some studies concluded that gains in muscle growth, induced by aerobic exercise, are comparable to the hypertrophic response observed following resistance training, and that endurance exercise can be considered an effective countermeasure for muscle loss with advancing age (Konopka and Harber, 2014).

The mechanisms, which underlie these positive effects on the muscle hypertrophic response, are diverse. Firstly, an upregulation of the transcription of genes involved in mitochondrial biogenesis, such as CAMK, AMPK, and PGC1α (Iolascon et al., 2014), increases the muscle mitochondrial content and function. This eventually results in an attenuated production of mtROS and thus oxidative stress, regarded as a potential mediator in skeletal muscle loss with aging (Jackson, 2016). Secondly, aerobic exercise training also has an impact on MPB, as basal FOXO3a (upstream of MAFbx and MuRF1) and myostatin (inhibitor of MPS upstream of Akt-mTORC1) mRNA levels were reduced in elderly following a 12 week training program, with concomitant muscle hypertrophy (Konopka et al., 2010). Another indirect mechanism involves the partial alleviation of age-related insulin resistance, which contributes to the age-related anabolic resistance to protein supplementation (Dickinson et al., 2014; Konopka and Harber, 2014). Finally, aerobic exercise also exhibits antiinflammatory capacities (Montero-Fernandez and Serra-Rexach, 2013). Transient elevations of IL-6, released in response to acute exercise, might stimulate in the long-term the expression of antiinflammatory mediators (such as IL-1 receptor antagonist and IL-10) and downregulate the expression of pro-inflammatory mediators (such as TNFα and IL-1β) (Pedersen et al., 2007). Together, these findings indicate that long-term aerobic training is effective to overcome anabolic resistance and decrease MPB, exhibiting a protective effect on muscle wasting.

### Resistance Exercise

Despite these promising findings, resistance exercise is generally accepted as the most effective approach to induce muscle hypertrophy in both young and old adults, as the relative gain in muscle size in response to resistance training was found to be similar in adult (<65 years) and young individuals (Narici et al., 2004; Slivka et al., 2008; Raue et al., 2009).

The hypertrophic response following resistance training is mediated through several mechanisms. The mechanical stimuli per se and growth factors released in response to mechanical stimuli are both able to independently activate the mTORC1 pathway. The activation of mTORC1 by growth factors (insulinlike growth factor, mechano growth factor) occurs via the "classic" PI3K/Akt signaling (Bodine et al., 2001). Mechanical stimuli have a more acute effect, which is, contrarily, not always dependent on PI3K/Akt signaling (Hornberger et al., 2004). It was suggested that mechanical signals can activate the mTORC1 pathway after they are converted in biological responses. The proposed mediators of these stimuli are phosphatidic acid, which is highly produced in response to stretching and activates p70S6K, and integrins, which connect the extracellular matrix to the muscle cell membrane (Zanchi and Lancha, 2008).

Despite positive findings in adults (<65 years), Peterson et al. (2011) reported less pronounced gains in lean body mass in response to resistance exercise in elderly. This indicates that the myocellular response to resistance training is blunted in advanced age (Narici et al., 2004; Slivka et al., 2008; Raue et al., 2009). The precise mechanism(s), underlying the development of this anabolic insensitivity in elderly, are not fully understood yet. It is likely, however, that the limited activation of the mTORC1 axis following an acute bout of exercise in elderly is an important contributor to this observation (Kumar et al., 2009; Fry et al., 2011). Accordingly, one study found an impaired downregulation of REDD1 (potent inhibitor of mTORC1) mRNA in elderly following resistance exercise (Greig et al., 2011). Nevertheless, it is conceivable that also inflammation is somehow involved in the blunted adaptation to resistance exercise, since the effect of resistance training on gains in muscle mass and muscle strength was reinforced when combined with NSAID administration in elderly (Trappe et al., 2011).

### Protein Supplementation

Among the nutritional interventions, supplementation with proteins/amino acids (leucine in particular) is widely applied to treat muscle loss in elderly. Proteins stimulate MPS, and are therefore generally expected to attenuate muscle wasting. The effects of protein supplementation on MPS are mediated through several mechanisms. Firstly, protein synthesis is induced by cell swelling per se (Lang et al., 1993). This swelling can be obtained through an increased cellular osmolarity, due to Na+-dependent transport into the cell or accumulation of metabolites, such as glutamate. Secondly, mTORC1 regulates protein translation by controlling the phosphorylation of its downstream targets p70S6K and 4E-BP1 (Laplante and Sabatini, 2012). Protein supplementation induces the mTORC1 activation via two independent ways. Amino acids can be directly sensed by GTPases, amino acid transporters and receptors, which transmit the signal to mTORC1 by different signaling pathways (Zheng et al., 2016). This cascade occurs without modulation of PI3K and its downstream effector Akt. Additionally, branched chain amino acids induce the release of insulin, which act upon the PI3K-Akt-mTORC1 axis. Due to its vasodilatory effect, insulin also facilitates the uptake of amino acids from the blood circulation by the skeletal muscle cells. It is accepted that the availability of insulin plays a rather permissive effect, as it was shown that only ∼10% of the post-prandial protein anabolism was dependent on insulin signaling, while ∼90% was related to the increased amino acid levels (Volpi et al., 1996). Furthermore, Abdulla et al. (2016) reported that, in healthy adults, the positive effects of insulin per se on MPS became significant, only when the amino acid delivery to the skeletal muscle increased. At fixed amino acids concentrations, supraphysiological levels of insulin did not induce further increments in MPS (Greenhaff et al., 2008). To summarize, the beneficial effect of physiological hyperinsulinemia per se on MPS can only be observed as long as it concomitantly increases amino acid delivery and availability to the muscle (Fujita et al., 2006).

Among the amino acids, the essential amino acids (EAAs) (e.g., leucine) are the most effective to stimulate MPS (Stipanuk, 2007). Along with its stimulatory effect, leucine in particular also decreases MPB (Buse and Reid, 1975). Another mechanism ascribed to leucine involves its capacity to reverse insulin insensitivity in elderly. The precise mechanism(s) of the insulin sensitizing effects of long-term leucine supplementation are not fully elucidated yet, but may act through leucineinduced decreased adiposity, hepatic glucose production, hepatic steatosis, and adipose tissue inflammation (Zhang et al., 2007; Macotela et al., 2011; Binder et al., 2013). Insulin resistance, which often occurs with aging, is an important contributor to the pathophysiology of age-related sarcopenia, as this negatively affects mTORC1 signaling (Cleasby et al., 2016). Therefore, interventions that might positively affect the insulin sensitivity, are very likely to increase MPS (Børsheim et al., 2006; Solerte et al., 2008). Recently, a downstream metabolite of leucine, βhydroxy β-methylbutyrate (HMB), gained much interest due to its broad ergogenic effects on muscle protein metabolism. HMB was shown to stimulate mTORC1 and the proliferation and differentiation of satellite cells (Cruz-Jentoft, 2017). Besides its anabolic actions, HMB was shown to exert anti-catabolic actions through the attenuation of proteasomal-mediated proteolysis and mitochondrial-mediated myonuclear apoptosis (Smith et al., 2005; Kovarik et al., 2010; Hao et al., 2011; Wilkinson et al., 2013).

Similar to resistance exercise, many elderly suffer from a blunted anabolic response to EAA intake. Therefore, it is generally suggested that larger doses may, at least partially, compensate for the blunted anabolic sensitivity (Paddon-Jones et al., 2004; Cuthbertson et al., 2005; Katsanos et al., 2006; Moore et al., 2015). Recent evidence-based recommendations state that 1.2–1.5 g.kg−<sup>1</sup> BW should be ingested on a daily base by elderly, while the RDA for young adults is 0.8 g.kg−<sup>1</sup> BW (Deer and Volpi, 2015; Devries and Phillips, 2015; Nowson and O'Connell, 2015; Moore et al., 2017). Furthermore, an intake of 25–30 g high quality proteins (∼10 g EAAs) each meal has been proven to maximally stimulate MPS in elderly for 24 h (Deer and Volpi, 2015; Devries and Phillips, 2015; Nowson and O'Connell, 2015).

Despite the well-established positive findings of acute protein/leucine supplementation on the stimulation of MPS, the effectiveness of long-term protein supplementation as a treatment for age-related sarcopenia is more equivocal. On the one hand, a recent meta-analysis demonstrated the beneficial effects of leucine supplementation on MPS in elderly (be it without changes in lean body mass or lean leg mass) (Xu et al., 2015), while other studies failed to show a consistent effect of protein/leucine supplementation on muscle mass, strength and/or function (Bonnefoy et al., 2003; Verhoeven et al., 2009; Tieland et al., 2012b). The review of Hickson (2015) concluded that both whole-protein and EAA supplementation failed to show consistent effects on muscle mass, strength or function. Accordingly, it was suggested that, despite its acute anabolic effects, leucine supplementation has no beneficial effects on skeletal muscle mass or function on the long-term in muscle wasting conditions (Ham et al., 2014). In contrast to leucine, studies that focused on HMB yielded more consistent results (Hickson, 2015; Wu et al., 2015). Consequently, HMB might be more effective for the preservation of muscle mass, strength and functionality in elderly. However, more research is required to further elucidate the efficacy of the long-term intake of amino acids and their metabolites to combat age-related sarcopenia. In addition, it might be of great interest to develop alternative treatments that sensitize the skeletal muscle to leucine and anabolic stimuli in general, such as anti-inflammatory strategies.

## Combined Protein Supplementation and Resistance Exercise

Since both protein supplementation and resistance exercise are suggested to induce muscle hypertrophy by stimulating the mTORC1 signaling pathway, it is not surprising that their combined effect was broadly researched. In healthy adults, evidence indicates that the combined intervention induces additional gains in muscle mass and muscle strength, when compared to resistance training as such (Morton et al., 2017). In elderly, however, the findings are less conclusive, as many studies reported no superiority of resistance exercise with protein supplementation compared to resistance exercise as such (Fiatarone et al., 1994; Godard et al., 2002; Candow et al., 2006; Kukuljan et al., 2009; Verdijk et al., 2009; Denison et al., 2015; Thomas et al., 2016). Denison et al. (2015) suggested that an additional effect due to protein supplementation was mainly to be expected in subjects with low basal protein intakes, whereas those with adequate basal intakes would benefit less from additional protein supplementation. Since adequate protein intake is often problematic in frail and institutionalized elderly, it is of importance to not only emphasize on protein supplementation during resistance training, but also to ensure a sufficient basal protein intake in these subpopulations (Tieland et al., 2012a; Thomas et al., 2016).

### INFLAMMATION-REDUCING APPROACHES TARGETING AGE-RELATED SARCOPENIA

As stated earlier, inflammation is closely involved in both the blunted anabolic response and increased catabolic processes in elderly. In the following section, nutritional strategies that attenuate muscle wasting in elderly, partially regulated through anti-inflammatory mechanisms, will be discussed. **Figure 1** gives an overview of the mechanisms through which LGI may indirectly affect age-related muscle wasting.

## Vitamin D Supplementation

Historically, vit D supplementation has been broadly applied in elderly, due to its well established effects on calcium (Ca2+) and bone homeostasis. It was shown that supplementation of vit D, preferably together with Ca2+, increases bone mineral density and decreases the risk of osteoporotic fractures in elderly. More recently, vit D has also been shown to play a regulatory role in metabolic pathways implicated in muscle wasting, and in the regulation of the immune system. It seems that vit D has beneficial effects on muscle strength (Muir and Montero-odasso, 2011; Beaudart et al., 2014), however, this only seems to apply to vit D-deficient elderly (Janssen et al., 2002; Stockton et al., 2011). In contrast, muscle mass seems not to be increased by vit D supplementation in elderly (Beaudart et al., 2014). Since the prevalence of vit D deficiency is very high among elderly (up to 42% in the U.S.), its supplementation should be recognized as an important intervention in elderly (Forrest and Stuhldreher, 2011).

The mechanisms through which vit D affects muscle cell functioning can be broadly divided into two categories, the genomic and non-genomic effects. For the genomic effects, binding of vit D to its nuclear receptor, a transcription factor, is required. This will result in changes in gene transcription, e.g., stimulation of cell proliferation and terminal differentiation, among others, both important in muscle development and growth (Boland, 1986; Ceglia and Harris, 2013). Indeed, decreases in muscle functioning related to vit D signaling with aging, can be mainly ascribed to the decreased vit D receptor expression in old skeletal muscle cells (Bischoff-Ferrari et al., 2004). Salles et al. (2013) also showed that vit D positively affected the expression of proteins involved in the insulin-AktmTORC1 pathway, along with an increased MPS in C2C<sup>12</sup> muscle cells. These findings were recently confirmed in vivo, as vit D supplementation restored the blunted anabolic response in vit D-deficient old rats (Chanet et al., 2017). To our knowledge, these mechanisms were not yet studied in humans. The nongenomic "rapid" effects are mediated through the membranebound vit D receptor. Firstly, this regulates the influx of Ca2<sup>+</sup> and inorganic phosphate in skeletal muscle cells, important for the regulation of muscle contractions and for the production of energy-rich phosphate groups, respectively. A second nongenomic effect involves the activation of protein kinase C. This kinase plays a crucial role in the regulation of MPS by mediating the anabolic signaling of both insulin (in an IRS1 and PI3K-dependent way) and leucine on protein synthesis (Selless and Boland, 1991; Mendez et al., 1997; Yagasaki et al., 2003).

As mentioned earlier, vit D also modulates the immune system and might therefore be an interesting target to combat LGI. Apart from muscle cells, the vit D receptor is expressed by cells which play a key role in immunity, such as macrophages and lymphocytes. Binding of vit D to its receptor in macrophages inhibits NF-κB activity (a transcriptional key regulator in the upregulation of pro-inflammatory mediators), and thereby attenuates TNFα production (Cohen-lahav et al., 2006). The same NF-κB-dependent inhibition of cytokine secretion through vit D signaling is present in lymphocytes (Calton et al., 2015). In murine models, vit D signaling also stimulates the production of lymphoid cell lineages with anti-inflammatory properties, such as Treg cells (Tian et al., 2015). Another mechanism involves a direct induction of relevant genes by vit D in specific liver cells, resulting in an elevated production of insulin-like growth factor-1 (IGF-1), which has previously been shown to engage in antiinflammatory actions (Bellini et al., 2011; Yu et al., 2012; Wang et al., 2017). Furthermore, IGF-1 is also engaged as a stimulator of cell growth and proliferation through activation of the Akt signaling pathway.

Compared to rodent studies, findings from human interventions with vit D supplementation are very hard to interpret. Most human studies, which link vit D signaling to anti-inflammatory actions, focused on several diseases, each with their specific inflammatory context, which results in ambiguous evidence. Consequently, there is a need for human trials, which study the effects of vit D supplementation in elderly with limited comorbidities. In general, it can be concluded that vit D supplementation might be useful in elderly, as they belong to a population that is prone to vit D deficiency due to reduced exposure to sunlight, low oral intake of vit D, intestinal malabsorption, and decreased vit D hydroxylase activity in the kidneys.

### N-3 Polyunsaturated Fatty Acids

N-3 poly unsaturated fatty acids (PUFAs) are characterized by a double bond at the third carbon from the methyl end of the carbon chain. Besides their role as a structural component, i.e., in membranes, PUFAs function in several cellular processes as regulatory or signaling molecules. In elderly, long-term n-3 PUFA supplementation increases muscle volume and muscle strength (Smith et al., 2015). Furthermore, n-3 PUFA/fish oil supplementation has additional beneficial effects on resistance training-induced muscle functionality in elderly women (Rodacki et al., 2012; Da Boit et al., 2017), though without changes in muscle mass (Da Boit et al., 2017). The anabolic effects of n-3 PUFAs can be ascribed to an increased MPS, mainly regulated through activation of the mTORC1 signaling pathway, as observed in both animal (Gingras et al., 2007; Wei et al., 2013) and human (Smith et al., 2011; Yoshino et al., 2016) studies. A possible link can be found with an increased insulin sensitivity, as it was shown that PUFA supplementation leads to increased phosphorylation of the insulin receptor and its downstream signaling (Kamolrat and Gray, 2013; Wei et al., 2013). Furthermore, enhanced insulin signaling induces vasodilation, which increases amino acid availability (Smith, 2016).

Besides their anabolic capacities, n-3 PUFAs possess wellestablished anti-inflammatory properties. Once transported into the cell, PUFAs are either stored, oxidized or incorporated into cellular membranes. PUFAs, released from the cell membranes, may form prostanoids, which are involved in direct inflammatory actions or regulate the production of other mediators such as inflammatory cytokines (Williams et al., 2000). When consuming a typical western diet, rather high rates of the n-6 PUFA arachidonic acid (ARA) are integrated in the membrane phospholipids of inflammatory cells, while proportions of the n-3 PUFA eicosapentaenoic acid (EPA) or docosahexaenoic acid (DHA) are low. Therefore, ARA is often the main substrate for the formation of prostanoid synthesis (Bagga et al., 2003). After mobilization from the membrane phospholipids by the enzyme phospholipase A2, ARA binds the cyclooxygenase enzymes (COX-1/2) from which downstream intermediates are further metabolized to finally synthesize pro-inflammatory prostanoids. When more n-3 PUFAs are integrated in the diet, less n-6 PUFAderived pro-inflammatory prostanoids will be formed.

Early findings indicate the association between muscle protein metabolism and a certain subclass of the prostanoids, i.e., prostaglandins (Rodemann and Goldberg, 1982; Rodemann et al., 1982). The research group of Goldberg observed that PGE2, produced by muscle upon ARA incubation, consistently stimulated protein degradation (+22%), while no significant changes in MPS were observed. PGF2α, derived from ARA, contrarily, had no effect on MPB but increased MPS. These findings are of great interest from an aging perspective, as an increased LGI was related to an overexpression of COX-2, among others, eventually evoking a pro-inflammatory status (e.g., elevated levels of PGE2) (Chung et al., 2006). More recently, Standley et al. (2013) suggested that COX-induced expression of PGE<sup>2</sup> increased the expression of IL-6 and MuRF-1, which are both related to MPB and might therefore explain the mechanism through which PGE<sup>2</sup> is linked to muscle proteolysis.

Despite promising associations between n-3 PUFAs and muscle metabolism, this area of research needs to be further explored. Animal studies revealed that the EPA was effective in attenuating the ubiquitin-proteasome pathway in several muscle wasting conditions, such as fasting, cachexia and arthritis (Whitehouse and Tisdale, 2001; Whitehouse et al., 2001; Castillero et al., 2009). Furthermore, in obese adults, with elevated inflammatory mediators (e.g., IL-6), n-3 PUFAs were effective in alleviating systemic inflammation (Itariu et al., 2012; Allaire et al., 2016). It would be of interest to expand this knowledge to elderly suffering from LGI in order to develop strategies that support/optimize the current interventions. Recently, a paper confirmed that n-3 PUFA supplementation in elderly downregulated pathways related to calpain- and ubiquitin-mediated proteolysis in skeletal muscle (Yoshino et al., 2016). This shows the anti-proteolytic potential of n-3 PUFAs in age-related muscle wasting.

### The COX-Pathway and Muscle Metabolism

There seems to be an interesting link between inflammation and muscle protein metabolism. A lower MPS was reported in old rats suffering from LGI (Balage et al., 2010) and their MPS was increased when the LGI was blunted by NSAID administration (Marzani et al., 2008; Rieu et al., 2009). Interestingly, LGI also impaired muscle protein anabolism in response to resistance training in elderly (Trappe et al., 2002). As mentioned before, these effects might be related to certain intermediates of the COX pathway, i.e., prostaglandins. COX-inhibiting interventions through NSAID administration might therefore be effective in improving muscle protein metabolism and thus for the treatment of muscle wasting (Greig et al., 2009). This idea was further strengthened by the cross-sectional study of Landi et al. (2013), which investigated the relationship between NSAID use and sarcopenia in community-dwelling elderly (>80 years). In this study only 9% of the subjects that chronically used NSAIDs were affected by sarcopenia, compared to 32% in the non-user group (Landi et al., 2013). Although it tantalizing to promote daily consumption of NSAIDs to overcome the age-related muscle anabolic resistance, the associated risks, such as gastric, cardiovascular and hepato-renal adverse events, cannot be denied (Greig et al., 2009). In this perspective, other anti-inflammatory compounds might be of great interest in the search for a physiological-holistic approach to combat age-related sarcopenia. It should be noted though that antiinflammatory compounds, inhibiting COX-2 and concomitant PG synthesis, might negatively affect skeletal muscle regeneration and hypertrophy. Some studies reported that COX-2 inhibition resulted in a decreased size of regenerating muscle fibers following injury and negatively affected ex vivo satellite cell proliferation, differentiation and fusion (Bondesen et al., 2004; Mendias et al., 2004). Furthermore, COX-2 inhibition decreased intramuscular macrophage accumulation and cell proliferation, and the consequent hypertrophic response in a synergist ablation model (Novak et al., 2009). Future studies should clarify whether these inhibitory effects on muscle regeneration also apply to sarcopenic elderly, and/or whether nutritional antiinflammatory strategies can enforce the positive effects of classic strategies.

### CONCLUSION

Plural mechanisms were shown to contribute to the etiology and/or progression of muscle wasting with advancing age. Somehow, many of these mechanisms interfere with inflammatory mediators. However, further research is required to determine through which mechanisms inflammation directly or indirectly affects MPB and MPS with aging. Classic interventions such as protein supplementation and resistance exercise are generally accepted to be the most appropriate to positively affect muscle protein metabolism in elderly. However, not all studies univocally support the effectiveness of these strategies for long-term treatment of age-related muscle wasting. Elderly, and very old/frail seniors in particular, might benefit from a strategy which primarily focusses on alleviating their muscle insensitivity to anabolic stimuli. In this regard, the treatment of LGI in these elderly might play an important role. Given the limited applicability of NSAIDs, other (non-pharmaceutical) approaches to attenuate LGI should gain more attention. Additionally, future research should also focus on possible interactions with the "classical" anabolic

### REFERENCES


treatments, in order to develop a holistic approach, that takes into account the personal capabilities and experiences of the old individual.

### AUTHOR CONTRIBUTIONS

Substantial contributions to the conception or design of the work: SD, LR, KK. Drafting the work or revising it critically for important intellectual content: SD, LR, KK. Final approval of the version to be submitted: SD, LR, KK. All authors agree to be accountable for all aspects of the work.

### FUNDING

LR is supported by a grant from the Ministry of Health of the Czech Republic (AZV 16-29182A).

### ACKNOWLEDGMENTS

The authors thank Prof. Dr. Vladimir Stich for his critical review during the preparation of this article.


Boland, R. (1986). Role of vitamin D in skeletal muscle function. Endocr. Rev. 7, 434–448. doi: 10.1210/edrv-7-4-434


as countermeasures to offset sarcopenia. BioFactors 40, 199–205. doi: 10.1002/biof.1138


systematic review and meta-analysis. J. Am. Geriatr. Soc. 59, 2291–2300. doi: 10.1111/j.1532-5415.2011.03733.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 Dalle, Rossmeislova and Koppo. 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.

# "Thinking" vs. "Talking": Differential Autocrine Inflammatory Networks in Isolated Primary Hepatic Stellate Cells and Hepatocytes under Hypoxic Stress

Yoram Vodovotz 1, 2, 3, Richard L. Simmons <sup>1</sup> , Chandrashekhar R. Gandhi 1†, Derek Barclay <sup>1</sup> , Bahiyyah S. Jefferson<sup>1</sup> , Chao Huang<sup>1</sup> , Rami Namas <sup>1</sup> , Fayten el-Dehaibi <sup>1</sup> , Qi Mi <sup>1</sup> , Timothy R. Billiar 1, 2, 3 and Ruben Zamora1, 2, 3 \*

*<sup>1</sup> Department of Surgery, University of Pittsburgh, Pittsburgh, PA, United States, <sup>2</sup> Center for Inflammation and Regenerative Modeling, McGowan Institute for Regenerative Medicine, University of Pittsburgh, Pittsburgh, PA, United States, <sup>3</sup> Pittsburgh Liver Research Center, University of Pittsburgh, Pittsburgh, PA, United States*

### Edited by:

*Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico*

#### Reviewed by:

*Savneet Kaur, Institute of Liver and Biliary Sciences, India David Sharkey, University of Adelaide, Australia*

> \*Correspondence: *Ruben Zamora zamorar@pitt.edu*

† Present Address: *Chandrashekhar R. Gandhi, Department of Surgery, University of Cincinnati, OH, United States*

#### Specialty section:

*This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology*

Received: *11 September 2017* Accepted: *14 December 2017* Published: *22 December 2017*

#### Citation:

*Vodovotz Y, Simmons RL, Gandhi CR, Barclay D, Jefferson BS, Huang C, Namas R, el-Dehaibi F, Mi Q, Billiar TR and Zamora R (2017) "Thinking" vs. "Talking": Differential Autocrine Inflammatory Networks in Isolated Primary Hepatic Stellate Cells and Hepatocytes under Hypoxic Stress. Front. Physiol. 8:1104. doi: 10.3389/fphys.2017.01104* We hypothesized that isolated primary mouse hepatic stellate cells (HSC) and hepatocytes (HC) would elaborate different inflammatory responses to hypoxia with or without reoxygenation. We further hypothesized that intracellular information processing ("thinking") differs from extracellular information transfer ("talking") in each of these two liver cell types. Finally, we hypothesized that the complexity of these autocrine responses might only be defined in the absence of other non-parenchymal cells or trafficking leukocytes. Accordingly, we assayed 19 inflammatory mediators in the cell culture media (CCM) and whole cell lysates (WCLs) of HSC and HC during hypoxia with and without reoxygenation. We applied a unique set of statistical and data-driven modeling techniques including Two-Way ANOVA, hierarchical clustering, Principal Component Analysis (PCA) and Network Analysis to define the inflammatory responses of these isolated cells to stress. HSC, under hypoxic and reoxygenation stresses, both expressed and secreted larger quantities of nearly all inflammatory mediators as compared to HC. These differential responses allowed for segregation of HSC from HC by hierarchical clustering. PCA suggested, and network analysis supported, the hypothesis that above a certain threshold of cellular stress, the inflammatory response becomes focused on a limited number of functions in both HSC and HC, but with distinct characteristics in each cell type. Network analysis of separate extracellular and intracellular inflammatory responses, as well as analysis of the combined data, also suggested the presence of more complex inflammatory "talking" (but not "thinking") networks in HSC than in HC. This combined network analysis also suggested an interplay between intracellular and extracellular mediators in HSC under more conditions than that observed in HC, though both cell types exhibited a qualitatively similar phenotype under hypoxia/reoxygenation. Our results thus suggest that a stepwise series of computational and statistical analyses may help decipher how cells respond to environmental stresses, both within the cell and in its secretory products, even in the absence of cooperation from other cells in the liver.

Keywords: liver, hepatocyte, stellate cell, cytokines, systems biology, hypoxia

### INTRODUCTION

In settings of stress, such as liver ischemia/reperfusion, hemorrhagic shock after trauma, and drug-induced liver injury, endogenous mediators released from liver cells are known to initiate and regulate sterile liver inflammation (Peitzman et al., 1995; Vodovotz et al., 2004; Sun et al., 2017; Woolbright and Jaeschke, 2017). There are two major types of cells in the liver: parenchymal cells [hepatocytes (HC)] and non-parenchymal cells [sinusoidal endothelial cells, phagocytic Kupffer cells, and hepatic stellate cells (HSC)]. Whereas 70–85% of the normal adult murine liver volume is occupied by HC (Kmiec, 2001 ´ ), Kupffer cells—which comprise approximately 35% of the nonparenchymal liver cells in normal adult mice (Bilzer et al., 2006)—play a central role in responses to early reperfusion injury following hypoxia/reoxygenation due to their macrophage-like properties (Bilzer and Gerbes, 2000). Studies over the past several decades have focused predominantly on the stress responses of HC, Kupffer cells (a heterogeneous cell population) and trafficking innate immune cells (de Groot, 1992; Peitzman et al., 1995; West and Wilson, 1996). Lesser attention has been paid to sinusoidal or vascular endothelial cells and HSC and the inherent autocrine complexity of the cellular components of the normal liver under stressful conditions have rarely been the subject of investigation.

Various studies have focused on the responses of HC cell lines to hypoxia/reoxygenation. For example, early studies showed that human HepG2 hepatoma cells express interferonγ (IFNγ), Tumor Necrosis Factor-α (TNF-α), Transforming Growth Factor-β1 (TGFβ1), Macrophage Colony-Stimulating Factor (M-CSF), Oncostatin-M (OSM), Intercellular Adhesion Molecule (ICAM-1), Interleukin 4 (IL-4), IL-5, IL-7, IL-10, IL-11, IL-12, and IL-6 receptor (IL-6R), while the expression of IL-1β, IL-2, IL-3, IL-6, CD40 ligand and IL-2R genes was not detected (Stonans et al., 1999). Normal HC seem to have a much narrower spectrum of response: for example, primary human HC were shown to release IFNγ, IL-12p40, IL-12p70, IL-17A, and IL-10 following exposure to hepatotoxic drugs (Ogese et al., 2017). We recently used LuminexTM technology coupled with computational analyses to study the in vitro response of mouse HC to hypoxia (Ziraldo et al., 2013). We found that many inflammatory mediators were changed significantly in both normoxic and hypoxic cultures, and MCP-1 was identified as central node in the inflammatory networks of HC and as an inducer of IL-6; segregating trauma patients based on their co-expression of MCP-1 and IL-6 allowed us to suggest MCP-1 as a potential biomarker for clinical outcomes in trauma/hemorrhagic shock (Ziraldo et al., 2013).

HSC represent only 5-8% of the total number of liver cells (Geerts, 2001). The inflammatory responses of this cell type to hypoxia/reoxygenation are less well studied, though HSC are considered important in the pathogenesis of liver fibrosis (de Oliveira da Silva et al., 2017); furthermore, protection of the liver cells from ischemia/reperfusion injury in HSC-depleted mice indicates that HSC are major contributors to liver damage (Stewart et al., 2014). DNA microarray analyses have shown that hypoxia regulates the expression of genes that may alter the sensitivity of HSC to chemotaxins and other stimuli, and regulates the expression of genes important for angiogenesis and collagen synthesis (Copple et al., 2011). Furthermore, in primary HSC, bacterial lipopolysaccharide (LPS) strongly upregulated numerous CC and CXC chemokines as well as IL-17F (Harvey et al., 2013). Other studies showed that HSC can express a number of other cytokines and chemokines such as Eotaxin, IFNγ, IL-6, IL-8, and IL-10 (Berardis et al., 2014).

Most studies of the effects of environmental stress are carried out in vivo, in which the interaction of various cell types within an organ cannot be discerned from one another. To our knowledge, no studies have utilized pure hepatic cell cultures to examine the autocrine effects, which themselves reflect complexity of response even in the absence of interaction with neighboring cell types. In addition, most studies explore the repertoire of mediator secretion ("talking") by stressed cells, without questioning the internal reactions that take place within the cell ("thinking"). The interpretation of such complex responses require a baseline of information about the inherent responsiveness of purified cells. In addition, mere description of the secretory output of a cell under stress does not adequately address the complexity of even the simplest of experimental systems. For this reason, and given the close proximity of HSC and HC (one HSC contacting two HC on either side of the central vein), this study was designed to compare the intracellular and secretory (autocrine) responses of pure HC and HSC in culture to hypoxic stress and reoxygenation using a unique stepwise series of statistical and data driven modeling techniques previously used to define dynamic molecular networks in both experimental and clinical trauma and acute systemic inflammation (Mi et al., 2011; Ziraldo et al., 2011, 2013; Zaaqoq et al., 2014; Almahmoud et al., 2015; Abboud et al., 2016; Namas et al., 2016). These approaches, including Two-Way ANOVA, hierarchical clustering, Principal Component Analysis (PCA) and Network Analysis were chosen to unmask the complexity of even simple cellular systems. Our results suggest that intracellular inflammatory networks and principal characteristics ("thinking") differ in several important respects from those in the conditioned medium ("talking") of both HSC and HC. Our results further suggest that a stepwise series of computational and statistical analyses may help decipher how cells respond to inflammatory stresses.

### MATERIALS AND METHODS

### Materials

Williams Medium E, penicillin, streptomycin, L-glutamine, and HEPES were purchased from Invitrogen (Carlsbad, CA). Insulin (Humulin <sup>R</sup> ) was purchased from Eli Lilly (Indianapolis, IN), and calf serum was obtained from HyClone Laboratories (Logan,

**Abbreviations:** GM-CSF, Granulocyte-Macrophage Colony Stimulating Factor; IFN, Interferon; IL, interleukin; IP-10, IFN-γ-inducible Protein of 10 kDa/CXCL10; KC, Keratinocyte-derived Cytokine/CXCL1; MCP-1, Monocyte Chemotactic Protein-1/CCL2; MIG, Monokine induced by γ-interferon/CXCL9; MIP-1α, Macrophage Inflammatory Protein-1α (MIP-1α/CCL3α); TNF-α, Tumor Necrosis Factor-α; VEGF, Vascular Endothelial Growth Factor; PCA, Principal Component Analysis.

UT). Tissue culture dishes were from Corning Glass Works (Corning, NY).

### Liver Cell Isolation and Culture

This study was carried out in accordance with the recommendations of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh. The liver cell isolation and culture procedures were as follows:



### Analysis of Inflammatory Mediators

Mouse inflammatory mediators were measured using a Luminex <sup>R</sup> 100 IS apparatus (Luminex, Austin, TX) and the BioSource 20-plexTMmouse cytokine bead kit (BioSource-Invitrogen, San Diego, CA) as per manufacturer's specifications. The antibody bead kit included: Basic fibroblast growth factor (FGF basic), Granulocyte-Macrophage Colony-Stimulating Factor (GM-CSF), Interferon-γ (IFN-γ), Interleukin (IL)- 1α, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-10, IL-12 (both p40 and p70 subunits), IL-13, IL-17A, Interferon-γ-inducible Protein 10 (IP-10/CXCL10), Keratinocyte-derived Cytokine (KC/CXCL1), Monocyte Chemoattractant Protein-1 (MCP-1/CCL2), Monokine induced by Interferon-γ (MIG/CXCL9), Macrophage Inflammatory Protein-1α (MIP-1α/CCL3), Tumor Necrosis Factor-α (TNF-α), and Vascular Endothelial Growth Factor (VEGF).

### Statistical and Computational Analyses

We applied a series of statistical and data-driven modeling techniques aimed at discovering principal drivers, interconnected networks, and potential key regulatory nodes of inflammation in HSC vs. HC following increasing levels of cellular stress. We applied these methods in a stepwise fashion based on our concept of the way liver cells respond to inflammatory stimuli (**Figure 1**, adapted from Namas et al., 2015). Importantly, we also sought to address intracellular information processing ("thinking") as well as extracellular inflammatory mediator production ("talking") carried out by each cell type individually, as well as inferring how these two processes may interact in each cell type. In brief, we hypothesize that low-level stresses associated with cell culture, as well as more stressful stimuli such as hypoxia with or without reoxygenation, will cause liver cells to produce a panoply of inflammatory mediators in order to restore homeostasis. We hypothesize that this dynamic production of inflammatory mediators takes the form of inflammatory networks, and thus can be elucidated using network inference algorithms. As these early inflammatory responses progress, and in the face of continued hypoxia or the onset of reoxygenation, these networks coalesce around an intermediate, core set of inflammation-associated cellular functions, which we hypothesize can be inferred via PCA and related methods. These intermediate processes evolve to form a differential set of inflammatory responses characterized by distinct mediators that characterize the autocrine response of each liver cell type. These differential responses can be discerned both by standard statistical approaches such as Twoway ANOVA as well as data-driven modeling tools such as hierarchical clustering (**Figure 1**). Accordingly, we followed the stepwise series of analyses detailed below:

1) Two-Way Analysis of Variance (ANOVA) was carried out to analyze the changes in inflammatory mediators in HC vs. HSC in both CCM and WCL, using SigmaPlot (Systat Software, San Jose, CA) as indicated.


each component, depicted as a stacked bar graph. This gives a measure of a given inflammatory mediator's contribution to the overall variance of the system. The mediators with the largest scores are the ones which contributed most to the variance of the process being studied (Mi et al., 2011).

4) Network Analysis was carried out to define the central inflammatory network nodes as a function of cell type (i.e., HC vs. HSC), experimental condition, and spatial localization (lysate vs. conditioned medium) using a modified version of our previously published algorithm for Dynamic Network Analysis (Mi et al., 2011; Ziraldo et al., 2013; Zamora et al., 2016). Connections between inflammatory mediators were created if the Pearson correlation coefficient between any two nodes (inflammatory mediators) was greater or equal to a threshold of 0.95, as indicated. The "network complexity" for each experimental condition was calculated using the following formula: Sum (N<sup>1</sup> + N<sup>2</sup> +. . .+ Nn)/n-1, where N represents the number of connections for each mediator and n is the total number of mediators analyzed. The total number of connections represents the sum of the number of connections in a given experimental sub-group.

### RESULTS

In the present study, we sought to compare, in a systematic fashion, the inflammatory responses of freshly isolated HC or HSC to hypoxia and hypoxia/reoxygenation. We reasoned that distinct statistical and computational analyses would elucidate different steps in the intracellular and extracellular responses of these cells to stress, shown schematically in **Figure 1** (see Materials and Methods for details). In the context of that schematic, we worked backwards from the most directly apparent/final responses (assessed by Two-Way ANOVA and hierarchical clustering) to the intermediate responses (assessed by PCA), and then to discerning the most proximal responses to stress (assessed by network analysis). Another key goal of our studies was to utilize computational modeling to assess the interactions between intracellular information processing ("thinking") vs. extracellular information transfer ("talking") in both HC and HSC. For that reason, we examined both the wholecell lysates (WCL) and the cell culture media (CCM) in each cell type.

### Differential Production and Release of Inflammatory Mediators in HSC vs. HC in Response to Hypoxia and Hypoxia/Reoxygenation

The production of inflammatory mediators in both CCM and WCL under the three experimental conditions, along with the corresponding P-values for **c**omparison of the changes in HSC vs. HC by Two-Way ANOVA, can be found in **Supplementary Figure 1**. Key examples of inflammatory mediators whose intracellular and extracellular levels are significantly different between HSC and HC are shown in **Figure 2**. In general, on a pg protein per mg total cell protein basis, HSC both produced and secreted much larger quantities of inflammatory mediators than the rather indolent response of HC. To identify those inflammatory mediators that showed similar expression or secretion behavior in each cell type, these patterns were then compared and grouped using hierarchical clustering. The resulting dendrograms showed a clear cluster separation between HC and HSC both in the CCM (**Figure 3A**) and WCL (**Figure 3B**) under the three experimental conditions studied. From this analysis, we infer that there is more heterogeneity in the intracellular ("thinking") vs. the extracellular ("talking") levels of inflammatory mediators in both cell types.

### Hypoxia and Hypoxia/Reoxygenation Induce a Differential Response in Hsc vs. HC Inferred from Principal Component Analysis (PCA)

Next, we sought to define an intermediate, core set of inflammation-associated characteristics of HSC and HC in response to increasing levels of stress, which we hypothesize represent different inflammation-related functions that ultimately become manifest in the expression and secretion of the defined subset of inflammatory mediators studied. To do so, we utilized PCA (**Figure 1**), a technique (or variants thereof) used by multiple investigators to delineate the core characteristics of a multivariate biological response (Janes et al., 2005; Janes and Yaffe, 2006; Mi et al., 2011; Nieman et al., 2012; Azhar et al., 2013; Ziraldo et al., 2013). We assessed several properties of the inflammatory responses under control, hypoxia, and hypoxia/reoxygenation: (1) the total number of mediators contributing to the total variance (i.e., mediators with sums of eigenvalues > 0; (2) the number of inferred principal components (represented as stacked bars), which we hypothesize represent distinct characteristics or inflammation-related functions; and (3) the sums of the eigenvalues (amplitude) for each experimental condition, which we hypothesize represents the magnitude of the inflammatory response to the indicated level of stress. The number of mediators and number of principal components expressed by each cell type under each condition is shown in **Figure 4**. An overall comparison of the inflammatory mediators showed that the inflammatory response of HSC was clearly different from that in HC both in CCM (**Figure 4A**) and WCL (**Figure 4B**) under each experimental condition. But certain similarities were seen. The major similarity was that the amplitude of the response was greatest when both HSC and HC were subjected to hypoxia/reoxygenation. Moreover, in general, the number of mediators involved as well as the inferred functions (number of principal components) in both WCL ("thinking") and CCM ("talking") were lowest in both HSC and HC subjected to hypoxia/reoxygenation as compared to control or hypoxia. A notable exception was the "thinking" response in HC WCL (**Figure 4B**), in which all three experimental conditions were characterized by two principal components, and the overall number of mediators was least in the hypoxia setting. Together, these analyses suggest that above a certain threshold of cellular stress, the inflammatory response generally becomes focused on a limited number of functions and associated mediators at a high intensity of response.

### Network Analysis Revealed the Presence of Distinct Autocrine Inflammatory Networks in HSC vs. HC

The final analysis we carried out was aimed at discovering the initial networks of inflammation induced under control conditions, hypoxia, or hypoxia/reoxygenation (**Figure 1**). We have previously employed Dynamic Network Analysis (DyNA) to define the dynamic interconnections among inflammatory mediators in a mouse model of trauma/hemorrhagic shock (Mi et al., 2011), in isolated mouse HC subjected to hypoxic stress (Ziraldo et al., 2013), and to define the dynamics of systemic inflammation after trauma (Abboud et al., 2016; Namas et al., 2016) or during acute liver failure (Zamora et al., 2016). For the current analyses, we used a modified version of this method, in which data from a single experimental condition (rather than a time interval) were used.

This network analysis suggested the presence of more complex extracellular inflammatory networks in HSC than in HC under all experimental conditions (**Figures 5A,B** and **Supplementary Figure 2A**) but roughly similarly complex intracellular networks (**Figures 5C,D** and **Supplementary Figure 2B**). Despite this difference in network complexity, multiple, shared inflammatory mediators were involved in both "thinking" and "talking" in HSC and HC. Secreted mediators shared by networks in both HSC and HC under control conditions included VEGF, MCP-1, KC, and IL-17A. Intracellularly under control conditions, only MIG and IL-4 were shared by HSC and HC networks. Under hypoxia, intracellular HSC and HC networks shared the mediators TNF-α, KC, MIG, and MCP-1, while intracellular networks only shared the mediators MIP-1α and IP-10. Under hypoxia/reoxygenation,

HC from C57BL/6 mice were cultured under 21% O2 for 6 h (control), hypoxia (1% O2) for 6 h or hypoxia (6 h) followed by reoxygenation for 18 h. Inflammatory mediators were measured by LuminexTMin both cell culture media and whole cell lysate as described in *Materials and Methods.* Box plots show the levels of IL-17A, IP-10, KC, MIP-1α, and TNF-α, where the line within the box marks the median and the lower and upper boundaries represent the 25th and 75th percentiles, respectively (\**P* < 0.001, HSC vs. HC, analyzed by Two-Way ANOVA).

secreted HSC and HC networks shared the greatest number of mediators: IL-1α, IL-2, IL-4, IL-6, IL-17A, MIG, KC, IP-10, GM-CSF, VEGF, and TNF-α; in contrast, shared intracellular mediators included TNF-α, MIG, MCP-1, IL-1α, and IL-17A. Control experiments comparing single network analysis of both isolated HC and HSC cultured under 21% O<sup>2</sup> for 24 h confirmed

the presence of more complex extracellular inflammatory networks in HSC than in HC under both control and hypoxia/reoxygenation ( **Supplementary Figure 3A**) and similar complex intracellular networks after hypoxia/reoxygenation (**Supplementary Figure 3B**).

Based on PCA (**Figure 4**), we hypothesized that network interactions would lead to a reduced set of principal mediators and inflammatory functions in both CCM and WCL of HSC as they transition from hypoxia to hypoxia + reoxygenation. In support of this hypothesis, the networks observed under hypoxia/reoxygenation in HSC CCM as well as WCL exhibited multiple anti-correlated nodes (indicated in red; **Figures 5A,C**). We interpret this finding to mean that both intracellular and extracellular HSC mediators induced in the setting of hypoxia followed by reoxygenation regulate each other in a negative fashion that leads ultimately to the evolution of a reduced number of inflammatory functions relative to hypoxia alone. In the extracellular milieu of HSC, the chemokine MIG; the cytokines TNF-α, IL-6, and IL-17A; and the hypoxia-inducible growth factor VEGF appear to coordinate some of these downregulatory responses (**Figure 5A**). VEGF, TNF-α, MCP-1, and IL-17A also appear to play a more limited negative intracellular role (**Figure 5B**).

The foregoing analyses suggested that there may be an interplay between intracellular responses ("thinking") and extracellular responses ("talking") in each cell type. We therefore next sought to use network analysis to gain insights into the autocrine information flow within each cell population. Accordingly, we repeated the network analysis, but this time using both CCM and WCL data together for HSC vs. HC (**Figure 6**). Surprisingly, this analysis suggested the presence of multiple negative interactions in control HSC. This analysis also inferred the presence of a large number of negative interactions in both HSC and HC exposed to hypoxia/reoxygenation (**Figure 6A**,**Table 1**). Network analysis also demonstrated a higher complexity of autocrine networks involving both intracellular ("thinking") and extracellular ("talking") in HSC as compared to HC (**Figure 6B**), similar to the results obtained separately in CCM and WCL (**Figures 5B,D**).

We next assessed the flow of information from the intracellular (WCL) to extracellular milieu (CCM) in HSC vs. HC in greater detail (**Figure 6** and **Supplementary Table 1**) Under control conditions, inflammatory networks in HC were comprised solely of extracellular mediators (IL-1β, IL-13, IL-17A, MCP-1, KC, VEGF, and TNF-α). In contrast, the combined network analysis inferred multiple positive and negative connections among a large number of intracellular and extracellular inflammatory mediators in HSC, suggesting a much more robust "thinking" and "talking" interplay in this cell type at baseline. Under hypoxia alone, both HSC and HC exhibited only positive, extracellular inflammatory connectivity, though the number of connections was higher in

HSC (a large number of mediators) as compared to HC (only TNF-α, IL-17A, KC, MCP-1, and MIG). Interestingly, when subjected to hypoxia/reoxygenation, both HSC and HC exhibited a large number of positive and negative connections in the intracellular as well as extracellular compartments. This suggests a major adaptation to stress in HC, while this phenotype in HSC remained qualitatively similar to that observed under control conditions.

## DISCUSSION

We report for the first time a detailed multiplex analysis of inflammatory mediators in isolated HSC and HC cultured under hypoxic conditions with and without reoxygenation, which suggests the presence of distinct, endogenous autocrine networks in isolated HSC and HC. The other cellular and structural components of the liver (vascular and sinusoidal endothelial cells, Kupffer cells, fibroblasts, biliary ducts, etc.) were excluded deliberately from these experiments in order to study the complexity of the autocrine responses of these two cellular components without the contribution of inflammatory mediators from contiguous cell types. Our datadriven computational work flow (see conceptual summary in **Figure 1**) was designed to elucidate how early inflammatory networks can lead to an intermediate set of responses that manifest ultimately in a generally restricted number of inflammatory mediators/functions that distinguish the responses of HSC from those of HC. Based on this work flow, we show that both cell types respond to hypoxic stress by releasing a somewhat similar, but not identical, repertoire of cytokines and chemokines in vitro. The similarities suggest the possible presence of a mutually supportive autocrine environment in the intact liver in vivo. The hypoxic stimulus evokes a much greater change in the secreted cellular inflammatory network of HSC than HC, indicating that the secretory machine in HC might be either more limited at baseline or disrupted following stress. Furthermore, the responses of HSC to hypoxia followed by reoxygenation are more vigorous and form more interactive networks than responses to hypoxia alone.

A significant role for HSC in regulating hepatic inflammatory and immunological responses by altering expression of numerous relevant genes has already been shown in LPSstimulated rat HSC (Harvey et al., 2013). We have also shown previously that HSC produce numerous inflammatory mediators constitutively, expression of which is increased upon stimulation by LPS (Uemura and Gandhi, 2001; Thirunavukkarasu et al., 2005) (Thirunavukkarasu et al., 2008; Jameel et al., 2010; Harvey et al., 2013; Dangi et al., 2016). Furthermore, activation of unstimulated HSC stimulate DNA synthesis in HC (Uemura and Gandhi, 2001), and the mediators produced by LPS-stimulated HSC induce autophagy in HC as a survival mechanism but also cause apoptosis of a subset of cells (Dangi et al., 2016). We also

reported a comprehensive analysis of LPS-induced expression of inflammatory mediators in HSC (Harvey et al., 2013). In the setting of hypoxia, we found that HSC-depleted mice are protected from ischemia/reperfusion injury, suggesting that under ischemic/hypoxic condition HSC release mediators that are injurious to HC (Stewart et al., 2014). Our analyses in the current study showed the presence of more complex autocrine inflammatory networks in HSC as compared to HC, which supports the notion that these two cell types not only coexist but also provide a microenvironment for each other in which HSC play the "messenger" role as opposed more of a "responder" role for HC. Experimental testing of this inference is part of ongoing and future experiments that will include other liver cells as well.

Our results suggest a complex interplay among chemokines (e.g., MCP-1, KC, MIG, and IP-10), cytokines (e.g., TNF-α and IL-17A, among many others), and growth factors (e.g., VEGF) in the response to hypoxia ± reoxygenation even in the absence of the contribution of other liver cells. Some of these cytokines have been reported to be induced by stress in HC and HSC previously (Kmiec´, 2001; Brenner et al., 2013; Rani et al., 2017). Our computational analysis, however, elucidate a novel dimension of information regarding how HSC and HC "think" and "talk": as stress increases, intracellular and extracellular information transfer appears to result in a reduction of inflammationrelated functions, a process we hypothesize is driven through negative network interactions among key inflammatory mediators.

Finally, we are aware of the limitations in translating these computational analyses in vivo. Under physiological conditions O<sup>2</sup> is a major effector of metabolic zonation and plays a major role in the liver. When compared to other somatic cells, liver cells are relatively hypoxic, both in the periportal

level 0.95) for each experimental condition combining the data in CCM and WCL (A). Black and red arrows represent positive and negative connections, respectively.

TABLE 1 | Total number of connections from Network Analysis.

(B) Compares the combined network complexity in HSC vs. HC.


*Freshly isolated hepatic stellate cells (HSC) and hepatocytes (HC) from C57BL/6 mice were cultured under 21% O*<sup>2</sup> *for 6 h (control), hypoxia (1% O*2*) for 6 h or hypoxia (6 h) followed by reoxygenation for 18 h. Inflammatory mediators released into cell culture media (CCM) and in whole cell lysate (WCL) were measured by LuminexTMand Network Analysis was performed as described in Materials and Methods (see Results in* Figure 6A*). Table shows the total number of positive (black arrows, see* Figure 6A*) and negative connections (red arrows, see* Figure 6A*) for each experimental condition in both cell types in CCM* + *WCL (stringency level 0.95).*

(zone 1, 9–11% O2) and the perivenous region (zone 3, 5– 7% O2) (Jungermann and Kietzmann, 2000). Changes in this complex O<sup>2</sup> gradient, as well as in the production of hormones and enzymes, affect the response of the liver in a number of pathophysiological processes. Thus, we agree that virtually all routine tissue culture experiments involving liver cells, such as the control conditions described herein are normally performed under relatively hyperoxic conditions that may not reflect in vivo physiology (Sullivan et al., 2006). Notably, the complex O<sup>2</sup> gradient in the liver is very difficult to duplicate using isolated cells in vitro. Furthermore, it is also difficult to identify (and quantify) the presence and activity of specific inflammatory or parenchymal cells during specific time periods, as is the identification of potential crosstalk among dynamic networks in vivo. Although it is very likely, whether or not these inflammatory networks change in both composition and complexity when HC and HSC are in close proximity, as it happens in vivo, has not yet been reported and it represents another interesting element in the complex relationship between liver cells. To elucidate this and other of our hypotheses will require a more detailed analysis of data obtained from co-cultures of HC with HSC both alone and in combination with other relevant cells in the liver such as Kupffer cells and endothelial cells. Future studies will be directed to address these limitations, specifically we have ongoing studies in primary mouse HC as well as in plasma and whole liver in a mouse model of ischemia/reperfusion.

In summary, our systems-based analysis suggests the presence of an inflammatory homeostatic environment in resting isolated liver cells, which is disrupted under stress such as hypoxia and hypoxia followed by reoxygenation. The vigorous production of secreted inflammatory mediators from HSC and the relative indolent response from isolated HC suggest that HSC act as resident transformers of external stimuli or blood-borne for other less responsive cells within the complex cellular structure of the liver. The relevance of these findings in vivo warrants further investigation.

### AUTHOR CONTRIBUTIONS

YV participated in computational and statistical analysis, data interpretation, and writing. RS participated in data interpretation and writing. CG participated in data interpretation and writing. DB participated in Luminex analysis. BJ participated in cell culture experiments (HC). Fe-D participated in data analysis. QM participated in statistical and computational analysis. CH participated in cell culture experiments (HSC). TB participated in data interpretation and writing. RZ participated in study design, computational and statistical analysis, data interpretation, and writing. RN participated in cell culture experiments (HSC).

## FUNDING

This work was supported by NIH grant RO1-GM107231.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2017.01104/full#supplementary-material

Supplementary Figure 1 | Production of inflammatory mediators in mouse hepatic stellate cells (HSC) and hepatocytes (HC). Freshly isolated HSC and HC from C57BL/6 mice were cultured under 21% O2 for 6 h (control), hypoxia (1% O2) for 6 h or hypoxia (6 h) followed by reoxygenation for 18 h. Inflammatory mediators were measured by LuminexTM in both cell culture media and whole cell lysate as described in *Materials and Methods*. Results shown represent mean ± SEM (∗*P* < 0.05, HSC vs. HC, analyzed by Two-Way ANOVA).

Supplementary Figure 2 | Network Analysis of inflammatory mediators in mouse hepatic stellate cells (HSC) and hepatocytes (HC). Freshly isolated HSC and HC from C57BL/6 mice were cultured under 21% O2 for 6 h (control), hypoxia (1% O2) for 6 h or hypoxia (6 h) followed by reoxygenation for 18 h. Inflammatory mediators released into cell culture media (CCM) and in whole cell lysate (WCL)

### REFERENCES


were measured by LuminexTM and Network Analysis was performed as described in *Materials and Methods*. Figure shows the networks (stringency level 0.95) for each experimental condition in both cell types in CCM (A) and WCL (B). Black and red arrows represent positive and negative connections, respectively.

Supplementary Figure 3 | Network Analysis of inflammatory mediators in mouse hepatic stellate cells (HSC) and hepatocytes (HC). Freshly isolated HSC and HC from C57BL/6 mice were cultured under 21% O2 for 24 h (control) or hypoxia (6 h) followed by reoxygenation for 18 h (as in Supplementary Table 1). Inflammatory mediators released into cell culture media (CCM) and in whole cell lysate (WCL) were measured by LuminexTM and Network Analysis was performed as described in *Materials and Methods*. Figure shows the networks (stringency level 0.95) for each experimental condition in both cell types in CCM (A) and WCL (B). Black and red arrows represent positive and negative connections, respectively.

Supplementary Table 1 | Network Analysis of inflammatory mediators using the combined data of both the cell culture media (CCM) and whole cell lysate (WCL) in mouse hepatic stellate cells (HSC) and hepatocytes (HC). Freshly isolated HSC and HC from C57BL/6 mice were cultured under 21% O2 for 6 h (control), hypoxia (1% O2) for 6 h or hypoxia (6 h) followed by reoxygenation for 18 h. Inflammatory mediators released into cell culture media (CCM) and in whole cell lysate (WCL) were measured by Luminex® and Network Analysis was performed as described in *Materials and Methods*. The figure shows the detailed connectivity (stringency level 0.95) for each experimental condition combining the data in CCM and WCL.

reversion as cellular connections that modulate liver fibrosis. Cell Biol. Int. 41, 946–959. doi: 10.1002/cbin.10790


**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 Vodovotz, Simmons, Gandhi, Barclay, Jefferson, Huang, Namas, el-Dehaibi, Mi, Billiar and Zamora. 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.

# Coactivation of TLR2 and TLR8 in Primary Human Monocytes Triggers a Distinct Inflammatory Signaling Response

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

#### Reviewed by:

Kevin Woollard, Imperial College London, United Kingdom Roland Lang, Universitätsklinikum Erlangen, Germany

#### \*Correspondence:

Richard K. Kandasamy richard.k.kandasamy@ntnu.no Bjarte Bergstrøm bjarte.bergstrom@nord.no †These authors have contributed

equally to this work.

#### ‡Present address:

Bjarte Bergstrøm, Faculty of Nursing and Health Sciences, Nord University, Bodø, Norway

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 17 November 2017 Accepted: 07 May 2018 Published: 29 May 2018

#### Citation:

Bösl K, Giambelluca M, Haug M, Bugge M, Espevik T, Kandasamy RK and Bergstrøm B (2018) Coactivation of TLR2 and TLR8 in Primary Human Monocytes Triggers a Distinct Inflammatory Signaling Response. Front. Physiol. 9:618. doi: 10.3389/fphys.2018.00618 Korbinian Bösl<sup>1</sup> , Miriam Giambelluca<sup>1</sup>† , Markus Haug1,2† , Marit Bugge<sup>1</sup> , Terje Espevik<sup>1</sup> , Richard K. Kandasamy1,3 \* and Bjarte Bergstrøm1,2 \* ‡

<sup>1</sup> Centre of Molecular Inflammation Research, Department of Clinical and Molecular Medicine, Norwegian University of Science and Technology, Trondheim, Norway, <sup>2</sup> Department of Infection, St. Olav's University Hospital, Trondheim, Norway, <sup>3</sup> Centre for Molecular Medicine Norway, Nordic EMBL Partnership, University of Oslo and Oslo University Hospital, Oslo, Norway

Innate immune signaling is essential to mount a fast and specific immune response to pathogens. Monocytes and macrophages are essential cells in the early response in their capacity as ubiquitous phagocytic cells. They phagocytose microorganisms or damaged cells and sense pathogen/damage-associated molecular patterns (PAMPs/DAMPs) through innate receptors such as Toll-like receptors (TLRs). We investigated a phenomenon where co-signaling from TLR2 and TLR8 in human primary monocytes provides a distinct immune activation profile compared to signaling from either TLR alone. We compare gene signatures induced by either stimulus alone or together and show that co-signaling results in downstream differences in regulation of signaling and gene transcription. We demonstrate that these differences result in altered cytokine profiles between single and multi-receptor signaling, and show how it can influence both T-cell and neutrophil responses. The end response is tailored to combat extracellular pathogens, possibly by modifying the regulation of IFNβ and IL12-family cytokines.

Keywords: innate immunity, Toll-like receptors, Toll-like receptor 2, Toll-like receptor 8, signaling, monocytes

### INTRODUCTION

Monocytes and macrophages constitute critical cell types in the innate immune response (Shi and Pamer, 2011). These cells are equipped with germline-encoded pattern recognition receptors/sensors (PRRs) that aid in the recognition of various microbial components from microbes termed pathogen-associated molecular patterns (PAMPs) and self-derived dangerassociated molecular patterns (DAMPs) released by damaged cells (Liston and Masters, 2017). Depending on the specific receptor-PAMP/DAMP match and whether multiple PRRs are engaged, various downstream effectors/pathways are activated, which prepare the cells to fend of the invading agents by activating degradation pathways and relaying signals such as cytokines to further alert innate and adaptive immune cells in the surrounding tissues and at distal sites (Ginhoux and Jung, 2014).

**112**

Toll-like receptors (TLRs), a major subgroup of PRRs, are type 1 transmembrane proteins with ligand binding extracellular domains composed of leucine rich repeats and cytoplasmic intracellular signaling domains known as the Toll/IL-1 receptor (TIR) domains (Botos et al., 2011). Currently there are 10 TLRs described in humans and 12 in mice, for most their ligands have been identified. TLRs can be broadly divided into two groups, depending on the subcellular location where they encounter their specific ligands. In humans, TLR1, TLR2, TLR4, TLR5 and TLR6 encounter their specific ligands at the cell surface, while TLR3, TLR7, TLR8 and TLR9 bind to their ligands in endosomes (Kaisho and Akira, 2006; O'Neill et al., 2013). Upon recognition of PAMPs and DAMPs, TLRs recruit adaptor proteins containing TIR domains such as MyD88 or TRIF, which acts as initiators of signal transduction pathways culminating in the activation of interferon regulatory factors (IRFs), NF-κB, or MAP kinases regulating the expression of cytokines, chemokines and type I Interferons (IFNs) (Kurt-Jones et al., 2000; Honda et al., 2005; Kawasaki and Kawai, 2014). For instance, TLR2 recognizes a variety of components derived from Gram-positive bacteria and downstream signaling requires TIRAP/Mal as an adaptor, which acts through the MyD88-dependent pathway, inducing activation of NF-κB and MAPK family members (Oliveira-Nascimento et al., 2012). TLR2 signaling is also dependent on the formation of a heterodimer, either of TLR1 and TLR2 or of TLR2 and TLR6. These hetereodimers have different affinities for lipopeptide ligands, but utilize the same adaptor molecules and have common signaling pathways (Farhat et al., 2008). On the other hand, TLR8 signaling is exclusively mediated through MyD88 after recognition single-stranded RNA and short double-stranded RNA, making it an important PRR for viral pathogens. In addition, TLR8 induces type I IFNs through IRF5 and IRF7 activation (Cervantes et al., 2012). Post-translational modifications (PTMs) plays a crucial role in the activation and regulation of this complex response (Liu et al., 2016). PTMs such as phosphorylation have distinct patterns depending on the type of activation (Kandasamy et al., 2016).

Several studies have shown that responses to bacterial and viral pathogens are not exclusively dependent on activation of individual TLRs, but results from complex TLR-TLR interactions (Underhill, 2007; Delaloye et al., 2009; Slater et al., 2010; Negishi et al., 2012). Nevertheless, the effects of engagement of more than one innate immune receptor in close temporal proximity are not well studied. Recently, it was reported that RNA derivate from bacteria could also activate TLR8, and TLR2 could be able to suppress IFNβ production induced by TLR8 activation (Mancuso et al., 2009; Bergstrøm et al., 2015; Kruger et al., 2015; Tanji et al., 2015; Lai et al., 2016; Gidon et al., 2017). To further understand the molecular mechanisms behind TLR2 and TLR8 interaction we studied changes in signaling pathways, gene expression and cytokines production in primary human macrophages stimulated with TLR2 and TLR8 ligands alone or in combination. Our results show that while prolonged PTMs on transcription factors could explain amplitude differences between TLR2 and TLR8 gene induction and cytokine production, suppression of IFNβ has wide-ranging consequences in shaping the overall immune response. Surface TLRs suppressing IFN β production via the TLR8-IRF5 axis after recognition of bacterial RNA can therefore be an important mechanism in shaping the adaptive immune response to pathogenic Gram-positive bacteria.

### MATERIALS AND METHODS

### Isolation of Leukocytes

Buffy coats and serum were obtained from healthy donors at the St. Olavs Hospital blood bank with approval by the Regional Committee for Medical and Health Research Ethics (REC Central, Norway, No. 2009/2245). Monocytes were isolated by LymphoprepTM (Axis-Shield) density gradient according to the manufacturer's instructions. Peripheral blood mononuclear cells (PBMCs) were selected by plastic adherence for 1 h in 6-well plates in RPMI 1640 supplemented with 2 mmol/L Lglutamine (Sigma-Aldrich) and 10% human serum, washed three times with Hanks balanced salt solution (Sigma-Aldrich) and rested for 1 h before stimulation. At specified time points cells were harvested and the cell pellet was stored at −80◦C for RNA isolation or protein analysis. Supernatants were collected and stored at −80◦C.

Purity of the isolated monocytes was validated using Flow Cytometry. Adhered monocytes were detached using Accutase (Sigma) solution (15 min, 4◦C) and transferred to flowtubes. Dead cells stained with Fixable Viability Dye eFluor780 (eBioscience). Fc receptor binding inhibitor (eBioscience) was used to block non-specific Fc receptor binding of antibodies. Cells were subsequently stained with fluorescence-labeled monoclonal antibodies to CD14 (PE/Cy7, BioLegend), CD11b (PE, BioLegend), CD3 (Brilliant Violet780, BioLegend), CD19 (Brilliant Violet510, BioLegend), CD56 (APC, eBioscience), CD303 (PerCP/Cy5.5, BioLegend). Flow-cytometry was performed on a BD LSRII flow-cytometer (BD Biosciences) and samples analyzed using FlowJo software (FlowJo, LLC).

A population of >85% CD14+/CD11b+ cells was observed, while surface markers for contaminating cell populations (CD3+, CD19+, CD56+, or CD303+) were detected with ≤0.6% each (**Supplementary Figure S1A**). mRNA expression of markers for non-monocytic cells types in all samples used for the Array-based RNA expression analysis were found to be low (**Supplementary Figure S1B**).

For viability analysis, monocytes were stimulated with FSL-1 (100 ng/ml) or CL075 (1 µg/ml) or both, unstimulated cells were used as a control. After 3 h of stimulation, calcein-AM (1 µg/ml, Invitrogen) and propidium iodide (1 µg/ml, Sigma) were added to stain viable and dead cells respectively. Subsequently, monocytes were detached and calcein-AM and propidium iodide fluorescence quantified by flow-cytometry on a BD LSRII flowcytometer (BD Biosciences); samples analysis using FlowJo software (FlowJo, LLC) (**Supplementary Figure S2A**). We also performed microscopy analysis of monocytes to test the viability using the EVOS system (Thermo Fisher) (**Supplementary Figure S2B**).

Untouched human primary CD4<sup>+</sup> T cells were isolated from PBMCs by negative selection using CD4<sup>+</sup> T Cell Isolation Kit (Miltenyi Biotec) according to the manufacturer's instructions.

Purity of CD4<sup>+</sup> T cells was assessed by flow cytometry with anti-CD4 (Alexa Fluor 700, eBioscience) and anti-CD3 [Brilliant Violet (BV) 785, BioLegend] antibody staining. Frequencies of contaminating cells were measured by antibody staining for CD8 (BV605, BioLegend), CD14 (PE/Cy7, eBioscience), and CD11c (PE, eBioscience). CD3 + CD4 + T cell purity was >95%.

Human neutrophils obtained from fresh peripheral blood drawn by venipuncture from healthy volunteers (approval by the Regional Committee for Medical and Health Research Ethics REC Central, Norway, No. 2009/2245) were isolated by double gradient centrifugation. 20 mL of heparinized blood were transferred to sterile polypropylene conical tubes containing a double gradient separation composed of equal volumes (10 mL) of Polymorphoprep (density, 1.113 g / mL) and Lymphoprep (density, 1.077 g / mL). Following centrifugation at 400 × g for 30 min, the layer of polymorphonuclear (PMN) cells was aspirated and contaminating erythrocytes were removed by 20 s of hypotonic lysis. Neutrophils were diluted in RPMI 1640 + 10% FCS to obtain a final cell concentration of 10<sup>6</sup> cells / mL, with the purity and viability of neutrophils equal to or more than 95 and 98%, respectively.

### Ligands and Antibodies

TLR8 ligand CL075 (used at 1 µg/mL) and TLR2 ligand FSL-1 (used at 100 ng/mL) were purchased from Invivogen. Antibodies used for Western blots in this study were: anti-phospho p38 (Cell Signaling Technology, No. 9211), anti-phospho JNK (Cell Signaling Technology, No. 4668), anti-phospho ERK1/2 (Cell Signaling Technology, No. 4370), p38 (Cell Signaling Technology, No. 9212), JNK (Cell Signaling Technology, No. 9252), ERK1/2 (Cell Signaling Technology, No. 4695), and GAPDH (Abcam, No. ab8245).

### Western Blotting

Cells were lysed in cell lysis buffer (Bergstrøm et al., 2015), centrifuged. The cleared lysates were then mixed with 4x LDS loading buffer (Life Technologies) and PAGE was performed with the NuPAGE system (Life Technologies) as per the manufacturer's recommendations. Immunoblotting was performed with the iBlot system (Life Technologies) as per the manufacturer's recommendations. Membranes were blocked with 5% BSA in Tris-buffered saline (TBS, pH 7,4) and incubated with primary antibodies overnight at 4 ◦C. Membranes were then washed in TBS with 0.5% tween-20 (TBS-T), incubated with HRP-conjugated secondary antibodies (Dako, No. P0399 and No. P0447) for 1 h at room temperature. Blots were developed with SuperSignal West Femto (Pierce) and imagined on a Li-Cor Odyssey system.

### Multiplexed Cytokine Profiling

Supernatants were diluted 1:20 and analyzed according to the manufacturer's protocol using the Human ProcartaPlex 34-plex panel 1A (ThermoFischer, No. EPX340-12167-901) on a Bio-Plex 200 instrument (BioRad).

### Array-Based RNA Expression Analysis

Gene expression was analyzed with an Illumina HT-12 v4 bead array as per the manufacturer's instruction and performed at the Genomics Core Facility at NTNU. RNA isolation was performed using RNeasy mini kit (Qiagen) according to the manufacturer's instructions. RNA integrity was examined by Bioanalyzer (Agilent). Hybridization to a HumanHT-12 v4.0 bead array was performed by the Genomics Core Facility (Department of Cancer Research and Molecular Medicine, NTNU). The Data was preprocessed using GenomeStudio v1.9.0 and imported to R/Bioconductor v3.3.2/3.4 (Huber et al., 2015). Probes with detection p-value > 0.05 and probes with reported unspecific binding were excluded. The expression data was background corrected using negative control probes and quantile normalized in limma v3.32.9 (Ritchie et al., 2015).

Inter-donor variation was included as one of the parameters for the linear model used in the analysis. Genes showing an expression log2-fold > 2.5 and FDR < 0.05 compared to the unstimulated samples for each time-point were considered differential expressed. Differential expressed genes were clustered using Manhattan distance matrix calculation and divided into 8 clusters for further analysis. Gene Ontology analysis was performed with clusterProfiler v3.4.4 (Yu et al., 2012), the selection of Gene Ontology Terms for visualization was curated manually. Raw data can be accessed from ArrayExpress database using the accession number E-MTAB-6222.

### T Cell Differentiation Assays

Human primary CD4<sup>+</sup> T cells (0.5 × 10<sup>6</sup> cells/well) were activated with anti-CD3 (plate-coated, 5 µg/mL, eBioscience) and anti-CD28 (1 µg/mL, eBioscience) on 96-well plates. CD4<sup>+</sup> T cells were differentiated for 2 – 8 days at 37◦C in 100 µL RPMI 1640 medium (Sigma) supplemented with 10% A+ serum and 100 µL supernatants from TLR2/TLR8 stimulated monocytes stimulated with TLR ligands. CD4<sup>+</sup> T cell effector cytokine production was analyzed 48 h post activation and on day 8 after 6 h short-term re-stimulation with Cell Stimulation Cocktail (eBioscience). Protein Transport Inhibitor Cocktail (eBioscience) was added for the last 4 h before harvest of the cells. Cell were stained with Fixable Viability Dye eFluor 780 (eBioscience) and surface-stained with fluorescent antibodies to CD3 (BV 785, BioLegend), CD4 (Alexa Fluor 700, eBioscience) before fixation and permeabilization (FOXP3 buffer set, BD Biosciences). Staining for intracellular cytokine production was performed with fluorescent antibodies to IFN-γ (FITC, Miltenyi Biotec), IL-17 (BV 510, BioLegend), IL-2 (PE/Cy7, eBioscience), IL-10 (APC, eBioscience) IL-4 (PE-Vio615, Miltenyi Biotec) and TNF-α (BV421, BioLegend) and Multicolour flow cytometry was performed on a BD LSRII flow cytometer and analyzed with FlowJo software (FlowJo, LLC).

### Neutrophil Migration Assay

Chemotaxis was measured as described previously (Gilbert et al., 2003), with some modifications. Briefly, neutrophils were resuspended in RPMI 1640 and 10% FBS at 10<sup>6</sup>

cells/mL and were preincubated with 5 µg/mL CellTracker Deep Red (Molecular Probes, Eugene, OR, United States). After 30 min incubation at 37◦C in the dark, cells were washed twice and re-suspended in RPMI/FCS at 3 × 10<sup>6</sup> cells/mL. Neutrophil migration was monitored using a 96 well chemoTX disposable chemotaxis system (NeuroProbe, Gaithersburg, MD, United States). The bottom wells were filled with 31 µL of macrophage supernatant or a dilution of the chemotactic agent. The polycarbonate filter was positioned on the plate, and neutrophils (25 µL) were seeded on the filter and allowed to migrate for 1 h at 37◦C in the presence of 5% CO2 in the dark. Non-migrated cells were gently removed by wiping the filters with a tissue. The fluorescence of the cells in the filters was measured with a microplate fluorescence reader (Excitation and emission wavelengths, 630 and 660 nm, respectively). Fluorescence was transformed to numbers of neutrophils based on a standard curve generated by seeding known numbers of neutrophils in the bottom of the chamber. The results were expressed as percentage of migrated neutrophils.

### Statistical Analysis

Statistical analysis was done in R v3.3.2. After testing for normal distribution a paired Wilcox test was performed. Significance levels are indicated as followed: <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

### RESULTS

### Transcriptome Analysis of Primary Human Monocytes Stimulated With TLR2/8 Ligands

We performed bead array-based transcriptomics analysis to identify genes induced by the stimulation or co-stimulation of TLR2 and TLR8. We used FSL-1 (Pam2CGDPKHPKSF a synthetic lipopeptide) as a TLR2/TLR6 agonist; and CL075 (a thiazoloquinolone derivative) as TLR8 agonist. Primary human monocytes from six different donors were isolated and stimulated with FSL-1 and/or CL075 for 1, 2, and 3 h; and were subjected to transcriptome analysis (**Figure 1A** and **Supplementary Figure S3**) to quantify the expression of 16,106 genes. Applying a log2(fold-change) of 2.5 and adjusted p-value < 0.05 cutoff, we identified 63, 194, and 132 genes differentially regulated at one or more time-points by FSL-1, CL075 and co-stimulation, respectively (**Figure 1B**). We found classic inflammatory genes (Turner et al., 2014) like TNFα, IL-6, CCL and CXCL family of cytokines highly upregulated by both TLR2 and TLR8 stimulation – alone and together. One of the highly induced genes specific to TLR8 stimulation at 1 h was Interferon β, as expected. The distribution of induced genes at the earliest time point (1 h) showed a large overlap between stimuli, where 11 out of 16 genes are induced by all stimuli alone or together. This distribution changed at later time points, where TLR8 stimuli induced an additional set of genes that were not induced by TLR2 (**Figure 1C**). Co-stimulation falls between TLR2 and TLR8 stimulation alone, suggesting that TLR2 signaling partially, but not completely, regulates TLR8 signaling. After 3 h of stimulation, the core response induced by all stimuli effectively triples compared to the earlier time points, consistent with the start of a secondary response. Together these results show that both stimuli share a core early response while TLR8 induced an additional set of genes as a separate and/or secondary response, and that upon co-stimulation TLR2 signaling modulates the TLR8 response. Macrophage differentiation markers such as CCR5, FCGR1 and CD68 are not affected during the tested timespan, while CD14 and MERTK show a down-regulation and CD40 and CD71 (TFRC) show an up-regulation from after 2 h, sharing the same pattern as the general inflammation markers (**Supplementary Figure S4**).

### Cluster Analysis of the Transcriptome Data Reveals Ligand Specific Transcriptional Activation

To investigate the temporal and ligand-specific activation profile, we carried out clustering analysis of all the 151 differentially regulated genes (**Figure 2** and **Supplementary Table S1**). This segregated the differentially regulated genes into 8 different clusters. Genes in cluster 6–8 are induced at later time-points, hence being indirect TLR signaling targets. Cluster 1, 2, and 4 included the upregulated classic inflammatory genes such as IL-6, CCL14, CXCL1, CXCL2, CCL20, TNFα and PTGS2, which were common to TLR2/8 stimulation and co-stimulation. Cluster 3 included 18 downregulated genes that were common to both TLR2/8 stimulation as well as co-stimulation. Clusters 5 and 6 included differentially upregulated genes specific to TLR8 stimulation. Interferon β was the lone gene in cluster 5 with a unique expression profile while cluster 6 included several interferon-induced genes such as IFIT1 and IFI27 (Rasmussen et al., 1993; Guiducci et al., 2013). Clusters 7 and 8 included several upregulated genes that were common to TLR2/8 stimulation and co-stimulation, though we see a stronger upregulation by TLR8 stimulation. Altogether, this analysis shows that there is a common and ligand specific transcriptional activation with a unique mechanism for regulation of interferon β as reported earlier (Bergstrøm et al., 2015).

We performed gene-ontology enrichment analysis to gain insights into the biological processes and signaling pathways that are affected by these differentially regulated genes. We identified several biological processes (**Supplementary Table S2**) that were significantly enriched (adjusted p-value < 0.05) such as regulation of cytokine production, JAK-STAT signaling, NFκB signaling, cell chemotaxis, regulation of MAPK signaling, and regulation of adaptive immune response (**Supplementary Figure S5**). MAPK signaling has been shown to be essential for transcription of specific antiinflammatory genes (Gottschalk et al., 2016). Corroborating with this data, our gene ontology analysis also pinpoints the

possible role of three canonical MAP kinases p38, ERK1/2 and JNK.

### TLR2/8 Co-stimulation Affects the Phosphorylation Dynamics of MAPK Signaling

MAPK signaling was one of the enriched biological processes regulated by the differentially expressed genes and it has been shown that TLR2 regulates MAPK signaling differently from TLR13 which is described as a murine TLR8 homolog (Choo et al., 2017). We probed phosphorylated forms of p38 (T180/Y182), JNK (T183/Y185) and ERK (ERK1/2-T202/Y204) in addition to their total protein levels by western blot (**Figure 3**). These phosphorylated forms are the functionally active versions that in turn activate various transcription factors such as AP-1 complexes. Upon TLR2 stimulation, the phosphorylation levels of p38, JNK and ERK peaks between 15 and 30 min and then starts to reduce. This suggests that TLR2 induces a feedback inhibition that reduces the level of MAPK signaling activity. TLR8 stimulation induces a lesser degree of MAPK phosphorylation at early time points but is prolonged compared to TLR2 signaling and does not show the same dephosphorylation kinetics as TLR2 signaling. Co-stimulation of TLR2 and TLR8 fits closely to the TLR2 signaling profile, showing that the same feedback mechanism limiting MAPK signaling after TLR2 stimulation also limits signaling induced by TLR8.

### Cytokine Profiling Shows Commonality and Specificity of TLR2/8 and Co-stimulation

The transcriptomics analysis showed transcriptional upregulation of several cytokines. We performed multiplexed cytokine profiling to measure the amount of secreted cytokines after 6 h of TLR2/8 activation and found 11 different cytokines that were significantly regulated (**Figure 4A**). We could divide these cytokines into four different groups: (1) induced by TLR8

but suppressed by TLR2 activation - TNFα, IL-1α, IL-1β, and IL12p70; (2) induced by both TLR2/8 stimulation and enhanced by co-stimulation – IL-8 and CXCL1; (3) induced by TLR8 and not suppressed by TLR2 stimulation – IL-6 and IL-23; and (4) induced by all and not affected by co-stimulation – CCL2, CCL3, and CCL4. To get a more global view of this cytokine response to TLR2/8 activation, we visualized the relative levels of cytokines in percentage as a spider plot (**Figure 4B**). We observed that TLR2

gives a limited response with a strong chemokine component related to the recruitment of phagocytosing cells while TLR8 signaling induces a mixed Th1/Th17-activating profile with a strong induction of IL12p70 and IL-1β. Co-stimulation enhances the release of neutrophil-recruiting cytokines such as IL-8 and CXCL2, but also provide a shift from IL12p70 to IL23 induction that would enable Th17 activation on the expense of Th1 activation. Thus, co-stimulation of monocytes should be able to induce both an acute innate and an adaptive neutrophil-based response through neutrophil-attracting chemokines and the Th17-polarizing cytokine IL-23.

Here co-stimulation is dominated by the de-phosphorylation induced by TLR2.

### TLR2/8 Co-stimulation Modulates Neutrophil Migration and T Cell Differentiation

In order to verify how co-stimulation can affect T cell and neutrophil response, we implemented two assays with supernatants from stimulated monocytes: (1) a neutrophil migration assay; and (2) a T-cell differentiation assay. The supernatant of primary human monocytes stimulated with TLR2 and TLR8 ligands, alone or in combination was used in a migration assay with peripheral human neutrophils from healthy donors. We found that migration was increased in neutrophils migrating toward the supernatant of TLR2 and TLR8-stimulated monocytes as compared to those migrating toward supernatant from unstimulated monocytes. The chemotaxis of human neutrophils was significantly higher with supernatants of TLR2/8 co-stimulated monocytes (**Figure 5A**). This effect is specifically attributed to the supernatants from the stimulated monocytes since the TLR2/8 ligands themselves does not have any significant effect on the neutrophil migration (**Supplementary Figure S6**).

Primary human CD4<sup>+</sup> T cells were isolated by negative selection and treated with a conditioned medium containing the supernatants derived from TLR2/8 stimulated monocytes. CD4<sup>+</sup> effector T cell differentiation was assessed by the percentage of IFNγ (Th1 subset) and IL-17 (Th17 subset) positive cells after 48 h and 8 days (**Figure 5B**). No significant difference was noted in the Th1 subset after 48 h of differentiation, but after 8 days a small but significant difference was noted where Th1 differentiation was suppressed by co-stimulation compared to TLR8 stimulation alone. Supernatant from TLR2-stimulated monocytes were more efficient in initiating Th17 differentiation as seen after 48h compared to supernatants from TLR8 and costimulated monocytes which showed a small inhibitory effect on Th17 differentiation. These differences were not evident after 8 days where all stimuli induced a robust Th17 differentiation. Together these experiments indicate that differential signaling induced by co-activation of TLR8 and TLR2 can have specific and emergent effects on the immune response.

### DISCUSSION

TLR8 is one of the lesser-studied TLRs, with a restricted expression pattern and an unresponsive murine homolog (Guiducci et al., 2013). It was previously established as an antiviral receptor based on its location in the endosome and its ability to induce IFNβ. More recently, TLR8 has been reported to be a general sensor of RNA breakdown products (Kruger et al., 2015) and shown to be important also for the recognition of

gives three overlapping but distinct cytokine profiles of the different stimulation status. Significance values are provided in Supplementary Table S3.

bacterial infections (Cervantes et al., 2013; Bergstrøm et al., 2015; Eigenbrod et al., 2015). The role of IFNβ production in response to bacterial infections is, however, controversial and depending on the nature of the pathogen and the host IFNβ can be both beneficial and detrimental in the clearance of bacterial infections (Boxx and Cheng, 2016).

Th1 phenotype was reduced by co-stimulation after 8 days. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001 paired Wilcox test.

TLR2 is a sensor of cell wall products and generally considered to be a sensor for Gram-positive bacterial infections via its ability to detect lipoteichoic acid from bacterial cell walls (Oliveira-Nascimento et al., 2012). TLR2 also has the ability to suppress IFNβ induction by TLR8 if signaling is initiated simultaneously or sequentially in short order by an unknown mechanism that is IRF5 dependent (Bergstrøm et al., 2015). We initially set out to identify possible other targets of this mechanism and were surprised when IFNβ appeared as a lone differentially regulated cluster (**Figure 2**). Although we cannot exclude that other differentially expressed target genes such as IL-6 (Steinhagen et al., 2013; Yasuda et al., 2013) or TNF (Krausgruber et al., 2010) seen in cluster 4 are also regulated by IRF5 (Steinhagen et al., 2013), regulation of mRNA stability could be a likely candidate to explain the differences when we take into account the differences seen for these genes on the protein secretion level (Deleault et al., 2008; Iwasaki et al., 2011). Several inflammatory response genes are reported to be regulated on the mRNA stability level, e.g., by Regnase-1 in an IKKβ-dependent loop (Iwasaki et al., 2011) or by Tristetraprolin (TTP) which again is regulated via a p38 MAPK-dependent signaling loop (Kratochvill et al., 2011).

It is well established that both TLR8 and TLR2 signaling is dependent on activation of MAPK pathways (Arthur and Ley, 2013) and that different TLRs propagate the MAPK-dependent signal with different amplitudes and temporalities (Choo et al., 2017). Thus, it is not surprising that MAPK signaling appears to be a differentially regulated process in the GO analysis and together with the reports on MAPK-dependent mRNA stabilization we found these differences to be interesting enough to study further. The fact that upon simultaneous stimulation the TLR2 signal (high amplitude, short duration) will override the TLR8 signal (slower, less amplitude, longer duration) has not been shown before. These properties are as important as the presence of the signal by itself (Atay and Skotheim, 2017). It is not clear how this affects TLR8/2 signaling, but it is evident that TLR2 signaling induces a potential feedback inhibitory loop that reduce the intensity of the signal, probably via the activation of one or several phosphatases (Kondoh and Nishida, 2007). This feedback loop should also affect mRNA stability and could at least partially explain the higher cytokine secreted levels seen by TLR8 signaling compared to TLR2 signaling. Since IRF5 is also activated by phosphorylation (Chang Foreman et al., 2012), suppression of IRF5 activation could also be a by-stander effect of MAPK phosphatase activation. Several phosphatases including dual specificity phosphatases (DUSP) have been reported to function as MAPK phosphatases that control inflammatory response (Chu et al., 1996; Zhang et al., 2004; Zhao et al., 2005; Lang et al., 2006). Although several phosphatases including dual specificity phosphatases such as DUSP1, DUSP2, DUSP4, DUSP5, DUSP9, DUSP10 and DUSP16 were upregulated (**Supplementary Figure S7**), we didn't find any difference between mRNA expression of these phosphatases in TLR8 stimulated and co-stimulated cells.

The ramifications of IFNβ suppression can be clearly observed in cluster 6 and 7, as TLR8-induced IFNβ autocrine or paracrine signaling induces a number of genes that are not or barely induced by TLR2 and are severely suppressed in co-stimulation.

further increasing neutrophil migration.

The functional difference between these stimuli is shown on the cytokine secretion level, where TLR8, TLR2 and co-stimulation induce overlapping but discrete cytokine profiles. TLR2 induce a limited but effective cytokine response alone, tailored to attract neutrophils. TLR8 induce a broader response partially dependent on IFNβ, particularly concerning IL-12-family cytokines IL-12p35 and IL-12p40. Together these form a heterodimeric cytokine named IL12-p70 which is a known activator of Th1 differentiation processes. Co-stimulation shifts this focus over to induction of IL-23p19 on the expense of IL12-p35. The resulting heterodimer of IL-12p40 and IL-23p19 is known as IL-23, a powerful enhancer of Th17 differentiation. These findings are in concordance with earlier reports that place IRF5 in a central position in inducing Th1/Th17-polarizing cytokines (Krausgruber et al., 2011), but also show that when IRF5 activation is suppressed it only affects the Th1 component of this response but not the Th17 component (**Figure 5B**). We were not able to observe increased Th17 differentiation in our T cell differentiation assay even in the presence of an increased concentration of IL23. This not surprising as Th17 differentiation require both IL-6 and TGFβ (Veldhoen et al., 2006) and we did not observe any differential expression of TGFβ at mRNA level in our monocyte experiments although is possible that TGFβ is regulated at the translational level. Induction of IL23 will, however, also activate a subset of so-called Type 17 cells which do not require TGFβ (Gaffen et al., 2014). IL23 could then provide two different paths to Type 17 immunity in vivo that would not be detected by our in vitro assay. The role of INFβ in bacterial infections are, however, still unclear, with evidence pointing to pathogen-specific roles that can be both beneficial and detrimental to the host (Eshleman and Lenz, 2014; Castiglia et al., 2016).

The observed changes in expression of some of the classical macrophage differentiation marker genes (**Supplementary Figure S4**) might indicate the start of monocyte/macrophage reprogramming, however, previous studies demonstrated that many of the underlying transcriptional changes occur at later time-points (Martinez et al., 2006).

In total, our gene expression analysis and cytokine profiling predict that co-signaling of TLR8 and TLR2 should increase the acute inflammatory response and provide a shift away from Th1 immunity toward increased Th17 immunity. The observed increase in neutrophil migration and decrease in Th1 differentiation upon co-stimulation underlines this hypothesis

(**Figure 6**). We can, however, not explain all the specific mechanisms behind this phenomenon, neither explain whether it is specific to tailor the immune response against pathogens that can stimulate TLR8 and TLR2 simultaneously or simply acts of interference in a complicated signaling environment. It is certainly of interest to note that a Gram-positive bacterial infection should lead to an enhanced neutrophilbased immune response through this mechanism, although it should be noted that most Gram-positive bacteria contain ligands for several other innate immune receptors leading to an even more complex immune signaling pattern. This could also be a mechanism through which monocytes sense an ongoing infection, as live or growing bacteria would contain more TLR8 ligands but the same amount of TLR2 ligands compared to a dead or stationary-phase bacteria (Sander et al., 2011). We also note that while not much is known about IRF5 and IL23 induction, both are heavily involved in autoimmune diseases and dysregulation of either one substantially increases the risk of an overzealous immune response (Eames et al., 2016; Pfeifle et al., 2017). That TLR2 signaling could be able to modulate both of these pathways, at least in human monocytes, might then present a possible target for clinical intervention in both infectious and autoimmune diseases.

### AUTHOR CONTRIBUTIONS

BB, MG, MH, and MB performed the experiments. BB conceived the study. KB, BB, MG, and MH analyzed the data. BB, MG, MH, and TE designed the experiments. RKK supervised the bioinformatics analysis and follow-up validations. RKK, BB, KB, and MG wrote the manuscript. All authors read and approved the manuscript.

### FUNDING

This work was funded by the Liaison Committee between the Central Norway Regional Health Authority and NTNU (to MH, TE, and BB), the Research Council of Norway (FRIMEDBIO "Young Research Talent" Grant 263168 to RKK; and Centres of Excellence Funding Scheme Project 223255/F50 to TE), and Onsager fellowship from NTNU (to RK).

### ACKNOWLEDGMENTS

We would like to thank Jørgen Stenvik, as well as Claire Louet, Liv Ryan, and Kai Sandvold Beckwith for help with the multiplexed cytokine profiling and microscopy.

### REFERENCES

Arthur, J. S., and Ley, S. C. (2013). Mitogen-activated protein kinases in innate immunity. Nat. Rev. Immunol. 13, 679–692. doi: 10.1038/ nri3495

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2018.00618/full#supplementary-material

FIGURE S1 | (A) Measurement of cell surface makers and potential contaminating cell types by Flow-cytometry. Adhered monocytes were detached and subsequently stained with fluorescence-labeled monoclonal antibodies to CD14 and CD11b to identify monocytes, as well as antibodies to CD3, CD19, CD56, and CD303 to identify contaminating cell types. Analysis in the plots was performed on ungated, viable singlet cells. (B) mRNA expression of cell type marker genes in samples used for the Array-based RNA expression analysis.

FIGURE S2 | Viability analysis of isolated monocytes. Monocytes were stimulated for 3 h with FSL-1, CL075 or both before staining with calcein-AM and propidium iodide. (A) Viability analysis by flow-cytometry. Dead cells can be seen in the upper left quadrant of the plots. (B) Viability analysis by fluorescence microscopy. Live cells in green, dead cells in red; scale bars 200 µm.

FIGURE S3 | Transcriptional changes were measured using Illumina HT-12 v4 bead array. Corresponding volcano plots of expression profiles from CL075, FSL-1 and FSL-1+CL075 stimulated monocytes for 1, 2, and 3 h show that TLR8 stimulation gives a larger magnitude of differentially expressed genes than TLR2 stimulation alone or co-stimulation compared to unstimulated samples at the same time-point.

FIGURE S4 | mRNA expression of macrophage differentiation marker genes from CL075, FSL-1 and FSL-1+CL075 stimulated monocytes for 1, 2, and 3 h. CCR5, FCGR1 and CD68 are unaffected, while CD14 and MERTK show a down-regulation; CD40 and CD71 (TFRC) show an up-regulation after 2 h.

FIGURE S5 | Enriched Gene Ontology "Biological Processes" terms from differential expressed genes 1, 2, and 3 h CL075, FSL-1 and FSL-1+CL075 stimulated monocytes, manually curated selection.

FIGURE S6 | Neutrophils migration toward medium (RPMI+10% serum), 100 ng/mL FSL-1, 1 µg/mL CL07 or combination of both, C5a 10 nM or IL8 10 nM. Neutrophils did not migrate toward TLR2 or TLR8 agonist alone or in combination while IL8 and C5a, two well-known chemotactic factors, significantly increased it. TLR2/8 ligands themselves does not have any significant effect on the neutrophil migration.

FIGURE S7 | Clustering and magnitude of mRNA expression of phosphatases from primary human monocytes stimulated with FSL-1, CL075 and FSL1+CL075 as compared to non-stimulated samples as log2 fold change. Hierarchical clustering using Manhattan distance matrix calculation was used to divide the expression matrix into 5 clusters. Genes in cluster 1, 3–5 are induced later than genes in Cluster 2. Cluster 2 includes DUSP1 and DUSP2 and which are induced immediately. Cluster 1 contains DUSP3, 4, 8, and 10 and which are induced in a secondary response. Cluster 3 and 5 show only weak changes in expression upon stimulation. Cluster 4 shows down regulation of PPM1M and CTDSP2 in the secondary response.

TABLE S1 | List of differentially regulated genes that was used for the clustering analysis.

TABLE S2 | List of significantly enriched gene ontology biological processes for the gene clusters.

TABLE S3 | p-values and significance levels for differential cytokine secretion after TLR8, TLR2 and co-stimulation.


and macrophages and induces IFN-beta production via a TAK1-IKKbeta-IRF5 signaling pathway. J. Immunol. 195, 1100–1111. doi: 10.4049/jimmunol. 1403176


**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 Bösl, Giambelluca, Haug, Bugge, Espevik, Kandasamy and Bergstrøm. 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.

fphys-09-00618 May 28, 2018 Time: 10:3 # 13

# Cellular Mechanisms of Myocardial Depression in Porcine Septic Shock

Dagmar Jarkovska1,2, Michaela Markova1,2, Jan Horak2,3, Lukas Nalos1,2, Jan Benes2,4 , Mahmoud Al-Obeidallah<sup>1</sup> , Zdenek Tuma<sup>2</sup> , Jitka Sviglerova1,2, Jitka Kuncova1,2 , Martin Matejovic2,3 and Milan Stengl1,2 \*

<sup>1</sup> Department of Physiology, Faculty of Medicine in Pilsen, Charles University, Pilsen, Czechia, <sup>2</sup> Biomedical Center, Faculty of Medicine in Pilsen, Charles University, Pilsen, Czechia, <sup>3</sup> Department of Internal Medicine I, Faculty of Medicine in Pilsen, Charles University, Pilsen, Czechia, <sup>4</sup> Department of Anesthesiology and Intensive Care Medicine, Faculty of Medicine in Pilsen, Charles University, Pilsen, Czechia

The complex pathogenesis of sepsis and septic shock involves myocardial depression, the pathophysiology of which, however, remains unclear. In this study, cellular mechanisms of myocardial depression were addressed in a clinically relevant, large animal (porcine) model of sepsis and septic shock. Sepsis was induced by fecal peritonitis in eight anesthetized, mechanically ventilated, and instrumented pigs of both sexes and continued for 24 h. In eight control pigs, an identical experiment but without sepsis induction was performed. In vitro analysis of cardiac function included measurements of action potentials and contractions in the right ventricle trabeculae, measurements of sarcomeric contractions, calcium transients and calcium current in isolated cardiac myocytes, and analysis of mitochondrial respiration by ultrasensitive oxygraphy. Increased values of modified sequential organ failure assessment score and serum lactate levels documented the development of sepsis/septic shock, accompanied by hyperdynamic circulation with high heart rate, increased cardiac output, peripheral vasodilation, and decreased stroke volume. In septic trabeculae, action potential duration was shortened and contraction force reduced. In septic cardiac myocytes, sarcomeric contractions, calcium transients, and L-type calcium current were all suppressed. Similar relaxation trajectory of the intracellular calcium-cell length phase-plane diagram indicated unchanged calcium responsiveness of myofilaments. Mitochondrial respiration was diminished through inhibition of Complex II and Complex IV. Defective calcium handling with reduced calcium current and transients, together with inhibition of mitochondrial respiration, appears to represent the dominant cellular mechanisms of myocardial depression in porcine septic shock.

Keywords: sepsis, pig, myocardial depression, calcium, mitochondria

### INTRODUCTION

Sepsis represents a well-recognized worldwide health problem. Based on the meta-analysis of studies from developed high-income countries, global annual estimates were 31.5 million sepsis and 19.4 million severe sepsis cases, with potentially 5.3 million deaths, in the hospital setting (Fleischmann et al., 2016). Sepsis plays a prominent role in hospital mortality: in two large

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

#### Reviewed by:

Jamil Assreuy, Universidade Federal de Santa Catarina, Brazil Kanigula Mubagwa, KU Leuven, Belgium

> \*Correspondence: Milan Stengl Milan.Stengl@lfp.cuni.cz

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 07 November 2017 Accepted: 25 May 2018 Published: 12 June 2018

#### Citation:

Jarkovska D, Markova M, Horak J, Nalos L, Benes J, Al -Obeidallah M, Tuma Z, Sviglerova J, Kuncova J, Matejovic M and Stengl M (2018) Cellular Mechanisms of Myocardial Depression in Porcine Septic Shock. Front. Physiol. 9:726. doi: 10.3389/fphys.2018.00726

**125**

complementary inpatient cohorts (total of 157,518 deaths/7,038,449 admissions) sepsis was found to contribute to 1 in every 2 to 3 deaths (Liu et al., 2014).

Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection (Singer et al., 2016). A common manifestation of sepsis/septic shock is myocardial depression with reversible biventricular dilatation and depressed ejection fraction (Antonucci et al., 2014). The precise mechanistic link between infection, sepsis, and myocardial depression remains unclear, although a number of possible pathways have been suggested (Merx and Weber, 2007; Sato and Nasu, 2015). An early theory of global myocardial ischemia due to inadequate coronary blood flow in sepsis was disproved by findings of high coronary blood flow and diminished coronary artery–coronary sinus oxygen difference in septic patients (Cunnion et al., 1986). Nowadays, there is a general consensus on the mechanism of a circulating depressant substance, as originally demonstrated by Parrillo et al. (1985). The exact nature of the circulating depressant substance, however, has not been sufficiently clarified yet. The list of possible candidates includes various cytokines, endotoxins, prostanoids, and nitric oxide (Merx and Weber, 2007; Sato and Nasu, 2015).

Similarly, the downstream cellular pathophysiology of myocardial depression is still obscure, although in experimental models, a number of contributing mechanisms have been reported. The cardiac cellular mechanisms include a reduction of L-type calcium current (ICaL) (Lew et al., 1996; Zhong et al., 1997; Stengl et al., 2010), altered calcium transients (Ren et al., 2002), increased calcium leakage from the sarcoplasmic reticulum (Zhu et al., 2005), impaired sarcolemmal diastolic calcium extrusion pathways (Wagner et al., 2015), oxidation and subsequent activation of calcium and calmodulin-dependent protein kinase with phosphorylation of the ryanodine receptor (Sepúlveda et al., 2017), altered phosphorylation and calcium sensitivity of cardiac myofibrillar proteins (Wu et al., 2001), and/or mitochondrial dysfunction (Levy et al., 2004; Watts et al., 2004).

Most data on the intrinsic cellular mechanisms of myocardial depression, however, were obtained in small animal (rodent) experimental models with limited clinical relevance and translatory potential (Poli-de-Figueiredo et al., 2008; Dyson and Singer, 2009). To overcome this limitation, cellular mechanisms of myocardial depression were examined in a clinically relevant porcine model of peritonitis-induced progressive septic shock, which, in contrast to rodent hypodynamic endotoxic shock, closely mimics human sepsis (hyperdynamic circulation with low systemic vascular resistance and multiple organ dysfunction). The myocardial functions were examined on several levels of biological complexity, from the in vivo experiment similar to the clinical scenario down to experiments in isolated cells and organelles, with special emphasis on calcium homeostasis and mitochondrial function.

### MATERIALS AND METHODS

Animal handling was in accordance with the European Directive for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (86/609/EU). The experiments were approved by the Committee for Experiments on Animals of the Charles University Faculty of Medicine in Pilsen and by the Ministry of Education, Youth and Sports of the Czech Republic (Protocol No. MSMT-24725/2014-05). All experiments were performed in the animal research laboratory at the Faculty of Medicine in Pilsen. Sixteen domestic pigs of both sexes and of similar weight (43.9 ± 5.8 kg) were used for experiments. Sepsis was induced by fecal peritonitis in eight pigs (seven boars, one sow) while control sham experiments (analogous procedure but without sepsis induction) were performed in another eight pigs (four boars, four sows).

### Anesthesia and Instrumentation

Anesthesia and instrumentation protocols were similar to those previously described (Jarkovska et al., 2016). Anesthesia was induced with intramuscular (IM) tiletamine (2.2 mg/kg), zolazepam (2.2 mg/kg), and xylazine (2.2 mg/kg), together with intravenous (IV) propofol 2% (1–2 mg/kg) and maintained with continuous IV propofol (1–4 mg/kg/h) and fentanyl (5– 10 µg/kg/h). Animals were mechanically ventilated (FiO<sup>2</sup> 0.3, PEEP 8 cm H2O, tidal volume 10 ml/kg, respiratory rate adjusted to maintain end/tidal pCO<sup>2</sup> between 4 and 5 kPa), and muscle paralysis was achieved with IV rocuronium (4 mg for induction, 0.2–0.4 mg/kg/h for maintenance). Ringerfundin solution (B. Braun Melsungen AG, Melsungen, Germany) was infused as maintenance fluid (7 ml/kg/h) and normoglycemia (arterial blood glucose level 4.5–7 mmol/L) was maintained using 10% glucose infusion (1–4 ml/kg/h).

All pigs were instrumented with a femoral artery catheter, triple lumen central venous catheter, and pulmonary artery catheter. Silicone drains directed into the anatomical spaces of Morison and Douglas were used for fecal inoculation.

### Experimental Protocol

Experimental protocols were identical to those previously described (Jarkovska et al., 2016). Peritonitis was induced by inoculating 1 g/kg of autologous feces (cultivated for 10 h in 100 ml isotonic saline at 37◦C) into the abdominal cavity. In addition to continuous crystalloid infusion, fluid boluses (10 ml/kg of Ringerfundin) were administered to maintain cardiac output and mean arterial pressure (MAP) in a goal-directed fashion. Continuous IV norepinephrine was administered if MAP fell below 65 mmHg despite fluid administration and titrated to maintain MAP above 70 mmHg. In total, the experiments lasted 34 h (4 h for surgical instrumentation, 6 h of recovery, and 24 h after induction of peritonitis). At the end of the experiment, the animals were euthanized by anesthetic overdose and excision of the heart.

### Measurements

Systemic and pulmonary hemodynamics were measured and electrocardiography (lead II) was performed as described previously (Stengl et al., 2010; Jarkovska et al., 2016). The modified sequential organ failure assessment (SOFA) score was determined according to the Third International Consensus Definitions for Sepsis and Septic Shock (Singer et al., 2016)

and modified by exclusion of the Glasgow Coma Scale-based neurologic component.

Action potentials and isometric contractions in the right ventricle trabeculae were recorded as described previously (Stengl et al., 2008, 2010, 2013). Action potentials were recorded with high-resistance (>20 M) glass microelectrodes filled with 3 M KCl at various stimulation frequencies (3, 2, 1, 0.5 Hz), and simultaneously, isometric contractions were recorded using an isometric force transducer (F30, Hugo Sachs, March-Hugstetten, Germany). All action potential [APD90, action potential duration at the 90% level of repolarization, action potential amplitude (APA), resting membrane potential (RMP)] and contraction (contraction force, time from resting tension to the peak of contraction, time to 90% relaxation) parameters were measured in 5 beats and averaged, and the mean values were used for further analyses and comparisons.

Cardiac myocytes were isolated from the left ventricle by enzymatic dissociation (collagenase A from Sigma-Aldrich, St. Louis, MO, United States) as previously reported (Stengl et al., 2010). ICaL was measured using the whole-cell configuration of the patch-clamp technique at 36◦C (Stengl et al., 2010).

Sarcomeric contractions and calcium transients of isolated cardiac myocytes were measured with Ionoptix HyperSwitch Myocyte Calcium and Contractility System (IonOptix LLC, Westwood, CA, United States), with the Sarclen sarcomere length acquisition module. Cells were loaded with Fura-2 (Molecular Probes, Invitrogen, Waltham, MA, United States). For stock solution Fura-2-am powder was dissolved in dimethyl sulfoxide (DMSO; Sigma-Aldrich, St. Louis, MO, United States) to reach a final concentration of 1 mmol/L. Cells were incubated for 20 min in normal calcium Tyrode solution with 2 µmol/L Fura-2-am and then repeatedly washed with normal calcium Tyrode solution. Measurements were performed in normal Tyrode solution at 37 ± 0.5◦C. Cells were stimulated with a field stimulator (MyoPacer Field Stimulator, IonOptix LLC, Westwood, CA, United States) at frequencies of 3, 2, 1, 0.5 Hz. For offline analysis of sarcomeric contractions and calcium transients, the IonWizard 6.5 software (IonOptix LLC, Westwood, CA, United States) was used.

Cardiac mitochondrial function was assessed using highresolution respirometry (oxygraph Oroboros; Oroboros Instruments, Innsbruck, Austria). Samples of left ventricular myocardium (1.5–2.0 mg) were permeabilized by saponin in BIOPS solution (Cantó and Garcia-Roves, 2015). The fibers were then washed with respiration medium containing catalase and placed into oxygraph chambers with MiR06 medium equilibrated with air. Mitochondrial oxygen consumption was measured at 37◦C after raising oxygen concentration to 400–500 µmol/L by titration of H2O<sup>2</sup> (200 mmol/L). In the titration protocol, different substrates and inhibitors of the mitochondrial respiratory system were sequentially added into the chambers to determine various respiratory states; nonphosphorylating LEAK state (L, oxygen consumption needed for electron transport compensating for proton leak across the inner mitochondrial membrane, induced by the addition of substrates providing electrons to Complex I – malate, 2 mmol/L, glutamate, 10 mmol/L, and pyruvate, 5 mmol/L), OXPHOS I (active phosphorylating respiration induced by 5 mmol/L ADP), OXPHOS Ic (the degree of cell membrane permeabilization verified with cytochrome c 10 µmol/L), OXPHOS I+II (mitochondrial respiration increased by succinate, Complex II substrate, 10 mmol/L), OXPHOS II (reflecting activity of Complex II, induced by inhibition of Complex I by rotenone, 0.5 µmol/L), ROX (residual oxygen consumption after complex III inhibition by antimycin A, 2.5 µmol/L), and Complex IV activity (respirometric assay for cytochrome c oxidase activity by simultaneous injection of N,N,N<sup>0</sup> ,N0 -tetramethyl-pphenylenediamine dihydrochloride, TMPD, 0.5 mmol/L, and ascorbate, 2 mmol/L).

The oxygen consumption was analyzed online by DatLab software (Oroboros Instruments, Innsbruck, Austria) as the negative time derivative of oxygen concentration in the chamber, expressed in pmol O2/(s·mg tissue wet weight), and corrected to ROX.

Citrate synthase activity serving as a marker of mitochondrial content was measured in all samples taken from the oxygraph chambers. The assay medium was mixed with homogenized chamber content and citrate synthase activity was measured spectrophotometrically at 412 nm and 30◦C, and expressed in IU per g tissue weight (Kuznetsov et al., 2002).

### Solutions and Drugs

The composition of the Tyrode solution was as follows (in mmol/L): NaCl 137, KCl 4.5, MgCl<sup>2</sup> 1, CaCl<sup>2</sup> 2, glucose 10, HEPES 5; pH was adjusted to 7.4 with NaOH. The patch-clamp pipette solution contained: cesium glutamate 125, tetraethylammonium chloride 25, MgCl<sup>2</sup> 1, Na2ATP 5, EGTA 1, HEPES 5; pH adjusted to 7.2 with CsOH. BIOPS solution was composed of CaK2EGTA 2.77, K2EGTA 7.23, Na2ATP 5.77, MgCl2·6H2O 6.56, taurine 20, Na<sup>2</sup> Phosphocreatine 15, imidazole 20, dithiothreitol 0.5, and MES hydrate 50, with pH adjusted to 7.1. (Pesta and Gnaiger, 2012). MiR06 respiration medium contained EGTA 0.5, MgCl2·6H2O 3, <sup>K</sup>-lactobionate 60, taurine 20, KH2PO<sup>4</sup> 10, HEPES 20, sucrose 110, fatty acid free bovine serum albumin 1 g/L, and catalase 280 U/mL at pH 7.0 (Gnaiger et al., 2000). The composition of the assay medium for determination of citrate synthase activity was 5,5-dithio-bis-(2 nitrobenzoic) acid 0.1, oxaloacetate 0.5, acetyl coenzyme A 0.31, triethanolamine hydrochloride 5, Tris-HCl 100, and 5 µmo/L EDTA, Triton-X 0.25%, pH adjusted to 8.1). All chemicals were from Sigma-Aldrich (St. Louis, MO, United States).

### Statistical Analysis

Results are presented as means ± SD. After testing for normality of distribution (Shapiro–Wilk test), statistical comparisons were made with the two-way mixed-design ANOVA (one repeatedmeasures factor for analysis of the progression of parameter in time and one between-groups factor for comparison between control and septic groups, in vivo data) followed by post hoc Tukey test or by unpaired t-test (control vs. septic group comparison of in vitro data). The analysis was performed using the software Origin 2017 (OriginLab, Corp., Northampton, MA, United States) or STATISTICA Cz 8 (StatSoft, Inc., Prague, Czechia). Values of p < 0.05 were considered significant.

### RESULTS

In the septic group, development of sepsis and septic shock with organ dysfunction was manifested by a significant increase of the total modified SOFA score (**Figure 1A**) and of serum lactate levels (**Figure 1B**). Out of eight animals in this group, septic shock requiring vasopressors and with lactate levels over 2 mmol/L developed in six animals; in two animals, vasopressor administration was not required, but the criteria for sepsis was fulfilled based on their SOFA scores. Concurrently, the septic animals developed hyperdynamic circulation with increased cardiac output and peripheral vasodilation (**Figures 2A,B**). The increase in cardiac output was predominantly due to elevated heart rate (**Figure 2C**), as the stroke volume was reduced (**Figure 2D**). The control animals did not show any signs of systemic inflammatory reaction, their SOFA scores remained normal (**Figure 1A**), their lactate levels remained low throughout the run of the experiment, and none of them needed vasopressor support. Their global hemodynamic parameters remained unchanged compared to the baseline (**Figures 2A–D**).

In cardiac trabeculae, sepsis induced a shortening of action potential duration (**Figures 3A,B**) at lower stimulation rates (1, 0.5 Hz) and reduction of contraction force (**Figures 3C,D**). Values of APA and RMP were similar in control and septic preparations (e.g., at 1 Hz APA of 101 ± 5 mV in control vs. of 100 ± 12 mV in sepsis; RMP of −71 ± 7 mV in control vs. −70 ± 7 mV in sepsis). Kinetics of trabecular contraction (time to peak, TTP; time to 90% relaxation, R90) were not affected by sepsis (e.g., at 1 Hz TTP of 184 ± 53 ms in control vs. 171 ± 57 ms in sepsis; R90 of 243 ± 80 ms in control vs. 240 ± 61 ms in sepsis).

In isolated myocytes, sarcomeric contractions (**Figures 4A,B**) and calcium transients (**Figures 4C,D**) were reduced in septic

FIGURE 3 | Action potential and contraction in cardiac trabeculae. (A) Action potential duration at 90% repolarization in cardiac trabeculae from control (n = 8) and septic (n = 8) animals. Empty squares, control; filled circles, sepsis. (B) Representative action potentials of cardiac trabeculae from control and septic animals. Stimulation frequency of 1 Hz. Black line, control; gray line, sepsis. (C) Contraction force in cardiac trabeculae from control (n = 8) and septic (n = 8) animals. Empty squares, control; filled circles, sepsis. (D) Representative contractions of cardiac trabeculae from control and septic animals. Stimulation frequency of 1 Hz. Black line, control; gray line, sepsis. <sup>∗</sup> Significantly different from control, p < 0.05.

cells at lower stimulation rates (1, 0.5 Hz). The resting intracellular calcium concentrations were not affected by sepsis at any stimulation frequency (e.g., at 1 Hz ratio of 0.588 ± 0.084 in control vs. 0.623 ± 0.115 in septic myocytes). Kinetic parameters of both sarcomeric contractions and calcium transients (time to 50% peak, TP50; time to 50% relaxation, TR50) were not affected by sepsis (e.g., for sarcomeric contractions at 1 Hz, TP50 of 78 ± 27 ms in control vs. 84 ± 29 ms in septic myocytes; TR50 of 365 ± 117 ms in control vs. 391 ± 104 ms in septic myocytes; for calcium transients at 1 Hz, TP50 of 12 ± 3 ms in control vs. 12 ± 3 ms in septic myocytes; TR50 of 316 ± 90 ms in control vs. 322 ± 110 ms in septic myocytes). ICaL was decreased in septic cardiac myocytes at membrane potentials between −20 and +40 mV (**Figures 5A,B**).

animals. Stimulation frequency of 1 Hz. Black line, control; gray line, sepsis. <sup>∗</sup> Significantly different from control, p < 0.05.

Since sarcomeric contractions and calcium transients were recorded simultaneously in each cardiac myocyte, phase-plane trajectories of sarcomeric length-intracellular Ca2<sup>+</sup> relationship were constructed (**Figure 6A**) and analyzed. The sarcomeric length-intracellular Ca2<sup>+</sup> trajectory during the relaxation phase of the twitch contraction was similar in control and septic myocytes (**Figure 6B**). Linear fitting of these relaxation phase trajectories revealed similar slopes (**Figure 6C**) for control and septic myocytes.

Mitochondrial respiration was suppressed in septic hearts (**Figure 7**). Oxygen consumption in the LEAK state was decreased (**Figure 7C**). Mitochondrial respiration in the presence of ADP and Complex I and II substrates (OXPHOS I+II) was reduced by sepsis, and this reduction was mainly due to inhibition of Complex II (OXPHOS II), while Complex I activity was

from control and septic animals. Left panel, control; right panel, sepsis.

FIGURE 6 | Phase-plane diagrams of sarcomeric length-intracellular Ca2<sup>+</sup> relationship. (A) Mean phase-plane trajectories of sarcomeric length-fluorescence ratio relationship in control (black line, n = 7) and septic (gray line, n = 8) myocytes. (B) Amplified relaxation phase of the mean trajectories with linear fits. Black lines, control; gray lines, sepsis. (C) Slopes of linear fits of relaxation portions of phase-plane trajectories. Empty column, control; filled column, sepsis.

not influenced significantly (**Figure 7D**). Cytochrome c oxidase (Complex IV) activity was also decreased in sepsis (**Figure 7E**).

Citrate synthase activity was not affected by sepsis, reaching 64.2 ± 12 and 62.1 ± 11 IU/g in control and septic samples, respectively.

### DISCUSSION

In our clinically relevant porcine model of peritonitis-induced sepsis/septic shock, we have managed to induce the typical hyperdynamic circulation pattern, with high heart rate but reduced stroke volume and low systemic vascular resistance. On the cellular level, this was associated with shortened action potential duration, decreased contraction force and calcium transient amplitude, and reduced ICaL. Analysis of phase-plane diagrams of sarcomeric length versus calcium concentration (fluorescence ratio) indicated no change in myofilament calcium sensitivity. Mitochondrial respiration was suppressed in septic hearts, predominantly due to an inhibition of Complexes II and IV.

To the best of our knowledge, this is the first complex analysis of cellular and subcellular mechanisms of septic myocardial depression in a clinically relevant large animal (porcine) experimental model. In earlier studies, the cellular effects of sepsis on excitation–contraction coupling in the myocardium were only studied in small animal (rodent) models. Consistent with our porcine data, septic peritonitis rat model cardiac myocytes exhibited a depression of both peak shortening and calcium transients (Ren et al., 2002). Similarly, decreased myocyte shortenings and peak systolic calcium levels, together with slowed-down kinetics of calcium transients, were described in isolated cardiac ventricular cardiomyocytes of rats with sepsis induced by cecal

ligation and puncture (Zhu et al., 2005). In mice with sepsis due to colon ascendens stent peritonitis, cardiac myocytes showed reduced cell shortening, calcium transient amplitude, and sarcoplasmic reticulum calcium content, which was associated with a significant increase in oxidation-dependent calcium and calmodulin-dependent protein kinase II activity (Sepúlveda et al., 2017). In general, the data obtained in rodent models of sepsis/endotoxemia indicate an important role of intracellular calcium homeostasis in septic myocardial depression.

In this study of porcine septic shock, the reduction of cardiac contractile force (documented in multicellular preparations of cardiac trabeculae, as well as in isolated cardiac myocytes) was associated with decreased amplitude of calcium transient, reduced ICaL, and shortened action potential duration. ICaL represents the main entry pathway of calcium into the cardiac myocyte and is crucial for both triggering calcium release from the sarcoplasmic reticulum and replenishing intracellular calcium stores during the plateau phase of the cardiac action potential (Eisner et al., 2017). Reduction of ICaL, together with the shortening of action potential duration, results in a substantial suppression of the calcium influx (Stengl et al., 2010). Similar kinetics of the rising phase of calcium transients in control and septic myocytes suggest that triggering of the sarcoplasmic reticulum calcium release was not significantly affected by the reduction of ICaL, leaving the diminished ICaL calcium influx for replenishing intracellular calcium stores and/or direct stimulation of contractile proteins as the most likely mechanism.

Another possible contributor to septic myocardial depression might be the altered functional properties of myofibrillar proteins. In rats with cecal ligation and puncture sepsis, the phosphorylation of both troponin I and of C protein was increased during the early phase but decreased during the late phase of sepsis (Wu et al., 2001). The decreases in the phosphorylation of troponin I and C protein during late sepsis coincided with the declines in the activities of myofibrillar ATPase and the calcium sensitivity of myofilaments. Similarly, in rabbit non-lethal endotoxemia, a phosphorylationdependent decrease in myofibrillar calcium sensitivity was documented (Tavernier et al., 2001). In contrast to these findings, septic plasma from a canine model of Escherichia coli sepsis failed to decrease isometric tension in the skinned trabecular preparations with chemically disrupted sarcolemmal, sarcoplasmic reticulum, and mitochondrial membranes, which ruled out a direct inhibition of the contractile apparatus by septic plasma (Gu et al., 1998). In line with the canine sepsis data, the phase-plane analysis of sarcomeric lengthintracellular Ca2<sup>+</sup> relationship revealed no significant change of myofilament responsiveness in porcine septic shock. The intracellular Ca2+-cell length trajectory during the relaxation phase of the twitch contraction in single cardiac myocytes defines a quasi-equilibrium of cytosolic calcium, myofilament Ca2<sup>+</sup> binding, mechanical force, and cell length (Spurgeon et al., 1992). Since the position and the slope of the relaxation phase trajectories were virtually identical in control and septic myocytes, the calcium responsiveness of myofilaments was

probably not affected in our experimental setting of porcine sepsis.

The growing consensus that mitochondrial dysfunction contributes to the pathogenesis of sepsis and development of septic cardiomyopathy is mainly based on studies performed in small laboratory animals (Cimolai et al., 2015). Decreased respiratory rates and/or reduced activities of respiratory mitochondrial complexes were reported in septic rabbits (Gellerich et al., 2002), rats (Vanasco et al., 2012), and mice (Piquereau et al., 2013). In majority of studies, inhibition of Complex I was documented, while inhibition of other Complexes was less consistent. Only two studies published so far addressed the myocardial mitochondrial respiratory dysfunction in porcine models of peritonitis-induced sepsis: one of them reported reduced activity of Complex I determined spectrophotometrically at 30◦C and expressed per citrate synthase activity (Li et al., 2007); the other documented no difference in myocardial mitochondrial oxygen consumption between control and septic animals treated with antibiotics for at least 48 h (Corrêa et al., 2012). In our current study, suppression of mitochondrial respiration related to Complexes I and II (OXPHOS I+II) could be attributed mainly to inhibition of Complex II. Oxygen consumption by artificially stimulated Complex IV was also reduced, while activity of Complex I was not significantly influenced. Expression of oxygen consumption per citrate synthase activity did not affect the pattern of changes. Decreased LEAK state with reduced mitochondrial respiration expressed per mg tissue wet weight would suggest reduced mitochondrial content and/or swelling of the tissue due to fluid resuscitation; however, unchanged citrate synthase activity argues against these options. The experimental discrepancies are probably related to species differences, as well as variable experimental protocols (models of sepsis, duration of sepsis, severity of insult). Is there a direct link between abnormal calcium handling and mitochondrial dysfunction? It is wellknown that cardiac sarcoplasmic reticulum and mitochondria closely interact, forming a mitochondrial calcium microdomain (Kohlhaas and Maack, 2013). The sarcoplasmic reticulum Ca2+- ATPase preferentially consumes mitochondrial ATP for active transport of cytosolic calcium back to the sarcoplasmic reticulum (Kaasik et al., 2001). On the other hand, calcium released from the sarcoplasmic reticulum enters the mitochondria, even on a beat-to-beat basis (Andrienko et al., 2009), and regulates mitochondrial enzymes of the tricarboxylic acid cycle, the proteins of the electron transport chain, and the ATP synthase (Williams et al., 2015), thus matching energy supply to demand. Accordingly, in our study, decreased contraction and calcium release were accompanied by decreased mitochondrial respiration, but the question remains as to what the primary event is. In our opinion, the facts that the kinetics of calcium transients decline and the myofilament calcium responsiveness were not affected by sepsis argue against the primary role of the mitochondria and insufficient energy supply, and rather indicate defective calcium transport (reduced ICaL with consequent alterations of sarcoplasmic reticulum calcium release) as the primary event of myocardial depression.

## CONCLUSION

fphys-09-00726 June 8, 2018 Time: 15:38 # 10

Defective calcium handling with reduced calcium current and transients, together with inhibition of mitochondrial respiration, appear to represent the dominant cellular mechanisms of myocardial depression in porcine septic shock. Consequently, these molecular mechanisms may help to identify potential therapeutic targets for preventing and/or reversing sepsisinduced myocardial dysfunction. In this line of thinking, calcium channel openers represent an obvious option. Bay K 8644, a dihydropyridine calcium channel agonist, was shown to enhance ICaL in cardiac myocytes from endotoxemic rats (Abi-Gerges et al., 1999). In vivo, BAY K 8644 elevated blood pressure in endotoxin-shocked rats (Ives et al., 1986) as well as in endotoxemic dogs (Preiser et al., 1991). On the other hand, in cardiac myocytes from endotoxemic guinea pigs BAY K 8644 did not reverse the endotoxin-induced reduction in peak ICaL, cell contraction, and systolic intracellular calcium concentration (Zhong et al., 1997). Mitochondrial respiration is another promising target for potential therapeutic interventions. In endotoxemic rat models, mitochondriatargeted antioxidants were demonstrated to reduce inflammatory responses, mitochondrial damage and organ dysfunction including cardiac depression (Supinski et al., 2009; Lowes et al., 2013).

### STUDY LIMITATIONS

The modified SOFA score was determined according to human sepsis criteria (Singer et al., 2016). Despite generally similar physiology and sepsis progression in this porcine model and humans, it is possible that some criteria are not completely compatible and will require further validation in the model.

### REFERENCES


Furthermore, sepsis was induced in young, healthy animals and the translation to elderly intensive care unit (ICU) patients with multiple comorbidities that may affect cardiac calcium handling and mitochondrial function themselves (e.g., heart failure) might be difficult.

The porcine peritonitis-induced sepsis model shows a clear hyperdynamic phenotype and therefore care should be taken to not generalize the results to the hypodynamic phenotypes. It remains, for future studies with more appropriate models, to determine whether or not and to what extent the above described mechanisms contribute to hypodynamic sepsis.

The upstream mechanisms that induce cardiac cellular alterations were not addressed in this study. Extensive research of possible candidates and signaling pathways (e.g., cytokines, oxidative and nitrosative stress, circulating histones) is clearly warranted.

### AUTHOR CONTRIBUTIONS

JH, LN, and JB conducted the in vivo experiments, DJ, JS, LN, MS, and MA participated in cellular and tissue cardiac experiments; MiM and JK conducted the mitochondrial experiments, and MaM and MS conceived of and designed the study and drafted the manuscript. All authors participated in interpretation of the studies, analysis of the data, review of the manuscript, and read and approved the final manuscript.

## FUNDING

This study was supported by the Grant Agency of the Czech Republic (Project No. 15-15716S).



**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 Jarkovska, Markova, Horak, Nalos, Benes, Al-Obeidallah, Tuma, Sviglerova, Kuncova, Matejovic and Stengl. 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.

# HMGB1 Increases IL-1β Production in Vascular Smooth Muscle Cells via NLRP3 Inflammasome

Eun Jung Kim1,2†, So Youn Park 1,2†, Seung Eun Baek 1,2, Min A. Jang1,2, Won Suk Lee<sup>1</sup> , Sun Sik Bae1,2, Koanhoi Kim<sup>1</sup> and Chi Dae Kim1,2 \*

<sup>1</sup> Department of Pharmacology, School of Medicine, Pusan National University, Yangsan, South Korea, <sup>2</sup> Gene and Cell Therapy Research Center for Vessel-associated Diseases, Pusan National University, Yangsan, South Korea

Vascular smooth muscle cells (VSMCs) are the major cell type in the blood vessel walls, and their phenotypic modulation is a key cellular event driving vascular remodeling. Although high mobility group box-1 (HMGB1) plays a pivotal role in inflammatory processes after vascular injuries, the importance of the links between VSMCs, HMGB1 and vascular inflammation has not been clarified. To prove the hypothesis that VSMCs might be active players in vascular inflammation by secreting inflammatory cytokines, we investigated the proinflammatory effects of HMGB1 and its intermediary signaling pathways in VSMCs. When cultured human VSMCs were stimulated with HMGB1 (10–500 ng/ml), IL-1β production was markedly increased. HMGB1 also increased the expression of NLRP3 inflammasome components including NLRP3, ASC and caspase-1. Among these components, HMGB1-induced expressions of NLRP3 and caspase-1 were markedly attenuated in TLR2 siRNA-transfected cells, whereas ASC and caspase-1 expressions were reduced in RAGE-deficient cells. In TLR4-deficient cells, HMGB1-induced caspase-1 expression was significantly attenuated. Moreover, IL-1β production in HMGB1-stimulated cells was significantly reduced in cells transfected with caspase-1 siRNA as well as in cells treated with monoclonal antibodies or siRNAs for TLR2, TLR4 and RAGE. Overall, this study identified a pivotal role for NLRP3 inflammasome and its receptor signaling involved in the production of IL-1β in VSMCs stimulated with HMGB1. Thus, targeting HMGB1 signaling in VSMCs offers a promising therapeutic strategy for treating vascular remodeling diseases.

#### Keywords: HMGB1, IL-1β, NLRP3 inflammasome, VSMC, TLRs, RAGE

### INTRODUCTION

Vascular inflammation plays an important role in the pathogenesis of vascular diseases, including vascular remodeling and atherosclerosis (Libby, 2002; Davis et al., 2003). During disease processes, inflammatory mediators are derived from inflammatory cells in vascular lesions, and promote the development of stenotic lesions through the proliferation and migration of vascular smooth muscle cells (VSMCs) (Galis and Khatri, 2002; Waitkus-Edwards et al., 2002). VSMCs are the major cell type in blood vessel walls and play pivotal roles in vascular disease processes by changing phenotype from the contractile to the synthetic phenotype, the latter exhibits distinct proliferative and migratory abilities and produces proinflammatory cytokines (Wang et al., 2012; Ackers-Johnson et al., 2015).

#### Edited by:

Maria Elena Soto, Instituto Nacional de Cardiología, Mexico

#### Reviewed by:

Karen Yvonne Stokes, LSU Health Sciences Center New Orleans, United States Adán Dagnino-Acosta, University of Colima, Mexico

> \*Correspondence: Chi Dae Kim chidkim@pusan.ac.kr

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Vascular Physiology, a section of the journal Frontiers in Physiology

Received: 18 September 2017 Accepted: 14 March 2018 Published: 28 March 2018

#### Citation:

Kim EJ, Park SY, Baek SE, Jang MA, Lee WS, Bae SS, Kim K and Kim CD (2018) HMGB1 Increases IL-1β Production in Vascular Smooth Muscle Cells via NLRP3 Inflammasome. Front. Physiol. 9:313. doi: 10.3389/fphys.2018.00313

**136**

The phenotypic modulation of VSMCs is a key cellular event that drives neointima formation and vascular remodeling. Reportedly, the phenotypic modulation of VSMCs induced by interferon-γ is mediated by high mobility group box-1 (HMGB1), a non-histone chromosomal protein (Wang et al., 2017). HMGB1 is released by monocytes/macrophages in response to inflammatory stimuli and then binds to DNA in a sequence-independent manner and modifies DNA structure, and thereby facilitates gene transcription (Kalinina et al., 2004; Stott et al., 2006; Yang et al., 2015). In blood vessels, high levels of extracellular HMGB1 have been detected in human atherosclerotic plaque, and reportedly are implicated in vascular inflammation by potentiating inflammatory responses (Kalinina et al., 2004; Cai et al., 2015). It has also been shown that HMGB1 modulates the phenotype of VSMCs toward the activated synthetic phenotype and stimulates MCP-1/CCL2 gene expression through toll-like receptor 4 (TLR4) (Cai et al., 2015; Wang et al., 2017). However, the nature of the links between VSMCs, HMGB1, and vascular inflammation have not been clarified.

Inflammasomes are key regulators of HMGB1-induced inflammation (Chi et al., 2015). These are cytoplasmic, highmolecular weight, multisubunit protein complexes capable of inducing inflammatory response by releasing IL-1β (Guo et al., 2015; Lamkanfi and Dixit, 2015), which is a pivotal player in vascular inflammatory processes (Kirii et al., 2003; Chamberlain et al., 2006). Inflammasomes are composed of an inflammasome sensor molecule, the adaptor protein ASC (apoptosis-associated speck-like protein), and procaspase-1 (Martinon et al., 2002; Chamberlain et al., 2006; Guo et al., 2015). Following activation, the inflammasome complex induces autocatalytic cleavage of procaspase-1 into its active form, which can cleave pro-IL-1β into its mature and released forms (Netea and Joosten, 2015). Thus, it is proposed that HMGB1 contributes to vascular inflammation by promoting the activation of NLRP3 (NLR family pyrin domain containing proteins 3) inflammasome and the processing of IL-1β in vascular cells.

Cell membrane pattern recognition receptors (PRRs), including TLR2, TLR4, and RAGE (receptor for advanced glycation end products), interact with HMGB1, and then mediate the production of pleiotropic cytokines in inflammatory cells (Park et al., 2003, 2006; Treutiger et al., 2003). In monocytes/macrophages, HMGB-1 binding to TLR4 enhanced inflammatory response through the synthesis of IL-1β (Andersson et al., 2002; Yang et al., 2010). which plays an essential role in the complex inflammatory process. Previous studies have also reported that TLR2 and TLR4 can activate NLRP3 inflammasome, and thus facilitate the secretions of inflammatory cytokines (Qi et al., 2014; Koch and Müller, 2015), which indicates all three PRRs may mediate HMGB1-induced inflammation. Although it was suggested that HMGB1 secretion by activated VSMCs seems to be critically involved in vascular inflammation, its role in the production of inflammatory cytokines in human VSMCs is not well clarified.

Inflammasomes are molecular platforms that trigger the maturation of proinflammatory cytokines in inflammatory cells when cells are exposed to stress. In the injured vasculatures, HMGB1 plays a pivotal role in the process of vascular inflammation. However, the links between HMGB1 and inflammasome in VSMCs has not been clarified. To prove the hypothesis that VSMCs might be active players in vascular inflammation by secreting inflammatory cytokines, we investigated the proinflammatory effects of HMGB1 and its intermediary signaling pathways in human VSMCs.

### MATERIALS AND METHODS

### Chemicals and Antibodies

Recombinant human high mobility group box 1 (HMGB1), anti-NLRP3, and anti-IgG antibodies were purchased from R&D Systems, Inc. (Minneapolis, MN). NLRP3-inflammasome inhibitor, MCC-950 was purchased from Invivogen (San Diego, CA). Anti-IL-1β, anti-TLR2, anti-TLR4, anti-RAGE, and anticaspase-1 antibodies were from Abcam (Cambridge, MA). β-Actin antibody was purchased from Santa Cruz Biotechnology Inc. (Beverly, MA), and anti-ASC antibody from Adipogen (San Diego, CA).

## Cell Culture

Human aortic smooth muscle cells were purchased from the ATCC (Manassas, VA). Cells were grown in culture dishes using smooth muscle cell growth medium (Gibco BRL, Grand Island, NY), smooth muscle growth supplement (Gibco BRL), and 10% fetal bovine serum (FBS), antibiotic antimycotic (Gibco BRL).

### RNA Isolation and RT-PCR

IL-1β and NLRP3 mRNA expressions in VSMCs were quantified by RT-PCR using GAPDH mRNA as an internal standard. Total RNA was isolated using Trizol reagent (Invitrogen, Carlsbad, NY), according to the manufacturer's instructions. RNA (1 µg) were reverse transcribed into cDNA using the ImProm-II reverse transcription system (Promega, Madison, WI). PCR amplification was performed using the IL-1β specific primers (forward, 5′ -GGG CCT CAA GGA AAA GAA TC-3′ ; reverse, 5′ - TTC TGC TTG AGA GGT GCT GA-3′ ) and the NLRP3 specific primers (forward, 5′ -GCG CCT CAG TTA GAG GAT GT-3′ ; reverse, 5′ -ACC AGC TAC AAA AAG CAT GGA-3′ ). Equal amounts of RT-PCR products were separated on 1% agarose gels and stained with ethidium bromide. Signals from bands were quantified using the Image J densitometry program, and data were expressed as relative GAPDH densities.

### Western Blot Assay

Cells lysates were prepared in lysis buffer (Thermo Scientific, Rockford, IL), and equal amounts of the proteins obtained were separated on 10∼15% polyacrylamide gels under reducing conditions, and then transferred to nitrocellulose membranes (Amersham-Pharmacia Biotech, Piscataway, NJ). Membranes were blocked with 5% skim milk in tris buffered saline Tween-20 (TBST), incubated overnight with primary antibody in 5% skim milk, washed with TBST, and incubated with HRP-conjugated secondary antibody for 2 h. Blots were developed using enhanced chemiluminescence (ECL) Western blot detection reagents (Amersham-Pharmacia Biotech).

### Small Interfering RNA (siRNA) Preparation and Transfection

TLR2 (GenBank accession no. NM\_003264.3), TLR4 (GenBank accession no. NM\_1385554.2), RAGE (GenBank accession no. NM\_001136.3), and caspase-1 (GenBank accession no. NM\_021571.2) siRNA oligonucleotides were synthesized by Bioneer (Daejeon, Korea). The siRNA negative control duplex was used as a control oligonucleotide. siRNA molecules were transfected into cells using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions.

### Enzyme-Linked Immunosorbent Assay

Levels of IL-1β in the culture media were measured using ELISA kits (Affymetrix, Inc., Santa Clara, CA) according to the manufacturer's instructions.

### Statistical Analysis

Results were expressed as means ± SEMs. The analysis was conducted using one-way analysis of variance (ANOVA) followed Dunnett Multiple Comparison Test. The Student's t-test was used to determine the significances of treatment effects. Statistical significance was accepted for p-values < 0.05

### RESULTS

### HMGB1 Increased IL-1β Production in Human Vascular Smooth Muscle Cells

To assess the effect of HMGB1 on IL-1β production in cultured human VSMCs, cells were serum starved for 24 h, and then stimulated with recombinant HMGB1 (HMGB1) at various concentrations. RT-PCR and Western blot analyses indicated that IL-1β mRNA and protein were expressed at low levels in VSMCs in the absence of HMGB1 stimulation, but when VSMCs were stimulated with 10–500 ng/ml of HMGB1, both IL-1β mRNA expression and protein production were markedly increased. In this experiment, dose-dependency was observed up to HMGB1 concentration of 100 ng/ml, and thus, HMGB1 was used at this concentration in subsequent experiments.

As shown in **Figures 1A,B**, RT-PCR analysis showed that IL-1β mRNA levels after treatment with HMGB1 (100 ng/ml, 6 h) significantly increased by 2.88 ± 0.25-fold (P < 0.01). We next analyzed the subsequent cleavage and secretion of active IL-1β, as VSMCs can be situationally an immune effector cells. When VSMCs were stimulated with HMGB1 (100 ng/ml), levels of active IL-1β protein significantly increased up to 24 h (**Figures 1C,D**). Similarly, levels of IL-1β in culture medium were significantly increased at 12–48 h and peaked at 24 h (**Figures 1E,F**).

### HMGB1 Increased the Expression of NLRP3 Inflammasome Components

NLRP3-dependent inflammasome is multiprotein complex, consisting of NLRP3, ASC and caspase-1, which processes proIL-1β into IL-1β (Dinarello, 2009). We then examined whether HMGB1 could affect the expression of the NLRP3 inflammasome in human VSMCs.

NLRP3 mRNA levels peaked after 12 h of stimulation with HMGB1 (100 ng/ml), and this induction was also observed when cells were stimulated with HMGB1 at various concentrations (0–500 ng/ml) (**Figures 2A,B**). Furthermore, stimulation of human VSMCs with HMGB1 (100 ng/ml) markedly increased NLRP3 (3.43 ± 0.62-fold, P < 0.01) and ASC (2.47 ± 0.34-fold, P < 0.01) at 24 h and caspase-1 (8.95 ± 2.01-fold, P < 0.01) at 48 h, respectively (**Figures 2C–E**).

### Dependence of HMGB1-Induced NLRP3 Inflammasome on TLR2, TLR4, and RAGE

It was reported that extracellular HMGB1 stimulates inflammatory cells by activating its receptors, including TLR2, TLR4, and RAGE that are involved in passive process of inflammation (Mitola et al., 2006). In the present study, TLR2, TLR4, and RAGE were found to be constitutively expressed on cultured human VSMCs (data not shown).

To identify the receptors that mediate NLRP3 inflammasome expression in HMGB1-stimulatd VSMCs, the expression of inflammasome components was determined in cells transfected with siRNAs for TLR2, TLR4, or RAGE. The transfection of target receptor-specific siRNAs (200 nM) reduced the protein expression of TLR2, TLR4, and RAGE to ∼59, ∼86, and ∼67% of the control level, respectively (**Figures 3A–C**). In cells transfected with scrambled siRNA duplex (negative controls), HMGB1 (100 ng/ml) significantly elevated the protein levels of NLRP3 (2.45 ± 0.20-fold, P < 0.01), ASC (2.19 ± 0.25-fold, P < 0.05) and caspase-1 (2.42 ± 0.35-fold, P < 0.05) (**Figures 3D–F**). TLR2 gene-knockdown cells exhibited significantly less protein levels of NLRP3 (1.10 ± 0.09-fold, P < 0.01) and caspase-1 (1.15 ± 0.13 fold, P < 0.05) in response to HMGB1 (100 ng/ml), while the protein levels of ASC (1.55 ± 0.13-fold) showed marginal effect (**Figure 3D**). TLR4 gene-knockdown cells showed a significant attenuation only in the caspase-1 protein expression increased by HMGB1 (1.03 ± 0.21-fold, P < 0.05; **Figure 3E**). In cells subjected to RAGE gene-knockdown, the expressions of ASC and caspase-1 protein were similarly decreased in response to HMGB1, while the expression of NLRP3 protein was not affected by RAGE gene-knockdown (**Figure 3F**).

### Dependences of HMGB1-Induced IL-1β Production on TLR2, TLR4, and RAGE

To further investigate the functional role of HMGB1 receptors on the production of inflammatory cytokines in VSMC, we measured IL-1β produced in cells deficient of TLR2, TLR4 and RAGE. As shown in **Figure 4A**, IL-1β production was markedly increased in HMGB1 (100 ng/ml)-stimulated VSMCs transfected with negative controls (19.94 ± 0.34 pg/ml, P < 0.01), which was significantly attenuated in cells transfected with siRNAs for TLR2, TLR4, and RAGE. In cells pretreated with neutralizing antibodies, as was expected, IL-1β induction by HMGB1 was significantly inhibited by anti-TLR2 antibody (8.36 ± 0.61-fold, P < 0.01), anti-TLR4 antibody (8.34 ± 0.69-fold, P < 0.01) and anti-RAGE antibody (8.38 ± 0.67-fold, P < 0.01), whereas anti-IgG antibody had no effect (**Figure 4B**). In addition, Western blot

FIGURE 1 | Effects of HMGB1 on IL-1β expression and its release from VSMCs. VSMCs were treated with HMGB1 (100 ng/ml) for 0–24 h, and were also treated with HMGB1 (0–500 ng/ml) for 6 h. (A,B) The mRNA levels of IL-1β were determined by RT-PCR. GAPDH was used as a control. Data are expressed as means ± SEMs of duplicates pooled from 4 independent experiments. (C,D) The protein levels of active IL-1β were determined by Western blot. β-Actin expression served as an internal control. Data are expressed as means ± SEMs of duplicates pooled from 4 independent experiments. (E,F) VSMCs were treated with HMGB1 (100 ng/ml) for 0–48 h, and were also treated with HMGB1 (0–500 ng/ml) for 24 h. The levels of IL-1β in the culture media were quantified by ELISA. Data are expressed as means ± SEMs of triplicates pooled from 4 independent experiments. \*P < 0.05, \*\*P < 0.01, and \*\*\*P < 0.001 vs. control (untreated cells).

analysis showed that the extent of induction of active IL-1β protein by HMGB1 (100 ng/ml) was significantly reduced by pretreating cells with anti-TLR2 antibody, anti-TLR4 antibody or anti-RAGE antibody (**Figure 4C**). These results suggest that HMGB1 increases the production of IL-1β via NLRP3 inflammasome activated by TLR2, TLR4, and RAGE signaling pathways.

## HMGB1 Increases the Production of IL-1β via Activation of NLRP3 Inflammasome

The stimulatory effect of HMGB1 on the expression of NLRP3 components in human VSMCs prompted us to investigate the effect of HMGB1 on the production of IL-1β. To determine the activity of NLRP3 inflammasome in HMGB1-stimulated cells, we measured cleaved caspase-1 (p20) protein levels in this study.

As shown in **Figure 5A**, stimulation of VSMCs with HMGB1 (100 ng/ml) increased the activity of caspase-1, as determined by cleaved caspase-1 (p20) levels. Likewise, the production of active IL-1β was markedly increased in cells stimulated with HMGB1.

To further determine the role of caspase-1 on the production of inflammatory cytokines, we measured IL-1β production in VSMCs transfected with caspase-1 siRNA. Transfection of VSMCs with caspase-1 siRNA (200 nM) reduced the expression of caspase-1 to ∼42% of the control level (data not shown). In caspase-1-deficient cells, active IL-1β was not increased in response to HMGB1 (100 ng/ml), whereas it was significantly increased in HMGB1-stimulated cells transfected with negative control (**Figure 5B**). In addition, the production of active IL-1β protein induced by HMBG1 was significantly suppressed when cells were pretreated with MCC-950, a selective inhibitor of the NLRP3 inflammasome (**Figure 5C**).

### DISCUSSION

Vascular injury initiates inflammatory cell infiltration into damaged tissues, and this is followed by increased production of inflammatory cytokines, which can cause vascular remodeling diseases (Eid et al., 2009; Li et al., 2012; Song et al., 2012). Previous studies have argued that IL-1β plays a crucial regulatory role in vascular inflammation, but little is known of the role played by VSMCs in damaged vasculatures. Here, we provide direct evidence that stimulation of human VSMCs with HMGB1 induced the release of IL-1β in association with an increased expression of NLRP3 inflammasome components, including NLRP3, ASC and caspase-1. In addition, HMGB1-induced IL-1β production by VSMCs was attenuated by inhibiting TLR2, TRL4 and RAGE signaling pathways or by inhibition of NLRP3 inflammasome. These observations suggest that VSMCs actively participate in vascular inflammatory processes by secreting inflammatory cytokines.

HMGB1 is one of the best characterized damage-associated molecular pattern, and activates NLRP3 inflammasome and NF-κB in inflammatory cells (Pisetsky et al., 2008; Chi et al., 2015). NLRP3 inflammasome is composed of NLRP3 protein, the adaptor molecule ASC, which contains two death-fold domains (one pyrin domain and one CARD), and procaspase-1 (Sutterwala et al., 2006; Chae et al., 2011). NLRP3 inflammasome can be activated by a variety of stimuli, such as extracellular ATP released by dying cells (Mariathasan et al., 2006), the phospholipid cardiolipin, mitochondrial DNA (Nakahira et al., 2011; Iyer et al., 2013), and bacterial toxins (Muñoz-Planillo et al., 2009). NLRP3 inflammasome activation requires two steps: priming and the inflammasome complex assembly (Sutterwala et al., 2014). The priming step is initiated by pattern recognition receptors, cytokine receptors, or any factor able to induce the activation of NF-κB, and these initiations result in the upregulation of NLRP3 to a functional level and pro-IL-1β expression (Sutterwala et al., 2014). The second step is posttranscriptional and enables the assembly of NLRP3 inflammasome complex (Bauernfeind et al., 2009; Sutterwala et al., 2014). In the present study, stimulation of VSMCs with HMGB1 upregulated the mRNA and protein expressions of NLRP3 components, including NLRP3, ASC and caspase-1, which suggests HMGB1 can initiate the activation of inflammasome in VSMCs. However, it remains to be determined

duplicates pooled from 3 independent experiments. \*P < 0.05, \*\*P < 0.01, and \*\*\*P < 0.001 vs. control (untreated cells). (D–F) VSMCs were transfected with siRNA (200 nM) for TLR2, TLR4, and RAGE, and then stimulated with HMGB1 for 24 h. The expressions of NLRP3, ASC, and Caspase-1 were determined by Western blot using β-actin as an internal control. Data are expressed as means ± SEMs of duplicates pooled from 4 to 6 independent experiments. \*P < 0.05, \*\*P < 0.01 vs. corresponding value in untreated cells. #P < 0.05 and ##P < 0.01 vs. corresponding value in negative control.

whether the assembly of NLRP3 inflammasome complex in VSMCs requires the presence of HMGB1.

HMGB1 primarily resides in the nuclei of quiescent cells, when cells are exposed to stress, HMGB1 can be translocated into the extracellular milieu, where it elicits the production of proinflammatory mediators and induces the infiltration of inflammatory cells (Harris et al., 2012; Yang et al., 2013). HMGB1 has been implicated in the pathogenesis of a variety of inflammatory diseases, and under these pathological conditions, the levels of HMGB1 are elevated in tissues and serum associated with the development of inflammation (Kalinina et al., 2004; Andrassy et al., 2008). Studies have showed that high levels of extracellular HMGB1 in atherosclerotic plaque were found in human blood vessels, and suggested its involvement in vascular inflammation via the potentiation of inflammatory processes (Kalinina et al., 2004; Cai et al., 2015). In addition to macrophages and endothelial cells, VSMCs have been identified as a major source of HMGB1 production in atherosclerotic lesions, and recent evidence suggests HMGB1 is required for the development of vascular inflammation and neointimal lesions following vascular injury (Chen et al., 2012; Zou et al., 2014). Furthermore, HMGB1 was reported to modulate the phenotype of VSMC toward activated synthetic type (Cai et al., 2015; Wang et al., 2017). Therefore, HMGB1 secretion by activated VSMCs might be critically involved in the process of vascular inflammation, and suggested as a potential therapeutic target to prevent vascular remodeling diseases.

Recently, it was suggested that VSMCs might be an active player in vascular inflammatory processes by secreting inflammatory cytokines (Wang et al., 2012; Ackers-Johnson et al., 2015). In a previous study, IL-1β was found to play an essential role in the complex inflammatory process by modulating the expression of genes induced by the transcription factors AP-1 and NF-κB, and also enhance VSMC proliferation and migration via P2Y2 receptor-mediated RAGE expression and HMGB1 release (Eun et al., 2015). Furthermore, it has been reported that arterial neointima formation and atherosclerotic lesion areas were lower in IL-1β knockout mice than in controls (Kirii et al., 2003; Chamberlain et al., 2006), suggesting IL-1β as a pivotal player in vascular inflammatory processes. Based on these reports and our present data in which stimulation of VSMC with HMGB1 increased NLRP3 inflammasome

FIGURE 4 | Functional role of HMGB1 receptors on the production of IL-1β in HMGB1-stimulated VSMCs. (A) VSMCs were transfected with siRNA (200 nM) for TLR2, TLR4, and RAGE, and then incubated with HMGB1 for 24 h. The levels of IL-1β in the culture media were quantified by ELISA. Data are expressed as means ± SEMs of triplicates pooled from 4 independent experiments. \*\*P < 0.01 vs. non-treated control. ##P < 0.01 vs. corresponding value in negative control. (B) VSMCs were pretreated with anti-IgG antibody (10µg/ml), anti-TLR2 antibody (10µg/ml), anti-TLR4 antibody (10µg/ml), or anti-RAGE antibody (10µg/ml) for 30 min, and then stimulated with HMGB1 (100 ng/ml) for 24 h. IL-1β release into culture media was quantified by ELISA. Data are expressed as means ± SEMs of triplicates pooled from 4 independent experiments. (C) Protein levels of active IL-1β were assessed by Western blot using β-actin as an internal control. Data are expressed as means ± SEMs of duplicates pooled from 4 independent experiments. \*\*P < 0.01 vs. non-treated control. #P < 0.05 and ##P < 0.01 vs. control in HMGB1-treated cells.

FIGURE 5 | Functional role of NLRP3 inflammasome on the production of IL-1β in HMGB1-stimulated VSMCs. (A) VSMCs were treated with HMGB1 (100 ng/ml) for 0–48 h. The protein levels of caspase-1, active caspase-1 (p20), pro- IL-1β, and active IL-1β were assessed by Western blot using β-actin as an internal control. (B) VSMCs were transfected with caspase-1 siRNA (200 nM), and then stimulated with HMGB1 for 24 h. The expression of active IL-1β was determined by Western blot using β-actin as an internal control. Data are expressed as means ± SEMs of duplicates pooled from 4 independent experiments. \*P < 0.05 vs. non-treated control. ##P < 0.01 vs. corresponding value in negative control. (C) VSMCs were pretreated with MCC-950 (100 nM; a NLRP3 inhibitor) for 30 min and then stimulated with HMGB1 for 24 h. The protein levels of active IL-1β were determined by Western blot using β-actin as an internal control. Data are expressed as means ± SEMs of duplicates pooled from 4 independent experiments. \*\*P < 0.01 vs. non-treated control in vehicle. ###P < 0.001 vs. corresponding value in HMGB1-treated cells.

components in association with an increase in IL-1β production, it was suggested that the HMGB1-stimulated VSMCs actively induce vascular inflammation by producing inflammatory cytokines.

The proinflammatory function of HMGB1 relies on its binding to certain cell membrane pattern recognition receptors (PRRs), including RAGE, TLR2, and TLR4 (van de Veerdonk et al., 2011). Previous studies have reported that the activations of RAGE, TLR2, or TLR4 can increase the expression of proinflammatory cytokines in different cell types (Szomolanyi-Tsuda et al., 2006; He et al., 2012; Rhee et al., 2013). Moreover, TLR2 and TLR4 can also activate NLRP3 inflammasomes, which then facilitate the maturation and secretion of inflammatory cytokines (Qi et al., 2014; Koch and Müller, 2015). Thus, all three PRRs were suggested as mediators involved in the HMGB1 induced production of inflammatory cytokines in keratinocytes. In line with these results, this study also showed the constitutive expression of HMGB1-binding PRRs including RAGE, TLR2, and TLR4 in cultured human VSMCs. Interestingly, the HMGB1-induced expression of NLRP3 and caspase-1 was markedly attenuated in TLR2-deficient cells, whereas ASC and caspase-1 expression was inhibited in RAGE-deficient cells. In TLR4-deficient cells, HMGB1-induced caspase-1 expression was markedly attenuated. Although the expression of inflammasome components was differentially affected by inhibition of various PRRs, caspase-1 expression induced in HMGB1-stimulated cells was markedly reduced in all cells transfected with siRNAs for TLR2, TLR4 and RAGE. Thus, caspase-1 was considered as a common component that mediates PRRs signals in HMGB1 stimulated VSMCs. On the basis of the importance of caspase-1 in the activation of inflammasome and the experimental results of this study in which IL-1β production in HMGB1-stimulated

### REFERENCES


cells was markedly reduced in cells transfected with caspase-1 siRNA, it is possible that IL-1β production in HMGB1-stimulated cells might be similarly reduced among cells treated with various inhibitors for TLR2, TLR4, and RAGE.

In conclusion, the principal finding made in this study was that HMGB1 induced IL-1β production in VSMCs via the increased expression of NLRP3 inflammasome components with subsequent activation of caspase-1. Moreover, this study identified a novel receptor signaling pathway involved in the expression of inflammasome components in VSMCs stimulated with HMGB1. Overall, this study provides new insights of innate responses that contribute to the pathogenesis of vascular inflammation, thus, targeting HMGB1 signaling in VSMCs offers a promising therapeutic strategy for treating vascular remodeling diseases.

### AUTHOR CONTRIBUTIONS

EK and SP designed and performed experiments, analyzed the experimental data, and wrote the manuscript. CK contributed to design and the writing. MJ and SEB performed experiments. WL, SSB, and KK approved manuscript.

### FUNDING

This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and future Planning (NRF-2016R1A2B2011509). This work was also supported by the Medical Research Center (MRC) Program through the NRF grant funded by the Korea government (MSIP) (NRF-2015R1A5A2009656).

circulating cells modulates the extent of neointima formation in mice. Am. J. Pathol. 168, 1396–1403. doi: 10.2353/ajpath.2006.051054


**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 Kim, Park, Baek, Jang, Lee, Bae, Kim and Kim. 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 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.

# Pyrrolidine Dithiocarbamate (PDTC) Attenuates Cancer Cachexia by Affecting Muscle Atrophy and Fat Lipolysis

Chunxiao Miao<sup>1</sup>† , Yuanyuan Lv<sup>1</sup>† , Wanli Zhang<sup>1</sup> , Xiaoping Chai<sup>1</sup> , Lixing Feng1,2 , Yanfen Fang1,3 \*, Xuan Liu<sup>2</sup> \* and Xiongwen Zhang<sup>1</sup> \*

<sup>1</sup> Shanghai Engineering Research Center of Molecular Therapeutics and New Drug Development, College of Chemistry and Molecular Engineering, East China Normal University, Shanghai, China, <sup>2</sup> Institute of Interdisciplinary Integrative Biomedical Research, Shanghai University of Traditional Chinese Medicine, Shanghai, China, <sup>3</sup> Division of Anti-tumor Pharmacology, Shanghai Institute of Materia Medica (CAS), Chinese Academy of Sciences, Shanghai, China

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

#### Reviewed by:

Satish Ramalingam, SRM University, India Jiiang-Huei Jeng, National Taiwan University, Taiwan

#### \*Correspondence:

Xiongwen Zhang xwzhang@sat.ecnu.edu.cn Xuan Liu 13764960370@163.com Yanfen Fang yffang@simm.ac.cn

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Inflammation Pharmacology, a section of the journal Frontiers in Pharmacology

Received: 19 October 2017 Accepted: 30 November 2017 Published: 12 December 2017

#### Citation:

Miao C, Lv Y, Zhang W, Chai X, Feng L, Fang Y, Liu X and Zhang X (2017) Pyrrolidine Dithiocarbamate (PDTC) Attenuates Cancer Cachexia by Affecting Muscle Atrophy and Fat Lipolysis. Front. Pharmacol. 8:915. doi: 10.3389/fphar.2017.00915 Cancer cachexia is a kind of whole body metabolic disorder syndrome accompanied with severe wasting of muscle and adipose tissue. NF-κB signaling plays an important role during skeletal muscle atrophy and fat lipolysis. As an inhibitor of NF-κB signaling, Pyrrolidine dithiocarbamate (PDTC) was reported to relieve cancer cachexia; however, its mechanism remains largely unknown. In our study, we showed that PDTC attenuated cancer cachexia symptom in C26 tumor bearing mice models in vivo without influencing tumor volume. What's more, PDTC inhibited muscle atrophy and lipolysis in cells models in vitro induced by TNFα and C26 tumor medium. PDTC suppressed atrophy of myotubes differentiated from C2C12 by reducing MyoD and upregulating MuRF1, and preserving the expression of perilipin as well as blocking the activation of HSL in 3T3-L1 mature adipocytes. Meaningfully, we observed that PDTC also inhibited p38 MAPK signaling besides the NF-κB signaling in cancer cachexia in vitro models. In addition, PDTC also influenced the protein synthesis of skeletal muscle by activating AKT signaling and regulated fat energy metabolism by inhibiting AMPK signaling. Therefore, PDTC primarily influenced different pathways in different tissues. The study not only established a simple and reliable screening drugs model of cancer cachexia in vitro but also provided new theoretical basis for future treatment of cancer cachexia.

#### Keywords: cancer cachexia, PDTC, C2C12 myotubes, 3T3-L1 adipocytes, muscle atrophy, fat lipolysis

### INTRODUCTION

Cachexia is a severe wasting syndrome accompanied with serious loss of body weight during a lot of chronic diseases such as cancer, AIDS, tuberculosis (Tisdale, 2009). Cancer cachexia affects about 50–80% of cancer patients and is mainly characterized by fatigue, loss of muscle and fat mass, excessive consumption of energy and systemic inflammation (Fearon et al., 2011; von Haehling and Anker, 2014). Cancer cachexia not only influences patients' quality of life, but also weakens the efficacy of chemotherapy and radiotherapy on tumor, therefore decreasing patients' survival time

**Abbreviations:** C26, colon-26; HSL, hormone sensitive lipase; MHC, myosin heavy chain; PDTC, pyrrolidine dithiocarbamate; TG, triglycerides.

seriously (Kumar et al., 2010). It is believed that cancer cachexia is responsible for death of more than 20% of cancer patients directly and indirectly (Fearon et al., 2013; von Haehling and Anker, 2014).

Given the detrimental clinical consequences, it is mandatory to relieve and/or delay the progression of cancer cachexia. At present, there is no approved therapeutic agent for the treatment or prevention of cancer cachexia. A variety of therapeutics including nutritional supplementation, appetite stimulation, and anti-inflammatory strategies has been used to manage cancer cachexia symptoms. Omega-3 Fatty Acids were investigated in clinical Phase I/II to test whether it could help body weight stabilization in cancer cachexia (Harle et al., 2005; Yeh et al., 2013). Anamorelin, a ghrelin receptor agonist, was applied to treat patients with non-small cell lung cancer (NSCLC) and cachexia–anorexia to enhance appetite and anabolic activity in clinical phase III (Garcia et al., 2015; Currow et al., 2017). MT-102 (Espindolol), a novel anabolic/catabolic transforming agent, was used to treat subjects with cachexia related to stage III and IV non-small cell lung cancer and colorectal cancer in clinical phase II. Infliximab, anti-TNFα monoclonal antibody, was applied to treat cancer-related cachexia in subjects with pancreatic cancer in clinical phase II (Wiedenmann et al., 2008; Arruda et al., 2010; Gueta et al., 2010; Miksza et al., 2013). Although these agents have entered into clinical evaluation, it is increasingly evident that a single therapy may not be sufficient to prevent or ameliorate cancer cachexia due to the complexity of this syndrome. Therefore, better understanding the molecular mechanisms of cancer cachexia will allow the identification of potential therapeutic targets and the development of promising drugs.

NFκB signaling plays an important role in skeletal muscle atrophy and fat lipolysis. NF-κB suppressed MyoD mRNA at the post-transcriptional level and upregulated the expression of MuRF1 in muscle decay and cachexia (Li and Reid, 2000; Bodine et al., 2001; Vallabhapurapu and Karin, 2009). And, TNF-α-mediated lipolysis was reduced in the presence of NF-κB inhibitor (Laurencikiene et al., 2007). Therefore, NF-κB inhibitors, such as Compound A, DHMEQ, curcumin, resveratrol, and SN50, were used to keep the mass of skeletal muscle and fat and even inhibit tumor growth. Compound A only partially rescues the phenotype of the cachectic gastrocnemius on the level of metabolism (Der-Torossian et al., 2013). DHMEQ could prevent the development of cachexia in JCA-1 tumor-bearing mice presumably through the inhibition of IL-6 secretion (Kuroda et al., 2005). SN50 inhibited the expression of proteasome induced by PIF (proteolysis-inducing factor) to relieve muscle wasting in cancer cachexia (Wyke et al., 2004). Curcumin completely attenuated total protein degradation in murine myotubes induced by PIF. However, it was ineffective in preventing loss of body weight of MAC16 tumor bearing mice (Wyke et al., 2004). Resveratrol was accompanied by inhibition of tumor growth while attenuating weight loss (Wyke et al., 2004).

Pyrrolidine dithiocarbamate (PDTC, **Figure 1**), a STAT/NFκB inhibitor and an antioxidant, is known to exert antiinflammation, antioxidant, and radical scavenger functions

(Tahata et al., 2014). Recently, the effect of PDTC on attenuating cachexia has attracted much attention. Nai et al. (2007) reported that PDTC could attenuate the development of cancer cachexia in C26 tumor-bearing mice by inhibiting the increase of IL-6 levels in serum and tumor tissue as well as inhibiting NF-κB activation in the tumor sites. In consistence, administration of PDTC also relieved cancer cachexia in Lewis lung carcinoma (LLC) tumor-bearing mice. PDTC reduced muscles STAT3 and p65 phosphorylation, but did not alter LLC-induced muscles AMPK or AKT phosphorylation (Puppa et al., 2014a). Moreover, in the study of Narsale et al., PDTC neither suppressed the cachexia induction of plasma IL-6, nor affected the cachexia-enhanced phosphorylation of NF-κB (S468) in skeletal muscle. The inhibitory effect of PDTC on cancer cachexia was further confirmed in APCMin/<sup>+</sup> mouse, which exhibited an IL-6-dependent cachexia and had long duration of cachexia development. PDTC suppressed the cachexia induction of STAT3 activation and increased mTORC1 signaling in muscle, while attenuated glycogen and lipid content depletion independent to the activation of STAT3 and mTORC1 signaling in liver. (Narsale et al., 2016). Collectively, these studies demonstrated that PDTC exhibited potential activity against cancer cachexia, but its mechanisms could not be simply attributed to the inhibitory effect of PDTC on NFκB signaling. Moreover, signaling responses to PDTC in different tissues might be different, which also deserved further evaluation.

In the present study, we systematically determined effects of PDTC on cancer cachexia in C26 tumor bearing mice in vivo and in vitro, and studied the signaling pathways involved in protein turnover in skeletal muscle atrophy and lipolysis in adipocytes to thoroughly elucidate the mechanisms of PDTC on relieving cancer cachexia.

### MATERIALS AND METHODS

### Reagents

PDTC purchased from Sigma–Aldrich (St. Louis, MO, United States) was reconstituted in sterile saline and stored at −20◦C. RIPA Lysis and Halt Protease and Phosphatase Inhibitor Cocktail (100×) were purchased from Thermo Scientific (Rockford, IL, United States) and stored at 4◦C. BCA protein assay kit used to quantify protein concentration were purchased from Beyotime (Shanghai, China) and stored at RT. DMEM (High Glucose), Penicillin/streptomycin and

Trypsin/EDTA were purchased from Hyclone (Los Angeles, CA, United States). Horse serum was purchased from Gbico (New York, NY, United States). Fetal bovine serum (FBS) was derived from Biological Industries (Kibbutz Beit Haemek, Israel). TNFα was purchased from PeproTech (Rocky Hill, CT, United States).

### Animals

All animal (purchased from Shanghai SLAC Laboratory Animal, Co., Ltd., Shanghai, China) care and experimental protocols for this study complied with the Chinese regulations and the Guidelines for the Care and Use of Laboratory Animals drawn up by the National Institutes of Health (United States) and were approved by the Institutional Animal Care and Use Committee of the East China Normal University. Male BALB/c mice (6–8 weeks old) were purchased from the Shanghai SLAC Laboratory Animal CO. LTD. Mice were maintained on a 12:12 light–dark cycle in a temperature-controlled (21∼23◦C) and specific pathogenfree (SPF) conditional room, and were provided standard rodent chow and water ad libitum. All animals were acclimatized for a week before beginning the study.

### Cancer Cachexia Model in Vivo

Male BALB/c mice with same initial body weight were randomly divided into four groups (12 mice per group): health group (without tumor), Colon-26 (C26, obtained from Shanghai Institute of Materia Medica, Chinese Academy of sciences) tumor-bearing mice group (C26 model group) and C26 tumorbearing mice treated with PDTC (50, 100 mg/kg) group (Nai et al., 2007). On day 0, mice were implanted subcutaneously in the right flank with 100 µl (1.0 × 10<sup>6</sup> ) C26 adenocarcinoma cells. Starting from the next day, C26 model group mice received daily intraperitoneal injections of sterile saline, while PDTC treated mice received daily intraperitoneal injections of PDTC (50, 100 mg/kg). Body weight, tumor volume, and food intake were measured daily from inoculation to completion of the study. On day 6, tumors were first noticed. Record the shortest diameter (x) and longest diameter (y) of tumor using calipers. Tumor volume was calculated following the formula: V = x ∗ x ∗ y ∗ 0.5. When the mice lost 10% of their body weight or when their tumor volumes reached 2,000 mm<sup>3</sup> , tumor, gastrocnemius muscles and eWAT (epididymal white fat) tissue were rapidly dissected, weighed, and frozen in liquid nitrogen, then stored at –80◦C until ready for further analyses, or fixed in 4% paraformaldehyde overnight and embedded in paraffin. All treatment groups were sacrificed by cervical dislocation under ether anaesthesia 6 h after the last treatment.

### Cell Culture

C26 adenocarcinoma cells were maintained in RMPI-1640 medium (Hyclone, Los Angeles, CA, United States) containing 10% FBS at 37◦C with 5% CO2.

C2C12 murine myoblast cell line, obtained from ATCC, were cultured in high-glucose DMEM with 10% FBS at 37◦C with 5% CO2. During differentiation, the medium of cells planted on culture plates coated with 0.1% gelatin was switched into differentiation medium (high-glucose DMEM containing 2% horse serum) when cell confluence reached 70%. After 5 days, multinuclear myotubes were formed.

3T3-L1 pre-adipocytes cells, obtained from Shanghai Institute of Materia Medica, Chinese Academy of Sciences, were cultured in adipocytes medium (AM, high-glucose DMEM with 10% FBS) at 37◦C with 5% CO2. During differentiation, the pre-adipocytes were planted on culture plates coated with 0.1% gelatin, with confluence reached 100% for 48 h in AM. Then they were induced to differentiate by treatment with differentiation media (DM I and DM II) for 48 h, respectively, DM I containing 10 µg/ml insulin (Solarbio, Beijing, China), 1 µM dexamethasone (DEX, Sigma–Aldrich, St. Louis, MO, United States) and 0.5 mM 3-isobutyl-1-methylxanthine (IBMX, Sigma–Aldrich, St. Louis, MO, United States) in AM and DM II (DEX- and IBMX-free DM I). Thereafter, the differentiated cells were maintained in AM changed in every 2 days until used (Chaiittianan et al., 2017).

All cells were negative for mycoplasma contamination before use.

## C26 Tumor Medium Collection

When C26 tumor cells confluence reached 70%, the medium was switched into new high-glucose DMEM medium for 48 h. Thereafter, medium was collected and centrifuged at 5000 g for 10 min at 4◦C. Medium from non-tumor cells (C2C12 cell or 3T3-L1 cell) was used as control medium. The final supernatant was filtered and stored at –20◦C or used immediately at a 1:1 dilution with fresh normal medium.

### Cancer Cachexia Models in Vitro

C2C12 myotubes cells were incubated with TNFα (100 ng/ml) or 50% C26 tumor medium in 2% horse serum in high-glucose DMEM for 48 h in the presence of PDTC or sterile saline as control reagents. Then cells were harvested for Western Blotting or used for morphological analysis.

3T3-L1 mature adipocytes were incubated with TNFα (50 ng/ml) or 50% C26 tumor medium in 10% FBS in highglucose DMEM for 48 h in the presence of PDTC or sterile saline as control reagents. Then cells were harvested for Western Blotting or used for morphological analysis.

### Western Blot

Western blots were performed as described previously (Hetzler et al., 2015). Briefly, C2C12 myotubes and 3T3-L1 mature adipocytes were homogenized in RIPA buffer plus a phosphatase protease inhibitor. The lysates were centrifuged at 13000 rpm for 30 min at 4◦C. The supernatant was quantified for protein concentration using the BCA Protein Assay Kit (Beyotime, Shanghai, China). Equal amounts of protein samples were separated by 10% SDS-PAGE gel electrophoresis and transferred to a polyvinylidene fluoride membrane. The PVDF membranes were blocked in 5% non-fat milk in phosphate buffered saline (PBST, containing 0.1% Tween 20) for 1 h at room temperature and then incubated with primary antibodies diluted in 5% BSA-TPBS at 4◦C overnight. The primary antibodies used were as follows:MuRF-1 (1:1000, Proteintech), P38, MyoD (1:1000, Cell Signaling Technology, Beverly, MA, United States), P65 (1:1000, Cell Signaling Technology), p-P65 (1:1000, Cell Signaling

Technology), AKT (1:500, Santa, Orange, CA, United States), p-AKT (1:2000, Cell Signaling Technology), AMPK (1:1000, Cell Signaling Technology), p-AMPK (1:1000, Cell Signaling Technology), P38 MAPK (1:1000, Cell Signaling Technology), p-P38 MAPK (1:1000, Cell Signaling Technology ), Peri A antibodies (1:1000, Cell Signaling Technology), MHC (1:1000, DSHB, Iowa City, IA, United States) and GAPDH-HRP (1:5000, Santa Cruz Biotechnology, Dallas, TX, United States). Anti-mouse (1:5000, Multi Sciences, Hangzhou, China) and anti-rabbit (1:5000, Multi Sciences, Hangzhou, China) IgG horseradish peroxidase-conjugated secondary antibody was incubated with membranes for 1 h in 5% non-fat milk in TPBS. ECL Chemiluminescent Kit (Thermo Fisher, Waltham, MA, United States) was used to visualize the antibody-antigen interaction and chemical luminescence of membranes was detected by Amersham Imager 600 (GE).

### Hematoxylin-Eosin (HE) Staining

Gastrocnemius muscle samples and epididymal white adipose tissue (eWAT) were freshly isolated and fixed in 4% paraformaldehyde (PFA) for 24 h. Paraffin-embedded tissues were cut in 10 µm sections stained with hematoxylin and eosin (H&E) by standard procedures.

### Immunofluorescent Staining

Differentiated C2C12 myotubes were fixed by 4% PFA for 30 min at room temperature, permeabilized with 0.5% Triton X-100 in PBS for 10 min, and then blocked with 5% bovine serum albumin (BSA) in PBS for 1 h at room temperature. Myotubes were incubated with anti-MHC (MF-20, 1:100, DSHB) diluted in 5% BSA overnight at 4◦C. Myotubes were incubated with secondary antibody Cy3-AffiniPure rabbit anti-mouse IgG (H+L) (1:500, Jackson) at room temperature. Images were captured by fluorescence microscope (Leica) and the diameter of myotubes was measured by Image J.

### Oil Red O Staining

Cells were washed three times with phosphate-buffered saline (PBS), fixed in 4% formalin for 30 min, and then washed three times with cold PBS. Cells were stained in the Oil Red O (Sigma– Aldrich, St. Louis, MO, United States) working solution (3:2, 0.5% Oil Red O dye in isopropanol: water) for 30 min at room temperature (25◦C) and washed three times with water. Staining was visualized by bright-field microscopy.

### Triglyceride Isolation and Determination

Triglycerides (TG) was assessed through commercial enzymatic kits. Differentiated 3T3-L1 adipocytes were harvested in 100 µl distilled water containing 5% Triton-X100; and the TG levels were determined using a commercial kit (Triglyceride Quantification Kit, Applygen, Beijing, China) following the manufacturer instructions. TG of serum was assessed with Automatic biochemical analyzer (HITACHI 7020).

### Lipolysis Assays in Vitro

For lipolysis experiments, glycerol accumulation in the media from 3T3-L1 mature adipocytes and serum of mice was measured using a Lipolysis Assay Kit (Applygen) following the manufacturer instructions. Briefly, 3T3-L1 mature adipocytes were washed three times with PBS and incubated with 100 µl phenol red-free DMEM supplemented with 1% fatty acid-free BSA containing 50 ng/ml TNFα or 50% C26 tumor medium with or without PDTC for 24 h. After incubation, the 100 µl medium was collected and centrifuged at 12000 g for 10 min to remove cell debris. The 50 µl supernatant or serum of mice and glycerol assay reagent (150 µl) were plated in a clean 96-well plate for 10 min at 37◦C and optical density of each well was measured at 550 nm.

### Statistical Analysis

Data are expressed as mean ± SEM. Two-tailed Student's t-test was used for comparisons between two groups. Oneway ANOVA test was performed to compare multiple groups followed by Bonferroni's post hoc test. A p-value of 0.05 or lower was considered significant in all experiments. All analyses were performed using GraphPad Prism 5.0. Values of p less than 0.05 were considered to be statistically significant and were presented as <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 or #p < 0.05, ##p < 0.01, ###p < 0.001.

### RESULTS

### PDTC Attenuates C26 Tumor-Induced Body Weight Loss in Vivo

The effect of PDTC to attenuate cachexia in C26 tumor bearing mice was systematically evaluated in our experiment. In line with previous study, PDTC effectively suppressed C26 tumorinduced body weight loss. Mice in C26 model group and PDTC (50 mg/kg)-treated group started to lose body weight on day 9. In contrast, the weight loss of mice with PDTC (100 mg/kg) treatment was delayed for two days (on day 11). At the end of the treatment (day 13), the body weight of mice treated with PDTC (100 mg/kg) was significantly higher than that of C26 model group. The body weight of mice treated with PDTC (50 mg/kg) was also higher than that of C26 model group, even though there was no statistical significance (**Figure 2A** and **Table 1**). To avoid the influence of tumor weight on body weight, we also analyzed the tumor-free body weight. The overall trend of tumor-free body weight recaptured the effect of PDTC presented by body weight (**Figure 2B** and **Table 1**). The changes of tumor-free body weight increased by 18.52% in healthy mice, decreased by 8.29 and 2.29% in C26 model group and in PDTC treatment group (50 mg/kg), respectively, but increased by 3.73% in PDTC treatment group (100 mg/kg) (**Figure 2C**). What's more, PDTC also increased food intake of mice. Specifically, the food intake in PDTC (100 mg/kg) group was a little higher than that in C26 model group (**Figure 2D**). As a result, the body weight of mice in PDTC (100 mg/kg) group started to increase on day 4 (**Figure 2A**). In addition, PDTC didn't influence C26 tumor growth in mice (**Figures 2E,F**). Together, these results demonstrated that PDTC effectively attenuated C26 tumor-induced body weight loss, and did not affect C26 tumor growth.

TABLE 1 | Effect of pyrrolidine dithiocarbamate (PDTC) treatment on parameters of healthy and C26-tumor bearing mice.


Data presented are the mean ± SE. #Versus health group mice; <sup>∗</sup> versus C26 tumor bearing group mice. ###p < 0.001, ∗∗∗p < 0.001.

### PDTC Reduces Loss of Skeletal Muscle and Adipose Tissue Mass in Vivo

As cancer cachexia-induced weight loss is primarily from loss of skeletal muscle and body fat, we then analyzed the effect of PDTC on C26 tumor-induced loss of skeletal muscle and adipose tissue. As expected, C26 tumor led to a significant decrease of gastrocnemius (GA) mass, which was relieved by the treatment of PDTC (**Figures 3A,B** and **Table 1**). Comparing to C26 model mice, the change of GA mass increased by 4.7% in PDTC treatment group (50 mg/kg) and by 14.0% in PDTC treatment group (100 mg/kg), respectively (**Figure 3B** and **Table 1**). What's more, PDTC also affected the myofibers size distribution. In healthy mice, a bell-like distribution of myofibers area was observed between 200 and1000 µm<sup>2</sup> . In contrast, the myofibers area of C26 model mice showed a smaller size distribution, with 80% cells distributed in less than 400 µm<sup>2</sup> . PDTC (50 and 100 mg/kg) treatment effectively reversed this shift and led the myofibers area redistributed between 200 and1000 µm<sup>2</sup> (**Figure 3C**). Similarly, PDTC also effectively inhibited the loss of body fat. Compared with healthy mice, C26 tumor caused a significant decrease of eWAT, which was relieved by the treatment of PDTC (**Figures 3D,E** and **Table 1**). Of note, the eWAT weights of PDTC treatment mice (100 mg/kg) were about 1.7-fold to that of C26 model mice (**Figure 3E** and **Table 1**). Moreover, PDTC also affected the size of adipocyte cell diameter. A bell-like distribution of adipocyte cell diameter was observed between 20 and 70 µm<sup>2</sup> in healthy mice. However, a left shift was observed in C26 model with more than 80% adipocyte cells distributed in less than 20 µm<sup>2</sup> . PDTC obviously reversed this shift, which was in a dose-dependent manner (**Figure 3F**). In addition, the glycerol and TG content in mice serum further confirmed the protection of PDTC on lipolysis (**Figures 3G,H**). All these results supported that PDTC treatment attenuated the loss of body weight in C26 tumor-bearing cachexia mice by

intraperitoneally daily (n = 12). (A,B) GA weight of each group mice. (C) H&E-stained sections of mice GA and quantify the myofiber area of GA cell. (D,E) eWAT weight of each group mice. (F) H&E-stained sections of mice eWAT and quantify the diameter of adipocyte cell. (G,H) Content of TG and glycerol in serum. Scale bar of C, 50 µm. Scale bar of F, 20 µm. Data presented are the mean ± SE of three independent experiments. GA, gastrocnemius. eWAT, epididymal white fat. #Versus health group mice; <sup>∗</sup> versus C26 tumor bearing group mice. One-way ANOVA test was performed followed by Bonferroni's post hoc test. #p < 0.05, ##p < 0.01, ###p < 0.001, <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

FIGURE 4 | The effect of PDTC on C2C12 myotubes atrophy in vitro. The myotubes atrophy in cancer cachexia model in vitro was induced by C26-tumor medium (1:1 dilution with fresh normal medium) or TNFα (100 ng/ml) for 48 h. (A) The myotubes atrophy in cancer cachexia model induced by C26-tumor medium. (B) Quantified diameter of myotubes. (C) Representative Western blot of MHC, MyoD and MuRF-1 in cachexia model induced with C26-tumor medium. (D) The quantification of (C). (E) The myotubes atrophy in cancer cachexia model induced by TNFα. (F) Quantified diameter of myotubes. (G) Representative Western blot of MHC, MyoD and MuRF-1 in cachexia model induced with TNFα. (H) The quantification of (G). Scale bar of A, 50 µm. Scale bar of E, 100 µm. Data presented are the mean ± SE of three independent experiments. #Versus non-tumor medium (3T3-L1 cell medium) or control group; <sup>∗</sup> versus C26-tumor medium or TNFα single treatment group. <sup>∗</sup>p < 0.05, ∗∗∗p < 0.001, #p < 0.05, ##p < 0.01, ###p < 0.001.

inhibiting GA atrophy and eWAT lipolysis in a dose-dependent manner (**Table 1**).

### PDTC Alleviates Muscle Atrophy in Cancer Cachexia Model in Vitro

Multiple factors, including inflammation cytokines, decreased food intake and neuroendocrine changes (Oliff et al., 1987; Scott et al., 1996; Rivadeneira et al., 1999; Zhou et al., 2010; Sishi and Engelbrecht, 2011; Winbanks et al., 2016), contribute to the occurrence of cancer cachexia in vivo, which makes the mechanisms of cancer cachexia remain largely unknown. To make the question simple, here we used the in vitro system to investigate the mechanisms of PDTC on attenuating cancer cachexia. We first used C26 medium to induce atrophy of C2C12 myotubes and observed the protective effect of PDTC on myotubes atrophy. As shown in **Figure 4A**, C26 medium caused an obvious decrease of C2C12 myotubes diameter, and PDTC at high concentration (25 and 50 µM) effectively inhibited this decrease. In detail, the myotubes diameter decreased from 14.55 ± 0.58 µm in control cells to 9.95 ± 0.41 µm in C26 medium treated cells, but increased to 12.60 ± 0.31 µm and 15.62 ± 0.59 µm in the presence of 25 and 50 µM PDTC, respectively. Myosin Heavy Chain (MHC), a myogenic differentiation marker protein, is a preferred target of multiple pro-cachectic factors inducing muscle atrophy (Acharyya et al., 2004). Here we observed that C26 medium decreased the expression of MHC in C2C12 myotubes, whereas PDTC effectively reversed the downregulation of MHC at concentration of 25 and 50 µM. Moreover, the muscle differentiation factor MyoD and the ubiquitin ligase MuRF1 which affected the transcription and degradation of MHC, respectively, were determined in our experiment. Interestingly, C26 mediuminduced downregulation of MyoD and upregulation of MuRF1 were suppressed by PDTC at concentration of 50 µM. The effect of PDTC on reversing myotubes atrophy was in a concentrationdependent manner (**Figures 4B–D**).

TNFα is one of the important factors involved in the pathogenesis of cancer cachexia; we then used TNFα to induce myotubes atrophy in vitro. Similarly, PDTC also efficiently inhibited TNFα-induced myotubes atrophy, which was in a concentration-dependent manner (**Figure 4E**). The myotubes diameter of cells treated with 50 µM PDTC was about 1.5-fold to that of TNFα treated cells (**Figure 4F**). PDTC also inhibited TNFα-induced downregulation of MHC and MyoD as well as up-regulation of MuRF1 (**Figures 4G,H**). The concentrations of PDTC used in these experiments had no cytotoxicity on viability of C2C12 myotubes (data not shown). These results demonstrated that PDTC protected the myotubes atrophy induced by C26 medium and TNF-α in vitro.

### Effect of PDTC on Inflammatory Signaling and Protein Synthesis of C2C12 Myotubes in Vitro

It is known that NF-κB upregulated the expression of MuRF1 (Li and Reid, 2000; Bodine et al., 2001; Vallabhapurapu and Karin, 2009) and suppressed MyoD mRNA at the posttranscriptional level in muscle decay and cachexia (Guttridge et al., 2000). Moreover, PDTC is an inhibitor of NF-κB and has different effect against the activity of NF-κB in different tissues. Therefore, we determined whether PDTC would inhibit the activation of NF-κB in C2C12 myotubes atrophy. The results showed that PDTC slightly inhibited C26 medium- induced phosphorylation of p65 (**Figures 6A,B**), which suggested that other signaling pathway was involved in the protective effect of PDTC against skeletal atrophy. Interestingly, we further found that PDTC significantly inhibited the enhanced phosphorylation of p38 MAPK in C26 medium-treated C2C12 myotubes. In addition, PDTC also increased the phosphorylation of AKT that was down-regulated after C26 medium treatment (**Figures 6C,D**). These results suggested that PDTC preserved the muscle mass by influencing the pathways of synthesis pathway and protein degradation.

### PDTC Attenuates Lipolysis in Cancer Cachexia Model in Vitro

In order to identify the effect of PDTC on lipolysis in cancer cachexia in vitro, we used C26 medium and TNFα to induce lipolysis of mature 3T3-L1 adipocytes. As shown by Oil Red O staining, the lipid of mature 3T3-L1 adipocytes with C26 medium treatment was much less than that with non-tumor medium (C2C12 cell medium). Meaningfully, C26 mediuminduced decrease of lipid was suppressed in the presence of PDTC (**Figure 5A**). Consistent with this finding, PDTC effectively inhibited the decrease of TG content in 3T3-L1 adipocytes induced by C26 medium. In detail, the adipocyte TG content relative to cell lysis protein decreased from 0.53 ± 0.05 µM/µg protein in cells treated with non-tumor medium (C2C12 cell medium) to 0.34 ± 0.04 µM/µg protein in cells treated with C26 medium, and was reversed back to 0.42 ± 0.05, 0.56 ± 0.03, 0.64 ± 0.03, and 0.62 ± 0.03 µM/µg protein by the treatment of PDTC (1, 10, 30, and 100 µM, respectively). The effect of PDTC to inhibit C26 medium-induced decrease of TG content was in a dose-dependent manner (**Figure 5B**). We then determined the expression of perilipin that is a critical regulator of lipid stores in adipocytes (McDonough et al., 2013). In line with above findings, the expression of perilipin was remarkably down-regulated in mature 3T3-L1 adipocytes with C26 medium treatment, and this down-regulation was effectively suppressed by PDTC. Hormone sensitive lipase (HSL) is a rate-limiting enzyme that regulates adipocytes lipolysis, and the phosphorylation of HSL on Ser559/660 is crucial for its activation (Anthonsen et al., 1998). Here we observed that the phosphorylation of HSL was enhanced by C26 medium, which was successfully inhibited in the presence of PDTC (**Figures 5C,D**).

Likewise, the protective effect of PDTC was observed in TNFα-induced lipolysis of mature 3T3-L1 adipocytes. In detail, PDTC inhibited TNFα-induced decrease of lipid in mature 3T3-L1 adipocytes and increase of glycerol release in culture medium of mature 3T3-L1 adipocytes (**Figures 5E,F**). Compared to control group, the glycerol content increased to 170% in culture medium of mature 3T3-L1 adipocytes treated with TNFα,

C26-tumor medium (1:1 dilution with fresh normal medium) or TNFα (50 ng/ml) for 48 h. (A) The lipid of 3T3-L1 mature adipocytes was detected by oil red O staining in cancer cachexia model induced by C26-tumor medium. (B) Quantified triglyceride (TG) with TG commercial kits. (C) Representative Western blot of phosphorylated HSL, total HSL and perilipin in cachexia model induced with C26-tumor medium. (D) The quantification of C. (E) The lipid of 3T3-L1 mature adipocytes was detected by oil red O staining in cancer cachexia model induced by TNFα. (F) Quantified glycerol release of medium with glycerol commercial kits. (G) Representative Western blot of phosphorylated HSL, total HSL and perilipin in cachexia model induced with TNFα. (H) The quantification of G. Scale bar, 20 µm. Data presented are the mean ± SE of three independent experiments. #Versus non-tumor medium (C2C12 cell medium) or control group; <sup>∗</sup> versus C26-tumor medium or TNFα treatment group. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, #p < 0.05, ##p < 0.01, ###p < 0.001.

while decreased to 150, 137, and 120% in the presence of PDTC (10, 30, and 100 µM, respectively), which was in a concentration dependent manner. Here we observed that the phosphorylation of HSL was enhanced by TNFα, which was then successfully inhibited in the presence of PDTC (**Figures 5G,H**). Furthermore, the concentration of PDTC and TNFα used in these experiments had no effect on 3T3-L1 mature adipocytes viability (data not shown). These results demonstrated that PDTC inhibited the lipolysis process in 3T3-L1 mature adipocytes in cancer cachexia condition.

### Effect of PDTC on Inflammatory Signaling and Energy Metabolism of 3T3-L1 Mature Adipocyte in Vitro

Previous studies demonstrated that NF-κB influenced human fat cell lipolysis and the expression pro-inflammatory adipokines (Laurencikiene et al., 2007; Hatano et al., 2014), so we wondered whether PDTC would affect the NF-κB signaling in lipolysis of mature 3T3-L1 adipocytes in vitro. Western Blot results showed that PDTC was able to inhibit the enhanced phosphorylation of NF-κB induced by C26 medium slightly (**Figures 6E,F**). Interestingly, we found that PDTC significantly reduced the phosphorylation of p38 and AMPK which were enhanced by C26 medium (**Figures 6G,H**). Overall, these results demonstrated that PDTC was able to inhibit lipolysis by suppressing p38 MAPK signaling and AMPK signaling.

### DISCUSSION

Cancer cachexia, characterized by severe wasting of muscle and fat, systemic inflammation, and energy metabolism hyperthyroidism, contributes to high mortality rate of cancer patients, especially for advanced solid tumor. There are a few clinical treatments to rescue cancer cachexia symptom, such as nutritional supplemental, which had been proven to be non-effective for cancer cachexia patients (Lainscak et al., 2008). Therefore, the discovery of effective anti-cancer cachexia drugs is very urgent and important. PDTC was reported to attenuate the development of cancer cachexia in mice bearing C26 and LLC tumor and in APCMin/<sup>+</sup> mouse (Nai et al., 2007; Puppa et al., 2014b; Narsale et al., 2016). However, the mechanism of PDTC on relieving cancer cachexia is largely unclear. Better understanding of the signaling pathways PDTC participating will allow the identification of potential therapeutic targets and is beneficial for the therapeutic of cancer cachexia.

Here we found that PDTC attenuated cancer cachexia symptom in C26 tumor bearing mice in vivo in our laboratory system, which was in consistence with previous studies (Nai et al., 2007; Puppa et al., 2014b; Narsale et al., 2016). PDTC significantly reduced body weight loss without influencing the tumor growth (**Figure 2**). Importantly, PDTC significantly attenuated the wasting of skeletal muscle and adipose tissue of the tumor-bearing mice as evidenced by the increased GA mass and myofiber area as well as the increased eWAT weight and diameter of adipocyte cells after PDTC treatment (**Figure 3**). Interestingly, these effects were further recaptured in in vitro system. PDTC blockaded C2C12 myotubes atrophy and 3T3-L1 mature adipose lipolysis induced by C26 tumor media or TNFα, suggesting that PDTC have direct effect on signaling pathways that mediated the wasting of skeletal muscle and adipose tissue (**Figures 4**, **5**).

Skeletal muscle is the most obvious tissue affected by cancer cachexia and MHC is a preferred target of multiple pro-cachectic factors inducing muscle atrophy (Cosper and Leinwand, 2012; Umeki et al., 2015). In our study, we found that C26 medium or TNFα caused the decrease of MHC in mature C2C12 myotubes, and PDTC obviously reversed the downregulation of MHC. MyoD has been shown to drive the transcription of MHC (Meissner et al., 2007; Daou et al., 2013). Here we observed that PDTC treatment inhibited C26 mediuminduced downregulation of MyoD, which might contribute to reverse the downregulation of MHC. In addition, the ubiquitin-dependent proteasome pathway has been reported to play important roles in muscle wasting process. The E3 ubiquitin ligase MuRF1 was involved in the degradation of MHC (Krawiec et al., 2005; White et al., 2011; Rom et al., 2015). Here we found that PDTC effectively reduced the expression of MuRF1, suggesting that PDTC also had effect on blocking the ubiquitin-dependent proteasome pathway. In our study, we observed that PDTC slightly inhibited the phosphorylation of p65 enhanced by C26 medium, suggesting other signaling pathways were employed by PDTC to relieve cancer cachexia. It is reported that the phosphorylation of AKT was inhibited in muscle atrophy (Quan-Jun et al., 2017), and the reduction of AKT phosphorylation led to increased MuRF1 transcription (Wadosky et al., 2014), so we wondered whether PDTC would affect the activation of AKT. Meaningfully, PDTC effectively increased the phosphorylation of AKT in C26 medium-treated C2C12 myotubes. Moreover, AKT signaling pathway also contributed to protein synthesis of skeletal muscle. Therefore, PDTC preserved the muscle mass by influencing the pathways of synthesis pathway and protein degradation. What's more, the p38 MAPK signaling has been demonstrated to play important roles in skeletal muscle atrophy. Endotoxininduced skeletal muscle wasting was reported to be through a p38 MAPK-dependent mechanism (Morales et al., 2015). Excessive fatty acid oxidation induces muscle atrophy in cancer cachexia by activating p38 MAPK pathway (Fukawa et al., 2016). Myostatin increased protein degradation and decreases protein synthesis of skeletal muscle by activation of the SMAD complex and by MAPKs and through PI3K/Akt pathway (Argiles et al., 2012). It has also been reported that p38 inhibitor could attenuate loss of skeletal muscle. P38 inhibitors, SB203580, blunted the expression of Atrogin1/MAFbx, and E3 ligases induced by TNF-α, and attenuated the protein degradation in C2C12 myotubes (Li et al., 2005). In addition, SB203580 also attenuated total protein degradation induced by TNF-α/IFN-γ and ANG II in murine myotubes (Eley et al., 2008). SB202190 (p38 inhibitors) administration blocks atrogin1/MAFbx upregulation and muscle protein loss in the muscle of LLC tumor-bearing mice (Zhang et al., 2011). In our study, we also found that C26 medium significantly enhanced the phosphorylation of p38 MAPK in C2C12 myotubes, and

induced with C26-tumor medium. (D) The quantification of C. (E) Representative Western blot of phosphorylated p65 and total p65 in 3T3-L1 adipocyte cachexia model induced with C26-tumor medium. (F) The quantification of E. (G) Representative Western blot of phosphorylated p38 MAPK, total p38 MAPK, phosphorylated AMPK, and total AMPK in 3T3-L1 adipocyte cachexia model induced with C26-tumor medium. (H) The quantification of I. Data presented are the mean ± SE of three independent experiments. #Versus non-tumor medium or control group; <sup>∗</sup> versus C26-tumor medium or TNFα treatment group. #p < 0.05, ##p < 0.01, ###p < 0.001, <sup>∗</sup>p < 0.05, ∗∗p < 0.01.

PDTC effectively inhibited this activation, indicating the effect of PDTC on relieving cancer cachexia was also through p38 MAPK signaling.

Previous researches on cancer cachexia mainly focus on muscle atrophy. But actually, fat is lost rapidly than skeletal muscle in cancer cachexia (Fouladiun et al., 2005; Das et al., 2011). In this study, we showed PDTC ameliorated cancer cachexia based on relieving not only muscle atrophy but also fat loss. The phosphorylation of HSL was upregulated by C26 medium to trigger adipocytes lipolysis, but PDTC effectively suppressed the increased phosphorylation of HSL, therefore inhibiting HSL-regulated adipocytes lipolysis. It is reported that HSL is a substrate for AMPK and activation of AMPK increases HSL phosphorylation (Garton and Yeaman, 1990; Holm et al., 2000; Holm, 2003; Carmen and Victor, 2006). It was also reported that the inhibitors of AMPK could alleviate lipolysis of fat. Ginsenoside Rh2 significantly activated AMPK and induced lipolysis in 3T3-L1 adipocytes, which was abolished by AMPK inhibitor treatment (Hwang et al., 2007). Compound C, an AMPK inhibitor, partially abrogated lipolysis of 3T3- L1 adipocytes by activating AMPK induced by thiacremonone (Kim et al., 2012). In addition, Compound C also attenuated lipolysis in isolated adipocytes induced by adrenaline (Koh et al., 2007). Here we found that PDTC significantly suppressed the phosphorylation of AMPK that was enhanced by C26 medium, suggesting the effect of PDTC on HSL might through AMPK. What's more, the p38 MAPK signaling was also reported to phosphorylate HSL in pancreatic cancer exosome-induced adipose tissue lipolysis (Sagar et al., 2016). The inhibitors of p38 MAPK attributed to attenuate lipolysis of fat. Lipolysis induced by AM (adrenomedullin) and PC-exosomes could be attenuated in the presence of p38 MAPK inhibitor (SB203580) in 3T3-L1 and human adipocytes (Sagar et al., 2016). Interestingly, we also found that the phosphorylation of p38 MAPK was enhanced by C26 medium and PDTC effectively decreased this activation, suggesting p38 MAPK might be an important target for PDTC to exert its effect against cancer cachexia. In addition, the increased phosphorylation of p65 was modestly suppressed by PDTC in 3T3-L1 mature adipocytes lipolysis which was in consistence with the observation in muscle atrophy. Together, our study suggested that the protective effect of PDTC against C26 medium induced adipocytes lipolysis was not only targeting the NF-κB pathway, but also affecting the AMPK and p38 MAPK signaling pathways.

Up to now, the effects of reagents on attenuating cancer cachexia were usually evaluated by using in vivo animal models; however, these models are expensive and time consuming which definitely delay the development of drugs against cancer cachexia. Here we used C26 tumor medium or TNFα to induce atrophy of mature C2C12 myotubes and lipolysis of mature 3T3-L1 adipose cells in vitro, which mimics the wasting of skeletal muscle and adipose tissue in vivo. By using these models in vitro, we recapitulated the protective effect of PDTC on muscle atrophy and adipose lipolysis, suggesting that these two in vitro models could be used as a simple and reliable platform for the screening of anti-cancer cachexia drugs.

In summary, our study showed that PDTC was sufficient to attenuate cancer cachexia-reduced loss of muscle and fat in vitro and in vivo. We further found that PDTC primarily influenced different pathways in different tissues. Specifically, PDTC regulated p38 MAPK signaling and AKT signaling to keep the mass of skeletal muscle, and regulated p38 MAPK signaling and AMPK signaling to reduce the loss of fat (**Figure 7**). Moreover, our study also established a simple and reliable in vitro cancer cachexia model for drug screening, which is definitely beneficial for identification of novel targets and development of new strategies for treatment of cancer cachexia.

### AUTHOR CONTRIBUTIONS

Designed the experiments: XZ, XL, YF, CM, and YL. Performed the experiments: CM, YL, WZ, XC, and LF. Analysis and interpretation of data: CM and YL. Drafting the manuscript: XZ, XL, YF, and CM.

### ACKNOWLEDGMENTS

fphar-08-00915 December 10, 2017 Time: 16:6 # 13

This work was supported in part by supports received by XL from the Shanghai Science and Technology Innovation Action Program (No. 15140904800), by XZ from the Science and Technology Commission of Shanghai

### REFERENCES


Municipality (Nos. 14431902700 and 16DZ2280100) and Open Funds of State Key Laboratory of Oncology in South China (No. HN2016-03), by YF from the National Natural Science Foundation of China (No. 81402953) and China Postdoctoral Science Foundation (No. 2015T80416).

palliative care–correlations with food intake, metabolism, exercise capacity, and hormones. Cancer 103, 2189–2198. doi: 10.1002/cncr.21013



**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 Miao, Lv, Zhang, Chai, Feng, Fang, Liu and Zhang. 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.

# Identification of Basic Fibroblast Growth Factor as the Dominant Protector of Laminar Shear Medium from the Modified Shear Device in Tumor Necrosis Factor-α Induced Endothelial Dysfunction

Huang-Joe Wang1, 2 and Wan-Yu Lo3, 4 \*

*<sup>1</sup> Department of Internal Medicine, School of Medicine, China Medical University, Taichung, Taiwan, <sup>2</sup> Cardiovascular Research Laboratory, Division of Cardiovascular Medicine, Department of Internal Medicine, China Medical University and Hospital, Taichung, Taiwan, <sup>3</sup> Cardiovascular and Translational Medicine Laboratory, Department of Biotechnology, Hungkuang University, Taichung, Taiwan, <sup>4</sup> Bachelor Degree Program in Animal Healthcare, Hungkuang University, Taichung, Taiwan*

#### Edited by:

*Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico*

#### Reviewed by:

*Jingyan Han, Boston University, United States Pascal Bernatchez, University of British Columbia, Canada*

> \*Correspondence: *Wan-Yu Lo drwanyu@gmail.com*

#### Specialty section:

*This article was submitted to Vascular Physiology, a section of the journal Frontiers in Physiology*

Received: *07 October 2017* Accepted: *13 December 2017* Published: *05 January 2018*

#### Citation:

*Wang H-J and Lo W-Y (2018) Identification of Basic Fibroblast Growth Factor as the Dominant Protector of Laminar Shear Medium from the Modified Shear Device in Tumor Necrosis Factor-*α *Induced Endothelial Dysfunction. Front. Physiol. 8:1095. doi: 10.3389/fphys.2017.01095* Background and Aims: Endothelial dysfunction is a hallmark of cardiovascular diseases. The straight region of an artery is protected from atherosclerosis via its laminar blood flow and high shear stress. This study investigated the cytoprotective effects of a new laminar shear medium (LSM) derived from a modified cone-and-plate shear device and identified basic fibroblast growth factor (bFGF) secreted by human aortic endothelial cells (HAECs) as the dominant protective factor in the LSM.

Methods: Based on a modified cone-and-plate shear device system, HAECs were exposed to laminar shear (15 dynes/cm<sup>2</sup> ) and static control for 24 h to produce a new supernatant LSM and static medium (SM). Evaluation of the protective effects of LSM and SM on endothelial dysfunction induced by tumor necrosis factor (TNF)-α (10 ng/mL), which leads to production of reactive oxygen species (ROS), inflammatory monocyte adhesion, and tissue factor activity. ROS induction-, inflammation-, and thrombosis-related genes and protein expression were evaluated by quantitative-PCR and western blotting. To identify the cytokines that played a key role in the cytoprotective action of the LSM, we used cytokine antibody arrays, selected an abundant marker cytokine, bFGF, and validated the different cytoprotective effects of recombinant bFGF (rbFGF) and neutralization by monoclonal antibody (rbFGF+Ab) co-treatment. Aortic and lung tissues from different groups of C57BL/6J mice were examined by immunohistochemistry. SB203580 (specific inhibitor of p38) and BIX02189 (specific inhibitor of MEK5) were used to identify bFGF as the main cytoprotective factor acting via p38/MAPK and MEK5-KLF2 pathways.

Results: Compared with traditional LSM, the new LSM not only significantly decreased TNF-α-induced intracellular adhesion molecule 1 and plasminogen activator inhibitor type 1 gene expression, but also significantly increased heme oxygenase 1 gene expression. The new LSM and bFGF attenuated TNF-α-induced ROS induction, inflammation, and tissue factor activity and inhibited the inflammatory- and thrombosis-related gene/protein overexpression both *in vitro* and *in vivo*. Mechanistically, the cytoprotective action of bFGF was mediated via the p38/MAPK and MEK5-KLF2 pathways.

Conclusion: bFGF was identified as the critical factor mediating the cytoprotective effects of LSM derived from the modified laminar shear system.

Keywords: laminar shear stress, basic fibroblast growth factor, endothelial dysfunction, tumor necrosis factor-α, cytokine antibody array

### INTRODUCTION

Endothelial cells are constantly exposed to blood flow; shear stress for blood flow within a vessel is defined as τ = 32•µ•Q/π•d 3 , where Q is the mean volumetric flow rate, µ is the mean velocity, and d is the vessel diameter (Papaioannou and Stefanadis, 2005). Previous studies revealed increased oxidative stress, impaired vasodilatation, proinflammatory, and prothrombotic effects on endothelial cells under disturbed low flow stress stresses (<4 dynes/cm<sup>2</sup> ). In contrast, laminar flow with high shear stresses (10–70 dynes/cm<sup>2</sup> ) are atheroprotective, and they are associated with the following phenotypes: low expression of adhesion molecules/inflammatory genes/chemokine genes, high expression of antioxidant genes, and inhibition of leukocyte adhesion/platelet aggregation/thrombosis (Chiu and Chien, 2011). Most in vitro and in vivo studies have focused on the effects of different shear stresses (static, laminar flow, and disturbed flow) on underlying endothelial cells (Chien, 2007; Chiu and Chien, 2011; Tarbell et al., 2014).

In in vitro shear stress systems, parallel-plate flow chambers and cone-and-plate shear devices are commonly used shear device systems (Rezvan et al., 2011). The original cone-and-plate system has been modified by numerous groups, including Dr. Hanjoong Jo, whose modification of the design included a 10-cm tissue culture disc on which human umbilical vein endothelial cells (HUVECs) were exposed to laminar or oscillating shear stress from a rotating cone, and the angular separation between the cone surface and culture disc was 0.5◦ . Using this traditional shear device (15 dynes/cm<sup>2</sup> laminar shear stress, LSS), we observed that human aortic endothelial cells (HAECs) in the center of the culture disc, compared with those at the periphery, were easily detached on the culture disc and did not show the typical high shear stress-induced alignment of endothelial cell shape, likely because of the non-uniform shear stress levels of this device (Rezvan et al., 2011). Therefore, we developed a modified cone-and-plate system to exclude centrally growing HAECs, in which the seeded endothelial cells did not fully cover the 10-cm culture dish, and retained a central circle devoid of endothelial cell growth to avoid interference from dead cells and metabolites contaminating the laminar shear medium (LSM).

For the past several decades many research groups have used various screening approaches to obtain the global genomic profile of the underlying endothelial cells under shear stress using microarray (Chen et al., 2001; McCormick et al., 2001; Dekker et al., 2002). These experiments have been invaluable for identifying novel target genes in cardiovascular studies. However, such approaches failed to reveal important post-transcriptional protein control mechanisms in the endothelium. In the past decade, some studies have conducted proteome analysis of cultured vascular endothelial cells from bovine, rat, mice, and humans under different shear stresses (Wang et al., 2007; Freed and Greene, 2010; Firasat et al., 2014). However, the biological role of the secretome in LSM from underlying endothelial cells exposed to the LSS remains unclear.

Cytokines have been widely studied in the field of biomedicine. In addition to playing critical roles in many normal cellular events, cytokines are involved in the initiation and development of nearly every major life-threatening disease. Cytokines have been explored as potential disease and physiological biomarkers, and cytokine antibody arrays are effective tools for biomarker discovery with high-throughput detection of many proteins simultaneously (Huang, 2007; Wilson, 2015).

Many previous studies reported that tumor necrosis factorα (TNF-α)-induction as the major indicator of atherogenesis and inflammation in endothelial cells (Blake and Ridker, 2002; Branen et al., 2004). In this study, we demonstrated that LSM derived from a new LSS system had unexpected protective effects in TNF-α-induced endothelial dysfunction [reactive oxygen species (ROS) induction, inflammation, and thrombosis]. Thus, we hypothesized that underlying HAECs secreting cytokines were dominant protectors of LSM, and they possess significant cytoprotective effects. The high abundance HAEC-secreted marker cytokines were selected using Human Cytokine Antibody Array, and autocrine basic fibroblast growth factor (bFGF) was identified as a critical protector in the LSM.

### MATERIALS AND METHODS

### Cell Culture, Shear Apparatus, and LSM/SM Collection

HAECs were purchased from Cell Applications, Inc. (San Diego, CA, USA) and cultured in endothelial cell growth medium (Cell Applications, Inc.) according to the manufacturer's recommendations. The human monocytic cell line THP-1 was obtained from the American Type Culture Collection (Manassas, VA, USA) and maintained as previously described (Wang et al., 2014). Using a cone-and-plate shear device, an LSM was obtained when LSS (15 dynes/cm<sup>2</sup> ) was applied to HAECs (between passages 2 and 5) for 24 h. A HAEC monolayer fully covering a 10-cm tissue culture dish represents the traditional shear system, whereas a monolayer grown in a 10-cm tissue culture dish

excluding a central circle (diameter of 5.41 cm) represents our new shear system. In this study, both systems were exposed to an arterial level of unidirectional laminar shear for 24 h by rotating a Teflon cone using the magnetic stirrer (Supplementary Data 1: video of laminar shear, 15 dynes/cm<sup>2</sup> ) and then traditional and new LSM were collected. The static medium (SM) was collected from the same cells exposed to static conditions for 24 h. The collected new LSM and SM were filtered through 0.45-µm filters and stored at −80◦C until analysis.

### MTS Cell Viability Assay

Cell viability was measured using the 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfo-phenyl)-2H-

tetrazolium (MTS) assay, CellTiter 96 <sup>R</sup> AQueous One Solution reagent (Promega, Madison, WI, USA), and absorbance measurement at 490 nm. The viabilities of the control cell groups incubated with fresh medium for 24 h were set to 100%. Cell viabilities of different percentages of LSM-treated cells were compared with those of the individual control group.

### In Vitro Experiments

For in vitro studies, co-treatment of the cells with recombinant human TNF-α (TNF-α) (10 ng/mL, Sigma–Aldrich, St. Louis, MO, USA) and 20% LSM (TNF-α+LSM group), TNF-α and 20% SM (TNF-α+SM group), and TNF-α and recombinant human bFGF (10 ng/mL, Thermo Fisher Scientific, Waltham, MA, USA) (TNF-α+rbFGF group) were performed. Additionally, for the TNF-α+LSM+Ab group, the LSM was pre-neutralized with antihuman bFGF antibody (10µg/mL, Santa Cruz Biotechnology, Dallas, TX, USA) at 37◦C for 1 h before co-treatment with TNF-α. LSM pre-neutralized with anti-IgG secondary antibody (10µg/mL, Jackson ImmunoResearch, West Grove, PA, USA) at 37◦C for 1 h before co-treatment with TNF-α, and this group was named as the TNF-α+LSM+IgG group. After 6 h incubation, genetic analysis, protein expression level measurement, and multiple biochemical experiments were carried out to explore the cytoprotective efficacies of the new LSM.

### Real-Time Quantitative Polymerase Chain Reaction (qPCR)

All mRNA transcript levels of HAECs were analyzed by qPCR. The individual Universal ProbeLibrary probe and primer sequences for glyceraldehyde-3-phosphate dehydrogenase (GAPDH), Kruppel-like factors (KLF2), NAD(P)H quinone dehydrogenase 1 (NQO-1), heme oxygenase-1 (HO-1), thrombomodulin (TM), Tissue factor (TF), plasminogen activator inhibitor type 1(PAI-1), Kelch-like ECH-associated protein 1 (Keap-1), vascular cell adhesion protein 1 (VCAM-1), intercellular adhesion molecule 1 (ICAM-1), monocyte chemo-attractant protein-1 (MCP-1), hepatocyte growth factor (HGF), granulocyte-colony stimulating factor (G-CSF), IL-17, granulocyte-macrophage colony-stimulating factor (GM-CSF), epidermal growth factor receptor (EGFR), monokine induced by gamma interferon (MIG), and basic fibroblast growth factor (bFGF) genes are provided in Supplementary Data 2. All PCRs were performed using the StepOnePlus Real-Time PCR instrument (Applied Biosystems, Foster City, CA, USA). Real time qPCR conditions were defined according to the manufacturer's recommendations. Gene expression levels were analyzed using StepOne software v2.2.

### Western Blot Analysis

Protein expression levels in the HAECs were analyzed by western blotting as previously described (Wang et al., 2010). Primary antibodies against HO-1 (1:1,000) (Abcam, Cambridge, UK), ICAM-1 (1:500) (Cell Signaling Technology, Danvers, MA, USA), PAI-1 (1:1,000) (Santa Cruz Biotechnology), bFGF (1:500) (Millipore, Billerica, MA, USA), and GAPDH (1:5,000) (Santa Cruz Biotechnology) were used. Immunostaining was visualized using SuperSignal West Pico Chemiluminescent Substrate for HO-1 and PAI-1 and SuperSignal West Femto Maximum Sensitivity Substrate for ICAM-1 and bFGF (Thermo Scientific).

### Measurement of ROS Induction

ROS accumulation was detected using 2′ ,7 ′ -dichlorofluorescin diacetate (H2DCFDA, Sigma). After different treatments, all groups were incubated with H2DCFDA (10µM) at 37◦C for 30 min in the dark. Unbound H2DCFDA was removed by washing with 1× phosphate-buffered saline, and H2DCFDA fluorescence was imaged using a fluorescent microscope equipped with a digital camera (Olympus DP72, Tokyo, Japan). The fluorescence intensity (ROS activity) was measured using ImageJ software (ImageJ, National Institute of Health, Bethesda, MD, USA) and expressed as fold-changes of the corresponding control.

### TF Activity Assay

Cellular TF-mediated procoagulant activity was measured using commercial Tissue Factor Human Chromogenic Activity Assay Kit (Cot. No. ab108906, Abcam) according to the manufacturer's instructions. Serum-starved HAECs (1 × 10<sup>5</sup> cells) were grown in six-well plates. After different treatments, cells were washed twice with 1× phosphate-buffered saline, followed by incubation with human factor VIIa (FVIIa) and factor X (FX) at 37◦C, which allowed the formation of a TF/FVIIa complex at the cell surface. The TF/FVIIa complex converted human FX to factor Xa, which was measured by its ability to metabolize a chromogenic substrate. A standard curve with lapidated human TF was obtained to ensure that measurements were acquired in the linear range of detection.

### Monocyte Adhesion Assay

In adhesion experiments, THP-1 cells were labeled with calcein acetoxymethyl ester (Calcein-AM; Molecular Probes, Eugene, OR, USA) as previously described (Lo et al., 2017).

### Cytokine Antibody Array

Semi-quantitative detection of 120 human cytokine and chemokine levels in the SM and LSM was performed using RayBio C-series Human Cytokine Antibody Array C1000 (AAH-CYT-1000-4, RayBiotech, Norcross, GA, USA) according to the manufacturer's instructions as previously described (Zhou et al., 2005). Detection of all spots was performed with the ChemiDoc MP Imaging Systems (Bio-Rad, Hercules, CA, USA), and the intensity of dots was quantified by densitometric analysis and ImageJ software. For each spot, the raw numerical densitometry data were extracted and subjected to background subtraction before normalizing the signal for each cytokine to the positive control spots.

### Animal Experimental Protocols

Male C57BL/6J mice (8 weeks of age) were used. All animal protocols were approved by the Institutional Animal Care and Use Committee of Hungkuang University (HK-105-29). Mice were randomized into four groups: TNF-α, TNF-α+LSM, TNF-α+LSM+Ab, and TNF-α+rbFGF groups. Intraperitoneal injection of recombinant mouse TNF-α (2 µg/100 µL saline/mouse) and 100 µL of LSM, LSM+Ab, or rbFGF (3 µg/mouse) were administered. After 24 h, mice were sacrificed by CO<sup>2</sup> narcosis, and the aortic and lung tissues were removed. Paraffin sections (5µm thickness) were prepared for immunohistochemistry staining.

### Immunohistochemical Staining

All aortic tissue cross-sections from the above experiments were prepared with a Bond-Max autostainer (Leica Microsystems, Wetzlar, Germany). Slides were stained with primary antibody on a fully automated Bond-Max system and VBS Refine polymer detection system (Leica Microsystems) as previously described (Lo et al., 2017). Negative controls did not contain primary antibody. Positive immunoreactivity signals in the endothelial layers were measured by ImageJ software. Immunoreactivity signals in endothelial layers from TNF-α and the TNF-α+LSM, TNF-α+LSM+Ab, or TNF-α+rbFGF groups were quantified as previously described (Federici et al., 2002).

### Identification of Protective Pathways

The cells were pretreated with 10µM SB203580 (specific inhibitor of p38) and 10µM BIX02189 (specific inhibitor of MEK5) (Cayman Chemical, Ann Arbor, MI, USA) for 1 h, followed by addition of TNF-α (10 ng/mL) and rbFGF (10 ng/mL) for 6 h. Gene expression was determined by q-PCR as described above.

### Statistical Analyses

Independent experiments were conducted to assess the significance of differences between the control group and other groups or TNF-α group and other groups. Significant differences were determined using Student's t-test and defined as p < 0.05.

### RESULTS

### Development of New LSS System and Evaluation of Protective Effects of LSM

We developed a new LSS system. HAECs (1 × 10<sup>6</sup> cells/dish) were seeded overnight into a 10-cm dish keeping a central empty area (diameter of 51.4 mm) using a 6-cm dish and 2% agarose (**Figure 1A**). After applying shear stress (15 dynes/cm<sup>2</sup> ) for 24 h, the media showed different phenotypes: LSM of traditional LSS was less clear (right), while the LSM of the new LSS system was relatively clear (left) (**Figure 1B**). HAECs exposed to LSS or static conditions for 24 h showed alignment of the cell shape in the direction of the laminar flow (right), while static cells showed the typical polygonal "cobblestone shape" (left) (**Figure 1C**). Previous studies suggested that laminar shear increases the expression of mechanosensitive genes to maintain endothelial hemostasis (Chiu and Chien, 2011). In our new LSS system, after HAECs were seeded and exposed to LSS for 24 h, the expression levels of mechanosensitive genes, TM, HO-1, NQO-1, and KLF-2, in underlying HAECS were upregulated significantly (**Figure 1D**). Other HAECs were also incubated in fresh media mixed with 2, 20, and 60% LSM for 24 h, but there was no significant induction of cell death (**Figure 1E**).

The cytokine TNF-α is an important mediator of acute inflammatory processes that occur during the progression of atherosclerosis. Examples include transcriptional regulation of various inflammation- and thrombosis-related genes (Matsumoto et al., 1998; Chiu et al., 2004). Numerous antioxidant pathways are involved in cellular redox homeostasis, among which the nuclear factor-E2-related factor 2 (Nrf2)/Kelchlike ECH-associated protein 1/antioxidant response element signaling pathway is perhaps the most prominent. Oxidative stress causes Nrf2 to dissociate from Kelch-like ECH-associated protein 1, and it translocates into the nucleus to bind to the antioxidant response element and regulate the transcription of downstream target antioxidant genes, such as HO-1 and NQO-1 (Chen et al., 2015). Thus, we treated the other strain of HAECs with TNF-α (10 ng/mL) for 6 h as a positive control platform of endothelial dysfunction. The new LSM not only decreased TNF-α-induced inflammation and thrombosis-related ICAM-1 and PAI-1 gene expression, but also significantly increased antioxidant HO-1 gene expression compared to the traditional LSM (**Figure 1F**). Thus, the following experiments used the new LSM collected from the modified cone-and-plate shear device.

### Evaluation of Protective Efficacy of LSM and SM on TNF-α-Induced Endothelial Dysfunction

A previous study showed that TNF-α induces a pro-oxidant environment in a cell that can be measured by H2DCFDA detection in the intracellular oxidative milieu (Shanmugam et al., 2016). Thus, we evaluated the protective efficacies of LSM and SM using the H2DCFDA assay. As shown in **Figure 2A**, untreated control cells displayed very low levels of H2DCFDA-dependent fluorescence. However, the TNF-α group showed a significantly higher average fluorescence intensity (1.95-fold) compared to the control group. The TNF-α+LSM group showed significant attenuation with an average intensity of 1.59-fold relative to the control group. The TNF-α+SM group showed an increased oxidative intracellular milieu, similar to that observed in the TNF-α group.

The adhesion of circulating monocytes to endothelial cells is an important event causing vascular inflammation (Hopkins, 2013). As shown in **Figure 2B**, TNF-α treatment for 6 h caused a 3.04-fold increase in THP-1/HAECs adhesion compared to that in the control group. The TNF-α+LSM group showed significant attenuation of THP-1/HAECs adhesion (1.44-fold) compared to the control group. However,

FIGURE 1 | HAECs exposed to the new LSS or static conditions for 24 h induced alignment of the cell shape in the direction of the laminar flow (right), while static cultured cells showed a typical polygonal "cobblestone shape" (left). (D) HAECs were exposed to the new LSS or static control for 24 h, and the expression levels of *TM*, *HO-1*, *NQO-1*, and *KLF-2* in the new LSS-induced cells were increased compared to those in the static control group. *n* = 5. \**p* < 0.05. (E) New LSM, 2, 20, or 60%, was mixed with fresh HAEC medium to test its cytotoxicity. An MTS assay was performed after 24 h incubation. *n* = 3. NS, not significant. (F) The new LSM not only significantly inhibited TNF-α (10 ng/mL, 6 h)-induced *ICAM-1* and *PAI-1* gene expression, but also significantly increased the expression of the antioxidant *HO-1* gene compared to traditional LSM. Data are expressed as the mean ± S.E.M. (*n* = 5). \**p* < 0.05 indicates a significant difference relative to the individual control group. #*p* < 0.05 indicates a significant difference relative to the individual new LSM group. SM, static medium; LSM, new laminar shear medium; T-LSM, traditional laminar shear medium.

attenuation of THP-1/HAEC adhesion in the TNF-α+SM group was not significant compared to that in the TNF-α group.

TF initiates the coagulation cascade upon vascular injury. TF-specific procoagulant activity is induced in endothelial cells to initiate coagulation and thrombosis (Steffel et al., 2006). Compared to the control group, TNF-α treatment for 6 h increased TF activity by 2.26-fold, while the TNF-α+LSM group showed significantly attenuated TF activity (1.16-fold). Similar to the monocyte adhesion assay, attenuation of TF activity in the TNF-α+SM group was not significant compared to in the TNF-α group (**Figure 2C**).

### Evaluation of Effects of LSM and SM on TNF-α-Induced Endothelial Dysfunction-Related Gene and Protein Expression

**Figure 3** shows that the gene expressions levels of inflammation-, thrombosis-, and ROS induction-related genes, ICAM-1, VCAM-1, MCP-1, HO-1, NQO-1, Keap-1, TF, TM, and PAI-1 of the TNF-α group increased significantly by an average of 1.48- , 1.56-, 1.68-, 1.3-, 1.19-, 1.43-, 2.15-, 1.09-, and 1.65 fold, respectively, relative to the control group (**Figures 3A–C**). Compared to the TNF-α group, the expression levels of inflammation- and thrombosis-related genes, ICAM-1, VCAM-1, MCP-1, TF, and PAI-1 in the TNF-α+LSM groups for 6 h were significantly attenuated (**Figures 3A,C**). Expression levels of the anti-thrombotic TM gene increased significantly by an average of 1.25-fold (1.36/1.09) relative to the TNF-α group (**Figure 3C**). Among the three ROS induction-related genes, Keap-1 gene expression was significantly attenuated in the TNFα+LSM groups. The expression levels of the other HO-1 and NQO-1 antioxidant genes increased significantly by an average of 1.08-fold (1.4/1.3) and 1.09-fold (1.3/1.19) relative to the TNF-α group (**Figure 3B**). However, gene expression in the TNF-α+SM groups was not significantly different from that in the TNF-α group.

Western blotting revealed that the protein levels of HO-1, ICAM-1, and PAI-1 increased significantly in the TNF-α groups compared to in the control group. Compared to the TNF-α group, the expressions level of the antioxidant protein HO-1 increased significantly in the TNF-α+LSM group, but did not show the same trend in the TNF-α+SM group. ICAM-1 and PAI-1 expression was significantly attenuated in the TNFα+LSM group compared to those in the TNF-α+SM group (**Figure 3D**). The data indicate consistent trends in gene and protein expression.

### Quantification of Cytokines Secreted by HAECs in SM and LSM by Cytokine Antibody Array

In **Figure 4A**, the left and right panels show SM and LSM data, respectively, and the array C6 and C7 are shown in upper and bottom panels, respectively. We quantified the raw numerical densitometry data and showed relative changes in the level of each cytokine in Supplementary Data 3. The top three over-expressed marker cytokines of LSM were HGF, G-CSF, and IL-17 (labeled in red), the top three under-expressed marker cytokines were GM-CSF, EGFR, and MIG (labeled in green), and the cytokine-spots are shown in the same color. Considering the dosage effects, we screened highly abundant marker cytokines (each densitometry data >5,000 in both the SM and LSM) from advanced experiments. Among all highabundance marker cytokines, secreted bFGF (1.72-fold, LSM relative to SM) and MCP-1 (0.75-fold, LSM relative to SM) were the two most abundant, showing a maximum fold-change as the most under and over-expressed biomarkers, respectively (in bold type in Supplementary Data 3). Thus, we investigated whether bFGF protects the LSM from endothelial dysfunction. To confirm whether the above over-expressed and underexpressed cytokines were secreted from LSS-exposed HAECs, we collected the cell pellets to determine the mRNA transcript levels by qPCR. The data indicated that the gene levels in the underlying HAECs exposed to LSS for 24 h were as follows: under-expressed cytokines (EGFR, GM-CSF, MIG, and MCP-1) were 0.63-, 0.64-, 0.54-, and 0.4-fold lower than the static control (**Figure 4B**) and over-expressed cytokines (HGF, G-CSF, IL-17, and bFGF) were 1.89-, 1.7-, 1.57-, and 1.45-fold higher than the static control (**Figure 4C**). These trends are consistent with those of cytokine expression between SM and LSM, as shown in **Figure 4A**. In western blotting, the secreted autocrine type 18 kDa bFGF protein was not only over-expressed in LSM compared to in SM (**Figure 4D**), but also over-expressed in the underlying HAECs exposed to LSS for 24 h compared to in the static control (**Figure 4E**).

### Evaluation of Protective Efficacy of LSM, rbFGF, and LSM+Ab on TNF-α-Induced Endothelial ROS Induction, Inflammation, and Thrombosis

To validate the HAEC-secreted autocrine rbFGF as a critical protector in the LSM, we evaluated the protective efficacy of rbFGF and LSM+Ab using the H2DCFDA assay for ROS activity. As shown in **Figure 5A**, the TNF-α group

S.E.M. (*n* = 4). \**p* < 0.05 indicates a significant difference relative to the control group. #*p* < 0.05 indicates a significant downregulation relative to the individual TNF-α group. (C) TF activity was significantly induced (average 2.26-fold) in the TNF-α group and attenuated significantly (average 1.16-fold) in the TNF-α+LSM group compared to that in the TNF-α group. Data are expressed as the mean ± S.E.M. (*n* = 3). \**p* < 0.05 indicates a significant difference relative to the control group. #*p* < 0.05 indicates a significant downregulation relative to the individual TNF-α group. SM, static medium. LSM, new laminar shear medium.

FIGURE 3 | Determination of inflammation, ROS induction, and thrombosis-related gene and protein expression. Expression of (A) inflammation*-*related *ICAM-1*, *VCAM-1*, and *MCP-1* genes, (B) ROS induction–related *HO-1*, *NQO-1*, and *Keap-1* genes and (C) thrombosis-related *TF*, *TM*, and *PAI-1* genes were all significantly increased in the TNF-α group compared to in the control group. Gene expression levels of *ICAM-1*, *V-CAM-1*, *MCP-1*, *Keap-1*, *TF*, and *PAI-1* were attenuated significantly by LSM treatment compared to those in the TNF-α group, but similar trends were not observed in the TNF-α+SM group. However, gene expression of *HO-1*, *NQO-1*, and *TM* were significantly enhanced by LSM treatment compared to those in the TNF-α group, but similar trends were not observed in the TNF-α+SM group. (A–C) Data are expressed as the mean ± S.E.M. (*n* = 3). \**p* < 0.05 indicates a significant difference relative to the control group. #*p* < 0.05 indicates a significant up- or downregulation, relative to the individual TNF-α group (excluding all control groups). (D) Compared with those in the control group, the protein levels of HO-1, ICAM-1, and PAI-1 were increased significantly in the TNF-α group. The expressions levels of the antioxidant protein HO-1 increased significantly in the TNF-α+LSM group compared to those in the TNF-α group. ICAM-1 and PAI-1 protein expression were attenuated significantly in the TNF-α+LSM group compared to those in the TNF-α group (*n* = 3). SM, static medium; LSM, new laminar shear medium. \**p* < 0.05 indicates a significant difference relative to the control group. *#p* < 0.05 indicates a significant up- or downregulation relative to the individual TNF-α group.

displayed high levels of H2DCFDA-dependent fluorescence, while the TNF-α+LSM group showed significantly attenuated levels of H2DCFDA-dependent fluorescence (average 0.82-fold) compared to the TNF-α group. The TNF-α+rbFGF group also showed significantly attenuated levels of H2DCFDA-dependent fluorescence (average 0.87-fold) compared to the TNF-α group.

*(Continued)*

FIGURE 4 | the 120 candidate cytokines, HAEC-secreted MCP-1 (C6) and bFGF (C7) were the two most abundant (O.D. level of each spot >5,000) and were accompanied by a maximum fold-change as the top under and over-expressed biomarkers. Secreted MCP-1 and bFGF proteins were 0.75- and 1.72-fold higher in LSM than in SM, respectively. The top three over-expressed marker cytokines in LSM are framed in red (HGF, G-CSF, and IL-17A), and the top three under-expressed cytokines (EGFR, GM-CSF, and MIG) in LSM are framed in green. A–N is X-axis and 1–10 is Y-axis, the XY combination indicated the location of each cytokine (Supplementary Data 3). (B,C) Comparison of gene expression levels of the above 8 marker cytokines in underlying HAECs between static control cells and cells exposed to LSS for 24 h by qPCR. Data are expressed as the mean ± S.E.M. (*n* = 3). \**p* < 0.05 indicates a significant difference relative to the static control groups. (D) The expressions levels of the secreted autocrine type, 18-kDa bFGF protein levels, increased significantly in the LSM compared to in the SM. (*n* = 3) (E) Expressions levels of the 18-kDa bFGF protein levels increased significantly in HAECs exposed to LSS for 24 h compared to the static control. (*n* = 3). \**p* < 0.05 indicates a significant difference relative to the SM (D) and static control (E) groups, respectively.

The TNF-α+LSM+Ab group displayed ROS activity similar to that of the TNF-α group (average 0.98-fold). As shown in **Figure 5B**, the TNF-α+LSM and TNF-α+rbFGF groups showed significantly reduced adhesiveness of THP-1/HAECs (0.34- and 0.4-fold) compared to the TNF-α group. The adhesiveness of THP-1/HAECs was not changed in the TNF-α+LSM+Ab group compared to that in the TNF-α group. As shown in **Figure 5C**, only TNF-α treatment for 6 h revealed a 2.66-fold increase in TF activity compared to that in the control group. The TNF-α+LSM and TNF-α+rbFGF groups showed significantly decreased TF activity by 1.32- and 1.42-fold compared to the control TF activity level. Similar to in the THP-1/HAECs adhesion assay, the TNFα+LSM+Ab group (2.35-fold) did not show a TF activity level that was significantly different from the level in the TNF-α group (**Figure 5C**).

### Evaluation of Effects of LSM, rbFGF, LSM+Ab, and LSM+IgG on TNF-α-Induced Endothelial Dysfunction-Related Gene and Protein Expression

Compared with the TNF-α group, the expression levels of inflammation, thrombosis, and ROS induction-related genes, ICAM-1, VCAM-1, MCP-1, HO-1, NQO-1, Keap-1, TF, TM, and PAI-1 in the TNF-α+LSM, TNF-α+LSM+IgG, and TNF-α+rbFGF groups differed significantly (**Figures 6A–C**). However, ICAM-1, VCAM-1, MCP-1, HO-1, NQO-1, Keap-1, TF, and TM gene expression levels in the TNF-α+LSM+Ab group did not significantly differ from those in the TNF-α group (**Figures 6A–C**). Unexpectedly, only PAI-1 gene expression levels in the TNF-α+LSM+Ab group decreased significantly compared to the TNF-α group. Western blotting revealed that the protein levels of ICAM-1 and PAI-1 were attenuated in the TNF-α+LSM, TNF-α+rbFGF, and TNF-α+LSM+IgG groups compared to that in the TNF-α group (**Figure 6D**). In contrast, protein expression levels of HO-1 increased significantly in the TNF-α+LSM, TNFα+rbFGF, and TNF-α+LSM+IgG groups compared to that in the TNF-α group. These data indicate that the pattern of gene and protein expression was nearly consistent in this study.

### Evaluation of Effects of LSM, rbFGF, and LSM+Ab on Acute TNF-α-Induced Endothelial Dysfunction and Lung Injury in C57BL/6J Mice

As shown in **Figure 7A**, immunohistochemical analysis showed that aortic endothelial ICAM-1 and PAI-1 immunoreactivity levels in the TNF-α+LSM and TNF-α+rbFGF groups were decreased dramatically compared to in the TNF-α and TNFα+LSM+Ab groups. No immunoreactivity was detected when the primary antibody was omitted (data not shown). Quantification of immunoreactivity signals in the endothelial layers revealed that ICAM-1 expression was decreased significantly to 0.21- and 0.15-fold in the TNF-α+LSM and TNF-α+rbFGF groups compared to that in the TNF-α group. PAI-1 expression in the endothelial layers decreased significantly to 0.22- and 0.26-fold in the TNF-α+LSM and TNF-α+rbFGF groups compared to in the TNF-α group (**Figure 7B**). The immunoreactivity of these two proteins did not differ significantly between the TNF-α group and TNFα+LSM+Ab group. In addition, 24 h after TNF-α injection, ICAM-1 and PAI-1 mRNA expression was significantly reduced by approximately to 0.55- and 0.64-fold from that in the TNF-α group in lung extractions by co-injection of rbFGF, respectively. Unexpectedly, the dramatic reductions in ICAM-1 and PAI-1 mRNA expression were not observed in the TNF-α+LSM group (**Figure 7C**). Interestingly, the highest serum bFGF levels were found in the TNF-α+rbFGF group (Supplementary Data 4) and the serum levels of bFGF increased by injection of rbFGF correlated with positive effects on lung tissues. These results suggest that intraperitoneal injection of rbFGF has more protective potential as a therapeutic drug than LSM for inhibiting inflammation and thrombosis in TNF-α-induced tissue injury by reversing the increased expression of ICAM-1 and PAI-1.

### rbFGF Inhibited Inflammatory- and Thrombosis-Related Gene and Protein Expression via Activation of MEK5-KLF2 and p38-MAPK Signal Pathway

Following the above in vivo study, we evaluated the protective mechanism of rbFGF. We examined whether the protective mechanism of bFGF also depended on the laminar shearresponsive transcription factor, KLF2, as the classical atheroprotection pathway of LSS (Dekker et al., 2002; Fledderus et al., 2008). As shown in **Figure 8A**, rbFGF did induce significant overexpression of the KLF2 gene compared to in the control and TNF-α groups and achieved similar expression levels as the KLF2 gene in the TNF-α+LSM and TNF-α+LSM+IgG groups. Moreover, we used BIX02159 (specific inhibitor of MEK5) and SB203580 (specific inhibitor of p38-MAPK) to test the protective mechanism of rbFGF in TNF-α-induced ICAM-1 and PAI-1 gene overexpression. As shown in **Figures 8B–D**, pretreatment with SB203580 and BIX02189 reversed the inhibition of

displayed a high level of H2DCFDA-dependent ROS activity, while the TNF-α+LSM and TNF-α+rbFGF groups showed significantly attenuated levels (average 0.81 and 0.86-fold) compared to the TNF-α group. (B) Compared to the TNF-α group, the high THP-1/HAECs adhesion ratios were attenuated significantly (average 0.34 and 0.4-fold) in the TNF-α+LSM and TNF-α+rbFGF groups. (A,B) The data are expressed as the mean ± S.E.M. (*n* = 4). \**p* < 0.05 indicates significant downregulation relative to the TNF-α group. (C) Compared to the control group, the highest level of TF activity was increased by 2.66-fold in the TNF-α group. TF activity decreased significantly (average 1.32- and 1.42-fold) in the TNF-α+LSM and TNF-α+rbFGF group compared to the TNF-α group, respectively. However, the TNF-α+LSM+Ab group (average 2.35-fold) was not significantly different from the TNF-α group. The data are expressed as the mean ± S.E.M. (*n* = 4). \**p* < 0.05 and *#p* < 0.05 indicate a significant difference relative to the control group and TNF-α group, respectively.

Frontiers in Physiology | www.frontiersin.org

FIGURE 6 | observed in the TNF-α+LSM+Ab groups. Additionally, ROS induction-related antioxidant HO-1, NQO-1 and anti-thrombotic TM gene levels were enhanced in the TNF-α+LSM, TNF-α+LSM+IgG, and TNF-α+rbFGF groups compared to in the TNF-α groups. (A–C) Data are expressed as the mean ± S.E.M. (*n* = 3). \**p* < 0.05 indicates a significant difference relative to the TNF-α treated groups. (D) Paralleling their gene expression levels, the expression levels of ICAM-1 and PAI-1 protein decreased in the TNF-α+LSM, TNF-α+rbFGF, and TNF-α+LSM+IgG groups compared to in the TNF-α group. In addition, the expression levels of HO-1 protein increased in the TNF-α+LSM, TNF-α+rbFGF, and TNF-α+LSM+IgG groups compared to in the TNF-α group (*n* = 3). \**p* < 0.05 indicates a significant different relative to the TNF-α group.

ICAM-1 and PAI-1 gene and protein overexpression from the TNF-α+bFGF groups. The inhibition assay revealed that the protective mechanism of bFGF on inflammation and thrombosis involved activation of the p38-MAPK and MEK5/ERK5-KLF2 signal pathway to inhibit nuclear factor-κB downstream target genes, ICAM1-1 and PAI-1, following TNF-α stimulation (**Figure 9**).

### DISCUSSION

In our modified cone-and-plate system, HAECs were seeded and exposed to LSS for 24 h. In the peripheral area of the 10 cm culture dish, HAECs exposed to LSS exhibited the typical alignment of cell shape in the direction of the laminar flow. In addition, expression levels of mechanosensitive genes of the underlying HAECs, TM, HO-1, NQO-1, and KLF-2, were also upregulated, similar to the results observed for the traditional LSS system. The MTS assay confirmed the non-toxicity of the new LSM. These data validated that our modified system also benefits the underlying HAECs, similar to the traditional coneand-plate system. Moreover, we showed that the new LSM of the modified system was more effective than the traditional system in preventing TNF-α-induced endothelial dysfunction based on qPCR of ICAM-1, HO-1, and PAI-1 gene levels. We also demonstrated that the LSM of the new LSS system has therapeutic potential to rescue TNF-α-induced dysfunction in HAECs, based on both quantitative and qualitative analyses. To our knowledge, this is the first study to report the application of a modified traditional cone-and-plate shear system in studies investigating LSS-mediated paracrine and autocrine effects.

In 2004, Chiu et al. showed that high shear stress (20 dynes/cm<sup>2</sup> ) enhanced the TNF-α-induced expression of ICAM-1 at the mRNA and surface-protein levels in underlying HUVECs, but suppressed TNF-α-induced expression of VCAM-1 and Eselectin proteins (Chiu et al., 2004). In 2015, Fan et al. also demonstrated that both high shear stress and oscillatory shear stress induced over-expression of ICAM-1 in the underlying HUVECs relative to the levels in the static control, although ICAM-1 levels were lower under higher shear stress (Fan et al., 2015). Interestingly, our data showed that LSM could overcome TNF-α-induced ICAM-1 gene and protein overexpression in the underlying HAECs. A possible explanation is that increased ICAM-1 levels have only been observed in venous endothelial cells, HUVECs. Thus, studies of aortic endothelial cells, HAECs, may show different results. Moreover, the increased ICAM-1 levels in underlying HUVECs may be induced by direct physical shear stress. Instead, the new LSM contained cytoprotective substances that functioned in an autocrine and paracrine manner by decreasing TNF-α-induced ICAM-levels in HAECs.

In the disturbed flow, Urschel et al. reported that HUVECs exposed to flow in bifurcating slides secreted increased levels of ICAM-1 and interleukin-8 in conditioned medium compared to cells grown under laminar flow (Urschel et al., 2012). Additionally, Bajari et al. reported that conditioned medium from chronic low shear stress (2.5 dynes/cm<sup>2</sup> ) enhanced the migration of vascular smooth muscle cells (Bajari et al., 2014). In 2015, Franzoni et al. also reported that conditioned medium from endothelial cells exposed to reciprocating flows could increase the proliferation of smooth muscle cells (Franzoni et al., 2016). In contrast, under laminar flow, Slater et al. reported that conditioned medium from glomerular endothelial cells under chronic LSS (10 dynes/cm<sup>2</sup> ) decreased podocyte monolayer resistance (Slater et al., 2012). These reports suggest the potential bio-functional flow-mediated effects of "conditioned media" when shear stress is applied to the underlying endothelial cells. Using the modified cone-and-plate shear device, we also found that LSM was effective in rescuing cells from TNF-α-induced endothelial dysfunction.

Previous studies indicated that upon exposure to environmental signals, cytokines in vascular endothelial cells undergo changes in gene expression and function that allow these cells to actively participate in inflammatory reactions, immunity, and thrombosis (Mantovani et al., 1992). In this study, the same commercial medium was used in both the static control and LSS system and the underlying HAECs did secrete diverse cytokines into LSM to provide significant protective effects compared to SM (**Figure 4**). Considering the effect of dosage, we isolated highly abundant cytokines (all OD > 5,000) as marker cytokines, and sorted all marker cytokines by relative changes in levels in the LSM and SM. Thus, the most over-expressed bFGF (LSM/SM is 1.72-fold) was chosen as the target marker cytokine in this study (Supplementary Data 3). Based on in vitro and in vivo experiments, bFGF was identified as a critical protective factor in LSM against endothelial dysfunction (**Figures 5**–**8**).

The FGF family is comprised secreted signaling proteins (secreted FGFs) that signal to receptor tyrosine kinases and intracellular non-signaling proteins (intracellular FGFs) (Ornitz and Itoh, 2015). bFGF (FGF-2) is the prototype member of a family of heparin-binding growth factors, and lacks a signal peptide that directs secretion through the classical secretory pathway (Bikfalvi et al., 1997; Ornitz and Itoh, 2015). It is also a multifunctional protein, translated from the same mRNA as its high molecular weight (21–24 kDa; Hi-bFGF) or low molecular weight (18 kDa; Lo-bFGF) isoforms. Hi-bFGF localizes preferentially to the cell nucleus and exerts exclusively intracrine activities. In contrast, autocrine or paracrine bFGF activities were demonstrated to represent the action of Lo-bFGF, also known as secreted-bFGF (Kardami et al., 2004). A previous

FIGURE 8 | Identification of cytoprotective mechanisms of bFGF. (A) Gene expression of *KLF2* was increased significantly by 1.2-fold by TNF-α, compared to the control group. In the TNF-α+LSM, TNF-α+LSM+IgG, and TNF-α+rbFGF groups, the gene expression of *KLF2* was further enhanced by 1.41-, 1.57-, and 1.53- fold compared to the control groups. The KLF2 level in the TNF-α+LSM+Ab group was not significantly different from that in the TNF-α group. Data are expressed as the mean ± S.E.M. (*n* = 4). \**p* < 0.05 and *#p* < 0.05 indicate a significant different relative to the control group and TNF-α group, respectively. (B) Compared to the TNF-α group, *ICAM-1* expression was significantly decreased to 0.71-fold in the TNF-α+bFGF group. However, the anti-inflammatory effect of bFGF on *ICAM-1* expression was reversed to 0.99- and 1.35-fold compared to in the TNF-α group by pretreatment with SB203580 (10µM) and BIX02189 (10µM), respectively. (C) Compared to the TNF-α group, the *PAI-1* expression level was significantly decreased to 0.76-fold in the TNF-α+bFGF group. However, the anti-thrombotic effect of bFGF on PAI-1 gene expression was reversed to 0.93- and 1.05-fold compared to in the TNF-α group by pretreatment with SB203580 and BIX02189, respectively. T: TNF-α (B,C) Data are expressed as the mean ± S.E.M. (*n* = 4). #*p* < 0.05 indicates a significant difference relative to the TNF-α group and \$ *p* < 0.05 indicates a significant difference relative to the TNF-α+rbFGF group. (D) Compared with the TNF-α group, ICAM-1 and PAI-1 protein levels were significantly decreased in the TNF-α+rbFGF group and were reverted by pretreatment with SB203580 and BIX02189, respectively (*n* = 3). #*p* < 0.05 indicates a significant difference relative to the TNF-α group and \$ *p* < 0.05 indicates a significant difference relative to the TNF-α+rbFGF group.

study suggested that endothelial cell-derived bFGF mediates angiogenesis in an autocrine manner in cancer (Seghezzi et al., 1998). This is the first study to demonstrate that LSS-induced aortic endothelial cells over-expressed and secreted autocrine type 18 kDa bFGF into the external environment (**Figures 4C,D**). In cultured microvascular endothelial cells, bFGF induced the development of an angiogenic phenotype including increased proliferation, migration, and proteinase production (Bikfalvi et al., 1997; Seghezzi et al., 1998). In addition, in cancer research, bFGF signaling was showed to possess powerful cardio-protective effects following stress and ischemic injury (Kardami et al., 2004; House et al., 2005; Liao et al., 2007). Mice lacking the bFGF gene develop normally, but show reduced vascular tone, impaired cardiac hypertrophy, reduced cortical neuron density, and defects in response to pulmonary or cardiac injury (Dono et al., 2002; House et al., 2010). No previous studies have shown that secreted bFGF from LLS-exposed HAECs can prevent endothelial dysfunction. In this study, we evaluated the protective efficacy of rbFGF and LSM+Ab (bFGF neutralization by monoclonal Ab) against TNF-α-induced HAEC dysfunction in vitro and in vivo (**Figures 5**–**7**). The results suggest significant protective effects of rbFGF (10 ng/mL), similar to LSM against TNF-α-induced endothelial dysfunction. Interestingly, the protective effects of LSM were only observed in aortic endothelial dysfunction, but not reproduced in lung extractions in vivo (**Figure 7B**). These results suggest that bFGF has more protective effects than LSM against inflammation and thrombosis following TNF-α stimulation in vivo.

Atheroprotective blood flow induces the expression of anti-inflammatory KLF2 expression, a transcriptional factor responsible for the physiological healthy, flow-exposed state of endothelial cells (Fledderus et al., 2008; Chiu and Chien, 2011). In a previous study, Parmar et al. documented that KLF2 is induced under laminar flow via MAPK/ERK kinase 5 (MEK5)-extracellular signal-regulated protein kinase 5 (ERK5)- MEF2 signaling pathway (Parmar et al., 2006). In addition, Flati et al. reported that bFGF can inhibit the activation of nuclear factor-κB and ICAM-1 elevation by inducing p38-MAPK (Flati et al., 2006). Thus, we tested KLF2 expression among the different treated groups and used the specific inhibitors BIX02159 (MEK5 inhibitor) and SB203580 (P38 inhibitor) to evaluate the individual reversions of anti-inflammatory and anti-thrombotic protections of bFGF. Similar to previous KLF2 and bFGF associated studies, we found that pretreatment with SB203580 and BIX02189 inhibitors significantly reversed the protection of bFGF on KLF2 downstream ICAM-1 and PAI-1 gene and protein reductions from the TNF-α+bFGF groups (**Figures 8B–D**).

This study implies three novel findings: (1) LSM collected from the new LSS system had more protective effects against TNF-α stimulated endothelial dysfunction, than traditional LSM. (2) LSS-exposed HAEC-secreted autocrine bFGF is the critical factor in the LSM that provides diverse cytoprotective effects against TNF-α-stimulated endothelial dysfunction. (3) The

### REFERENCES


protective mechanism of bFGF via activation of the p38-MAPK and MEK5/ERK5-KLF2 signal pathway inhibits nuclear factorκB downstream target genes (**Figure 9**). Based on previously reported variable atheroprotective effects of LSS and our in vitro and in vivo results, autocrine-type secreted bFGF is a critical protector in LSM, providing a new perspective for vascular endothelial shear stress studies.

### AUTHOR CONTRIBUTIONS

H-JW and W-YL designed and performed the experiments; analyzed the data and wrote the manuscript; provided the funding. W-YL supervised the study.

### FUNDING

This study was supported in part by a grant from China Medical University and Hospital (DMR-106-010) and Hungkuang University (HK-KTOH-105-01 and HK105-102).

### ACKNOWLEDGMENTS

The authors thanks Dr. Hanjoong Jo (Department of Biomedical Engineering and Division of Cardiology, Emory University) for kindly providing the cone-and-plate shear device. The authors thank Jen-Duo Liou and Wan-Yu Pai for technical assistance. The animal study was supported by Laboratory Animal Center of Hungkuang University.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2017.01095/full#supplementary-material


proteins in human coronary endothelial cells. Circulation 106, 466–472. doi: 10.1161/01.CIR.0000023043.02648.51


**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 Wang and Lo. 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.

# MicroRNA-146a-5p Mediates High Glucose-Induced Endothelial Inflammation via Targeting Interleukin-1 Receptor-Associated Kinase 1 Expression

Wan-Yu Lo<sup>1</sup> , Ching-Tien Peng2, 3 and Huang-Joe Wang4, 5, 6 \*

<sup>1</sup> Cardiovascular & Translational Medicine Laboratory, Department of Biotechnology, Hungkuang University, Taichung, Taiwan, <sup>2</sup> Department of Pediatrics, Children's Hospital, China Medical University and Hospital, Taichung, Taiwan, <sup>3</sup> Department of Biotechnology, Asia University, Taichung, Taiwan, <sup>4</sup> School of Medicine, China Medical University, Taichung, Taiwan, <sup>5</sup> Division of Cardiovascular Medicine, Department of Medicine, China Medical University and Hospital, Taichung, Taiwan, <sup>6</sup> Cardiovascular Research Laboratory, China Medical University and Hospital, Taichung, Taiwan

Background and Aims: Interleukin-1 receptor-associated kinase-1 (IRAK-1) is critical for mediating toll-like receptor and interleukin-1 receptor signaling. In this study, we have examined whether IRAK-1 expression is altered in high glucose (HG)-stimulated human aortic endothelial cells (HAECs), and whether microRNAs (miRs) target IRAK-1 to regulate HG-induced endothelial inflammation.

### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

### Reviewed by:

Adán Dagnino-Acosta, University of Colima, Mexico Ricardo Espinosa-Tanguma, Universidad Autónoma de San Luis Potosí, Mexico

> \*Correspondence: Huang-Joe Wang joe5977@ms32.hinet.net

#### Specialty section:

This article was submitted to Vascular Physiology, a section of the journal Frontiers in Physiology

Received: 13 April 2017 Accepted: 17 July 2017 Published: 02 August 2017

### Citation:

Lo W-Y, Peng C-T and Wang H-J (2017) MicroRNA-146a-5p Mediates High Glucose-Induced Endothelial Inflammation via Targeting Interleukin-1 Receptor-Associated Kinase 1 Expression. Front. Physiol. 8:551. doi: 10.3389/fphys.2017.00551 Methods: HAECs were treated with HG for 24 and 48 h. Real-time PCR, Western blot, monocyte adhesion assay, bioinformatics analysis, TaqMan® arrays, microRNA mimic or inhibitor transfection, luciferase reporter assay and siRNA IRAK-1 transfection were performed. The aortic tissues from db/db type 2 diabetic mice were examined by immunohistochemistry staining.

Results: HG time-dependently increased IRAK-1 mRNA and protein levels in HAECs, and was associated with increased VCAM-1/ICAM-1 gene expression and monocyte adhesion. Bioinformatic analysis, TaqMan® arrays, and real-time PCR were used to confirm that miR-146a-5p, miR-339-5p, and miR-874-3p were significantly downregulated in HG-stimulated HAECs, suggesting impaired feedback restraints on HG-induced endothelial inflammation via IRAK-1. However, only miR-146a-5p mimic transfection reduced the HG-induced upregulation of IRAK-1 expression, VCAM-1/ICAM-1 expression, and monocyte adhesion. Additionally, IRAK-1 depletion reduced HG-induced VCAM-1/ICAM-1 gene expression, and monocyte adhesion, indicating that HG-induced endothelial inflammation was mediated partially through IRAK-1. In vivo, intravenous injections of miR-146a-5p mimic prevented endothelial IRAK-1 and ICAM-1 expression in db/db mice.

Conclusion: These results suggest that miR-146a-5p is involved in the regulation of HG-induced endothelial inflammation via modulation of IRAK-1; indicating that miR-146a-5p may be a novel target for the treatment of diabetic vascular complications.

Keywords: diabetes, high glucose, endothelial inflammation, Interleukin-1 receptor-associated kinase-1, miR-146a-5p

## INTRODUCTION

The increasing number of people with obesity, advanced age, and physically inactive lifestyles contributes to the increased incidence of diabetes, which had an estimated global prevalence of 6.4% in 2010 (Shaw et al., 2010). Diabetic vascular disease is a chronic inflammatory disease that accounts for the majority of morbidity and mortality in diabetic patients. Hyperglycemia in diabetes causes endothelial dysfunction (Nakagami et al., 2005), a condition characterized by impaired vasodilatation, proinflammation, prothrombosis, and impaired endothelial repairs that precipitate atherosclerotic progression and atherothrombotic complications (Sena et al., 2013).

Interleukin-1 receptor activated kinases (IRAKs) are the key mediators of innate immunity. The mammalian IRAK family consists of four members: IRAK-1, IRAK-2, IRAK-M, and IRAK-4. Of these, IRAK-1 was the first to be identified; its role as an adaptor and kinase is essential for the toll-like receptor (TLR) and interleukin-1 receptor (IL-1R) signaling pathways, which regulate cellular inflammation (Gottipati et al., 2008). TLR signaling is initiated by ligand-induced dimerization of receptors (e.g., lipopolysaccharide ligand (LPS)–TLR4 receptors). Following the recruitment of adaptor molecules, including myeloid differentiation primary response protein (MYD88), the downstream signaling pathways involve the interaction of interleukin-1 receptor-associated kinase 1 (IRAK-1) and TNF receptor-associated factor 6 (TRAF-6) (Flannery and Bowie, 2010). Downstream from TRAF-6, transcription factors, including nuclear factor κB (NF-κB), interferon-regulatory factor 5, cyclic AMP response element-binding protein, and activator protein 1, are activated to induce the production of inflammatory cytokines and type 1 interferon (O'Neill et al., 2013; Jain et al., 2014). In addition, IRAK-1 signaling can be initiated through IL-1R upon ligand binding (Jain et al., 2014). IL-1R activation can drive a variety of inflammatory mediators (e.g., interleukin 6, tumor necrosis factor α) to induce host protection (Gottipati et al., 2008). Hyperglycemia has been suggested to cause dysregulated innate immunity (Jafar et al., 2016; Kousathana et al., 2017). In addition, innate immunity-induced inflammation is important for the pathogenesis and disease progression of both type 1 and type 2 diabetes (Prajapati et al., 2014; Cabrera et al., 2016; Wada and Makino, 2016; Mistry et al., 2017). However, little is known about the role of IRAK-1 in diabetes.

MicroRNAs (miRs) are important post-transcriptional regulators of the endothelial oxidative and inflammatory responses (Marin et al., 2013). Increasing evidence suggests that miRs are involved in the pathogenesis of diabetes (Guay et al., 2011). Additionally, diabetic complications can be predicted from the circulating levels of certain miRs (Guay and Regazzi, 2013). Although utilizing differential miRs as therapeutic targets in diabetes is a potentially promising strategy, the human genome encodes more than 1,600 miR precursors, making the identification of potential miR targets difficult (Kolfschoten et al., 2009; Guay and Regazzi, 2013).

In this study, we have investigated the potential miRs that regulate IRAK-1 expression in high glucose (HG)-stimulated human aortic endothelial cells (HAECs), and performed an in vivo examination of aortic endothelial cells from db/db type 2 diabetic mice.

### MATERIALS AND METHODS

### Cell Culture

HAECs were purchased from Cell Applications, Inc. (San Diego, CA, USA) and cultured in endothelial cell growth medium (Cell Applications, Inc.) according to the manufacturer's recommendations. All chemicals were obtained from Sigma-Aldrich (St. Louis, MO, USA) unless otherwise specified. High glucose (HG, 25 mM) was added to HAECs for 24 and 48 h in the different experiments. Mannitol (25 mM) was the osmotic control. The human monocytic cell line THP-1 was obtained from the American Type Culture Collection (Rockville, MD, USA), and maintained in RPMI 1640 culture medium supplemented with 10% FBS, L-glutamine, and penicillin.

### Real-Time Polymerase Chain Reaction (PCR)

Expression of mRNA in the HAECs was analyzed by real-time PCR as previously described (Wang et al., 2010). The primer sequences for glyceraldehyde-3-phosphate dehydrogenase (GAPDH), vascular cell adhesion protein 1 (VCAM-1), and intercellular adhesion molecule 1 (ICAM-1) are provided in Supplemental Data 1.

### Western Blot Analysis

Protein expression levels in the HAECs were analyzed by western blot as previously described (Wang et al., 2010). Antibodies against IRAK-1 (Cell Signaling Technology, Danvers, MA, USA), and GAPDH (Santa Cruz Biotechnology, Santa Cruz, CA, USA) were used at 1:1,000.

### Monocyte Adhesion Assay

In adhesion experiments, THP-1 cells were labeled with calcein acetoxymethyl ester (Calcein-AM; Molecular Probes, Eugene, OR, USA) as previously described (Wang, H. J. et al.,, 2014). Briefly, THP-1 cells were stained with the dye at a concentration of 7.5 µM for 30 min immediately preceding the adhesion assay. HAECs were maintained in 12 well-plates until 90% confluence. The HAECs (10<sup>5</sup> cells/well) were then treated with HG for 24 and 48 h and incubated with culture medium containing the labeled THP-1 cells (THP-1/HAECs = 7) for 10 min. Nonadherent THP-1 cells were removed by washing with PBS for 20 s. Adherent THP-1 cells were visualized and quantified in 10 randomly viewed fields by the fluorescent microscope (OLYMPUS, Japan).

### TaqMan Array Human MicroRNA Card Analysis

The TaqMan <sup>R</sup> Array Human MicroRNA A Card V2 (Applied Biosystems, Foster, CA, USA) was used to analyze miR expression profiles. The card contains 377 preloaded human miR targets and four endogenous controls. For each sample, 500 ng of total RNA was used for reverse-transcription, using Megaplex RT primer Pool A and a TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems). The resulting cDNA was diluted, mixed with TaqMan Gene Expression Master Mix (Applied Biosystems), and loaded into the ports on microfluidic cards. The cards were briefly centrifuged for 1 min at 1,600 × g to distribute samples to the multiple wells, sealed to prevent well-to-well contamination, and analyzed using a 7900 HT Real-Time PCR System (Applied Biosystems).

### Extraction and Analysis of HAEC miRs

The protocols for miR extraction and determination of different miR expression levels from HAECs were as previously described (Wang et al., 2013). The reverse transcription and PCR primer sequences are provided in Supplemental Data 1.

### Transfection of miR Mimics and Inhibitors

Selected miR mimics, inhibitors, and a negative control (NC) were transfected into HAECs as previously described (Wang, H. J. et al.,, 2014). After transfection, HAECs were treated with HG for 48 h, after which the expression levels of IRAK-1 mRNA, IRAK-1 protein, VCAM-1 mRNA, and ICAM-1 mRNA were determined. THP-1 adhesion assays were also performed after miR-146a-5p mimic transfection.

### Luciferase Reporter Assay

A partial IRAK-1 mRNA 3′ -UTR containing the miR-146a-5p target site was constructed into a pGL-3-promoter vector (Promega, Madison, WI). HAECs were cotransfected with 1 µg of constructed plasmids and 100 nM of miR-146a-5p mimic and the negative control using LipofectamineTM 2000 (Invitrogen, Carlsbad, CA). Empty vector was used as blank control. After 24 h of transfection, cells were harvested to measure luciferase activity using the Luciferase Assay System Kit (Promega, E1500), according to the manufacturer's instructions.

### IRAK-1 Gene Silencing

The HAECs were transfected with 100 nM of either ON-TARGETplus SMARTpool Human IRAK-1 small interfering RNA (siRNA; Dharmacon, Thermo Scientific, Lafayette, CO, USA) or a negative control, as previously described (Wang et al., 2016). The IRAK-1 siRNA target sequences were shown in Supplemental Data 2. Briefly, HAECs were transfected using LipofectamineTM 2000 transfection reagent (Invitrogen, Carlsbad, CA, USA) in M-199 medium for 2 h. After transfection, the medium was changed to endothelial cell growth medium, and the HAECs were treated with HG for 48 h. After HG treatment, expression of IRAK-1 mRNA, IRAK-1 protein, VCAM-1 mRNA, and ICAM-1 mRNA were determined. THP-1 adhesion assays were also performed after IRAK-1 siRNA transfection.

### Type 2 Diabetic Mouse Model Studies

The animal study was conducted with the ethical standards of the field and performed in accordance with the ethical guidelines provided by the Hungkuang University Institutional Animal Care and Use Committee (Permit Number: HK 105-02). Male db/db diabetic mice were purchased from National Laboratory Animal Center (Nangang, Taipei, Taiwan). Eleven-week db/db mice were administered (100 µL) miR-146a-5p mimic or a negative control (13 µg per week, 3 times) by tail-vein injection, using equal volume mixtures of LipofectamineTM 2000 and miR-146a-5p mimic or negative control. The control db/db group received equal volume mixtures of vehicle (LipofectamineTM 2000) and PBS. Three weeks later, mice were euthanized by CO<sup>2</sup> narcosis. The aortic tissue were carefully excised and fixed with 10% formalin solution. Paraffin sections (5 µm thickness) of aorta were prepared for immunohistochemistry (IHC) staining.

### IHC Staining

For aortic tissue sections from db/db mice, 3,3′ diaminobenzidine staining was performed using a Bond-Max autostainer (Leica Microsystems). Briefly, paraffin-embedded aortic tissue sections were placed in Tris buffered saline with Tween-20, then rehydrated through serial dilutions of alcohol, and washed with PBS (pH 7.2). Slides were then stained with primary antibodies against IRAK-1 (dilution 1:50, mouse monoclonal antibody, Santa Cruz), or ICAM-1 (dilution 1:50, mouse monoclonal antibody, Thermo Fisher), or incubated with PBS (as a negative control) on a fully automated Bond-Max system using onboard heat-induced antigen retrieval and a VBS Refine polymer detection system (Leica Microsystems).

### Statistical Analysis

Statistical analysis was performed using the SPSS 12.0 statistical software package for Windows (SPSS Inc., Chicago, IL, USA). All data are presented as the mean ± SEM. Independent experiments were performed to evaluate significant differences between the control and other experimental groups. Significant differences were determined using one-way analysis of variance (ANOVA) with post-hoc Tukey test or Student's t-tests, where appropriate. Significant differences were defined as p < 0.05.

## RESULTS

### HG Induced Endothelial IRAK-1 Expression and Inflammatory Phenotypes

We first determined the effects of HG on endothelial IRAK-1 expression. After 24 and 48 h stimulation, HG caused significant (1.12- and 1.29-fold) increases in IRAK-1 gene expression in HEACs, compared to the unstimulated control (**Figure 1A**). The expression of IRAK-1 protein also displayed a time-dependent increase, with 1.67- and 1.96-fold increases after 24 and 48 h of HG stimulation, respectively (**Figure 1B**). The osmotic control experiments showed that mannitol treatment did not modulate the expression levels of IRAK-1 (Supplemental Data 3). The adhesion of monocytes to the inflamed endothelium is a hallmark of the initiation of atherosclerotic plaques (Tuttolomondo et al., 2012). Vascular cell adhesion protein 1 (VCAM-1) and intercellular adhesion molecule 1 (ICAM-1) are essential molecules for this adhesive process. As shown in **Figure 1C**, HG stimulation for 24 and 48 h caused 1.48- and 1.88-fold increases of VCAM-1 gene expression, respectively. Similarly, HG caused 1.34- and 1.77-fold increases in ICAM-1 gene expression (**Figure 1D**). The increases of VCAM-1 and ICAM-1 expression levels were associated with increased adhesion of THP-1 monocytic cells to HAECs, as HG stimulation for 24 and

### FIGURE 1 | Continued

control (ANOVA). (B) Stimulation of HAECs with HG for 24 and 48 h induced 1.67- and 1.96-fold increases, respectively, in IRAK-1 protein expression. n = 5, \*p < 0.05, compared with the control (ANOVA). (C) Stimulation of HAECs with HG for 24 and 48 h induced 1.48- and 1.88-fold increases, respectively, in VCAM-1 gene expression levels. n = 5. \*\*\*p < 0.001, compared with the control (ANOVA). (D) Stimulation of HAECs with HG for 24 and 48 h induced 1.33- and 1.77-fold increases, respectively, in ICAM-1 gene expression levels. n = 5. \*p < 0.05 and \*\*\*p < 0.001, compared with the control (ANOVA). (E) Stimulation of HAECs with HG for 24 and 48 h induced 1.50- and 2.86-fold increases, respectively, in THP-1 adhesion to HAECs. n = 6. \*p <0.05 and \*\*p < 0.01, compared with the control (ANOVA).

48 h caused 1.50- and 2.86-fold increases in THP-1 adhesion to HAECs, respectively (**Figure 1E**).

### MiR-146a-5p, miR-339-5p, and miR-874-3p were Downregulated in HG-Treated HAECs

To determine potential binding partners of the 3′ -UTR of human IRAK-1 mRNA, in silico analyses using miRanda-mirSVR (http://www.microrna.org) were performed. This online database incorporated a mirSVR scoring system to improve predictions of the effects that a miR may have on gene expression (Betel et al., 2010). A total of 13 miRs were predicted to possess homology with the 3′ -UTR of the human IRAK-1 mRNA (Supplemental Data 4). The TaqMan <sup>R</sup> Array Human MicroRNA Card contained all 13 possible miRs predicted to target IRAK-1, and 24 h HG-stimulation caused the downregulation of seven endothelial miRs: miR-146a-5p, miR-339-5p, miR-874-3p, miR-125-3p, miR-431-5p, miR-192-5p, and miR-215-5p (**Figure 2A**). Real-time PCR analyses of 24 h HG-stimulated HAECs showed that only miR-146a-5p, miR-339-5p, miR-874-3p, and miR-125- 3p expression were significantly downregulated compared to the unstimulated control (**Figure 2B**). These four miRs were further selected for 48 h HG stimulation experiments. When HAECs were stimulated with HG for 48 h, the expression levels of miR-146a-5p, miR-339-5p, and miR-874-3p were significantly downregulated compared to the unstimulated control (**Figure 2C**).

### MiR-146a-5p Mimic Inhibited HG-Induced Endothelial IRAK-1 Expression and Inflammatory Phenotypes

The miR target analysis showed homologies between the 3′ - UTR of the human IRAK-1 mRNA and miR-146a-5p (two binding sites), miR-339-5p, and miR-874-3p, indicating potential regulation of IRAK-1 (**Figure 3A**). We performed a miR mimic competitive transfection assay to determine whether mimics of these three miRs could decrease IRAK-1 gene expression. As shown in **Figure 3B**, only miR-146a-5p transfection significantly decreased IRAK-1 gene expression in the 48 h HG-stimulated HAECs. The osmotic control experiments showed that mannitol treatment did not modulate the expression levels of miR-146a-5p (Supplemental Data 3). To investigate whether miR-146a-5p can interact with the IRAK-1 mRNA 3′ -UTR, a luciferase reporter assay was performed. As shown in **Figure 3C**, cotransfection of pGL3-IRAK-1-3′ -UTR and the miR-146a-5p mimic resulted in

24 h revealed that seven miRs, miR-146a-5p, miR-339-5p, miR-874-3p, miR-125-3p, miR-431-5p, miR-192-5p, and miR-215-5p, were downregulated by HG, as compared with unstimulated control. The 13 miR expression levels are expressed as log Relative Quantity (HG/control). N.D., not detected. (B) Seven downregulated miRs found via TaqMan® Array were confirmed by real-time PCR. HG stimulation for 24 h decreased miR-146a-5p, miR-339-5p, miR-874-3p, and miR-125-3p expression levels to 18, 20, 28, and 54% of the control level, respectively. n = 3. \*p < 0.05, \*\*p <0.01, and \*\*\*p < 0.001, compared with the control (t-test). The levels of the miRs are expressed as the ratios of U6 levels. (C) Four downregulated miRs, as shown by 24 h real-time PCR, were further examined in the 48 h experiment. HG stimulation for 48 h decreased miR-146a-5p, miR-339-5p, and miR-874-3p to 68, 60, and 49% of the control level, respectively. n = 3. \*p < 0.05 and \*\*p < 0.01, compared with the control (t-test). The levels of the miRs are expressed as the ratios of U6 levels.

a decrease in luciferase signal to 65% of that in the negative control, which confirmed direct binding of miR-146a-5p to the IRAK-1 3′ -UTR. Transfection of the miR-146a-5p mimic also significantly attenuated IRAK-1 protein expression in HGstimulated HAECs (**Figure 3D**). By contrast, the stimulatory effects of HG on IRAK-1 expression were potentiated by

FIGURE 3 | (A–G) The miR-146a-5p mimic inhibited HG-induced endothelial IRAK-1 expression and inflammatory phenotypes. (A) Bioinformatic miR target analysis identified homologies between miR-146a-5p (two binding sites), miR-339-5p, miR-874-3p, and the 3′ -UTR of human IRAK-1 mRNA, indicating potential regulation of IRAK-1 by these three miRs. (B) Three miR mimic competition assays determined that the stimulatory effect of HG on IRAK-1 mRNA expression was inhibited only in miR-146a-5p mimic-transfected HAECs. n = 4. \*\*\*p < 0.001, as compared with HG treatment transfected with a negative control (HG+NC) (ANOVA). (C) A luciferase reporter assay showed that the miR-146a-5p mimic could downregulate the relative luciferase activity of pGL3-IRAK-1 3′ -UTR. n = 5, \*p < 0.05, compared with the negative (Continued)

#### FIGURE 3 | Continued

control (NC) (ANOVA). (D) The stimulatory effect of HG on IRAK-1 protein expression was inhibited in miR-146a-5p mimic-transfected HAECs. n = 4. \*\*p < 0.01, compared with the NC (t-test). (E) The stimulatory effect of HG on IRAK gene expression was enhanced by the miR-146a-5p inhibitor. n = 3. \*\*\*p < 0.001, compared with the NC (t-test). (F) The stimulatory effects of HG on the gene expression of VCAM-1 and ICAM-1 was inhibited in miR-146a-5p mimic-transfected, HG-stimulated HAECs. n = 3. \*\*p < 0.01, as compared with the NC (t-test). (G) The stimulatory effect of HG on THP-1 adhesion to HAECs was inhibited in miR-146a-5p mimic-transfected, HG-stimulated HAECs. n = 3. \*\*\*p < 0.001, as compared with the NC (t-test).

a miR-146a-5p inhibitor (**Figure 3E**). To determine whether reduced IRAK-1 expression was associated with reduced endothelial inflammation in miR-146a-5p mimic-transfected, HG-stimulated HAECs, we measured VCAM-1 and ICAM-1 expression levels and THP-1 adhesion. As anticipated, both VCAM-1 and ICAM-1 levels were attenuated by transfection of the miR-146a mimic (**Figure 3F**). Furthermore, THP-1 adhesion to HAECs was also significantly reduced (**Figure 3G**).

### IRAK-1 siRNA Depletion Inhibited HG-Induced Endothelial Inflammation

To determine whether enhanced IRAK-1 expression was associated with HG-induced endothelial inflammation, an IRAK-1 siRNA transfection experiment was performed. After IRAK-1 siRNA transfection, both IRAK-1 mRNA (**Figure 4A**) and protein levels (**Figure 4B**) were significantly downregulated in HG-stimulated HAECs. Furthermore, similar to the effects of the miR-146a mimic, IRAK-1 siRNA significantly inhibited the expression of HG-stimulated VCAM-1/ICAM-1 gene expression (**Figure 4C**) and THP-1 adhesion to HAECs (**Figure 4D**), indicating that HG-induced endothelial inflammation was mediated partially through IRAK-1.

### The miR-146a-5p Mimic Decreased Endothelial IRAK-1 and ICAM-1 Expression in Type 2 Diabetic Mice

To examine the effect of the miR-146a-5p mimic on the expression of endothelial IRAK-1 and ICAM-1 in vivo, we performed IHC on aortic tissues from db/db type 2 diabetic mice. As shown in **Figure 5A**, aortic endothelial IRAK-1 and ICAM-1 protein levels from miR-146a-5p mimic-treated mice were dramatically decreased, as compared to negative control or vehicle-treated mice. This demonstrates that miR-146a-5p may have therapeutic potential, mitigating endothelial dysfunction through the downregulation of both IRAK-1 and ICAM-1. The proposed role of miR-146a-5p in regulating HG-induced endothelial inflammation via IRAK-1 is shown in **Figure 5B**.

### DISCUSSION

This study demonstrates that HG induced IRAK-1 expression and endothelial inflammation in HAECs via the downregulation of miR-146a-5p expression. Among multiple mechanisms responsible for HG-induced endothelial inflammation, the

IRAK-1-transfected HAECs. n = 3. \*p < 0.05, compared with the NC (t-test).

induction of IRAK-1 plays a proinflammatory role in HGstimulated HAECs, and represents an important mediator that maintains chronic inflammation in diabetic vascular diseases. Both in vitro and in vivo experiments identified miR-146a-5p as a target in treating diabetic vascular complications.

Among the different target prediction tools, the miRandamirSVR database possesses the advantageous properties of being easy to use, containing relatively up-to-date information, and possessing a large range of capacity (Peterson et al., 2014). Target genes can be predicted with miRanda by considering seed region-weighted algorithms, free energy analyses, and crossspecies sequence conservation (Enright et al., 2003). In addition, the pre-computed mirSVR scores are useful in representing the effects of a specific miR on gene expression (Betel et al., 2010). Generally, a higher absolute value of mirSVR score indicates greater downregulation at the mRNA or protein levels. In our study, although miR-339-5p and miR-874-3p had mirSVR scores that ranked them fourth and fifth among our 13 miRandapredicted miRs that would interact with the 3′ -UTR of IRAK-1 mRNA, in the transfection assays, these miRs did not regulate IRAK-1 expression. Interestingly, miR-146a-5p had the highest pre-computed mirSVR score among the 13 predicated miRs. Furthermore, the 3′ -UTR of IRAK-1 possessed two miR-146a-5p 7-mer seed region binding sites; these findings support the hypothesis that the interactions of the miR-146a-5p::IRAK-1 duplex are functional (Brennecke et al., 2005).

The role of IRAK-1 in diabetes is not clear. Recent studies have documented the involvement of TLR signaling in the metabolic aberrations of diabetes. In human microvascular endothelial cells incubated with HG, the expression levels of TLR4, MyD88, and IL-1β were increased (Wang et al., 2015), implying that both TLR-signaling and IL-1R-signaling are activated. In the

monocytes of type 1 (Devaraj et al., 2008) and type 2 (Dasu et al., 2010) diabetes patients, levels of TLR2, TLR4, and other TLRsignaling components (e.g., MyoD88 and NF-κB) were increased. Our data demonstrated that HG induced the expression of an essential TLR/IL-1R signaling component: IRAK-1. Importantly, the increase in IRAK-1 expression was involved in the regulation of downstream endothelial inflammatory phenotypes, as HGenhanced VCAM-1/ICAM-1 gene expression and monocyte adhesion was partially reduced after IRAK-1 depletion by siRNA.

In LPS-stimulated human monocytes, miR-146a-5p was identified as a negative regulator of the NF-κB pathway, targeting IRAK-1 and TRAF-6 expression (Taganov et al., 2006). In addition, several groups have reported that IL-1β, TNFα, IL-8, and ox-LDL could stimulate increased expression of miR-146a-5p in different cell types (Chen et al., 2011; Li et al., 2012; Cheng et al., 2013). The miR-146a-5p promoter contains two NF-κB binding sites; these were responsible for the LPS/IL-1β/TNF-α-stimulated expression of miR-146a-5p (Taganov et al., 2006). Although the activation of endothelial NF-κB by HG has been reported (Ho et al., 2006), HG did not induce endothelial miR-146a-5p expression. In our previous work, we found that in HAECs, a 5 h HG-stimulation could downregulate miR-146a-5p expression to 83% of that observed in the control (Wang, H. J. et al.,, 2014). We also reported that glycated albumin can downregulate endothelial miR-146a-5p expression (Wang et al., 2013). Different mouse tissues, including retina, heart, kidney, diabetic wound, and dorsal root ganglion neuron, also displayed decreased expression of miR-146a-5p, suggesting that diabetes-associated injuries, including those due to hyperglycemia, contributed to the prolonged reduction of miR-146a-5p expression observed in vivo (Feng et al., 2011, 2017; Xu et al., 2012; Wang, L. et al.,, 2014). In this study, we extended HG stimulation of HAECs for 24 and 48 h, and the results display sustained downregulation of miR-146a-5p, to 18 and 68% of the control levels, respectively. At 24 and 48 h, VCAM-1/ICAM-1 gene expression levels and THP-1 adhesion were enhanced, suggesting that the role of miR-146a-5p as an anti-inflammatory brake had been impaired.

Although the 3′ -UTRs of VCAM-1 and ICAM-1 mRNA are not predicted to bind to miR-146a-5p, we observed that HGinduced VCAM-1 and ICAM-1 expression levels and THP-1 adhesion were effectively inhibited by the miR-146a-5p mimic. These results suggest that the expression of VCAM-1 and ICAM-1 is indirectly regulated by miR-146a-5p in HG-stimulated HAECs. Both the VCAM-1 and ICAM-1 promoters require NF-κB for maximal levels of induction (Collins et al., 1995); therefore, the inhibitory effects of the miR-146a-5p mimic were most likely mediated through NF-κB. Apart from the indirect regulation of inflammatory molecules through the NF-κB pathway, miR-146a-5p has multiple direct targets that modulate different inflammatory pathways. At the receptor level, TLR-4 is a direct miR-146a-5p target in oxidized lowdensity lipoprotein-stimulated macrophages (Yang et al., 2011). Downstream from TLR-4, both IRAK-1 and TRAF-6 are known miR-146a-5p targets that dampen LPS-induced inflammation in monocytes (Taganov et al., 2006). NADPH oxidase 4 (NOX4) is an important mediator responsible for diabetic complications, and our previous work revealed that miR-146a-5p was a direct regulator of NOX4 in HG/thrombin-stimulated HAECs (Wang, H. J. et al.,, 2014). In gastric cancer, in addition to its role in TLR/IL-1R signaling, miR-146a-5p has been reported to directly control the G-protein-coupled receptor-mediated activation of NF-κB, via caspase recruitment domain-containing protein 10 (CARD10) and COP9 signalosome complex subunit 8 (COPS8) (Crone et al., 2012). Overall, multiple inflammatory mediators are regulated by miR-146a-5p, emphasizing the evolutionary effectiveness of restraining excessive inflammation via a single mediator.

### AUTHOR CONTRIBUTIONS

HW, CP, and WL conceived the project. HW and WL wrote the manuscript. HW, CP, and WL provided funding. HW and

### REFERENCES


WL performed critical experiments. WL and HW supervised the study.

### FUNDING

This study was supported by a grant from China Medical University Hospital (DMR-105-012).

### ACKNOWLEDGMENTS

The authors thank Jen-Duo Liou for his technical assistance. The animal study was supported by Laboratory Animal Center of Hungkuang University, Taiwan, R.O.C.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fphys. 2017.00551/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 Lo, Peng and Wang. 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.

# Prognostic Values of Long Noncoding RNA GAS5 in Various Carcinomas: An Updated Systematic Review and Meta-Analysis

Qunjun Gao1, 2†, Haibiao Xie1, 3†, Hengji Zhan1, 4†, Jianfa Li 1†, Yuchen Liu<sup>1</sup> \* and Weiren Huang<sup>1</sup> \*

<sup>1</sup> Key Laboratory of Medical Reprogramming Technology, Shenzhen Second People's Hospital, Graduate School of Guangzhou Medical University, Shenzhen, China, <sup>2</sup> Guangzhou Medical University, Guangzhou, China, <sup>3</sup> Shantou University Medical College, Shantou, China, <sup>4</sup> Shenzhen University, Shenzhen, China

#### Edited by:

Carlos Rosales, National Autonomous University of Mexico, Mexico

#### Reviewed by:

Antonio Longo, Università degli Studi di Catania, Italy Neha Nagpal, Independent Researcher, Germany

#### \*Correspondence:

Yuchen Liu liuyuchenmdcg@163.com Weiren Huang pony8980@163.com

† These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 19 July 2017 Accepted: 03 October 2017 Published: 02 November 2017

#### Citation:

Gao Q, Xie H, Zhan H, Li J, Liu Y and Huang W (2017) Prognostic Values of Long Noncoding RNA GAS5 in Various Carcinomas: An Updated Systematic Review and Meta-Analysis. Front. Physiol. 8:814. doi: 10.3389/fphys.2017.00814 The growth arrest-specific transcript 5 (GAS5) is a long noncoding RNA with low expression in multiple cancers. This meta-analysis aims to explore the association between GAS5 expression levels and cancer patients' prognosis. We collected all the relevant literatures about GAS5 expression levels associated with overall survival (OS), lymph node metastasis (LNM) and high tumor stage (II/III/IV) (HTS) from the PubMed and Web of Science. The hazard ratio (HR) and the corresponding 95% confidence interval (CI) were calculated to evaluate the link strength between GAS5 and cancer prognosis. A total of 934 patients from 14 studies were included to the present meta-analysis, according to the inclusion and exclusion criteria. The results demonstrated that low expression of GAS5 could predict poor OS in cancer patients (HR = 1.955, 95% CI: 1.551–2.465, P < 0.001). Meanwhile we also analyzed the following cancers independently: hepatocellular carcinoma (HR = 1.893, 95% CI: 1.103–3.249, P = 0.021) and urothelial carcinoma (HR = 1.653, 95% CI: 1.185–2.306, P = 0.003). Compared to the high GAS5 expression group, additionally, patients with low GAS5 expression in tumor tissues were more prone to lymph node metastasis (OR = 0.234, 95%CI: 0.153–0.358, P < 0.001) and high tumor stage (OR = 0.185, 95% CI:0.102–0.333, P < 0.001). In conclusion, this meta-analysis showed that GAS5 might be served as a novel biomarker for predicting prognosis in various types of cancers.

Keywords: lncRNA, GAS5, cancer, prognosis, lymph node metastasis, high tumor stage, meta-analysis

### INTRODUCTION

Cancer has become a global health problem. In recent years, the incidence of cancer has been increased year by year. According to WHO estimates, 14.1 million new cancer patients and 8.2 million deaths from cancer occurred worldwide in 2012 and more than 20 million new cases of cancer will be expected as early as 2025 (Ferlay et al., 2015). At present, cancer treatment includes surgery,radiotherapy, chemotherapy andetc., but the 5 years survival rate is still not ideal, especially some patients with lymph node metastasis or high stage tumor (Saika and Sobue, 2013). Therefore, it is important to find a new biological target that plays a guiding role in the carcinogenesis to detect cancer. It is also more conducive to early detection, early diagnosis and early treatment of tumor patients.

Long non-coding RNAs (lncRNAs) are noncoding RNAs with a length of more than 200 nucleotides that regulate gene expression (Mattick and Makunin, 2006). They were described as "noise," and did not attract much attention in the past few decades (Ponjavic et al., 2007). With the application of whole genome sequencing and microarray, lncRNAs have attracted more and more attentions (Batista and Chang, 2013; Tang et al., 2013). The increasing evidence show that lncRNAs play a pivotal role in the development and progression of tumors, which means that they can be used as biomarkers for some tumors (Fang et al., 2017; Liu et al., 2017; Sun et al., 2017; Yang et al., 2017). However, only a few number of lncRNAs have corresponding functional features, and most of the functions of lncRNAs remain unclear.

The growth arrest-specific transcript 5 (GAS5) is a rising star among tumor-suppressive lncRNAs among all the kinds of lncRNAs (Ma et al., 2016). Recent studies have shown that GAS5 plays a key role in a variety of human diseases and participates a variety of biological processes, such as cell proliferation, cell apoptosis, epithelial-mesenchymal transition and etc. (Tan et al., 2017; Tao et al., 2017; Wen et al., 2017; Yang et al., 2017). Meanwhile, GAS5 is also involved in the progression of many types of cancer, such as bladder cancer (BC) (Zhang et al., 2017), colorectal cancer (CRC) (Yin et al., 2014; Li et al., 2017), nonsmall cell lung cancer (NSCLC) (Shi et al., 2015; Wu et al., 2016), breast cancer (BRC) (Li W. et al., 2016), hepatocellular carcinoma (HCC) (Tu et al., 2014; Chang et al., 2016; Hu et al., 2016), epithelial ovarian cancer (EOC) (Gao et al., 2015), gastric cancer (GC) (Sun et al., 2014), cervical cancer (CEC) (Cao et al., 2014), and head and neck squamous cell carcinoma (HNSCC) (Gee et al., 2011). The clinic pathological features of the patients, such as overall survival (OS), lymph node metastasis (LNM) and high tumor stage (II/III/IV) (HTS), are also highly correlated with the level of GAS5 expression in these cancers (Gee et al., 2011; Cao et al., 2014; Sun et al., 2014; Tu et al., 2014; Yin et al., 2014; Gao et al., 2015; Shi et al., 2015; Chang et al., 2016; Hu et al., 2016; Li J. et al., 2016; Wu et al., 2016; Droop et al., 2017; Li et al., 2017; Zhang et al., 2017). All these indicate that GAS5 can be a novel prognostic biomarker in unique cancer. To shed light on the relationship between GAS5 and cancer prognosis, the metaanalysis on the association between the expression of GAS5 and the prognosis of cancer is required. Although a meta-analysis has reported that the expression of GAS5 predicts poorer survival outcomes, only 4 literatures have been included in that work and the results may be incidental (Song et al., 2016). To verify the accuracy of the previous results, the present meta-analysis with 14 studies may provide a more accurate conclusion.

### MATERIALS AND METHODS

### Literature Collection

We searched potentially eligible literatures through PubMed, Web of Science to locate articles (published during March 2011 to April 2017), including articles referenced in the publications. We used "GAS5 or growth arrest specific 5" AND "cancer or tumor or carcinomas or neoplasm" as the keywords, in order to identify potentially relevant studies. Citation lists of retrieved articles were searched manually to ensure sensitivity of the search strategy.

### Inclusion and Exclusion Criteria

All the eligible study data elements were independently assessed and extracted by two investigators. For inclusion in this metaanalysis, the studies met the following criteria: the association between GAS5 and cancer prognosis (OS) was investigated; patients were grouped according to the expression levels of GAS5;related clinic pathologic parameters were described, such as LNM, TNM and sufficient original data for calculating a hazard ratio (HR) with its 95% confidence interval (CI). Exclusion criteria are as the following: Duplicate publications; irrelevant to cancer, GAS5, or cancer prognosis; animal studies, letters, editorials, expert opinions, abstracts, case reports and reviews; studies without usable data.

### Data Extraction

According to the inclusion and exclusion criteria, two investigators extracted and reviewed the data independently (GQJ, XHB), and disagreements were discussed with two investigators (ZHJ, LYC) in conference. The following data were extracted: first author, publication date, country of origin, tumor type, total number of patients, number of high GAS5 expression group and low GAS5 expression group, number of patients with LNM, number of patients with HTS, detection method of GAS5 expression levels, follow-up month and cut-off values, multivariate analysis, hazard ratios (HRs), and corresponding 95% CI for OS.

### Statistical Methods

Meta-analysis was performed using Stata12.0 software. Pooled hazard ratios (HRs) were extracted from the included studies; the log HR and standard error (SE) were used for aggregation of the survival results (Tierney et al., 2007). To determine the heterogeneity among the included studies, chi-square-based Q test and I<sup>2</sup> statistics were used (Higgins et al., 2003). If the P < 0.1 or I <sup>2</sup> > 50%, it means that significant heterogeneity existed among the included studies, thus the random-effects model was adopted to analyze the results. The fixed-effects model was applied when between-study heterogeneity was absent (P > 0.1 and I<sup>2</sup> < 50%). The potential publication bias was assessed using the Eegg'stest and P < 0.05 was considered representative of statistically significant publication bias. Sensitivity analysis was performed by sequential omission of each individual study in order to validate the stability of outcomes in the present meta-analysis.

### Quality Assessment of Primary Studies

Two investigators (GQJ, XHB) performed the quality assessment of primary studies independently. We evaluated all eligible studies' quality by using the Newcastle-Ottawa Scale (NOS) for assessing the quality of studies in meta-analyses (Zeng et al., 2015). The higher scores indicated better methodological quality.

### RESULTS

### Characteristics and Eligible Studies

The initial search of the electronic database retrieved 137 literatures. After removing the duplicates, 104 articles were remained. Then we carefully screened the title and abstract, 25 literatures were excluded because the studies were irrelevant. Upon further review of the full articles, the articles with no survival outcomes, lymph node metastasis, TNM stage, animal testing and other factors were excluded. 14 articles were eventually selected for the present meta-analysis (**Figure 1**). A total of 934 patients were included among these studies, with a maximum sample size of 106 and a minimum sample size of 24 patients (Mean 67). The publication years of the included studies were between 2011 and 2017. In these studies, one was from UK, one was from Germany and 8 were from China. A total of 9 different types of cancer were evaluated in studies of this meta-analysis (3 hepatocellular carcinoma, 2 colorectal cancer, 2 non-small cell lung cancer, 2 urothelial carcinoma, 1 breast cancer, 1 epithelial ovarian cancer, 1 gastric cancer, 1 cervical cancer and 1 head and neck squamous cell carcinoma). The expression of GAS5 was detected by qRT-PCR and the cut-off values included in the studies were inconsistent. All diagnoses of LNM and TNM were based on pathology. Hazard ratios with the corresponding 95% CIs were extracted from the graphical survival plots and the articles. The main characteristics of the eligible studies were summarized in **Table 1**. The Newcastle-Ottawa Scale (NOS) confirmed that all the studies were of high quality (**Table 2**).

### Meta-Analysis Result

### Association between GAS5 and OS in Seven Types of Cancers

Among the included studies, 10 reported the overall survival (OS) of 730 patients according to GAS5 expression levels. In order to study the relationship between GAS5 expression level and prognosis, the fixed-effect model was used to calculate the pooled HR with corresponding 95% CI because heterogeneity analysis showed that low between-study heterogeneity among those nine studies for GAS5 expression was found (I<sup>2</sup> = 0.0%, P(H) = 0.728). We found an inverse relationship that low expression of GAS5 might be associated with poor overall survival outcome (HR = 1.955, 95% CI:1.551–2.465, P < 0.001, fixed-effect model) (**Figure 2**). In a subgroup analysis of cancer sites, we also found the similar significant adverse association between levels of GAS5 and OS in the following cancers (low/high): HCC (HR = 1.893, 95% CI: 1.103–3.249, P = 0.021, P(H) = 0.902), UC (HR = 1.653, 95% CI: 1.185–2.306, P = 0.003, P(H) = 0.268) and HR for the subgroup of other cancers was 2.641 (95%CI: 1.625–4.204, P < 0.001, P(H) = 0.730). We didn't perform subgroup analyses for CRC, BRC, EOC, GC, CEC, and HNSCC, because there is only one paper investigating these associations between GAS5 and OS (**Figure 2**) in each cancer type. Compared with the high


 HNSCC, head and neck squamous cell carcinoma; UK, United Kingdom of Great Britain and Northern Ireland; HTS, high tumor stage(II/III/IV); LNM, lymph node metastasis; DM, distant metastasis; qRT-PCR, quantitative real-timepolymerase chain reaction; OS, overall survival; NA, not available; Rep, reported; SC, survival curve; L/H, low expression of GAS5/high expression of GAS5.



expression group, the low GAS5 expression group indicates a poorer OS which was confirmed statistically significant.

### Association between GAS5 and LNM

Based on the differential expression levels of GAS5, seven studies reported 443 patients with lymph node metastasis. Because of the significant between-study heterogeneity (I<sup>2</sup> = 59.6%, p = 0.021), the random-effects model was adopted to calculate the odds ratio (high GAS5 expression group vs. low GAS5 expression group; OR = 0.234, 95% CI: 0.153–0.358, P < 0.001). It demonstrated that patients with low GAS5 expression in tumor tissues were more prone to lymph node metastasis (**Figure 3**). In a subgroup analysis of cancer sites, we found the similar outcomes in CRC (OR = 0.353, 95% CI: 0.151–0.831, P = 0.017). OR for the subgroup of other cancers was 0.115 (95% CI: 0.06–0.221, P < 0.001). But the expression of GAS5 in NSCLC tumor tissues might not be a direct evidence of LNM (OR = 0.516, 95% CI: 0.229–1.164, P = 0.111). We didn't perform subgroup analyses for UC, BRC, HCC, EOC, GC, CEC, and HNSCC, there is only one paper investigating these associations between GAS5 and LNM (**Figure 3**) in each cancer type.

### Association between GAS5 and HTS

Five studies reported the HTS of 335 patients based on variousGAS5 expression levels. The fixed-effect model was adopted because there was no heterogeneity (I<sup>2</sup> = 0.0%, p = 0.691). The odds ratio, expressed as high GAS5 expression group vs. low GAS5 expression group, was 0.185 (95% CI: 0.102–0.333, P < 0.001). The result showed that patients with low GAS5 expression in cancerous tissues were more prone to high tumor stage (**Figure 4**). All the results were listed in the **Table 3**.

### Sensitivity Analysis and Publication Bias

To test the stability of the results of GAS5 and OS, we performed sensitivity analyses by sequentially removing each eligible study and the result was not significantly affected (**Figure 5**). We also performed a sensitivity analysis of lymph node metastasis and GAS5, and got similar results (**Figure 5**). We used Eegg's test to evaluate potential publication biases of the GAS5 and OS, and the result did not display obvious publication bias for the HR evaluations of OS (p = 0.996) (**Figure 6**).

### DISCUSSION

GAS5 has been reported to be down-regulated in multiple cancers, leading to changes in tumor cell production, proliferation, apoptosis, metastasis, and survival time (Chang et al., 2016; Hu et al., 2016; Zhang et al., 2017). In our metaanalysis, we explored the relationship between the level of GAS5 expression and cancer prognostic parameters. The results demonstrated that low expression levels of GAS5 predicted poor OS in various cancers and patients with low GAS5 expression in tumor tissues were more prone to lymph node metastasis.

FIGURE 3 | Meta-analysis of the LNM of different types of cancer with the level of GAS5 expression. (A) Forest plot for the correlation between GAS5 expression levels and LNM in different cancer patients. (B) Subgroup analysis of lymph node metastasis by factor of different types of cancer.

Meanwhile, we found patients with low GAS5 expression in cancerous tissues were more prone to high tumor stage. Our results showed that low expression levels of GAS5 could be a molecular biomarker of poor prognosis in cancer patients.

As shown in **Figure 2**, GAS5 and OS are positively related in OS analysis without heterogeneity and publication bias: the low expression of GAS5 predicts poorer survival outcomes. To investigate whether the above analysis was applicable in separate cancers, we made a subgroup analysis. The results was HCC (HR = 1.893, 95% CI: 1.103–3.249, P = 0.021, P(H) = 0.902), UC (HR = 1.653, 95% CI: 1.185–2.306, P = 0.003, P(H) = 0.268) which meant that the above conclusions applied equally in HCC and UC. Meanwhile, we investigated the association between the GAS5 expression levels and LNM and HTS, and we found that low GAS5 expression in cancerous tissues were


OS, overall survival; LNM, lymph node metastasis; HTS, high tumor stage (II/III/IV); HCC, hepatocellular carcinoma; UC, urothelial carcinoma; CRC, colorectalcancer; NSCLC, nonsmall cell lung cancer; others, other cancer types; HR, hazard ratios; OR, odds ratios; No, number; CI, confidence interval.

more prone to LNM and HTS (**Figures 3**, **4**). However, in LNM analysis we found that the included studies existed significantly heterogeneity. So we performed a subgroup analysis according to tumor type, and the results showed that the heterogeneity disappeared obviously in CRC(P for heterogeneity = 0.211, I<sup>2</sup> = 36.1%, random-effects model), and other types of cancer(P for heterogeneity = 0.516, I<sup>2</sup> = 0.0%, random-effects model), while the heterogeneity still existed in NSCLC(P for heterogeneity = 0.107, I<sup>2</sup> = 61.6%, random-effects model) which might be caused by the different cut-off value methods which were adopted to define the high GAS5 expression group or low GAS5 expression group. In conclusion, all these results provided strong evidence for GAS5 as a potential biomarker for the prognosis of various cancers.

Nowadays, many lncRNAs have been found to be abnormally expressed in cancer. Therefore, many meta-analysis articles, like our study, have been used to reveal the correlation of lncRNAs and cancer prognosis. Several lines of studies, meanwhile, have revealed that a lot of lncRNAs play a important role in cancer prognosis, such as TUG1, SPRY4, MALAT1 (Wang et al., 2015, 2017; Yu et al., 2017). For instance, Wang et al. found that SPRY4 is remarkably upregulated in various cancer. Thus, they performed the meta-analysis to examine the association between the SPRY4-IT1 expression level and prognosis in cancer patients. Finally, they suggested the prognostic role of SPRY4-IT1 in human cancers, and increased SPRY4-IT1 expression was closely associated with advanced features of human cancers (Wang et al., 2017). Likewise, NEAT1, as a novel lncRNA, has been recently found to be up-regulated in several cancers, contributing to tumor proliferation, apoptosis, metastasis and survival. Chen et al. conduct a meta-analysis to clarify the association between high NEAT1 expression and poor prognosis. Eventually, they concluded that NEAT1 may serve as a molecular marker and a prognostic factor for patients with various cancers (Chen et al., 2017). Additionally, among these studies, it can be found that

(B) Sensitivity analysis of effect of individual studies on ORs for GAS5 and lymph node metastasis of patients.



NA, not available.

different lncRNA has specific signaling pathways in cancers. They move the extracellular signaling molecules into the cell and then, in some way, further affect cell phenotypic changes, such as cell metabolism, proliferation, invasion, apoptosis, and so on (Wang et al., 2015, 2017; Yu et al., 2017). To further investigate the value of GAS5, we analyzed and screened the signaling pathways and mechanisms of action from all GAS5 related literatures, which will be useful for future studies on tumorigenesis (**Table 4**).

There are several limitations in our study that should be acknowledged. Firstly, the present study used the summary data rather than a specific patient data. Secondly, the methods for distinguishing the cut-off value of GAS5 in high and low expression groups were inconsonant which inevitably could cause heterogeneity. Thirdly, most of the HR values were not directly reported in these included studies. We extracted and calculated them according to the survival curves, so inevitably there might be errors. Fourthly, different treatment methods for different types of cancer patients after surgery might have great influence on the survival time, which leaded to the heterogeneity of the researches. Fifthly, we only included English related literatures that could not be so comprehensive. Sixthly, most of the studies were from China, so the conclusion might not necessarily apply in other areas. Seventhly, we only included related studies reporting OS, LNM and HTS, and the articles on other prognostic indicators were thus excluded. In the light of the above deficiencies, a more comprehensive study covering larger samples, more regions, and more indicators will be needed to confirm our results.

In conclusion, our meta-analysis found that lncRNA GAS5 could sever as a molecular biomarker to predict the prognosis of

### REFERENCES


various cancers and the low GAS5 expression could indicate the poor prognosis.

### AUTHOR CONTRIBUTIONS

QG and HX performed Data extraction, HZ and JL did the data analysis. YL and WH designed the project and QG wrote the paper. YL supervised the project. WH provided financial support for the project.

### ACKNOWLEDGMENTS

We are very grateful to all the colleagues who had contributed to the work and were not listed on the author list. This work was supported by the National Key Basic Research Program of China (973 Program) (2014CB745201), the Chinese High-Tech (863) Program (2014AA020607), National Natural Science Foundation of China (81402103), International S&T Cooperation program of China (ISTCP) (2014DFA31050), The National Science Foundation Projects of Guangdong Province (2014A030313717), the Shenzhen Municipal Government of China (ZDSYS201504301722174, JCYJ20150330102720130, GJHZ20150316154912494), and Special Support Funds of Shenzhen for Introduced High-Level Medical Team.

J. Physiol. Gastrointest. Liver Physiol. 313:ajpgi.00426.02016. doi: 10.1152/ajpgi. 00426.2016


colorectal cancer by inhibiting interleukin-10 and vascular endothelial growth factor expression. Oncotarget 8, 13690–13702 doi: 10.18632/oncotarget.14625


**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 Gao, Xie, Zhan, Li, Liu 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.

# Integrative Computational Network Analysis Reveals Site-Specific Mediators of Inflammation in Alzheimer's Disease

Srikanth Ravichandran<sup>1</sup> , Alessandro Michelucci 2,3 and Antonio del Sol 1,4 \*

<sup>1</sup> Computational Biology Group, Luxembourg Centre for Systems Biomedicine, University of Luxembourg, Luxembourg, Luxembourg, <sup>2</sup> NORLUX Neuro-Oncology Laboratory, Department of Oncology, Luxembourg Institute of Health, Luxembourg, Luxembourg, <sup>3</sup> Luxembourg Centre for Systems Biomedicine, University of Luxembourg, Luxembourg, Luxembourg, <sup>4</sup> Moscow Institute of Physics and Technology, Dolgoprudny, Russia

Alzheimer's disease (AD) is a major neurodegenerative disease and is one of the most common cause of dementia in older adults. Among several factors, neuroinflammation is known to play a critical role in the pathogenesis of chronic neurodegenerative diseases. In particular, studies of brains affected by AD show a clear involvement of several inflammatory pathways. Furthermore, depending on the brain regions affected by the disease, the nature and the effect of inflammation can vary. Here, in order to shed more light on distinct and common features of inflammation in different brain regions affected by AD, we employed a computational approach to analyze gene expression data of six site-specific neuronal populations from AD patients. Our network based computational approach is driven by the concept that a sustained inflammatory environment could result in neurotoxicity leading to the disease. Thus, our method aims to infer intracellular signaling pathways/networks that are likely to be constantly activated or inhibited due to persistent inflammatory conditions. The computational analysis identified several inflammatory mediators, such as tumor necrosis factor alpha (TNF-a)-associated pathway, as key upstream receptors/ligands that are likely to transmit sustained inflammatory signals. Further, the analysis revealed that several inflammatory mediators were mainly region specific with few commonalities across different brain regions. Taken together, our results show that our integrative approach aids identification of inflammation-related signaling pathways that could be responsible for the onset or the progression of AD and can be applied to study other neurodegenerative diseases. Furthermore, such computational approaches can enable the translation of clinical omics data toward the development of novel therapeutic strategies for neurodegenerative diseases.

Keywords: neuroinflammation, Integrative approach, computational modeling, signaling network, sustained inflammatory response

## INTRODUCTION

Alzheimer's disease (AD) is one of the most prevalent chronic neurodegenerative disease and is responsible for 60–70% of cases of dementia, thus laying important healthcare problems in countries with aging populations (Burns and Iliffe, 2009). Although the precise cause of the disease is not yet understood, several biochemical and neuropathological studies of brains from individuals

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

#### Reviewed by:

Jesús Espinal-Enríquez, National Institute of Genomic Medicine, Mexico Beth J. Allison, Hudson Institute of Medical Research, Australia

> \*Correspondence: Antonio del Sol antonio.delsol@uni.lu

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 17 October 2017 Accepted: 14 February 2018 Published: 02 March 2018

#### Citation:

Ravichandran S, Michelucci A and del Sol A (2018) Integrative Computational Network Analysis Reveals Site-Specific Mediators of Inflammation in Alzheimer's Disease. Front. Physiol. 9:154. doi: 10.3389/fphys.2018.00154

**197**

with AD provide clear evidences for the involvement of inflammatory pathways (Wyss-Coray and Rogers, 2012). The neurodegenerative processes during the course of the disease essentially render neurons unable to fulfill essential functions, such as signal transmission and network integration in the central nervous system (CNS), thus affecting essential daily activities, such as thinking and moving (Burns and Iliffe, 2009). Importantly, local CNS environment contributes to neurodegeneration and supportive cells, such as glia and endothelial cells, are responsible to maintain an ideal surrounding for neuronal functions (Glass et al., 2010). Several accumulating evidences suggest that neurodegeneration occurs in part because the environment is affected during the disease in a cascade of processes collectively termed neuroinflammation (Morales et al., 2014; Ransohoff, 2016). Sustained or chronic inflammation resulting in neuronal death implies persistence of an inflammatory stimulus or a failure in normal resolution mechanisms. A persistent stimulus may result from environmental factors or due to the formation of endogenous factors, such as protein aggregates, that are perceived by resident immune cells, e.g., microglia, as "stranger" or "danger" signals (Glass et al., 2010). Inflammatory responses that induce autocrine and paracrine neuronal feed-forward/feedback loops as well as influence the neuronal crosstalk with microglia and other CNS cell types may hinder normal resolution mechanisms (Glass et al., 2010; Morales et al., 2014). Although certain inflammatory stimuli are associated to beneficial effects, such as phagocytosis of debris and apoptotic cells where inflammation is linked to beneficial tissue repair processes, uncontrolled and sustained inflammation may result in the production of neurotoxic factors that amplify the underlying disease state (Glass et al., 2010).

The pathological hallmarks of AD in the brain include extracellular amyloid plaques comprising aggregated, cleaved products of the amyloid precursor protein (APP) and intracellular neurofibrillary tangles (NFTs) resulting from hyperphosphorylation of the microtubule-binding protein tau (O'Brien and Wong, 2011). Evidence of an inflammatory response in AD includes changes in microglia morphology from ramified (resting) to amoeboid (active)—and astrogliosis (manifested by an increase in the number, size, and motility of astrocytes) surrounding the senile plaques (Akiyama et al., 2000; Liang et al., 2008). Elevation of inflammatory factors in culture and animal models are known to typically result in neurodegeneration, and have been reported to be elevated in pathologically vulnerable regions of the AD brain (Wyss-Coray and Rogers, 2012). Several existing genetic, cellular, and molecular changes associated with AD provide clear support for the role of immune and inflammatory processes in the disease (Wyss-Coray and Rogers, 2012).

Omics technologies such as, transcriptomics and proteomics, have enabled the identification of key factors that exhibit differential expression patterns in disease conditions compared to homeostatic states (Dendrou et al., 2016). The related experimental datasets contain rich source of molecular profiles under different disease conditions. In particular, clinical data from patients and age-matched controls offer a wealth of information to be analyzed in order to get insights into the role of specific deregulated processes involved in the disease onset and progression (Dendrou et al., 2016). Due to the recent technological advances, it is now possible to generate and analyze high-throughput data from numerous individuals, even down to the level of single cells (Glass et al., 2010; Ransohoff, 2016). However, the enormous complexity of the molecular and cellular pathways involved in neuroinflammation necessitates parallel implementation of computational analyses to investigate pathophysiological mechanisms across different neurological disorders (Dendrou et al., 2016; Hasin et al., 2017). Means to translate wealth of clinical omics information into practical medical benefit is, however, a fundamental challenge that requires the development and application of novel computational methods for data analyses and interpretation.

In the context of AD, a significant amount of high-throughput omics data can be used to further understand the different deregulated processes that contribute to the disease (Hasin et al., 2017). Importantly, it has been observed that upregulated inflammatory mechanisms co-localize in the AD brain with those regions that exhibit high levels of AD pathology (e.g., frontal neocortex, limbic cortex) and are absent or minimal in brain regions with low AD pathologic susceptibility (e.g., cerebellum) (Akiyama et al., 2000). Furthermore, major efforts have been directed toward the identification of specific molecular players of inflammation and their contribution to the disease, but lack an integrative perspective of common and specific features of inflammation across different brain regions in AD (Morales et al., 2014). For this, datasets obtained by tissue or region-specific molecular profiling of AD post-mortem brain serves as a rich resource to study and analyze the involvement of inflammation in different brain regions affected by AD. In this study, in order to shed more light on neuronal inflammatory signaling pathways associated with AD, we analyzed publicly available gene expression data of six different neuronal populations isolated from post-mortem brains of AD patients. Specifically, the objective of the current study was to identify region-specific signaling pathways likely to be involved in inflammation and how these pathways are distributed across different regions of the brain affected by AD.

### MATERIALS AND METHODS

### Rationale

Sustained inflammation resulting in tissue pathology can imply persistence of an inflammatory stimulus due to failure in normal clearance mechanisms. A persistent inflammatory stimulus may result from environmental factors or due to the formation of endogenous factors, (e.g., protein aggregates) that eventually cause a sustained activation of certain key intracellular signaling events in the cognate cells, possibly affecting their homeostatic states (**Figure 1A**). Consequently, certain signaling pathways that were upregulated under normal homeostatic conditions can get downregulated or inhibited under chronic inflammatory conditions, and certain pathways originally inhibited or downregulated under normal conditions can be activated due to chronic inflammation (**Figure 1A**). In order to infer such

both conditions. However, P1 is active (upregulated) only in healthy microenvironment while it gets inhibited (downregulated) under chronic inflammation. On the other hand, P3 is activated under chronic inflammation while P1 gets downregulated. The goal in this study is to identify such constantly activated/inhibited signaling pathways as consequence of inflammation in AD affected in different regions of brain. (B) Represents the overall schematic if the analysis employed in the study.

constantly activated/inhibited signaling pathways/subnetworks possibly mediated by chronic inflammatory conditions in AD, we adapted a method that we originally developed to identify constantly activated/inhibited signaling subnetworks due to sustained effect of the niche or microenvironment on stem cell state (Ravichandran et al., 2016; Ravichandran and Del Sol, 2017). Briefly, the methodology combines gene expression data with signaling interactome, and identifies sparsest signaling sub-networks by connecting the receptors/ligands with the transcription factors (TFs) that best explain the differential gene expression pattern. For this, we assigned differential weights for interactions (edges) based on expression data and employed a Prize Collecting Steiner Tree (PCST) algorithm to infer minimal subgraphs of the signaling interactome (**Figure 1B**).

### Gene Expression Data Sources

Since we were interested in studying the role of chronic inflammation in different brain regions affected by AD, we analyzed gene expression datasets obtained from six different neuronal populations located in different areas of the brain. Specifically, neuronal populations from entorhinal cortex, hippocampus, middle temporal gyrus, posterior cingulate cortex, superior frontal gyrus, and primary visual cortex were collected (Liang et al., 2008). Gene expression datasets with appropriate controls are available from the Gene Expression Omnibus (GEO) with Accession No. GSE5281. The samples consisted of 13 control subjects and 10 AD cases for entorhinal cortex, 13 control subjects and 10 AD cases for hippocampus, 12 control subjects and 16 AD cases for middle temporal gyrus, 13 control subjects and 9 AD cases for posterior cingulate, 11 control subjects and 23 AD cases for superior frontal gyrus, and 12 control subjects and 19 AD cases for primary visual cortex (Liang et al., 2008). Further, gene expression was profiled by microarray using Affymetrix Human Genome U133 Plus 2.0 Array platform (Liang et al., 2008).

### Identification and Classification of Differentially Expressed Genes

We used the lists of differentially expressed genes (DEGs) directly from the original study (Liang et al., 2008). Supplementary Table 1 lists the DEGs for different neuronal subpopulations used in this study. From the lists of DEGs specific for different brain regions, we identified a set of differentially expressed TFs (DETFs) and transcriptional regulators based on the annotation available at Animal TFDB (http://bioinfo.life.hust.edu.cn/AnimalTFDB/; Hasin et al., 2017). For differentially expressed receptors (DERs), since a complete database of receptor molecules is currently unavailable, we used Gene Ontology classification of receptor activity and plasma membrane (GO:0004872, GO:0005886) to identify DEGs with possible receptor activity. The set of DERs served as the potential sensors of the environment and, specifically, chronic inflammatory stimuli. Further, secreted molecules, such as cytokines and chemokines, can act as regulators of downstream signaling pathways by activating their cognate receptors. In order to infer differentially expressed ligands, we took advantage of the classification of potential ligands reported in a recent study (Ramilowski et al., 2015). For the purpose of our analysis, we discretized the expression data based on differential gene expression and considered genes identified as upregulated and downregulated (based on the above mentioned cutoff) as "1" and "−1", while the non-DEGs were considered "0." Supplementary Table 2 contains the classification of the DEGs for different brain regions.

### Compilation of Interactions to Build a Background Signaling Interactome

In addition to gene expression data, we also required a compiled list of potential signaling interactions with direction (sourcetarget relationship) and sign (activation or inhibition) as an input for our method. For this, we used publicly available signaling interactions databases OmniPath and ReactomeFI (Wu et al., 2010; Turei et al., 2016). We combined them by removing the redundant interactions commonly present in both databases (by removing the duplicate entry) to acquire only unique interactions. We chose these two databases as they are well curated and contain directionality and signs (positive or negative regulation) of signal flow.

### Capturing the Effect of Chronic Inflammatory Signals

In order to recapitulate the effect of inflammatory signals on the cells under different diseased conditions, we consider that the upregulated receptors/ligands for the particular disease are under direct influence of the diseased environment (or niche). Since the exact mechanisms of chronic inflammation are not known, we represent them by introducing a dummy inflammation node in the raw signaling network. This dummy node is then connected to all upregulated receptors for each phenotype under consideration. Therefore, signal transduction from the inflammatory niche to DETFs must be propagated through at least one of the upregulated receptors. Such a representation of unknown external influence by a dummy node has been applied earlier (Tuncbag et al., 2013). Therefore, according to our consideration, signal transduction due to inflammation must be propagated through one of the upregulated receptors in order to reach the downstream TFs.

### Calculation of Differential Edge Weights

The edges in the signaling interactome were weighted using the gene expression data. This weighting scheme was implemented in order to maximize the compatibility between the expression data and interaction sign. By compatible, we mean consistency between the sign of the interactions (i.e., positive when activating and negative when inhibiting) and the effect (i.e., activation or inhibition) that the receptor has on its downstream target TFs. For example, sign of a signaling path from a receptor to a TF that is up regulated or overexpressed must be positive (activation), while it must be negative (inhibition) for down regulated or under-expressed TF. We calculated the differential edge weight such that for a given phenotype (for example disease) it reflected the probability of the target gene of the specific interaction to be relatively more active when compared to the other phenotype (for example healthy control) by considering the interaction sign and booleanized expression state of the interacting nodes. For example, considering an interaction A activating B with booleanized expression state of both the nodes being 1 (upregulated) for first phenotype and consequently −1 (downregulated) for the other phenotype. Here, since both nodes are upregulated in the first phenotype, the probability of B being differentially active in first phenotype will be higher in comparison to the second phenotype where both nodes are downregulated. However, when we consider an example of A inhibiting B, the probability of B being differentially active across the two phenotypes is low, and in such cases we consider equally low edge weights for the interaction. Since we worked with booleanized expression values, we considered the fixed probabilities, where an edge was assigned a probability of 0.9 if it was classified as high probability interaction and 0.1 when it was classified as low. Based on such differential edge-weighting scheme, we calculated differential edge weights for the two phenotypes under consideration by accounting for interaction sign and booleanized expression status of the interacting nodes.

### Identification of Signaling Subnetworks

In this weighted raw signaling interactome, we aimed to identify signaling paths that are potentially affected by chronic inflammation and are responsible for the observed expression pattern of the DETFs. Our method considers that microenvironment maintains the cells in a stable state by a sustained effect on their TFs via constantly activated/inhibited intracellular signaling pathways compatible with the phenotypespecific TRN state. The fact that the cells exhibit differences in their phenotype (diseased vs. healthy) due to differential effect of the microenvironment suggests that the intracellular signaling events controlling the specific GRN for maintain the specific phenotype are also differentially active. For identifying such signaling sub-networks, we employed PCST formalism to infer sub-networks with the dummy inflammation node as the root or origin node and the DETFs as the terminal nodes employing a heuristic algorithm MSGSTEINER (Bailly-Bechet et al., 2011). The Steiner Tree formalism have been used earlier to reconstruct active signaling pathways (Bailly-Bechet et al., 2011). Here, the objective is to infer the sparsest sub-networks that connect the root node (dummy node) and all the terminal nodes (TFs), that is also compatible with the differential expression states of the nodes inferred from the data. Since the dummy node is connected only to the upregulated receptors/ligands, the inferred sub-networks (Steiner trees) will encompass only those receptors that are both topologically favorable and compatible in the expression state to link the DETFs. Therefore, from several hundreds of upregulated receptors/ligands, we could narrow down to the few key ones linking the DETFs based on their unique network topological features. Importantly, as these sub-networks are more topologically favorable to explain the downstream gene expression pattern of TFs, they are likely to represent constantly activated/inhibited signaling sub-networks due to effect of chronic inflammation. In fact, our computational approach attempts to infer the sparsest subnetwork that connects the DERs/ligands and the downstream DETFs, and attempts to include as many differentially expressed intermediates (linker molecules) as possible (by maximizing the edge weights) which are also consistent with the sign of interaction. However, in cases where there are no such intermediates that are differentially expressed, certain non-differentially expressed intermediates are chosen as linker molecules depending on the network topology. Therefore, these intermediates are necessary for signal transduction from the DERs/ligands to DETFs. However, it is important to note that, this does not imply that differential expression of the downstream genes are caused by the intermediates that are not differentially expressed. Further, since we are attempting to capture sustained signaling relying only on gene expression data, some of the molecules that function via post-translational modifications are not differentially expressed but can still act as intermediate linker molecules.

The method infers receptors and associated signaling subnetworks that are crucial for influencing the DETFs in a sustained manner, and does not attempt to rank all the inferred signaling intermediates (other downstream molecules like associated kinases and phosphatases) or the entire pathway as a whole.

### RESULTS

We employed our computational method to identify signaling networks that are likely to be constantly perturbed in AD patients when compared to healthy elderly control individuals. Here, we discuss certain key receptors and their associated signaling components that were identified by our method. Although several of the identified receptors/ligands had a direct link with inflammatory immune responses, the method also inferred molecules not directly related to immune responses, but associated to the disease through other mechanisms. This could be possibly due to the fact that other processes that are responsible of AD progression can also exert sustained influence apart from inflammatory stimuli. Since the method attempts to infer sustained signaling components, irrespective of whether they are derived by inflammatory means or any other process linked to AD, such as deposition of amyloidbeta (Aβ) plaques or neuronal death, it can therefore gather components that are not associated to an inflammatory immune response, but linked to the disease itself. Further, it must be mentioned that we focused predominantly on characterizing the region specific inducers (receptors/ligands) of inflammation and not on all the downstream intermediates and target TFs that transmit the inflammatory signals based on the inferred signaling subnetworks. **Table 1** lists the receptors/ligands identified for six different brain regions in the signaling subnetworks with evidences for their involvement in AD focusing on neuronal inflammatory immune responses.

In the original study, the analysis was conducted comparing non-tangle-bearing neurons from AD patients with healthy neurons from control elderly subjects. In fact, the brain regions analyzed have been previously observed to show characteristic pathological differences in the brains of individuals afflicted with AD compared to healthy individuals. The entorhinal cortex and the hippocampus are two regions that have been found to be susceptible to early NFT formation (Bouras et al., 1994). The mid temporal gyrus and the posterior cingulate cortex have been found to exhibit an elevated susceptibility to amyloid plaque formation (Blesa et al., 1996). The superior frontal gyrus has been observed to show metabolic changes relative to normal aging (Blesa et al., 1996), and the primary visual cortex has been found to be relatively unaffected from any form of age-related or disease-related neurodegeneration (Liang et al., 2008). These two regions are also known to represent late stages of AD (Liang et al., 2008).

### Hippocampus

The hippocampus is critical for learning and memory, is specifically vulnerable to damage at early stages of AD (Mu and Gage, 2011). In fact, perturbed neurogenesis in the adult hippocampus indicates an early critical event in the onset and progression of AD. From a functional point of view, hippocampal neurogenesis plays an important role in structural plasticity and network maintenance (Mu and Gage, 2011). The CA1 region was selected in the original study (from where the gene expression data for the computational analysis was obtained) because this area is the earliest (Braak stages I–IV) and most heavily affected region of the hippocampus in terms of tangle formation (Liang et al., 2008).

We identified fibroblast growth factor receptor 1 (FGFR1) mediated signaling activity as a key upregulated component in neurons isolated from the hippocampus of AD patients (**Figure 2**). FGFR1 signaling is known to transmit inflammatory signals through regulation of other surface proteins (Woodbury and Ikezu, 2014). In fact, due to its importance in adult neurogenesis and neuroinflammation, manipulation of fibroblast growth factor 2 (FGF2)/FGFR1 signaling has been a focus of therapeutic development for neurodegenerative disorders, such as AD, multiple sclerosis (MS), Parkinson's disease and traumatic brain injury (Woodbury and Ikezu, 2014). In our sub-network, FGFR1 signals through CREB binding protein (CREBBP) induction, a protein involved in the transcriptional co-activation of many different TFs, such as the inflammatory mediators

#### TABLE 1 | Identification of AD specific factors.


interferon regulatory factor 1 (IRF1) and E2F transcription factor 1 (E2F1), the latest involved in the modulation of neuronal apoptosis (Hou et al., 2000). High mobility group box 1 (HMGB1) is a mediator of inflammation that is released extracellularly during cells death or secreted by activated cells (Jiao et al., 2016). In line with its involvement in inflammatory processes, HMGB1 upregulation in our sub-network is linked to tumor protein p53 (TP53) and nuclear factor kappa b subunit 1 (NFKB1), two main inflammatory signaling pathways.

Among the downregulated receptors, ADAM metallopeptidase domain 17 (ADAM17), a metalloprotease involved in the processing of tumor necrosis factor alpha (TNF) that has been described to counteract inflammation and further neuronal damage, was identified (**Figure 2**; Qian et al., 2016).

### Posterior Cingulate Cortex

The precise function of the posterior cingulate cortex is not yet clearly established. However, it is known to be involved in cognitive tasks (Leech et al., 2012). This region is known to acquire early amyloid deposition, reduced metabolism in AD (Leech and Sharp, 2014), and therefore serves as a key region to be studied despite the lack of clarity in its functionality in the brain.

TNF receptor superfamily member 1 (TNFRSF1A), a known pro-inflammatory signaling component, was a major player in the posterior cingulate cortex sub-network (**Figure 3**). Excess of inflammatory mediators in the brain are associated, at least partly, to activated microglia, which accumulate around amyloidbeta (Aβ) plaques in AD brains, showing chronic activation and therefore signaling constantly. Elevated levels of proinflammatory cytokines, such as TNF, could potentially inhibit phagocytosis of Aβ in AD brains thereby hindering efficient plaque removal by resident microglia (McAlpine and Tansey, 2008). Moreover, FGFR1, also upregulated in the hippocampal neurons from AD patients, was identified in the posterior cingulate cortex region.

Among the downregulated signaling networks, we identified sortilin 1 (SORT1), a pro-neurotrophin receptor which plays a major role in the clearance of apolipoprotein E (APOE)/Aβ complexes in neurons (**Figure 3**; Carlo et al., 2013). APOE sequesters neurotoxic Aβ peptides and deliver them for cellular catabolism via neuronal APOE receptors (Carlo et al., 2013). SORT1 binds APOE with high affinity and lack of receptor expression in mice results in accumulation of APOE and Aβ in the brain and in aggravated plaque burden, thus suggesting a link between Aβ catabolism and pro-neurotrophin signaling converging to this receptor (Carlo et al., 2013). Another receptor identified in the network was protein phosphatase 2 catalytic subunit alpha (PPP2CA). It binds to tau and is the primary tau phosphatase (Sontag and Sontag, 2014). Its deregulation correlates with increased tau phosphorylation likely contributing to tau deregulation in AD (Sontag and Sontag, 2014). The signaling network controlled by SORT1 was also inferred in the middle temporal gyrus network, while PPP2CA was inferred in the superior frontal gyrus network (**Table 1**), thereby implicating their role in other affected brain regions.

### Middle Temporal Gyrus

This brain region is known to be involved in cognitive processes including, language and semantic memory processing, visual perception, and multimodal sensory integration (Onitsuka et al., 2004). Further, the middle temporal gyrus is known to exhibit reduced metabolic activity in AD (Liang et al., 2008).

Syndecan 2 (SDC2) and ERB-B2 receptor tyrosine kinase 4 (ERRB4) mediated signaling were identified in our sub-networks to play key roles in the middle temporal gyrus area (**Table 1**; **Figure 4**). In addition, we found Notch1 mediated signaling to be crucial in that region. Notch1 signaling is essential for various CNS functioning from brain development to adult brain function (Brai et al., 2016). Reduction in Notch1 expression affects synaptic plasticity, memory and olfaction. On the contrary, Notch1 over-activation after brain injury is detrimental for neuronal survival (Brai et al., 2016). Some familial AD mutations in presenilins can affect Notch1 processing/activation (Brai et al., 2016). Further reports described Notch1 overexpression in sporadic AD.

In the downregulated signaling networks, calreticulin (CALR) and its downstream network was identified by our method (**Figure 4**). Calreticulin is found in a complex with APP and Aβ and levels of the calreticulin mRNA and protein are reduced in patients with AD. This suggests that calreticulin is implicated in the proteolytic processing of APP and, thus, in AD pathogenesis (Stemmer et al., 2013).

### Entorhinal Cortex

The entorhinal cortex is thought to be a major input and output structure of the hippocampal formation, acting as the nodal point of cortico-hippocampal circuits (Canto et al., 2008). This is one of the most vulnerable brain regions that is attacked during the early stages of AD (Van Hoesen et al., 1991) and is thought to spread from here to other regions of brain. Further, there are emerging roles of inflammation in promoting neurodegeneration in the entorhinal cortex (Criscuolo et al., 2017).

Epidermal growth factor receptor (EGFR)-mediated signaling was identified as a key upregulated component of the entorhinal cortical neurons isolated from AD patients compared to healthy subjects (**Figure 5**). Although EGFR is not directly implicated in neuroinflammation, it is known to play a central role in neurometabolic aging. EGFR acts as a signaling entity for several ligand mediated mechanisms and cellular stress responses directly related to aging and degeneration (Siddiqui et al., 2012). Further, EGFR signaling has been implicated in a spectrum of neurometabolic conditions, such as metabolic syndrome, diabetes, AD, cancer, and cardiorespiratory function (Siddiqui et al., 2012). More recently, it has been observed that inhibition of EGFR enables rescue of memory loss in both mouse and drosophila.

LDL receptor related protein associated protein 1 (LRPAP1) and its downstream signaling network was found to be downregulated in this brain region (**Figure 5**). Interestingly, LRPAP1 levels have been found to be low in patients with increased susceptibility to AD, which implicates a link of this receptor with Aβ clearance. Co-localization of Aβ, APOE and LRPAP1 on senile plaques suggests its involvement in the clearance of APOE/Aβ complex (Pandey et al., 2008). In recent years, the lipoprotein receptor lowdensity lipoprotein receptor-related protein 1 (LRP1), the downstream effector of LRPAP1 in our sub-network, emerged as an important regulator of the inflammatory response (May, 2013).

figure legends are the same as that in Figure 2. Supplementary Figure 3 contains the other subnetworks identified for middle temporal gyrus that did not have known role in AD.

### Primary Visual Cortex and Superior Frontal Gyrus

The primary function of the early visual cortex is visual perception (Petro et al., 2017) and the superior frontal gyrus is thought to contribute to higher cognitive functions, in particularly to working memory (du Boisgueheneuc et al., 2006). These two brain regions have been included in the original study as they are known to be later (Braak stages V and VI) and less affected by the disease, and consist of least number of DEGs among the analyzed brain regions (Supplementary Table 1). The major signaling pathway upregulated in these two regions involved EGFR in both areas and FGFR1 in the primary visual cortex. These receptors were also identified by our analysis in other affected brain regions (**Table 1**; **Figure 6**).

Among the downregulated effectors, the phosphatase PPP2CA, already identified as decreased in the posterior cingulate cortex and upregulated in the hippocampus, was identified in both areas. Furthermore, we found brain-derived neurotrophic factor (BDNF) and its downstream signaling to be specifically downregulated in the superior frontal gyrus area (**Figure 7**). Importantly, BDNF is known to protect against tau-related neurodegeneration in Alzheimer's disease (Jiao et al., 2016).

### Brain Regions Sharing Common Inflammatory Mediators

After identifying the signaling sub-networks likely to be constantly activated/inhibited in different neuronal subpopulations of AD affected brain, we identified the common and specific factors among different brain regions. As it could be seen from **Figure 8**, only few identified factors were shared across different brain regions, while the majority of the factors were specific for the respective brain region. Notably, EGFR signaling was identified to be active for four regions, namely, middle temporal gyrus, entorhinal cortex, primary visual cortex and superior frontal gyrus. In fact, recent studies in AD mouse models have observed that EGFR is a preferred target for treating Aβ-induced memory loss, adding value to our computational inference (Wang et al., 2012). Another key receptor found to be commonly active in posterior cingulate cortex and entorhinal

cortex is TNFRSF1A, which is a well-known pro-inflammatory factor (Carlo et al., 2013). More importantly, blocking TFN signaling either via genetic manipulation or using chemical inhibitors, reduced the accumulation of intraneuronal amyloidassociated proteins triggered by chronic systemic inflammation, and could possibly act as a valid therapeutic target to modify disease progression during the early stages of AD (McAlpine et al., 2009). FGFR1 mediated signaling pathway, which is known to have profound roles in neurogenesis, was found to be present in entorhinal cortex, hippocampus and primary visual cortex (Woodbury and Ikezu, 2014). Further, activation of this signaling by FGF2 has proven to be highly efficient for the regeneration of neurons in multiple experimental animal models (Woodbury and Ikezu, 2014). In fact, several groups have shown the potential use of FGF2 as a therapeutic for neurodegenerative conditions including AD and PD. FGF2 gene transfer in AD transgenic mouse models is known to significantly restore spatial learning, hippocampal CA1 long-term potentiation, and neurogenesis in the SGZ (Kiyota et al., 2011). Another active signaling pathway commonly present in middle temporal gyrus and primary visual cortex were mediated by ADAM10 and this signaling has known roles in AD (Yuan et al., 2017).

In the case of inhibited signaling pathways (**Table 1**), only SORT1 and PPP2CA were identified to be commonly present in multiple brain regions. SORT1 was present in the middle temporal gyrus and primary visual cortex, and this molecule has recently been found to act as a novel receptor for apolipoprotein E (APOE) (Carlo, 2013; Carlo et al., 2013). Importantly, ablation of sortilin expression in mice results in accumulation of APOE and Aβ in the brain resulting in AD like physiology the mice (Carlo, 2013). PPP2CA was another key inhibited signaling identified to be present in three different brain regions namely, posterior cingulate cortex, primary visual cortex and superior frontal gyrus. Alterations in this phosphatase activity have been

reported in AD-affected brain regions and has been linked to tau hyperphosphorylation, amyloidogenesis and synaptic deficits (Sontag and Sontag, 2014).

### DISCUSSION

Technical advancements in sequencing allow the analysis of thousands of molecular profiles from clinical samples with high quality. These high-throughput techniques open-up opportunities for the development of computational analysis tools to infer meaningful patterns from big data. In this study, we have analyzed gene expression profiling data of specific neuronal populations collected by laser capture microdissection from postmortem samples of AD patients with the aim of identifying brain region specific signaling subnetworks affected by chronic inflammation. The network analysis we carried out enabled the identification of a fraction of receptor/ligands (and the associated downstream signaling networks) as key inflammatory mediators from a large number of DERs and ligands. In fact, on average, there were about 100 DERs/ligand molecules in the datasets we analyzed, and our method was able to identify crucial inflammation mediators. This highlights the utility of network approaches to refine and extract accurate and relevant information from high-throughput datasets. Furthermore, computational analyses are amenable to consider more than one diseased region simultaneously and therefore can be used to evaluate the commonalities and distinct features of regulatory processes involved in different regions in an integrative manner. In addition, when data are available, approaches like ours can be employed to study differences and commonalities of a specific disease phenotype in the context of different regions or tissues that are affected due the disease.

EGFR is one of the key molecule identified from our analysis to be common for four different regions namely, middle temporal gyrus, entorhinal cortex, primary visual cortex and superior frontal cortex of AD brain. Importantly, recent evidences suggest a crucial role for EGFR in AD pathogenesis, where potential interactions between Aβ oligomers and EGFR were found (Wang et al., 2012). Furthermore, inhibition of EGFR led to reversal of memory loss in AD mouse models (Wang et al., 2012). Notably, brain regions known to be less affected by AD, such as primary visual cortex and superior frontal gyrus, contained lesser number of sub-networks compared to regions highly perturbed by the disease, such as entorhinal cortex and middle temporal gyrus. This probably suggests that changes in AD brains are more pronounced in the main affected areas when compared to less influenced regions. This feature could also be partly attributed to the less number of DEGs in these two regions of AD brain (Supplementary Table 1). Despite the lower number of DEGs in the primary visual cortex and the superior frontal gyrus, EGFR was found to be crucial in these two regions in addition to its role in the middle temporal gyrus and the entorhinal cortex. This could also possibly reflect a crucial role for EGFR in the progression of AD in both early and late stages of the disease. It has to be mentioned that EGFR was not upregulated in two other important regions, i.e., the hippocampus and the posterior

cingulate cortex at least in the dataset we analyzed. It could possibly reflect region specific perturbation of key molecules and it also might relate to the fact that EGFR can also be regulated post-translationally and not always at the gene expression level.

Our computational analysis of AD samples from six different brain regions identified certain common regulators, such as FGFR1 and PPP2CA, known to be key neuroinflammatory factors (Tuncbag et al., 2013; Rajendran et al., 2017). In addition, certain region specific factors, such as CALR and EPHB1 in the middle temporal gyrus, were also identified by our analysis. In addition to several receptor/ligand molecules with direct role in inflammation, our analysis identified several novel candidates whose direct link to inflammation and the disease is not clearly known. For instance, we identified the Notch receptor, NOTCH1, in the middle temporal gyrus network. Although this receptor is not directly related to neuroinflammation, the role of Notch signaling pathway in promoting pro-inflammatory responses in the cells expressing NOTCH1 has been observed (Brai et al., 2016). Such molecules with no direct evidences in inflammation are novel predictions from our analysis, and will be interesting to experimentally test if they are important to transmit inflammatory signaling in these disease conditions.

Overall, our analysis revealed mostly distinct and few common inflammatory signaling components across the different neuronal populations analyzed. However, despite the predominance of region specific factors, the analysis identified certain shared inflammatory factors, reflecting some environmental similarities across the affected brain regions, such as the presence of Aβ plaques. Further, we observed that not all receptors/ligands that were identified by our analyses were directly related to inflammation, but were known to be implicated in the disease via some other processes, such

as metabolic dysregulation (for example, SORT1). This is because, although we considered inflammation to be one of the predominant cause of sustained signaling activities, our computational approach infers constantly activated/inhibited signaling pathways irrespective of whether these pathways are induced by inflammation or any other cellular processes. Although this aspect could represent a limitation of our approach, it rather demonstrates that it can be applied in cases where sustained signaling pathways need to be identified (Ravichandran et al., 2016; Ravichandran and Del Sol, 2017). Further, certain classical inflammatory mediators, such as FAS- and IL6-related pathways, were not identified. This could be explained by the fact that our approach requires that the modulation of the factors has to occur at gene expression level, and that our approach does not take into account specific receptors and TFs that are not differentially expressed under the disease condition. However, these mediators can still be active during inflammatory processes, for instance via post translation modifications, and our approach cannot detect such mediators since it currently relies only on mRNA level changes. In general, these limitations can be addressed when we consider proteomics or phosphoproteomics datasets for the inference of sustained signaling. However, availability of such protein level datasets are still limited compared to transcriptomics datasets and despite such limitations, predominantly due to incomplete data, we observed that our computational approach revealed the involvement of several disease related pro- and antiinflammatory factors highlighting the value of such analysis.

To our knowledge, this is one of the first study, which computationally analyzed the potential implication of chronic inflammation on AD to infer the induced signaling networks/pathways. In this regard, it must be mentioned that two recent studies have attempted to infer common and unique pathways/signatures in different neurodegenerative diseases (Li et al., 2014, 2015). However, their focus was more on "generic" pathways affected due to the disease and not specifically related to inflammation or sustained signaling. Further, in both studies, the analysis was predominantly based on differential gene expression signatures and pathway enrichment, and did not attempt to capture dysregulated network components as described in our study (Li et al., 2014, 2015). Although these studies are very useful to understand the overall pathways that are dysregulated in the disease condition and to define disease signatures, they do not capture specific molecular features that can relate the pathways to the disease itself. Alternatively, our analysis focus on inference of sustained signaling subnetworks to capture specific molecular features of chronic inflammation, which was not the focus of the other studies. Consequently, our current analysis revealed specific factors that were up- or down- regulated in neuronal populations from different brain regions affected by AD that were implicated in inflammation and the disease.

In the context of neurodegenerative diseases, integrative approaches enable to obtain a holistic understanding of the processes and factors that initiate and sustain specific disease pathologies that act primarily at the cellular level. For instance, protective and pathogenic roles of glial cells, such as microglia and astrocytes, in addition to the activation of common inflammatory pathways in these cells in several neurodegenerative diseases, support the concept that gliainduced inflammation can possibly sustain the disease pathology (Burns and Iliffe, 2009; Wyss-Coray and Rogers, 2012). In this context, integrative computational approaches like ours enable the identification of factors whose perturbation (by either activating or inhibiting) could reduce the production of factors that contribute to neurotoxicity, thereby potentially resulting in clinical benefit in specific neurodegenerative diseases.

### AUTHOR CONTRIBUTIONS

SR and AdS conceived the idea. SR performed the computational analysis. All authors contributed to the writing of the manuscript and approved the final version of the manuscript. AdS coordinated the overall project.

### REFERENCES


### ACKNOWLEDGMENTS

The authors would like to acknowledge the funding from University of Luxembourg, for ASTROSYS (R-AGR-3227) Internal Research Project (IRP grant).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2018.00154/full#supplementary-material


disease. A rational framework for the search of novel therapeutic approaches. Front Cell Neurosci 8:112. doi: 10.3389/fncel.2014.00112


**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 JEE and handling Editor declared their shared affiliation.

Copyright © 2018 Ravichandran, Michelucci and del Sol. 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.

# Exploration of the Anti-Inflammatory Drug Space Through Network Pharmacology: Applications for Drug Repurposing

#### Guillermo de Anda-Jáuregui, Kai Guo, Brett A. McGregor and Junguk Hur\*

Department of Biomedical Sciences, School of Medicine & Health Sciences, University of North Dakota, Grand Forks, ND, United States

#### Edited by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

### Reviewed by:

Antonio Longo, Università degli Studi di Catania, Italy Shao Li, Tsinghua University, China

> \*Correspondence: Junguk Hur junguk.hur@med.und.edu

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 12 September 2017 Accepted: 13 February 2018 Published: 01 March 2018

#### Citation:

de Anda-Jáuregui G, Guo K, McGregor BA and Hur J (2018) Exploration of the Anti-Inflammatory Drug Space Through Network Pharmacology: Applications for Drug Repurposing. Front. Physiol. 9:151. doi: 10.3389/fphys.2018.00151 The quintessential biological response to disease is inflammation. It is a driver and an important element in a wide range of pathological states. Pharmacological management of inflammation is therefore central in the clinical setting. Anti-inflammatory drugs modulate specific molecules involved in the inflammatory response; these drugs are traditionally classified as steroidal and non-steroidal drugs. However, the effects of these drugs are rarely limited to their canonical targets, affecting other molecules and altering biological functions with system-wide effects that can lead to the emergence of secondary therapeutic applications or adverse drug reactions (ADRs). In this study, relationships among anti-inflammatory drugs, functional pathways, and ADRs were explored through network models. We integrated structural drug information, experimental anti-inflammatory drug perturbation gene expression profiles obtained from the Connectivity Map and Library of Integrated Network-Based Cellular Signatures, functional pathways in the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome databases, as well as adverse reaction information from the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS). The network models comprise nodes representing anti-inflammatory drugs, functional pathways, and adverse effects. We identified structural and gene perturbation similarities linking anti-inflammatory drugs. Functional pathways were connected to drugs by implementing Gene Set Enrichment Analysis (GSEA). Drugs and adverse effects were connected based on the proportional reporting ratio (PRR) of an adverse effect in response to a given drug. Through these network models, relationships among anti-inflammatory drugs, their functional effects at the pathway level, and their adverse effects were explored. These networks comprise 70 different anti-inflammatory drugs, 462 functional pathways, and 1,175 ADRs. Network-based properties, such as degree, clustering coefficient, and node strength, were used to identify new therapeutic applications within and beyond the anti-inflammatory context, as well as ADR risk for these drugs, helping to select better repurposing candidates. Based on these parameters, we identified naproxen, meloxicam, etodolac, tenoxicam, flufenamic acid, fenoprofen, and nabumetone as candidates for drug repurposing with lower ADR risk. This network-based analysis pipeline provides a novel way to explore the effects of drugs in a therapeutic space.

Keywords: anti-inflammatory drugs, network pharmacology, adverse drug reactions, pathways, systems pharmacology, drug repurposing

### INTRODUCTION

Inflammation is a complex phenomenon involving immune cell recruitment in response to harmful stimuli as a protective measure by the body. The recruitment process is achieved by a wide variety of cytokines released by resident immune cells that can be either pro- or anti-inflammatory. Inflammation begins as an acute response aimed toward clearance of stimuli as well as mediators in an effort to restore normal function. Inflammatory mediators involved in an acute response are often short-lived, leading to resolution of inflammation once the stimuli are cleared (Cotran et al., 2015). However, acute inflammation can shift to chronic inflammation if the stimuli are not removed. An immune response resulting in an inflamed condition is triggered by a wide variety of stimuli, such as pathogens, damaged cells, or irritants (Ferrero-Miliani et al., 2007). As such, inflammation is central to a myriad of pathological manifestations that result in a collection of complex and nonlinear biological processes involved in an organism's response to stimuli (Vodovotz et al., 2010). Management of these processes with anti-inflammatory drugs is an important part of medical practice.

Even though interactions between molecules involved in the inflammatory response give way to a complex system (Vodovotz et al., 2008), most pharmacological drugs that deal with the response do so by modulating concrete molecular targets along the biochemical pathway of arachidonic acid to eicosanoids (Haeggström et al., 2010). Traditionally, two classes of antiinflammatory drugs exist: corticosteroids and non-steroidal antiinflammatory drugs (NSAIDS) (Agambar and Flower, 1990; Dinarello, 2010). The canonical targets through which these drugs exert anti-inflammatory effects are phospholipase A2 and cyclooxygenase (COX) 1 and 2 (Bozimowski, 2015).

However, anti-inflammatory drugs may have effects on molecules beyond these canonical targets (Barnes, 2006; Palayoor et al., 2012), and these off-site effects may have fortunate and unfortunate consequences. The effects of drugs on unintended targets can be the origin of adverse drug reactions (ADRs) and side effects (Rudmann, 2013); however, effects on alternative molecular and functional targets can lead to the repurposing of a drug, in which the drug is used in an alternative therapeutic application (Chartier et al., 2017). A successful pharmaceutical drug will strike an adequate balance between its therapeutic and unintended toxic effects. Drug repositioning can overcome limitations in development pipelines by dealing with drugs that have already been tested and marketed, reducing the risk of failure at the development stage. Anti-inflammatory drugs have been identified in drug repurposing studies (Strittmatter, 2014); however, there has been no systematic analysis of their functional and adverse effects.

Large-scale datasets related to drugs are common and contain information on chemical structures, adverse effects in the clinical setting, and system-wide effects from high-throughput drug screening projects. The magnitude of these datasets requires alternative analytical strategies from those of traditional pharmacological approaches, which can generate hypotheses and guide adequate experimental designs. In this context, networkbased models can be useful (Hopkins, 2008).

A biological network is a mathematical model in which biological elements, such as molecules, are represented as nodes (also known as vertices), and defined relationships between these elements are represented as links (or edges) (de Anda-Jáuregui et al., 2015). As a mathematical object, networks (also known as graphs) can be analyzed using well-established algorithms (Albert and Barabási, 2002; Barabási et al., 2011); in this sense, they offer a general framework for the study of natural phenomena. A biological network will have, based on the existing relationships between the elements that compose them, structural and topological characteristics that reflect underlying biological properties (Barabasi et al., 2004).

Network pharmacology expands on the network biology paradigm to address the problem of identifying pairs of drugs and targets that are clinically successful, maximizing therapeutic effects and minimizing toxicity (Harrold et al., 2013). It provides a framework which may be used to overcome limitations of other methods for drug exploration such as those based on phenotypic effects or those based only on chemical structure; it serves as a tool to integrate knowledge from the pharmacological and genomic spaces (Zhao and Li, 2010). Network pharmacology is becoming more relevant as the traditional single target pharmacological model shifts toward a model that (1) considers the perturbation of multiple biological entities in a disease (Kibble et al., 2016) and (2) considers functional targets as more suitable than molecular targets for effective drug therapies (Hopkins, 2008).

Pharmacological phenomena fundamentally involve interactions between elements of different origin and nature; for example, interactions between a drug and its biological targets or observable biological effects, either therapeutic or toxic. With this in mind, a suitable network model to study the pharmacological space can be a bipartite network, in which nodes represent elements of two distinct classes and relationships exist only between elements of different classes (Guillaume and Latapy, 2006).

In this work, we modeled the relationships among antiinflammatory drugs, their effects at the gene perturbation and pathway perturbation level, and their associated adverse reactions from a network perspective. Based on the topological properties derived from these networks, we propose strategies to prioritize anti-inflammatory drug repurposing, quantify potential side effect risk, and identify possible pathway perturbation-related mechanisms associated with side effects in the context of specific anti-inflammatory drug sets This provides a toolset to identify anti-inflammatory drugs that are candidates with better repurposing opportunities.

### MATERIALS AND METHODS

In this work, anti-inflammatory drugs were identified, and chemical structures, gene perturbation profiles, and ADR report data were collected. Using these data, similarity matrices and bipartite networks were generated. **Figure 1** illustrates our overall workflow.

### Data

### List of Anti-inflammatory Drugs

Anti-inflammatory drugs were identified using the anatomical therapeutic chemical (ATC) classification system. Drugs with ATC codes M01A (anti-inflammatory and antirheumatic products, non-steroids), H02AB (corticosteroids for systemic use, plain), and N02BA (salicylic acid and derivatives) were selected. Drug names were normalized to the official DrugBank name or generic names to be consistent and comparable across various datasets in the current study. Our in-house drug-name normalization pipeline was used as previously described (Hur et al., 2014). The chemical structures of the anti-inflammatory drugs of interest were retrieved from PubChem (Kim et al., 2016) and DrugBank (Wishart et al., 2008) databases in simplified molecular-input line-entry system (SMILES) format.

### Drug-Gene Perturbation Profiles

Two large-scale drug-perturbation gene expression datasets containing genome-wide transcriptional expression data from cultured human cell lines treated with bioactive small-molecules were included. The first dataset is the Connectivity Map (CMap, version 02), with over 7,000 expression profiles representing 1,309 compounds in five cultured human cell lines that were measured using Affymetrix human genome U133A (HGU133A) arrays. The second is the L1000 dataset from the Library of Integrated Network-based Cellular Signatures (LINCS) project (Duan et al., 2014), which measured gene expression changes after treatment of 83 human cells with over 20,000 smallmolecule compounds using the L1000 platform (https://www. ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GPL20573). The LINCS L1000 dataset is composed of two publicly available releases with Gene Expression Omnibus (GEO; https://www.ncbi.nlm. nih.gov/geo/) accession numbers GSE92742 (LINCS phase I) and GSE70138 (LINCS phase II), which were merged into a single LINCS dataset in our study.

Probe data were aggregated to the gene level when applicable (HGU133A platform). In cases of more than one experimental condition involving a drug of interest, we merged perturbation profiles using the Kruskal–Borda merging algorithm (Iorio et al., 2010) to generate a consensus profile. Considering the use of different quantification platforms for both datasets, analyses using these were conducted independently for the first CMap dataset (hereafter referred to as CMap) and the dataset belonging to the LINCS project (hereafter referred to as LINCS).

### Drug Adverse Reaction Information

Pharmacovigilance information from the Food and Drug Administration (FDA) adverse event reporting system (FAERS) (FDA, 2014) from 2012 to the third quarter of 2016 was used. The FAERS data were downloaded and processed using the FAERS package (https://github.com/mlbernauer/FAERS). Drug names were normalized to generic names using the FAERS package as well as our name-normalization pipeline. Drugs with multiple single active ingredients were excluded from further analyses. ADRs were normalized to the MedDRA release 20.0 preferred

terms (PTs) and mapped to MedDRA higher level terms (HLT) (https://www.meddra.org/).

### Pathway Collection

Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (Kanehisa et al., 2014), as well as Reactome (Fabregat et al., 2016) and BioCarta (Nishimura, 2001) databases retrieved from the Broad Institute's Molecular Signature database (Subramanian et al., 2005) category C3 were used.

### Similarity Matrices Construction Drug-Drug Structural Similarity

Drug structures in SMILES format were used to obtain a unique atom pair library (Chen and Reynolds, 2002) for each anti-inflammatory drug. For each pair of anti-inflammatory drugs, Tanimoto similarity was calculated using the ChemmineR package in R (Cao et al., 2008).

### Drug-Drug Gene Perturbation Profile Similarity

For each collection of drug perturbation profiles (CMap and LINCS), similarity in gene-level effects between drugs was analyzed. To do so, Spearman similarity between the ranked gene perturbation profiles was calculated for each pair of drugs. The CMap set and the LINCS set were analyzed separately.

### Non-supervised Hierarchical Clustering

Drugs were clustered hierarchically in both the structural and perturbation profile similarity matrices. Briefly, a set of dissimilarities were generated from the matrices, and each element (drug) was assigned to a cluster; iteratively, each pair of most similar clusters was merged, until a single cluster was obtained. Clustering was performed using the complete linkage method with the hclust function in the R stats package (version 3.4.2).

## Bipartite Network Construction

### Drug-Pathway Perturbation Network

Any drug may potentially have effects beyond their canonical targets. Some of these effects manifest in the perturbation of system-scale biological pathways, which are evident as a significantly coordinated change in the genetic expression of the molecules involved in it. To identify possible system-wide effects of anti-inflammatory drugs, bipartite networks of drugs and pathway perturbations were constructed. For each gene perturbation profile associated with a drug in our datasets, we used a gene set enrichment analysis (GSEA) (Subramanian et al., 2005) to identify the pathways exhibiting significant alterations attributable to anti-inflammatory drug treatment. The fast GSEA (FGSEA) (Sergushichev, 2016) implementation was used with 50,000 permutations per analysis, with a significance cut-off set as an adjusted p < 0.1.

Results of the enrichment analysis were integrated into a bipartite network composed of anti-inflammatory drugs and significantly enriched pathways in at least one drug. An edge was established between a drug and a pathway if the drug treatment resulted in significant enrichment of the pathway (adjusted p < 0.1). The edges in this network were unweighted, as any pathway with an adjusted p value below the threshold was determined to be significantly enriched.

### Drug-ADR Network

ADRs were linked to those drugs most likely to produce them. The proportional reporting ratio (PRR) (Evans et al., 2001) is a statistical measure of relative risk for a given ADR extracted from pharmacovigilance data that compare the specific frequency of an ADR under a given condition (i.e., drug treatment) against the overall frequency of the ADR. PRRs for ADRs between anti-inflammatory drugs and non-anti-inflammatory drugs were first calculated. ADRs with PRRs > 1 were then selected: these included ADRs more likely to be associated with antiinflammatory drugs than with other drugs. ADRs more likely to be associated with specific anti-inflammatory drugs were identified by calculating the PRR for the previously selected ADRs among anti-inflammatory drugs only. Calculations were done using the R package PhViD (Ahmed et al., 2010).

The FAERS dataset presents ADRs in terms of MedDRA Preferred Terms. Considering that there are 22,500 different PTs, these were mapped to their corresponding MedDRA HLT. A bipartite network was constructed in which the first set of nodes are drugs and the second set of nodes are ADRs represented as HLTs. An edge was established between a drug and an HLT if there was at least one PT associated with the HLT and a PRR > 1. Since PRR is a measure of relative risk, the edges in this network were considered to have a weight; the weight of an edge is the sum of the PRRs between the drug and the set of PTs that can be associated with a particular HLT.

### Pathway-ADR Network

Through the merging and projection of the drug-pathway network and the drug-ADR network, a pathway-ADR network was generated. In this network, a pathway was connected to an ADR if there was at least one drug connected to both. Edges in the network have a weight, which represents the number of drugs through which a pathway and an ADR are connected.

### Network Analysis

For each network, basic network properties such as number nodes, number of edges, and network density were calculated. Centrality measures such as degree, clustering coefficient, and redundancy coefficient (Latapy et al., 2006) were calculated for each set of nodes using the R package Igraph (Csardi and Nepusz, 2006) and the Python package NetworkX (Hagberg et al., 2008). **Supplementary File 1** contains GML files for each network.

## RESULTS

### Drug-Drug Structural Similarity

A total of 114 anti-inflammatory drugs were identified based on our ATC inclusion criteria as of July 2017. We obtained drug structure data for 110 of these drugs from DrugBank and PubChem. **Figure 2** shows a heatmap representing the structural similarities among these compounds. **Supplementary File 2** contains the corresponding data matrix. In this heatmap, drugs are arranged through non-supervised hierarchical clustering. These clusters are composed of drugs with similar structures, such as drugs that are derived from a lead molecule. For instance, there is a large cluster which contains steroidal anti-inflammatory drugs that are similar in structure to hydrocortisone; this

cluster also has the highest similarity values. Non-steroidal antiinflammatory drugs are grouped in smaller clusters, consistent with the more diverse structures found in this group.

### Gene Perturbation Similarity of Anti-inflammatory Drugs

Approximately 40% of the 114 anti-inflammatory drugs were included in the CMap and LINCS datasets (47 and 45 drugs, respectively). Only 25 drugs were found in both datasets. The heatmaps in **Figure 3** illustrate the similarities between drugs in terms of gene perturbation based on the Spearman correlation of their ranked gene perturbation profiles; color intensity is proportional to the correlation value. **Supplementary File 1** contains the corresponding data matrix.

In the CMap dataset, a group of 31 drugs formed a large cluster based on their gene perturbation effects. The rest of the drugs are in smaller clusters of 9 and 7 drugs. These drugs belong to both steroidal and non-steroidal classes of anti-inflammatory drugs, showing a similarity in perturbation effects not limited by structural features. Although we identified similar clusters using the LINCS dataset, the overall similarities between drug profiles were lower, as visualized in the heatmap via less color intensity.

FIGURE 3 | Heatmap of gene perturbation similarity. These heatmaps represent similarities in gene expression profiles induced by drugs obtained from the (A) Connectivity Map (CMap) and (B) Library of Integrated Network-based Cellular Signatures (LINCS) datasets. The color intensity is proportional to the similarity between the gene expression profiles of two drugs measured using the Spearman correlation. Drugs are ordered using non-supervised hierarchical clustering. In both panels, a larger cluster containing most drugs and two smaller clusters are shown. The clusters comprise drugs that do not necessarily share structural similarities.

### Drug-Pathway Perturbation Network

Networks linking drugs to functional pathways were constructed based on their effects at the gene perturbation level. These networks allow the identification of biological functions that can be affected by a given drug and the identification of drugs that can affect a given biological function. The GSEA algorithm was used to identify which functional pathways are affected by each anti-inflammatory drug. **Supplementary Figure 1** shows the resulting bipartite networks using a hive plot visualization. In this visualization, nodes are arranged along axes and edges and are represented by Bezier curves. **Supplementary Figure 1A** shows the CMap-derived drugpathway perturbation network, which is dominated by the largest connected component (a subgraph composed of a set of nodes in which any pair of nodes is connected by a path, and there is no path connecting them to a node outside the subgraph), represented by a drug-containing axis and a pathway-containing axis densely populated with edges between them; this component contains most pathways perturbed by anti-inflammatory drugs. A second component containing pathways and two drugs is also shown. Twenty drugs with no pathway effects were found and are arranged along an axis with no edges. **Supplementary Figure 1B** shows the LINCS-derived drug-pathway perturbation network. Again, this network was dominated by the largest connected component concentrating most of the perturbed pathways, along with four smaller pathway-containing components and 7 disconnected drugs. The properties of these networks are summarized in **Table 1**.

In both networks, most drugs with at least one pathway target belong to the same largest connected component. TABLE 1 | Graph parameters for drug-pathway perturbation networks.


The most noticeable exceptions are azapropazone and acemetacin in the CMap-based network. These two drugs form a lone connected component where they both target 11 pathways related to signaling [including g protein-coupled receptor (GPCR) and calcium signaling] as well as xenobiotic metabolism, setting them apart from the rest of the antiinflammatory drug space in terms of system-wide effects. In the case of the LINCS-based network, there were three small components comprising single drug-pathway pairs [celecoxib and the trefoil factor (TFF) pathway; bufexamac and valine, leucine, and isoleucine degradation; and nabumetone and WNT signaling], as well as a component formed by a drug and two pathways [valdecoxib and the extracellular signalregulated kinase (ERK) pathway and acyl chain remodeling of phosphatidylglycerol].

**Figure 4** shows the degree distribution in these networks. The degree of a node refers to its number of adjacent edges; it is one of the defining measures of any network. The degree value in these drug-pathway perturbation networks has different

meanings for each type of node. A "drug degree" represents the capacity of a given drug to have multiple functional effects at the pathway level. Meanwhile, a "pathways degree" represents the susceptibility of a given pathway to be targeted by antiinflammatory drugs. It should be noted that the drug degree distribution and the pathway degree distribution are different in each network: first, the drug degree distribution has a larger range than the pathway degree distribution; second, the frequency of drug nodes with a single neighbor is higher than that of pathway nodes with a single neighbor. These plots also show that these networks are predominantly populated with pathways and drugs that have few neighbors, whereas highly connected nodes are scarce.

Pathway degrees range from 1 to 13 in the LINCS network and 17 in the CMap network; there are no pathways with a degree of 0, as expected given the network construction model. Interestingly, the most connected pathway was the "cell cycle" pathway from the Reactome database in both networks, making it the most susceptible pathway to anti-inflammatory drugs. However, there was not a single pathway that was perturbed at the gene expression level by all the anti-inflammatory drugs. Drug degrees range from 0 to 65 in the LINCS-based network and from 0 to 180 in the CMap-based network; drugs with the highest degrees were azapropazone and rimexolone, respectively. It must be noted that in the case of drug nodes, the network construction model allows for the presence of nodes with a degree of 0.

**Supplementary Figure 2** shows clustering coefficient distributions for each network. The clustering coefficient of a node in a bipartite network is a measure of how likely it is, on average, for a given node to share neighbors with others (Latapy et al., 2006). In the case of these networks, a higher clustering coefficient for a drug node can be interpreted as the likelihood that pathways affected by a given drug are affected by another drug. In contrast, for pathway nodes, a higher clustering coefficient represents the likelihood of finding another pathway that is susceptible to a similar set of drugs. The clustering coefficient for drugs in both networks is below 0.2 for all drugs in the largest connected component. A related concept is node redundancy. This parameter measures whether the removal of a given node from the bipartite graph leads to the disconnection of two of their neighbors. Drug node redundancy can represent the uniqueness of pathway effects, with highly non-redundant drugs being those that are able to affect pathways untargeted by other anti-inflammatory drugs. The distribution of redundancy values is shown in **Supplementary Figure 3**.

### Drug-ADR Network

Relationships between anti-inflammatory drugs and ADRs contained in the FAERS dataset were used to generate a bipartite network. This network allows the identification of possible adverse effects for a given drug and drugs that are associated to a given adverse effect. We first identified those side effects more likely to be associated with anti-inflammatory drugs; individual drugs more likely to be associated with particular side effects were then identified. The resulting bipartite network is presented in **Supplementary Figure 4** as a hive plot comprising a single large component containing all anti-inflammatory drugs studied and their side effects. The transparency of the Bezier lines is proportional to the weight of the edge, representing the ADR risk for the associated drug. The statistics describing this network are found in **Table 2**.

**Figure 5** shows the cumulative frequency distribution for degree. The ADR degrees ranged from 1 to 38; the highest degree was associated with a non-specific "general signs and symptoms, not elsewhere classified." Further, the drug degrees ranged from 3 for flufenamic acid to 635 in the case of prednisolone, suggesting that all anti-inflammatory drugs are more frequently associated with side effects than other drugs. **Supplementary Figures 5**, **6** show the previously discussed parameters of clustering coefficient and redundancy. In the case of drugs, the clustering coefficient increased up to a value of 0.18, while for ADR nodes, the maximum clustering coefficient was 0.23.

Another centrality measure only defined in weighted networks is node strength. The strength of a node in a weighted network is defined as the sum of weights for all adjacent edges (Barrat et al., 2004). In this network, weights represent the risk of a given ADR for a drug. Node strength for each drug can be used as a general

TABLE 2 | Graph parameters for the drug-adverse drug reaction (ADR) network.


measure of relative risk for any side effect, regardless of severity, to manifest in a treatment with a given drug. The strength distribution for drug nodes is shown in **Supplementary Figure 7**.

### Pathway-ADR Network

This network represents links between functional pathways and ADRs in the anti-inflammatory setting resulting from a projection of the drug-pathway and drug-ADR networks. Here, functional pathways and ADRs are linked if both are associated with at least one shared anti-inflammatory drug; the edge weight in these networks is the number of shared drugs between a pathway and an ADR. **Table 3** contains descriptors related to these networks.

These networks help identify possible associations between functional pathways and ADRs in the context of antiinflammatory drugs that can be used to generate hypotheses regarding the underlying mechanisms of the side effects associated with these drugs. For instance, "eye and eyelid infection" was a side effect identified with a high number of connections to three Reactome pathways, including cell cycle, cell cycle mitotic, and DNA replication using the CMapderived pathway-ADR network. There were four drugs through which these ADR and pathways are connected: hydrocortisone, ketoprofen, methylprednisolone, and rimexolone. Each of these drugs can be found in ophthalmic medications. The connection between these drugs and this ADR can therefore be explained by the therapeutic indication of these drugs. Meanwhile, the mechanism through which these drugs may affect these pathways is not known, although it is important to notice that cell cycle pathways are affected by several anti-inflammatory drugs, as mentioned previously in section Drug-Pathway Perturbation Network. Without further experimental confirmation, further discussion on the role of these pathways regarding the ADR would be speculative. However, a putative role of the perturbation of these pathways in the ophthalmic infection condition, at least in the context of treatment with this set of drugs, is an example

TABLE 3 | Graph parameters for pathway-adverse drug reaction (ADR) networks.


of a non-trivial insight that can be generated through a networkbased analysis and which may drive novel experimental research.

### DISCUSSION

In this work, we explored relationships among anti-inflammatory drugs based on parameters such as chemical structure similarity, gene perturbation, functional pathway perturbation, and ADRs. Network models were constructed and examined based on these features, which can provide further insights into the relationship of these drugs with their pharmacological effects; we found that functional features, such as perturbed pathways and ADRs, are more informative for this purpose. Through the exploration of drug centrality and functional features in bipartite networks, we identified drugs that could be prioritized for potential repurposing. Finally, we found associations between drug effects at the pathway level and side effects, which may point to underlying mechanisms for these effects in the anti-inflammatory setting.

Structural and perturbation similarities are good comparison parameters for the general drug space (Iorio et al., 2010; Lo et al., 2015). As the anti-inflammatory drug space is already a subset of the greater drug space that shares common therapeutic applications, differences (structural and at the raw gene perturbation level) are probably much less likely to be observed in the anti-inflammatory drug space than in the general drug space. It is difficult to determine a nonarbitrary cut-off value to generate an insightful network model based on similarities in the structural or gene perturbation profile. However, the bipartite models proposed in this work do not have this limitation, leading to a more descriptive landscape of the functional effects of these therapeutic drugs. Around 300 functional pathways that are potentially susceptible to modulation through anti-inflammatory drug use were identified, along with nearly 1,200 ADRs. Therefore, this network-based approach is presented as more suitable for the exploration of a limited section of the pharmacological space than studies based on chemical structural features alone.

The structure of these networks provides insight regarding the similarities and differences between anti-inflammatory drugs. Drug-pathway networks are dominated by the largest connected component in each network. Drugs that are part of the smaller components affect pathways that are not susceptible to the effects of other drugs. Drugs that have common effects that are shared with many other drugs in the network have higher clustering coefficients and redundancy values. Drugs whose pathway effects are completely redundant such that there is another drug with the exact same set of pathway effects are rare, which is consistent with the coexistence of these drugs in current medical practice. The drug-ADR network exclusively comprises a single connected component, indicating a widespread overlap in potential sideeffects between these drugs.

The connectivity of a drug in the networks presented in this manuscript is indicative of the drug's effects, either therapeutic, in the context of the pathway networks, or toxic, in the case of the ADR network. In the case of drug-pathway networks, node degree is a measure of connectivity; high-degree drugs in these networks affect more functional pathways. Some drugs were found to have no significant effects on pathways, appearing as drugs with a degree of 0. Evidently, this does not mean that these drugs have no pharmacological effects; rather, it is indicative of no observable activity at the gene expression level (as tested in cell cultures), making them comparatively less likely to exhibit other system-wide effects than other more connected anti-inflammatory drugs. In the context of ADRs, there are two complementary connectivity measures for each drug: degree indicates the number of possible side effects of a given drug, while node strength is a measure of relative risk for any side effect. The use of network-based metrics allows for a simple and generalizable categorization of these drugs based on their possible biological effects.

Drug connectivity in the networks presented in this manuscript can be useful in the context of drug repurposing. It is possible to use the degree of a drug node in the drug-pathway network as a measure of the number of potential alternative therapeutic targets. The node strength of a drug in the drug-ADR network is an indicator of its general risk for generating a side effect. Using these two axes, the anti-inflammatory space can be divided into four groups as shown in **Figure 6**. The drug space with the highest pathway targets and the lowest side effect strengths would be the best suited to identify drug candidates for repurposing. For illustration purposes, each axis is divided by its median value; drugs with higher than median pathway effects and lower than median side effect strength appear in the upper, left quadrant. Arguably, these drugs are more likely to be successfully repurposed, and there are literature reports that support this hypothesis. For instance, the use of naproxen as a cytotoxic drug in urinary bladder cancer has been reported (Kim et al., 2014). Other repurposed applications have been proposed for meloxicam in the treatment of non-Hodgkin's lymphoma (Nugent et al., 2016; Chartier et al., 2017), etodolac for the treatment of breast cancer (Yang et al., 2014), tenoxicam for the treatment of tuberculosis (Maitra et al., 2016), hydrocortisone in the treatment of Alzheimer's Disease (Zhang et al., 2015), flufenamic acid for the treatment of Salmonella infection (Ekins et al., 2011; Preethi et al., 2016), fenoprofen as a melanocortin receptor allosteric enhancer (Montero-Melendez et al., 2017), and nabumetone (Shameer et al., 2017).

The approach described in this work is currently limited by constraints related to the availability of drug effect information. The CMap and LINCS drug perturbation datasets, which are

some of the most extensive and have robust and reproducible generation methodologies, were used. Nonetheless, the coverage of tested drugs is far from complete, which is reflected in the composition of the resulting networks. In the case of pharmacovigilance information, it is important to consider the inherent limitations associated with drug monitoring, as well as limitations in the reporting infrastructure and changes in prescription habits, which may affect the availability of adverse effect information for drugs. As such, it is important to note that the approach presented here is implemented in an exploratory setting. The aforementioned limitations may lead to the identification of false positive leads, which can only be overcome with further, well-defined experimental research.

Since our focus in this work was a well-defined subset of the therapeutic drug space, we suggest that the relationships identified to the functional features identified here may be specific only to this subset of drugs. We do not know whether the topologies of similarly constructed bipartite networks for other drug classes, or for the larger drug space, will share similarities, as this was beyond the scope of the present work.

### CONCLUSIONS

In this work, we explored a subset of the pharmacological space integrated by anti-inflammatory drugs. We identified relationships between these drugs based on their chemical structure and effects on gene expression, as well as physiological effects such as alteration of functional pathways and the onset of adverse reactions. We integrated these into bipartite network models that we analyzed to identify topological properties related to these relationships. We showed that using the bipartite network model provides advantages for the exploration of the anti-inflammatory drug space that are not possible by using other analysis strategies. We expect to expand our model as new high-throughput drug screening protocols generate further information regarding drug effects.

We suggest that the present work provides a framework to explore functional effects of certain therapeutic classes. We focused on the anti-inflammatory space, considering its notable clinical importance. We demonstrated that it is possible to gain insights relevant to pharmaceutical research using these models, which can be integrated to drug repurposing and drug combination pipelines, as well as to the clinical setting. This will provide further criteria for the selection of optimal antiinflammatory therapies. Finally, we exemplified an application integrating different sources of pharmacological information into network models for drug repurposing.

## AUTHOR CONTRIBUTIONS

GdA-J: developed and implemented the network models and analyses; KG: implemented pipelines for database management and visualization; GdA-J, BM, KG, and JH: contributed to the analysis and discussion of results. All authors approve of the final manuscript.

### FUNDING

KG was partially supported by the University of North Dakota Post-Doctoral Pilot Grant (#21230).

### ACKNOWLEDGMENTS

The authors thank Editage by Cactus Communications co. Ltd. for their professional editing service.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2018.00151/full#supplementary-material

Supplementary File 1 | A compressed file containing all networks as graph markup language (GML) files.

Supplementary File 2 | A compressed file containing all similarity matrices as comma-delimited files.

Supplementary Figure 1 | Graph visualization of the drug-pathway perturbation networks. In these visualizations, networks are presented as hive plots. Nodes representing drugs and pathways are arranged along axes, and edges representing perturbation of a pathway by a drug are shown as Bezier curves. (A) shows a network derived from CMap data containing two connected components; the largest one (comprising the orange drug axis and the yellow pathway axis) contains the majority of pathway perturbing drugs and perturbed pathways. A second component contains two drugs (pink axis) perturbing 11 pathways. Drugs for which no pathway effects were found are arranged along the green axis. (B) shows a network derived from LINCS data that is dominated by a large connected component with 4 smaller components containing drugs and perturbed pathways. Drugs for which no pathway effects were found are arranged along the purple axis.

Supplementary Figure 2 | Clustering coefficient in drug-pathway networks. (A) Drug nodes in the CMap-based network, (B) pathway nodes in the CMap-based network, (C) drug nodes in the LINCS-based network, and (D) pathway nodes in the LINCS-based network. In each panel, the normalized, cumulative clustering coefficient frequency distribution is presented. Drug nodes have degree values that range from 0 to 0.45 in the CMap-based network and 0 to 0.19 in the LINCS-based network. Pathways have degree values that range from 0 to 0.75 in the CMap-based network and 0 to 1 in the LINCS-based network.

### REFERENCES


Supplementary Figure 3 | Redundancy coefficient in drug-pathway networks. (A) Drug nodes in the CMap-based network, (B) pathway nodes in the CMap-based network, (C) drug nodes in the LINCS-based network, and (D) pathway nodes in the LINCS-based network. In each panel, the normalized, cumulative redundancy frequency distribution is presented.

Supplementary Figure 4 | Graph visualization of the drug-adverse drug reaction (ADR) network. A hive plot representation of the drug-ADR network is shown. This network comprises 52 drug nodes and 1,227 ADR nodes, with 9,597 links between them. Drugs nodes are arranged along the blue axis, while ADR nodes are arranged along the red axis. Edges between them are shown as Bezier curves. The transparency of each line is proportional to the edge strength. The network comprises a single large connected component containing all drugs and pathways.

Supplementary Figure 5 | Clustering coefficient in drug-adverse drug reaction (ADR) networks. This figure illustrates the clustering coefficient distribution of (A) drug nodes and (B) ADR nodes. In each panel, the normalized, cumulative clustering coefficient frequency distribution is presented. Drug nodes have a maximum value of 0.18, while pathway nodes have a maximum value of 0.24.

Supplementary Figure 6 | Redundancy coefficient in drug-adverse drug reaction (ADR) networks. This figure illustrates the redundancy coefficient distribution of drug nodes (A) drug nodes and (B) ADR nodes. In each panel, the normalized, cumulative redundancy frequency distribution is presented for drug and ADR nodes, respectively.

Supplementary Figure 7 | Node strength in drug-adverse drug reaction (ADR) networks for drug nodes. The normalized, cumulative strength distribution for drug nodes is shown, with a range of 0 to 85,591.61. A node's strength in a weighted network is the sum of the weights of all adjacent edges. In this network, an edge weight represents the risk of an ADR for a given drug. Therefore, a drug's node strength is a measure of general risk for a drug.


**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 de Anda-Jáuregui, Guo, McGregor and Hur. 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.

# Inhibition of Toll-Like Receptor Signaling as a Promising Therapy for Inflammatory Diseases: A Journey from Molecular to Nano Therapeutics

Wei Gao<sup>1</sup> , Ye Xiong<sup>2</sup> , Qiang Li <sup>1</sup> \* and Hong Yang<sup>1</sup> \*

*<sup>1</sup> Department of Respiratory Medicine, Shanghai First People's Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China, <sup>2</sup> Department of Respiratory Medicine, Changhai Hospital, Second Military Medical University, Shanghai, China*

The recognition of invading pathogens and endogenous molecules from damaged tissues by toll-like receptors (TLRs) triggers protective self-defense mechanisms. However, excessive TLR activation disrupts the immune homeostasis by sustained pro-inflammatory cytokines and chemokines production and consequently contributes to the development of many inflammatory and autoimmune diseases, such as systemic lupus erythematosus (SLE), infection-associated sepsis, atherosclerosis, and asthma. Therefore, inhibitors/antagonists targeting TLR signals may be beneficial to treat these disorders. In this article, we first briefly summarize the pathophysiological role of TLRs in the inflammatory diseases. We then focus on reviewing the current knowledge in both preclinical and clinical studies of various TLR antagonists/inhibitors for the prevention and treatment of inflammatory diseases. These compounds range from conventional small molecules to therapeutic biologics and nanodevices. In particular, nanodevices are emerging as a new class of potent TLR inhibitors for their unique properties in desired bio-distribution, sustained circulation, and preferred pharmacodynamic and pharmacokinetic profiles. More interestingly, the inhibitory activity of these nanodevices can be regulated through precise nano-functionalization, making them the next generation therapeutics or "nano-drugs." Although, significant efforts have been made in developing different kinds of new TLR inhibitors/antagonists, only limited numbers of them have undergone clinical trials, and none have been approved for clinical uses to date. Nevertheless, these findings and continuous studies of TLR inhibition highlight the pharmacological regulation of TLR signaling, especially on multiple TLR pathways, as future promising therapeutic strategy for various inflammatory and autoimmune diseases.

Keywords: toll-like receptor (TLR), autoimmune diseases, inflammatory diseases, inflammation, TLR antagonist, TLR inhibitor, nanotherapeutics

## INTRODUCTION

Toll-like receptors (TLRs) are a family of pattern recognition receptors (PRRs) that can recognize and respond to a unique repertoire of distinct molecules referred to as pathogen-associated molecular patterns (PAMPs) and danger-associated molecular patterns (DAMPs; Kawai and Akira, 2010; Moresco et al., 2011). These conserved receptors belong to type I transmembrane

#### Edited by:

*Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico*

#### Reviewed by:

*Olalekan Michael Ogundele, Louisiana State University, United States Yasser Mohamed El-Wazir, Suez Canal University, Egypt*

#### \*Correspondence:

*Qiang Li liqressh@hotmail.com Hong Yang hongyang36@gmail.com*

#### Specialty section:

*This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology*

> Received: *23 April 2017* Accepted: *03 July 2017* Published: *19 July 2017*

#### Citation:

*Gao W, Xiong Y, Li Q and Yang H (2017) Inhibition of Toll-Like Receptor Signaling as a Promising Therapy for Inflammatory Diseases: A Journey from Molecular to Nano Therapeutics. Front. Physiol. 8:508. doi: 10.3389/fphys.2017.00508*

**224**

glycoproteins and are made of three structurally important components: (1) the leucine-rich repeat motifs for ligand recognition at N-terminus; (2) a single transmembrane helix; and (3) a conserved cytoplasmic Toll/Interleukin-1 (IL-1) receptor (TIR) domain at C-terminus for intracellular signaling transduction (Botos et al., 2011; Bryant et al., 2015). A representative TLR structure is shown in **Figure 1**. Currently, a total of 13 TLRs have been identified, among which TLRs 1–10 are expressed in human despite the function of TLR10 is still unclear (Moresco et al., 2011). The expression of TLRs can be found in a variety of immune cells (e.g., dendritic cells, monocytes, macrophages, and B lymphocytes) and non-immune cells (e.g., epithelial cells, endothelial cells, and fibroblasts; Takeda et al., 2003). While TLR1, TLR2, and TLR4–6 are mainly found on the cell surface, TLR3 and TLR7–9 are primarily expressed in the endosomes (Gay et al., 2014). Two TLRs form a hetero- or homo-dimer that allow them to recognize and bind to different molecular patterns (Gay et al., 2014).

When engaging with PAMPs or DAMPs, TLRs can transduce the signaling to initiate innate and adaptive immune responses (Akira and Takeda, 2004; Palm and Medzhitov, 2009). The molecular pathways of TLR signal transduction can be simply categorized into two cascades (Akira and Takeda, 2004; Kawasaki and Kawai, 2014). One is through the main adaptor protein, myeloid differentiation factor 88 (MyD88). The ligation of TLR ligands results in the recruitment of MyD88 to the TIR domain and subsequently leads to activation of nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) and mitogen-associated protein kinase (MAPK), which drive the expression of pro-inflammatory genes. The other is through the adaptor TIR-domain-containing adaptor-inducing interferon-β (TRIF), also known as the MyD88-independent pathway, to activate interferon (IFN) regulatory factors (IRFs) and NF-κB for the production of type I IFNs. Among all human TLRs, TLR3 exclusively signals through TRIF-dependent pathway while others use MyD88-dependent pathway; (Uematsu and Akira, 2007) notably, TLR4 can signal through both pathways in a timedependent manner (Sato et al., 2003; Yamamoto et al., 2003).

TLR activation is one of the first defensive mechanisms utilized by the host to mount innate and subsequent adaptive immune responses to fight the invading pathogens and repair the damaged tissues (Takeda and Akira, 2005; Beutler, 2009; Palm and Medzhitov, 2009). However, dysregulated TLR signaling could disrupt the immune homeostasis (i.e., a self-regulating process to remain physiological balance) in the host by sustained pro-inflammatory cytokines and chemokines secretion. This often contributes to the development of many inflammatory and autoimmune diseases, such as systemic lupus erythematosus (SLE), sepsis, atherosclerosis, and asthma (Marshak-Rothstein, 2006; Tsujimoto et al., 2008; Drexler and Foxwell, 2010). Considering the pathological role of TLRs in these inflammatory diseases, inhibitors targeting TLR signaling might be beneficial in treating these disorders (Hennessy et al., 2010).

Despite many efforts have been put in developing (bio)molecular inhibitors targeting TLR signaling pathways, unfortunately, very few compounds are currently available for clinical uses. Therefore, it is important to search for novel TLR inhibitors that can be utilized both as therapeutic agents and as experimental reagents for dissecting innate immune molecular pathways. Manipulation of immune responsiveness using nanodevices provides a new strategy to treat human diseases, because nanotechnology offers many beneficial features for therapeutics development, including cell penetrating and targeting capabilities, desired biodistribution, improved pharmacokinetics, enhanced therapeutic activity, better biocompatibility, and in vivo tracking (bioimaging). Accordingly, novel nanodevices have emerged with bio-activities to potently regulate TLR signaling, aiming for inhibiting disease-associated TLR activation.

In this review, we first briefly summarize the pathophysiological role of TLRs in many diseases of acute and chronic inflammation as well as autoimmunity. These diseases include SLE in autoimmunity, infection-induced sepsis for acute inflammation, vascular system disorders across both acute and chronic inflammation, and other inflammationassociated diseases such as psoriasis, gastrointestinal cancer, asthma, and chronic obstructive pulmonary disease (COPD; **Table 1**). We then focus on the current knowledge in developing TLR antagonists and inhibitors at various stages from preclinical evaluation to clinical trials, including the most recent development of nano-based modulators on inhibition of TLR signaling, and discuss their potential uses in treating various inflammatory diseases. Particularly, we describe the design of these nanodevices and discuss their mechanism(s) of action in TLR inhibition. Due to the unique feature of nanodevices, these nano-inhibitors hold great promises in treating various inflammatory and autoimmune diseases.

### TOLL-LIKE RECEPTOR ACTIVATION IN AUTOIMMUNE AND INFLAMMATORY DISEASES

### Systemic Lupus Erythematosus (SLE)

Systemic lupus erythematosus (SLE) is a complex autoimmune disease characterized by the loss of tolerance to self-nuclear antigens and the production of autoantibodies, leading to the attack of the immune system on the self-components and healthy tissues. Recent evidences have suggested a close relationship between the endosomal TLR activation and the onset of SLE (Celhar and Fairhurst, 2014; Wu et al., 2015). It has been found that in B cells and plasmacytoid dendritic cells (pDCs), endosomal TLRs play an essential role in the generation of anti-nuclear antibodies and type I IFNs (Clancy et al., 2016). Particularly, in lupus-prone mouse models, TLR7 overexpression is associated with the production of autoantibodies and the development of autoimmune phenotypes (Pisitkun et al., 2006; Subramanian et al., 2006); on the other hand, in the situation of TLR7 deletion the levels of circulating autoantibodies and inflammatory cytokines, such as IL-6 and INF-α, are significantly decreased with improved disease symptoms (Christensen et al., 2006; Lee et al., 2008; Kono et al., 2009). In contrast to TLR7, the role of TLR9 in SLE is not well-defined. It has been shown that TLR9 is indispensable in the autoantibody production in

B cells from several mouse studies (Christensen et al., 2006; Nickerson et al., 2010; Chen et al., 2011). However, the absence of TLR9 in these lupus-prone models does not help improve disease conditions, but leads to disease exacerbation (Christensen et al., 2006; Nickerson et al., 2010; Jackson et al., 2014). This suggests that TLR9 may have a regulatory role in the progression of SLE. Nevertheless, the expression of both TLR7 and TLR9 is up-regulated in the peripheral blood mononuclear cells (PBMC) from the SLE patients, and their levels are correlated with the production of IFN-α (Lyn-Cook et al., 2014), making them potential therapeutic targets of SLE.

In addition to endosomal TLRs, TLR2 and TLR4 have been found to participate in the pathogenesis of SLE based on the following evidences. First, their mRNA levels in PBMC of SLE patients are much higher than those in healthy controls (Komatsuda et al., 2008; Lee et al., 2016). Second, in a lupus mouse model, the deletion of TLR4 or TLR2 (to a lesser extent) decreases the levels of autoantibodies, and subsequently ameliorates the disease symptoms (Lartigue et al., 2009). Third, the elevation of TLR4 can induce lupus-like autoimmune disease (Liu et al., 2006). All these studies clearly demonstrated the importance of TLR signaling in SLE development and progression.

### Infection-Associated Sepsis

Due to their capability of recognizing a variety of PAMPs, TLRs play a critical role in fighting against pathogen infection (Mogensen, 2009). The first identified TLR4 can detect a wide range of Gram-negative bacteria by recognizing and responding to the coating molecules on the bacteria cell wall lipopolysaccharide (LPS). TLR2 (in conjunction with either TLR1 or TLR6) primarily recognizes lipoproteins/lipopeptides and glycolipids from various pathogens including microbes and fungi. TLR5 is the only TLR that identifies a protein ligand—flagellin from nearly all flagellated bacteria. On the other hand, the endosomal TLRs (TLR3, and TLR7-TLR9) are mainly responsible for the recognition of nucleic acid ligands, including dsRNA, ssRNA, and CpG-DNA, from viral and intracellular bacterial infection. Therefore, activation of TLRs by engaging with these PAMPs allows the host to combat invading pathogens. In fact, TLR agonists have long been considered as vaccine adjuvants to boost the immune system to eliminate viral infections and most recently fight cancer (O'Neill et al., 2009; Hedayat et al., 2011).

However, excessive and uncontrolled TLR activation can lead to overwhelming systemic inflammation and multiple organ injury, characterized as sepsis. Among all TLRs, TLR2, and TLR4 are the two notable contributors to the pathogenesis of sepsis (Tsujimoto et al., 2008). Evidence came from the observation that the expression of both TLR2 and TLR4 was up-regulated during their ligand stimulation in monocytes from healthy subjects (Wittebole et al., 2005), as well as in PBMC from sepsis patients (Harter et al., 2004). In addition, TLR4 knockout mice are resistant to Gram-negative bacteria-induced septic shock (Roger et al., 2009), and TLR2-deficient mice have increased survival rates compared to wild type mice in a polymicrobial sepsis model (Bergt et al., 2013). Moreover, blockade of TLR2 and TLR4 signaling by antagonistic antibodies successfully decreases disease severity in sepsis models of Gram-positive and Gramnegative bacteria, respectively (Meng et al., 2004; Daubeuf et al., 2007). Interestingly, the protective effect can also be seen in the TLR3- and TLR9-deficient mice against microbial challenge;


TABLE 1 | The role of TLR activation in the pathogenesis of inflammatory diseases.

*SLE, systemic lupus erythematosus; IR, ischemia-reperfusion; HSP, heat shock protein; COPD, chronic obstructive pulmonary disease; IBD, inflammatory bowel diseases.*

(Cavassani et al., 2008; Hu et al., 2015) and the pharmacological inhibitors of TLR9 can reduce the mortality of mice in severe sepsis and protect the host from serious complications, such as acute kidney injury (Yasuda et al., 2008; Liu et al., 2012). Since sepsis is one of the leading causes of death in the intensive care units worldwide, much effort has been put to develop therapeutic agents blocking TLR activation, especially TLR2 and TLR4, although none has yet been clinically successful (see Section Toll-Like Receptor Antagonists/Inhibitors and Their Clinical Applications).

### Vascular System Disorders

Not only can TLRs detect PAMPs in defending pathogen invasion, but they can also recognize circulating host-derived DAMPs released from the dying cells and damaged tissues to trigger the process of self-healing and tissue repair. Till recent years, studies have revealed that sensing the danger signals by TLRs could also contribute to the pathogenesis of many cardiovascular diseases, including atherosclerosis, hypertension, and stroke (Goulopoulou et al., 2016).

Atherosclerosis nowadays is considered as a chronic, progressive inflammatory condition corresponding to lipid accumulation in the vascular system than simply a lipid buildup disease. It has been found that the onset and ongoing of inflammation is mainly caused by the recognition of oxidized lipoproteins (as DAMPs) by TLRs expressed on innate immune cells (e.g., macrophages) and endothelial cells (Miller, 2005); TLR are also involved in the progression of the disease by lesion orientation (Edfeldt et al., 2002), collagen degradation (Monaco et al., 2009), and plaque destruction (Ishibashi et al., 2013). Specifically, it has been shown that deficiency in TLR2 or TLR4 results in diminishing inflammation in the prominent mouse models of atherosclerosis [apolipoprotein E (ApoE)- and LDL receptor-deficient mice; (Michelsen et al., 2004; Mullick et al., 2005, 2008; Higashimori et al., 2011)]. The activation of endosomal TLRs, on the other hand, seems to have controversial effects on atherosclerosis. TLR3 activation has been found to promote atherogenic inflammation, especially in mediating plaque instability (Ishibashi et al., 2013), while the absence of TLR9 exacerbates atherosclerosis in ApoE-deficient mice on a high-fat diet (Koulis et al., 2014).

In the case of hypertension, TLR4 has been well-documented to mediate aberrant immune and inflammation in vasculature, kidneys and autonomic nervous system (McCarthy et al., 2014). This is mainly because TLR4 can respond to angiotensin II, a DAMP, and cause subsequent vascular dysfunction and high blood pressure (Bomfim et al., 2015; Hernanz et al., 2015). Surprisingly, the evidence of TLR2 involved in hypertension is limited to inflammation in the renal system, where TLR2 activated NF-κB signaling is significantly correlated with renal ischemia/reperfusion injury (Khan et al., 2012). Recently, attentions have been drawn to TLR9 for its role in hypertension and blood pressure regulation. Studies have shown that TLR9 activation could elevate blood pressure in normotensive rats (McCarthy et al., 2015), and could also serve as a negative modulator of cardiac autonomic tone and baroreflex function (Rodrigues et al., 2015).

TLR2, TLR4, and endosomal TLRs have all been found to play an important role in stroke and cerebrovascular injury. Interestingly, most of them have a dual function along the time course of the disease: activation of TLRs in post-ischemia state would mediate neuroinflammation and neurodegeneration, but preconditioning with TLR stimulation could be neuroprotective from hypoxia and nutrient deprivation (Wang et al., 2011; Gesuete et al., 2014). For example, TLR7 and TLR8 expression is correlated to the aggravated neuroinflammation and poor outcome in acute ischemic stroke patients (Brea et al., 2011). On the other hand, TLR7 preconditioning [i.e., pretreatment with a TLR7 agonist, gardiquimod (GDQ), prior to ischemia] has shown protection against subsequent stroke injury through type I IFN-mediated mechanism (Leung et al., 2012).

### Other Inflammatory Diseases

Activation of TLRs is also involved in many other inflammationassociated diseases, such as psoriasis, gastrointestinal malignancies, and inflammatory airway diseases. Their relationships are briefly described hereafter.

Psoriasis is a long-lasting, relapsing inflammatory disease of dermis and epidermis. It has been observed that the expression of transforming growth factor (TGF)-α and heat shock proteins (e.g., HSP60) from keratinocytes can upregulate TLR5 and TLR9 (Miller et al., 2005), and activate TLR2 and TLR4, respectively (Seung et al., 2007). In addition, activation of TLR7/8 by its specific agonist imiquimod could also aggravate the disease symptoms (Gilliet et al., 2004).

It is known nowadays that chronic inflammation is often linked to carcinogenesis. In particular, chronic gastritis due to Helicobacer pylori infection and severe colitis of patients with inflammatory bowel diseases (IBD) greatly increase the risk of having gastrointestinal malignancies (Itzkowitz and Yio, 2004; Houghton and Wang, 2005), and TLR signaling is evidently involved in such inflammatory complications (Fukata and Abreu, 2008). It has been found that TLR4/MyD88 pathway can promote the development of colitis-associated colorectal tumors (Fukata et al., 2007). Since tumor cells also express TLR4, TLR4-induced uncontrolled production of immunosuppressive cytokines has been suggested to contribute to the tumor progression (Sato et al., 2009).

In inflammatory airway diseases, such as asthma and COPD, TLR signaling pathways are closely linked to the disease pathophysiology (Bezemer et al., 2012). For examples, house dust mites, one of the main risk factors for allergic asthma has been suggested to trigger airway neutrophils and monocytes through a TLR4-dependent mechanism (Li et al., 2010). Furthermore, the TLR4 agonist LPS is able to induce either Th1 or Th2 immune response in asthma in a dose dependent manner (Dong et al., 2009). In the case of COPD, TLR2, TLR4, and TLR9 have been shown to participate in cigarette smoke-induced inflammatory responses (Droemann et al., 2005; Karimi et al., 2006; Mortaz et al., 2010). During the exacerbation of the airway diseases, overactive TLR signaling can contribute to excessive inflammatory responses and lung tissue destruction (Tokairin et al., 2008; Stowell et al., 2009).

### TOLL-LIKE RECEPTOR ANTAGONISTS/INHIBITORS AND THEIR CLINICAL APPLICATIONS

Although, evolutionarily TLRs are essential elements in innate immune system and play a critical role in the host defensive mechanism against invading pathogens, overactivation of TLRs is inevitably involved in the pathogenesis of many inflammatory diseases. Thus, inhibition of TLR signaling pathways has been predicted to be an effective therapeutic strategy to suppress unwanted, disease-associated inflammatory responses. In general, TLR inhibition can be achieved by two major strategies: (1) blocking the binding of TLR ligands to the receptor; (2) interfering the intracellular signaling pathways to stop the signal transduction (**Figure 2**). Accordingly, various therapeutic agents for inhibiting TLR signaling have been developed to control excessive inflammation; they can be classified as small molecule inhibitors, antibodies, oligonucleotides, lipid-A analogs, microRNAs, and new emerging nano-inhibitors. The current advances of each class are summarized and discussed hereafter (**Tables 2**–**5**).

### Small Molecule Inhibitors (SMIs)

Small molecule inhibitors (SMIs) are synthetic or naturally derived chemical agents with the activity to inhibit TLR signal transduction. Their amphipathic property and small size allow them to cross cell membranes and act on specific intracellular adapter proteins or compartments along the TLR signaling pathways. They usually have good bioavailability but poor specificity and targeting capability. The ease of manufacturing and handling of SMIs makes them popular drug candidates in pharmaceutical industries (**Table 2**). In the clinical practice, a group of antimalarial drugs, including hydroxychloroquine sulfate (HCQ), chloroquine (CQ), and quinacrine (or mepacrine), have been used to treat autoimmune diseases (arthritis and SLE) after World War II (Fox, 1996; Ruiz-Irastorza et al., 2010). It was not until recent years that their additional mechanisms of action on endosomal TLR signaling (TLR7/8/9) was identified (Kuznik et al., 2011; Lee et al., 2011). These SMIs are all weak bases and tend to accumulate in the acidic intracellular compartments like endosomes and lysosomes, and are able to modulate the pH in these vesicles. The pH modulation can lead to suppression of autoantigen presentation, blockade of endosomal TLR signaling, and decrease in cytokine production (Fox, 1996; Wozniacka et al., 2006; Kuznik et al., 2011). Other mechanisms include inhibition of MAPK signaling and phospholipase A2, antiproliferation, photoprotection as well as reduction of matrix metalloproteinase-9 (MMP-9) activity (Loffler et al., 1985; Weber et al., 2002; Kim et al., 2006; Wozniacka et al., 2008; Lesiak et al., 2010). These mechanisms of action altogether suggest their anti-inflammatory and immuno-suppressive activity.

In addition to autoimmune diseases, these SMIs of TLR7/8/9 are also under investigation in other diseases associated with uncontrolled acute or chronic inflammation. For example, the antimalarial drug CQ has shown its clinical potential in treating severe sepsis, especially in reducing the risk of sepsis-induced acute renal failure (Yasuda et al., 2008). For its capability of blocking endosomal acidification and anti-inflammation, CQ has been applied to prevent and treat various viral infections, such as HIV, influenza, and dengue (Paton et al., 2011; Borges et al., 2013). In pre-clinical studies, CQ pretreatment can significantly improve the cerebral ischemia symptoms in a transient global cerebral ischemia rat model, which suggests that these SMIs of endosomal TLRs could be beneficial for patients with cardiovascular diseases (Cui et al., 2013). Moreover, longterm administration of HCQ has been found to ameliorate

agonists to the corresponding TLRs, and (2) inhibiting the intracellular signaling of the TLR pathways. The antibodies, lipid A analogs and oligonucleotides primarily target at the ligand-receptor binding, whereas the microRNAs (miRNAs) mainly act on the intracellular signaling cascades of TLR pathways; the small molecule inhibitors (SMIs) can inhibitor TLR signaling through both strategies.



*ARB, angiotensin II receptor blocker; CQ, chloroquine; HCQ, hydroxychloroquine; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; IND, investigational new drug application. <sup>a</sup>Study terminated.*

*<sup>b</sup>Unknown status.*

*<sup>c</sup>Completed with unknown results.*

<sup>1</sup>*Clinical drugs with new applications.*


*RA, rheumatoid arthritis.*

TABLE 4 | The development status of TLR antagonists/inhibitors: class of oligonucleotides.


*SLE, systemic lupus erythematosus.*

*<sup>a</sup>Study discontinued.*

*<sup>b</sup>Completed with unknown results.*



*SLE. systemic lupus erythematosus; NAHNP, non-anticoagulant heparin nanoparticle; HDL, high-density lipoprotein; NP, nanoparticle; GNP, gold nanoparticle. <sup>a</sup>Study discontinued.*

hypertension and aortic endothelial dysfunction in a lupus animal model (Gomez-Guzman et al., 2014). These evidences highlight the great potential of the early developed antimalarial drugs and their derivatives as endosomal TLR SMIs in treating a wide range of inflammatory and autoimmune diseases.

Following the success of these early antimalarial drugs, new SMIs have been developed to target endosomal TLR7/8/9 signaling. Among these newly developments, CpG-52364 is a quinacrine derivative, a SMI of TLR7/8/9 (Lipford et al., 2007). Comparing to HCQ, CpG-52364 is more therapeutically effective and has less side effects from animal studies; a phase I clinical trial was completed in 2009 for SLE treatment (Pfizer: NCT00547014), but the results have not been released since. Another exciting advancement is SM934, a novel analog of artemisinin (or "qinghao su," the new antimalarial drug identified in the early-70's in China), targeting TLR7/9. Studies have shown the therapeutic effects of the orally administered SM934 on lupus-prone MRL/lpr mice by inhibiting Th1 and Th17 cell responses as well as suppressing the B cell activation and plasma cell production (Hou et al., 2011; Wu et al., 2016). ST2825 is a peptidomimetic compound with an inhibitory activity on the MyD88 dimerization and recruitment of interleukin-1 receptor-associated kinase 1 (IRAK1) and IRAK4 (Loiarro et al., 2007); it has been found to suppress TLR9-induced B cell proliferation and differentiation into plasma cells as well as autoantibody production (Capolunghi et al., 2010). Although ST2825 is primarily used as an experimental agent, it provides evidence of inhibiting MyD88 (the universal adaptor used by most TLRs except TLR3) as a potential therapeutic target to treat patients with SLE and possible other autoimmune diseases.

In addition to endosomal TLR signaling, TLR2 and TLR4 are other two major targets of therapeutic SMIs. TAK-242 (Resatorvid) is an anti-sepsis SMI that targets TLR4 signaling pathways (Matsunaga et al., 2011). This compound binds to cysteine 747 in the intracellular TIR domain of TLR4, which blocks the interaction between TLR4 and the adaptor proteins-TIR domain containing adaptor protein (TIRAP) and TRIF-related adaptor molecule (TRAM), thereby diminishing LPS-induced TLR4 signaling and inflammation. Based on its preclinical success, TAK-242 advanced into clinical investigations. Two phase III clinical trials NCT00143611 for severe sepsis and NCT00633477 for sepsis-induced cardiovascular and respiratory failure were launched. In the first trial, the results were unfortunately not satisfactory due to failure to effectively suppress serum cytokine levels (IL-6, IL-8, and TNFα) comparing to controls, even though the drug was well-tolerated (Rice et al., 2010). The second trial, however, was terminated due to a business decision, and no further clinical development of TAK-242 has been conducted ever since.

Like antimalarial drugs, angiotensin II receptor blockers (ARBs), and statins are among the early developed drugs with newly discovered inhibitory activity on TLR2 and TLR4 signaling. For example, the ARB family valsartan can decrease proinflammatory cytokines release and infarct size by inhibiting TLR4 signaling (Yang et al., 2009) while candesartan can suppress Pam3CSK4 and LPS induced TLR2 and TLR4 activation, respectively (Dasu et al., 2009). Statins, also known as 3 hydroxy-3-methyl-glutaryl-coenzyme A (HMG-CoA) reductase inhibitors, are a class of lipid-lowering drugs. Among the statin family, fluvastatin, simvastatin, and atorvastatin all have shown potent inhibitory activity on TLR4 and subsequent inflammatory pathways to reduce inflammation in vascular systems (Methe et al., 2005; Foldes et al., 2008; Fang et al., 2014).

### Antibodies

Due to their high specificity, antibodies have been widely used as biological tools to precisely probe certain molecules. For therapeutic applications, they can be designed to neutralize soluble effectors, block the binding of receptors to their ligands to stop signal transduction, or induce targeted cytotoxicity (Suzuki et al., 2015). The most attractive feature of therapeutic antibodies is their superior specificity to the drug targets comparing to other type of drugs, and hence they are getting great attentions especially for immunotherapy nowadays. However, antibodies still possess a number of limitations including high costs in manufacturing, poor cellular and tissue penetration, and risks of immunogenicity (Chames et al., 2009).

In terms of inhibiting TLR signaling, antibodies are designed (as antagonists) to block the binding of ligands to the specific TLRs (**Table 3**). One promising advancement is OPN-305, which is the first fully humanized IgG4 monoclonal TLR2-specific antibody. Pre-clinical studies have shown the potency of this compound in blocking TLR2-mediated proinflammatory cytokine production in vitro and in ischemiareperfusion (IR) injury animal models (Ultaigh et al., 2011; Arslan et al., 2012; Farrar et al., 2012). The success of OPN-305 in phase I study suggests that OPN-305 is an effective and well-tolerated treatment for the prevention of IR injury in solid organ transplantation (Reilly et al., 2013). It is currently under a phase II clinical trial (NCT01794663) in renal transplant patients with a high risk of developing delayed graft function.

T2.5 is another anti-TLR2 antibody with therapeutic potential. It was found to increase animal survival and protect against severe shock-like syndrome in a mouse model challenged with Pam3CSK4 (a TLR1/2 ligand) or lethally challenged with Bacillus subtilis (Meng et al., 2004). A combination therapy of T2.5 with 1A6 (an anti-TLR4 antibody, see below) have shown beneficial effects in a sepsis mouse model challenged by S. enterica and E. coli (Spiller et al., 2008). In addition to sepsis models, T2.5 treatment was able to decrease neural death and inflammatory responses in a model of transient brain ischemia (Ziegler et al., 2011). Moreover, T2.5 was found to prevent angiotensin II-induced cardiac fibrosis through suppressing macrophage recruitment and inflammation in the heart (Wang et al., 2014). Altogether, these proof-of-principle studies suggest that blocking TLR2 signaling is therapeutically beneficial to sepsis and vascular system disorders. However, this compound was not moved forward into clinical development but remains as a widely used experimental tool now.

NI-0101 is the first monoclonal anti-TLR4 antibody entering clinical development. It can potently inhibit TLR4 signaling by blocking TLR4 dimerization, which is independent on the ligand type and concentration (Monnet et al., 2015). NI-0101 just completed a phase I clinical trial (NCT01808469) and the results were very encouraging (Monnet et al., 2017). NI-0101 was well-tolerated without safety concerns at various doses; it could successfully block cytokine release ex vivo and in vivo, and prevent flu-like symptoms upon LPS administration. Recent data demonstrated that NI-0101 could also block rheumatoid arthritis (RA) synovial fluids-induced pro-inflammatory cytokine production in monocytes isolated from RA patients (Hatterer et al., 2016). Now, the company is preparing for a phase II clinical trial in RA patients with a TLR4-driven pathotype in 2017.

Another monoclonal antibody of TLR4 is 1A6, which targets the extracellular portion of the TLR4-myeloid differentiation factor 2 (MD2, a co-receptor for TLR4 activation) complex (Spiller et al., 2008). In preclinical study, 1A6 showed protective effect on microbial-induced septic shock in vivo (Spiller et al., 2008; Lima et al., 2015). Interestingly, in a mouse model of dextran sulfate sodium (DSS)-induced colitis, 1A6 could ameliorate inflammation and prevent the progression of the intestine inflammation; however, the blockade of TLR4 signaling by 1A6 could also cause a defect in the mucosal healing during the recovery stage (Ungaro et al., 2009). This study perhaps sent a warning message of blocking early TLR4 signaling as potential therapy for inflammatory bowel diseases (IBD); on the other hand, it provided a strong evidence of multiple roles of TLRs in disease development, progression and remission, guiding the development of next generation of therapeutic TLR inhibitors.

### Oligonucleotides

Since endosomal TLRs primarily recognize nucleic acid structures from dsRNA, ssRNA, and CpG-DNA, oligonucleotides with certain sequences are predicted to function as antagonists of endosomal TLRs. They can interfere the binding of endosomal TLRs to their ligands, and hence block the TLR signal transduction. Accordingly, several oligonucleotidebased TLR antagonists have been developed particularly to treat inflammatory diseases associated with endosomal TLR activation, such as SLE; each is discussed hereafter (**Table 4**).

One of the leading developments of oligonucleotide-based TLR antagonists is named immunoregulatory DNA sequence (IRS; by Dynavax Technologies). Advancing from the earlier encouraging prototypes, IRS-661 (TLR7 specific) and IRS-869 (TLR9 specific), a dual-function targeting both TLR7 and TLR9 antagonist, IRS-954 was developed for the treatment of SLE (Barrat et al., 2005). IRS-954 was found to inhibit IFNα production by human plasmacytoid predendritic cells in response to DNA or RNA of viruses and immune complexes from SLE patients. In addition, IRS-954 treatment showed a reduction of autoantibodies and improvement of disease symptoms in lupus-prone mice (Barrat et al., 2007). Furthermore, it could also reverse TLR7/9-mediated glucocorticoid resistance of SLE, and thus could be potentially used as corticosteroid-sparing drug (Guiducci et al., 2010). Although IRS-954 has shown promising pre-clinical therapeutic efficacy, there is no further advancement documented in the clinical development to date. On the other hand, DV-1179, another TLR7/9 dual antagonist developed by the same company, entered a phase Ib/IIa study (in partner with GSK) for its safety in both healthy volunteers and patients with active SLE (Dynavax, 2014). Although DV-1179 administration was well tolerated, it failed to achieve the endpoints of reducing the IFN-α-regulated genes<sup>1</sup> . Such an unfortunate outcome ended the clinical development of this compound ever since.

Another development of oligonucleotide-based antagonists (also termed as immune modulatory oligonucleotide IMO) has made some success into clinical trials (by Idera Pharmaceuticals). The compound IMO-3100 is also a TLR7/9 dual antagonist. In preclinical studies, IMO-3100 could significantly reduce the expression of inflammatory genes, such as IL-17A, β-defensin,

<sup>1</sup>http://www.streetinsider.com/Corporate+News/...DV1179.../9729788.html

CXCL1, and keratin 16 in a mice model of IL-23-induced psoriasis (Suarez-Farinas et al., 2013). Moreover, this compound showed an inhibitory activity on disease progression in a lupusprone mice model (Zhu et al., 2011). A phase II trial for the treatment of psoriasis was completed in 2012 (NCT01622348), but the results have not yet been disclosed. IMO-8400 is another candidate under current development. Different from IMO-3100, IMO-8400 is capable of inhibiting all TLR7, TLR8, and TLR9. This compound was very effective in preventing inflammation and disease development in both lupus and psoriasis mouse models (Jiang et al., 2012; Zhu et al., 2012; Suarez-Farinas et al., 2013). Currently, two phase II clinical trials are underway for the treatment of plaque psoriasis (NCT01899729) and dermatomyositis (NCT02612857). Similar to IMO-8400, a newly developed IMO-9200 can also target TLR7/8/9. From a phase I clinical trial (in partner with Vivelix), it showed a safe and well-tolerated profile in healthy subjects<sup>2</sup> .

There are other immunoregulatory oligonucleotides under preclinical developments. The inhibitory oligonucleotide (IHN-ODN) 2088 is a potent TLR9 antagonist. In spontaneously hypertensive rats, IHN-ODN 2088 treatment showed positive effects on systolic blood pressure reduction (McCarthy et al., 2015). Based on IHN-ODN 2088, a guanine-modified inhibitory oligonucleotide, INH-ODN-24888, was designed; this compound demonstrated a promising therapeutic effect in SLE through the suppression of both TLR7 and TLR9 (Rommler et al., 2015). The guanine modification potentiates the inhibitory activity of INH-ODN-24888 and efficiently improves its capacity to reduce TLR7/9-mediated immune signaling in human immune cells (Rommler et al., 2013). These results indicates that IHN-ODN-24888 may be further developed into a promising clinical treatment for SLE.

### Lipid A Analogs

A special group of TLR antagonists is the lipid A analog (**Table 5**), specifically targeting TLR4. Lipid A is a lipid component of LPS (or endotoxin) contributing to the toxicity of an endotoxin molecule. Because the structure of lipid A is highly conserved in endotoxins, it becomes an attractive therapeutic target to regulate TLR4 signaling (Rietschel et al., 1994).

A synthetic lipid A analog of Rhodobacter sphaeroides, eritoran (E5564), is probably the most exciting but also disappointing clinical development of TLR4 antagonist (by Eisai Research Institute of Boston Inc.) (Barochia et al., 2011). Eritoran is the 2nd generation of its own kind derived from the first developed E5531 (Christ et al., 1995). The design principle of these compounds was to competitively bind to the MD2 pocket by mimicking the lipid A structure, which in turn prevents the LPS binding and the induction of TLR4 signaling (Park et al., 2009). The structural improvement of eritoran makes it superior to E5531 for its higher potency due to extended duration of action, cost-effective manufacturing because of a simpler structure, greater water solubility, and better chemical stability. Preclinically, eritoran could significantly reduce LPS-induced NF-κB activation and pro-inflammatory cytokine production (TNF-α, IL-1β, IL-6, and IL-8) in vitro and in animal models (Mullarkey et al., 2003; Savov et al., 2005). In a phase I clinical study on LPS challenged healthy volunteers, eritoran treatment effectively decreased TNF-α and IL-6 levels in the blood, reduced the white blood cell counts and C-reactive protein levels, and diminished sepsis-associate clinical symptoms; the only observed adverse response was the dose-dependent phlebitis (Lynn et al., 2004; Rossignol et al., 2004, 2008). It then went on a randomized, double-blind, multicenter phase II trial (NCT00046072) to treat critically ill septic patients with high predicted risk of mortality; the results were positive as the intravenous treatment of eritoran at the dose of 105 mg led to a trend toward a lower mortality rate (Tidswell et al., 2010). Sadly, the phase III trial (ACCESS, NCT00334828) of eritoran on severe sepsis patients failed because eritoran did not reduce the patient mortality at 28 days and 1 year from the statistical analysis (Opal et al., 2013). Accordingly, the company decided to give up launching the drug to the market.

Although the development of eritoran led to a disappointing end, valuable lessons can be learned from this failed practice for developing the next generation of TLR inhibitors. First, the patient recruitment is key to objectively evaluate the drug efficacy in a clinical trial. One major reason that eritoran failed in the ACCESS was probably because the recruited patients were not well-assessed according to their disease pathogenesis, particularly the circulating levels of LPS in the patients. This is a critical factor that should be included in the recruiting criteria. Since the mechanism of eritoran is to block LPS binding to TLR4, those patients who had low levels of circulating LPS at the time of studies may not respond well to the treatment, and thus contributed to an overall null efficacy of the drug. Second, the time and route of drug administration need to be optimized. Considering the acute inflammatory responses in sepsis, the timing of eritoran administration could be too late to be therapeutic effective when other inflammatory mediators started to take over. Third, the genetic backgrounds, the infection status, and the disease severity all contributed to the heterogeneity in the patients, which ultimately resulted in different responses toward the treatment. Finally, the inflammatory responses of the septic patients could be triggered by multiple TLR pathways, and blocking one of them (even though TLR4 has been considered the major one) may not be enough to stop the overwhelming inflammation in severe sepsis. To ensure future success in developing new TLR inhibitors for inflammatory diseases like sepsis, perhaps new inhibitory strategy on multiple TLR pathways should be considered; and clearly in the clinical trials, patients should be carefully selected and examined based on appropriate criteria with well-defined biomarkers to obtain the fair evaluation and interpretation of the therapeutic efficacy of the tested drugs.

Apart from treating sepsis, eritoran could still be therapeutically beneficial for other inflammatory diseases. In fact, one study has shown that eritoran could prevent influenza-induced death in a mouse model, where it improved the clinical symptoms, inhibited cytokine production (TNF-α, IL-1β, CXCL1, and IL-6), and reduced viral titers (Shirey et al., 2013). Other studies have shown the therapeutic efficacy

<sup>2</sup>http://vivelix.com/research-and-development/

of eritoran in several cardiovascular disease models. For example, eritoran could reduce cardiac hypertrophy in an aortic constriction mouse model (Ehrentraut et al., 2011), and attenuate inflammatory cytokine production and myocardial IR injury in a rat model (Shimamoto et al., 2006). In addition, eritoran has also provided protective effects on kidney IR injury (Liu et al., 2010). Unexpectedly, it has also been suggested that eritoran could potentially overcome the blood-brain barrier under certain inflamed conditions (like stoke) and play an advantageous role in treating cerebral infarction (Buchanan et al., 2010).

### miRNA Inhibitors

MicroRNAs (miRNAs) are small (∼22 nucleotides), endogenous non-coding RNAs with post-transcriptional regulatory functions to finely tune gene expression (Bartel, 2009). They bind to a target sequence in the 3′ -untranslated region (UTR) of the messenger RNA (mRNA) to facilitate mRNA degradation or inhibit its translation. Since miRNAs are involved in the regulation of diverse biological processes and associated with pathogenesis of many diseases, they have emerged as new therapeutic targets.

Till now, there are about 20 miRNAs identified to be involved in the regulation of TLR signaling pathways (He et al., 2014). Among them, miR-146a, miR-155, and miR-21 are the three miRNAs received extensive attention for their regulatory roles in TLR signaling and autoimmune diseases (**Table 5**; Quinn and O'Neill, 2011; Shen et al., 2012). MiR-146a was identified from a screen of LPS-responsive genes regulated by NF-κB; it was found to inhibit the translation of the TNF receptorassociated family 6 (TRAF6) and IRAK1, both of which are in the downstream pathway of TLR4 signaling (Taganov et al., 2006). Further studies have shown that miR-146a could play a negative regulatory role in autoimmunity (Boldin et al., 2011), particularly in blocking the type I IFN production in human lupus by targeting TLR-type I IFN associated genes TRAF6, IRAK1, IFN regulatory factor 5 (IRF5), and signal transducer and activator of transcription 1 (STAT1; Tang et al., 2009). Given the fact that miR-146a acts on multiple targets in the signaling cascades of TLR-mediated induction of type I IFN, the integrated effects of negative regulation by miR-146a on the pathway are expected to be beneficial in treating SLE and perhaps RA. To translate miR-146a into clinic uses, one possible way is to apply a delivery vehicle to help miR-146a cross the cell membranes in order to function intracellularly (Pan et al., 2012).

Both miR-21 and miR-155 were originally identified as "onco-miR," but later discovered to have a new role in regulating TLR signaling and inflammation. They were found to be upregulated in TLR-associated inflammatory conditions (O'Connell et al., 2007; Sheedy et al., 2010). Specifically, miR-21 could enhance the LPS-induced anti-inflammatory cytokine IL-10 production by lowering programed cell death 4 gene (PDCD4) expression, serving as a negative regulator of the inflammatory response (Sheedy et al., 2010). MiR-155, on the other hand, appears to have controversial roles in controlling inflammation. In one study, miR-155 targeted TGFβ activated kinase 1 (TAK1) binding protein 2 (TAB2) and inhibited TAK1 activation, which subsequently blocked the activation of downstream NF-κB and MAPK pathways, and dampened the inflammatory responses (Ceppi et al., 2009). Another study reported that miR-155 expression contributed to the production of the proinflammatory cytokine TNF-α and increased the susceptibility to septic shock (Tili et al., 2007). In addition, miR-155 was found to have a positive effect in the pathogenesis of SLE, as miR-155 knockout mice had phenotypes similar to some human systemic autoimmune disorders, and were highly resistant to experimental autoimmune encephalomyelitis, a SLE mouse models (O'Connell et al., 2010). Accordingly, a miR-155 antagonist, MRG-107, is currently under preclinical development (by miRagen Therapeutics) to treat neuro-inflammation.

### New Emerging Nano-Inhibitors

In addition to above described inhibitors/antagonists, nanodevices are emerging as a new class of potent TLR inhibitors that can target a single or multiple TLR pathways (**Table 5**). Due to their nanoscale sizes, nano-inhibitors are expected to have better bio-distribution and sustained circulation (He et al., 2010; Blanco et al., 2015). They can be functionalized to meet the desired pharmacodynamic and pharmacokinetic profiles (Ernsting et al., 2013; Lin and Tam, 2015). One distinct feature of these new nano-inhibitors is that the bio-activity comes from their self-properties, which can be tailored to different medical complications, rather from a carried therapeutic agent. This characteristic makes them a special class of drug (or "nanodrug") as the next generation nano-therapeutics. Currently, they are at the preclinical investigation stage. **Figure 3** summarizes their targets in the TLR pathways, whereas their structures, physicochemical properties and mechanisms of action are listed in **Table 6**.

One interesting immunomodulatory nanodevice is the lipidconjugated non-anticoagulant heparin nanoparticle (NAHNP; Babazada et al., 2014a,b). In addition to its well-known anticoagulant function, heparin is known to have anti-inflammatory activity for its ability to bind and inhibit various cytokines, growth factors and enzymes that are involved in inflammation (Tyrrell et al., 1999); especially, the 6-O sulfation of glucosamine residues of heparin can block the binding of P- and L-selectins for the recruitment of leukocytes at inflammatory sites (Wang et al., 2002). Being intrigued by the intrinsic anti-inflammatory properties of heparin, Babazada et al. synthesized the glycol-split non-anticoagulant heparin with D-erythro-sphingosine grafts, which can self-assemble into NAHNP (Babazada et al., 2014b). These nanoparticles (NPs) were found to suppress LPS-induced MyD88-dependent NF-κB activation and inhibit subsequent cytokines production in mouse macrophages. They further discovered that the length of the alkyl chains of conjugated D-erythro-sphingosine was critical for the anti-inflammatory activity as shortening the alkyl chain of NAHNP resulted in loss of activity. Again, the 6-O-sulfate groups of d-glucosamine residue were important for such an inhibitory activity. By removing the anti-coagulant activity of heparin, NAHNP could serve as new anti-inflammatory nano-therapeutics.

Another novel nanodevice, the high-density lipoprotein (HDL)-like nanoparticle (HDL-like NP), was developed as a TLR4 antagonist by sequestering LPS (Foit and Thaxton, 2016). HDL has been known to naturally bind to and neutralize LPS (Brandenburg et al., 2002), and has been applied to reduce LPSinduced inflammation in vivo (Pajkrt et al., 1996; Guo et al., 2013). The HDL-like NP was designed with a gold NP core and a HDL coating, where the lipid components of HDL can be modified (Foit and Thaxton, 2016). Through screening five variants of the lipid coatings, one HDL-like NP was identified to be able to potently inhibit TLR4 signaling triggered by various sources of LPS and Gram-negative bacteria on human cell lines and PBMC. As expected, the acting mechanism of this HDLlike NP was through scavenging the LPS molecules. It is worth noting that this newly synthesized HDL-like NP exhibits much higher anti-inflammatory activity than the HDL alone, making it a promising candidate as a next generation nano-antagonist of TLR4.

Gold nanoparticles (GNPs) have caught much attention as emerging nanotheranostics in nanomedicine, owing to their ease of fabrication and functionalization, chemical stability, excellent photothermal physical properties, and good biocompatibility (Pedrosa et al., 2015). The fact that a TNF-α-bound GNP agent (CYT-6091 by CytImmune Sciences) showed very-well tolerated profile in a phase I clinical trial warrants the potential clinical uses of GNP-based therapeutics in the near future (Libutti et al., 2010). The exciting advancement in the field is the development of GNP-based "nano-inhibitors" for TLR signaling. In one study, bare GNP was applied as the posttreatment topically to ameliorate LPS-induced eye inflammatory responses and reduce the oxidative damage in irides (Pereira et al., 2012). The possible mechanism was through downregulating TLR4 expression and the corresponding LPS-induced NF-κB activation. This discovery is quite surprising as the bare GNP is physiologically unstable, and is usually thought to be inert without biological activity. It would be interesting to see the follow-up study and the detailed working mechanisms of bare GNP. In another study, a cationic glycolipid-coated GNP system was developed as a new TLR4 antagonist (Rodriguez Lavado et al., 2014). A new class of glycolipids was synthesized to have a structure mimicking lipid A binding to CD14 and the TLR-MD2 pocket. The biological analysis showed that the GNP coated with a specific glycolipid (11-NP) was able to inhibit the LPS-induced TLR4-MD2 activation in human and murine cells; such an effect was similar to the very efficient TLR4 antagonist eritoran. Further studies on the structure-activity relationship analysis indicated that the presence of fatty ester chains and the facial amphiphilicity of the glycolipids are required to maintain TLR4 antagonistic activity. Preliminary in vivo studies showed that 11-NP was very effective in reducing TNF-α level in the blood in a LPS mouse model.

One exciting invention is the development of a completely novel peptide-GNP hybrid system as a next generation nanoinhibitor led by Yang et al. (2015, 2016). The system is made of a GNP core coated with a hexapeptide layer to enhance the physiological stability of the GNP, change the surface physicochemical characteristics, and enable the biological activity (Yang et al., 2011, 2013). Through, screening a small library of the established, physiologically stable peptide-GNP hybrids, they discovered a potent multi-TLRs inhibitor, P12, which not only suppressed both arms of TLR4 signaling (i.e., MyD88 dependent NF-κB and TRIF-dependent IRF3 activation), but also inhibited TLR2, TLR3, and TLR5 pathways. The structureactivity relationship analysis revealed that the activity index is strongly dependent on the surface properties of the hybrid, where the hydrophobicity and the aromatic ring-like structure are the two key factors dictating the inhibitory activity (Yang et al., 2015). Further studies conferred the anti-inflammatory activity of P12 in correcting LPS-induced gene expressions, reducing the subsequent pro-inflammatory cytokine production (IL-12p40, MCP-1, and IFN-γ), and increasing the anti-inflammatory cytokine IL-1RA. In searching for the mechanism of action, delicate experiments were performed to show that the potent inhibitory activity of P12 is mainly due to its endosomal pH modulation capability (Yang et al., 2016). Like chloroquine (CQ), a lysotropic agent that can elevate endosome/lysosome pH, P12 was able to prevent the endosomal acidification process, which in turn attenuated downstream signal transduction of endosomal-dependent TLR signaling. More importantly, the inhibitory activity of the "nano-drug" can be well-tuned by

cytokines and type I interferons.


*NAHNP, non-anticoagulant heparin nanoparticle; HDL, high-density lipoprotein; NP, nanoparticle; GNP, gold nanoparticle; Apo A1, apolipoprotein A1; N/A, not available.*

selecting different amino acids that are displayed on the GNPs. The in vivo efficacy of P12 was investigated using a murine model of intestinal inflammation; it was found that P12 treatment could reduce the animal weight loss, improve the disease activity index, and ameliorate colonic inflammation.

All the above nanoparticles have minimum acute toxicity in culture cells as well as in the limited animal studies. However, since the gold nanoparticle is not biodegradable, the longterm safety issue of using GNP-based therapeutics is still a concern, even though they are well-tolerated in a phase I study. Therefore, addressing the long-term safety problem of GNPbased TLR inhibitors is critical in order to move this line of research into clinic uses. Alternatively, one could replace the GNP core with a biodegradable NP while maintaining the inhibitory activity, preferably targeting multiple TLRs like peptide-gold nanoparticle hybrids, to accelerate the translation of these novel agents into clinical treatments for inflammatory diseases.

Note that the majority of these developed TLR nanoinhibitors act on TLR4 signaling, except the peptide-gold nanoparticle hybrid P12 which has shown inhibitory activity on multiple TLRs. In the current literatures reviewed, LPS has been used universally as the TLR4 agonist. However, other purinergic receptors, such as P2X7 receptor (encoded by P2RX7 gene), can also recognize LPS and induce the secretion of inflammatory mediators in macrophages (Denlinger et al., 2001). Interestingly, the recognition of LPS by P2X7 requires a coreceptor CD14, which facilitates LPS internalization and binding to the intracellular C-terminal domain of P2X7 (Dagvadorj et al., 2015). Therefore, the specificity of these nanoparticles in inhibition of TLR4 vs. other membrane receptors that also recognize LPS needs to be further addressed. It will also be interesting to see whether these nano-inhibitors can also modulate signaling of other TLRs in addition to TLR4.

Currently, these TLR nano-inhibitors are still at the preclinical investigation stage. Their pharmacokinetics and pharmacodynamics profiles in small and large animals need to be accomplished before launching any clinical trials. However, a few nanoparticles/nano-formulations have already been approved by FDA to deliver water-insoluble drugs or biologics, and have shown enhanced efficacy and safety profiles in cancer treatments. Some well-known examples include the polyethylene glycol (PEG) formulated L-asparaginase (Oncaspar by Sigma-Tau Pharmaceuticals, Inc.) and liposome formulated doxorubicin (Doxil by Jansen Pharmaceutical). Furthermore, several inorganic nanoparticles have been approved by FDA as diagnostic imaging agents. For instance, the iron oxide nanoparticles (Feridex by Bayer HealthCare Pharmaceuticals Inc.) have been used in medical magnetic resonance imaging (MRI). Recent records of clinically used nanoparticles are summarized in two comprehensive reviews in 2016 (Anselmo and Mitragotri, 2016; Bobo et al., 2016). Different from the clinically used nano-therapeutics, the reviewed TLR nanoinhibitors herein would represent a new class of "nano-drug" modulating TLR signaling, and hold great promise for clinical use in the future.

## CONCLUDING REMARKS

TLRs are very important PRRs in the first-line defense system of our bodies. They can recognize both invading pathogens and endogenous danger molecules released from dying cells and damaged tissues, and play a key role in linking innate and adaptive immunity. Despite their wellknown function in sensing pathogens and mounting defense mechanisms, overactivation of TLRs can ultimately lead to disruption of immune homeostasis, and thus increase the risk for inflammatory diseases and autoimmune disorders, such as SLE, RA, atherosclerosis, stoke, IR-induced injury, COPD, asthma, gastrointestinal inflammation, and even cancer. Evidences have shown the involvement of TLRs in the development and progression of these diseases. Accordingly, antagonists/inhibitors targeting TLR signaling pathways have emerged as novel therapeutics to treat these diseases. Although, this therapeutic strategy has been well-accepted experimentally, there are only limited numbers of TLR inhibitors/antagonists that have undergone clinical trials to date; many of them failed during the trials, and none have been approved for clinical uses. Lessons have been learned from these failed developments. The efficacy, safety, and stability of the new therapeutic agents, as well as whether these therapies could impair local immune defenses should be considered comprehensively in the clinical development. More importantly, given the facts that pathogen infection or danger signals often activate multiple TLRs, and that the failure of TLR4 specific antagonist/inhibitor in the clinical trials despite its success in preclinical studies, developing more potent antagonists/inhibitors that target multiple TLR signaling pathways perhaps could shine the way to facilitate the translation of such a promising strategy into clinical uses in the future. Certainly, nano-devices will play an important role in the development of next generation TLR inhibitors.

### AUTHOR CONTRIBUTIONS

QL and HY proposed and designed the scope of the review. WG and HY wrote the manuscript. YX and QL critically revised and commented on the manuscript. All the authors read and approved the final manuscript and agreed to be accountable for the accuracy and integrity of any part of the manuscript.

### ACKNOWLEDGMENTS

QL would like to thank the support from the National Natural Science Foundation of China (Grant No. 81170060/H0111). HY would like to acknowledge the support from the Program for Professor of Special Appointment (Eastern Scholar) at Shanghai Institutions of Higher Learning (TP2016014), the starting fund from Shanghai First People's Hospital, Gaofeng Clinical Medicine Grant support from Shanghai Jiaotong University School of Medicine, and the funding from the Crohn's and Colitis Foundation of Canada (CCFC).

### REFERENCES


from sepsis-induced acute kidney injury. Am. J. Physiol. Renal Physiol. 294, F1050–F1058. doi: 10.1152/ajprenal.00461.2007


of inflammatory cells and neuronal injury in experimental stroke. J. Cereb. Blood Flow Metab. 31, 757–766. doi: 10.1038/jcbfm.2010.161

**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 Gao, Xiong, Li and Yang. 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.

# LASSBio-897 Reduces Lung Injury Induced by Silica Particles in Mice: Potential Interaction with the A2A Receptor

Vinicius F. Carvalho<sup>1</sup> \*, Tatiana P. T. Ferreira<sup>1</sup> , Ana C. S. de Arantes<sup>1</sup> , François Noël2,3 , Roberta Tesch3,4, Carlos M. R. Sant'Anna<sup>5</sup> , Eliezer J. L. Barreiro3,4, Carlos A. M. Fraga<sup>3</sup> , Patrícia M. Rodrigues e Silva1,3 and Marco A. Martins1,3

<sup>1</sup> Laboratório de Inflamação, Instituto Oswaldo Cruz, Fundação Oswaldo Cruz, Rio de Janeiro, Brazil, <sup>2</sup> Laboratório de Farmacologia Bioquímica e Molecular, Instituto de Ciências Biomédicas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil, <sup>3</sup> Programa de Pós-Graduação em Farmacologia e Química Medicinal, Instituto de Ciências Biomédicas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil, <sup>4</sup> Laboratório de Avaliação e Síntese de Substâncias Bioativas, Instituto de Ciências Biomédicas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil, <sup>5</sup> Departamento de Química, Instituto de Ciências Exatas, Universidade Federal Rural do Rio de Janeiro, Rio de Janeiro, Brazil

#### Edited by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

#### Reviewed by:

Sergey V. Ryzhov, Maine Medical Center, United States Pallavi R. Devchand, Icahn Institute for Genomics and Multiscale Biology at Mount Sinai, United States

> \*Correspondence: Vinicius F. Carvalho vfrias@ioc.fiocruz.br; viniciusfrias@hotmail.com

#### Specialty section:

This article was submitted to Inflammation Pharmacology, a section of the journal Frontiers in Pharmacology

Received: 11 July 2017 Accepted: 16 October 2017 Published: 31 October 2017

#### Citation:

Carvalho VF, Ferreira TPT, de Arantes ACS, Noël F, Tesch R, Sant'Anna CMR, Barreiro EJL, Fraga CAM, Rodrigues e Silva PM and Martins MA (2017) LASSBio-897 Reduces Lung Injury Induced by Silica Particles in Mice: Potential Interaction with the A2A Receptor. Front. Pharmacol. 8:778. doi: 10.3389/fphar.2017.00778 Silicosis is a lethal fibro-granulomatous pulmonary disease highly prevalent in developing countries, for which no proper therapy is available. Among a small series of N-acylhydrazones, the safrole-derived compound LASSBio-897 (3-thienylidene-3, 4 methylenedioxybenzoylhydrazide) raised interest due to its ability to bind to the adenosine A2A receptor. Here, we evaluated the anti-inflammatory and anti-fibrotic potential of LASSBio-897, exploring translation to a mouse model of silicosis and the A2A receptor as a site of action. Pulmonary mechanics, inflammatory, and fibrotic changes were assessed 28 days after intranasal instillation of silica particles in Swiss–Webster mice. Glosensor cAMP HEK293G cells, CHO cells stably expressing human adenosine receptors and ligand binding assay were used to evaluate the pharmacological properties of LASSBio-897 in vitro. Molecular docking studies of LASSBio-897 were performed using the genetic algorithm software GOLD 5.2. We found that the interventional treatment with the A2A receptor agonist CGS 21680 reversed silica particle-induced airway hyper-reactivity as revealed by increased responses of airway resistance and lung elastance following aerosolized methacholine. LASSBio-897 (2 and 5 mg/kg, oral) similarly reversed pivotal lung pathological features of silicosis in this model, reducing levels of airway resistance and lung elastance, granuloma formation and collagen deposition. In competition assays, LASSBio-897 decreased the binding of the selective A2A receptor agonist [3H]-CGS21680 (IC<sup>50</sup> = 9.3 µM). LASSBio-897 (50 µM) induced modest cAMP production in HEK293G cells, but it clearly synergized the cAMP production by adenosine in a mechanism sensitive to the A2A antagonist SCH 58261. This synergism was also seen in CHO cells expressing the A2A, but not those expressing A2B, A<sup>1</sup> or A<sup>3</sup> receptors. Based on the evidence that LASSBio-897 binds to A2A receptor, molecular docking studies were performed using the A2A receptor crystal structure and revealed possible binding modes of LASSBio-897 at the orthosteric and allosteric sites. These findings highlight LASSBio-897 as a lead compound in drug development for silicosis, emphasizing the role of the A2A receptor as its putative site of action.

Keywords: A2A receptor, cAMP, fibrosis, LASSBio-897, silicosis

## INTRODUCTION

fphar-08-00778 October 28, 2017 Time: 16:21 # 2

Silicosis is one of the most important occupational diseases in both developed and developing countries, and is characterized by an irreversible inflammatory process into the lungs caused by inhalation of free crystalline silica (Leung et al., 2012). Although the prevalence of silicosis is decreasing in some developed countries (Bang et al., 2015), this disease is still very common in developing nations (Nelson et al., 2010), particularly in mining countries (Rees and Murray, 2007; Tse et al., 2007). In Brazil, a prevalence of 20 new annual cases of silicosis per 100,000 exposed people was reported (Ministry of Health, 2010). Silicosis pathogenesis is described as a complex interaction between different cell types and inflammatory mediators, leading to nodular lesions formation, fibrosis and reduction of lung elasticity (Ferreira et al., 2013; Franklin et al., 2016). After phagocytosis of silica particles, macrophages orchestrate this complex response through the release of inflammatory and fibrogenic cytokines. These mediators stimulate granuloma formation and lung fibroblasts to produce extracellular matrix components, culminating to the development of fibrosis (Fazzi et al., 2014). Once fibrosis has developed, there is no effective treatment for silicosis. Thus, the search for new compounds becomes indispensable for the management of this disease.

3,4-Methylenedioxybenzoyl-3-thienylhydrazone (LASSBio-897) (**Figure 1**) is a new N-acylhydrazone bioactive derivative synthesized from safrole, a Brazilian natural product that could be obtained in high yield from sassafras oil. LASSBio-897 was designed as a regiomeric analog of 3,4-methylenedioxybenzoyl-2-thienylhydrazone (LASSBio-294) (Gonzalez-Serratos et al., 2001), through a substitution of 2-thienyl ring to the bioisosteric 3-thienyl ring (Zapata-Sudo et al., 2010). LASSBio-294 induces vasodilatation in rat aortic rings (Silva et al., 2002), reduces blood pressure in spontaneously hypertensive rats (SHR), improves diastolic activity in myocardial infarction rats and inhibits collagen deposition and heart fibrosis (Costa et al., 2010). LASSBio-897 was about 200-fold more potent than LASSBio-294 causing vasodilation of rat aortic rings and significantly reduced blood pressure values in SHR (Zapata-Sudo et al., 2010). While trying to assess the spectrum of pharmacological activity in multiple targets, we submitted 10 µM LASSBio-897 to a screening of targets in the Cerep's "Diversity Profile" platform (Poitiers, France). Under this condition, it displayed an inhibitory activity ≥ 30% at only 3 of the 98 potential targets investigated, which were A2A receptor (72%), 5-HT transporter (56%), and NE transporter (42%).

The A2A receptor is an adenosine receptor that is coupled to Gs-protein and its activation results in increased levels of intracellular cAMP (Antonioli et al., 2015). Adenosine is a purine nucleoside involved with preservation and restoration of tissue homeostasis, in several sites, including lung (Chen et al., 2013). Adenosine receptors, such as A2A receptors, are widely distributed in various organs, including lungs, as well as fibroblasts and immune cells (Factor et al., 2007; Koupenova and Ravid, 2013; Eisenstein and Ravid, 2014). Activation of

A2A receptors induces a strong anti-inflammatory response by reduction of adherence, chemotaxis and phagocytosis of neutrophils, and reduction of pro-inflammatory cytokines generation and oxidative stress (Lappas et al., 2005). The activation of A2A receptors inhibits inflammatory response in different models of lung diseases, including murine models of asthma, lipopolysaccharide-driven airway neutrophilic infiltration and cigarette smoke-induced chronic obstructive pulmonary disease (Fozard et al., 2002; Bonneau et al., 2006). In the current study, we evaluated the effect of LASSBio-897 on silica-induced lung inflammation and fibrosis. We also investigated the interaction of this bioactive compound with A2A receptors.

### MATERIALS AND METHODS

### Synthesis of 3,4-Methylenedioxybenzoyl-3-Thienylhydrazone (LASSBio-897)

LASSBio-897 was synthesized at Laboratório de Avaliação de Substâncias Bioativas of Federal University of Rio de Janeiro, Brazil. The methodology used in the synthesis of LASSBio-897 was previously reported by our research group (Zapata-Sudo et al., 2010) and was reproduced in this work.

### Silicosis Induction and Treatment

Male Swiss mice weighing 18–20 g were obtained from the breeding colonies of Oswaldo Cruz Foundation and used in accordance with the guidelines of the Committee on Use of Laboratory Animals of the Oswaldo Cruz Foundation (CEUA-FIOCRUZ, license L-034/09). Mice were housed in groups of five at 25oC on a 12 h day/night cycle and fed with a standard sterile diet of mouse chow and water was allowed ad libitum. Silicosis was induced by intranasal instillation with 10 mg crystalline silica particles (SiO2; Sigma Chemical Co, St. Louis, MO, United States; particle size 0.5–10 µm) under halothane volatile anesthesia (Cristália, São Paulo, Brazil). A control group received sterile saline (0.9% NaCl) intranasally. All analyses were performed 28 days post-silica instillation. LASSBio-897 was given by gavage and CGS 21680 was given intraperitoneally, once a day for 7 days, starting at the 21st day post-silica particle provocation. Non-treated silicotic mice received an equal volume of vehicle (0.1% DMSO, oral route). All in vivo analyses were

repeated twice with the exception of the one involving CGS 21680 treatment.

### Invasive Assessment of Respiratory Mechanics

Following general anesthesia with nembutal <sup>R</sup> (60 mg/kg, i.p.), tracheostomy and neuromuscular blockade (pancuronium bromide, 1 mg/kg, i.v.), the mechanic ventilator was connected to the mouse through an endotracheal tube. Then, transpulmonary resistance and elastance were assessed 28 days after silica particle intranasal instillation by using invasive whole-body plethysmography (Buxco Electronics, United States). Mice were allowed to stabilize for 5 min and increasing concentrations of methacholine (3, 9, and 27 mg/ml) were aerosolized for 5 min each. Baseline pulmonary mechanics parameters were assessed with aerosolized phosphate-buffered saline (PBS) (Ferreira et al., 2013).

### Lung Histology

Left lungs from each experimental group were dissected and placed in Millonig fixative solution (pH = 7.4) with 4% paraformaldehyde for 48 h to preserve tissue architecture. Tissues were dehydrated and clarified in xylene before paraffin embedding. Lung sections of 4 µm were stained with hematoxylin and eosin (H&E), to analyze granulomas, or Picrosirius, to evaluate collagen deposition, in light microscope using 3DHISTECH–Pannoramic MIDI whole slide scanner (capture with a 20× objective lens) and the resulting images analyzed with CaseViewer 3.3, Pannoramic Viewer 1.15.4, and HistoQuant softwares (3DHISTECH). Silica crystals were analyzed, in 15 independent fields, with a light microscope (Olympus BX51) equipped with polarizing attachment for detecting birefringent particles and the quantification were performed using the software Image-Pro Plus Version 6.2 (Media Cybernetics Inc., Rockville, MD, United States).

### Lung Cytokine and Chemokine Quantification

Murine IL-6, TNF-α, and macrophage inflammatory protein-2 (MIP-2/CXCL-2) levels were measured in right lung samples by means of ELISA kit (R&D Systems, Minneapolis, MN, United States) as previously described (Ferreira et al., 2016). The results were expressed as picograms of cytokine per right lung.

### Collagen Evaluation

For quantification of collagen, the right lung was homogenized in 0.05 M Tris + 1 M NaCl (pH = 7.4), containing protease inhibitor (Hoffmann-La Roche Ltd., Switzerland). Total soluble collagen was extracted overnight at room temperature and centrifuged for 1 h at 15,000 × g. The supernatant was used to evaluate collagen levels by means of SircolTM kit (Biocolor Ltd., Newtownabbey, United Kingdom) following manufacturer's guidelines. The results were expressed as mg collagen per right lung.

### Immunohistochemistry Staining

Paraffin-embedded sections of lungs were deparaffinized, rehydrated, and boiled in 10 mM urea for 15 min to enhance antigen retrieval. Tissue sections were incubated with 3% H2O<sup>2</sup> for 10 min to block endogenous peroxidase. To prevent non-specific binding, the sections were then incubated for 2 h with a solution containing 5% bovine serum albumin (BSA) dissolved in Tris-buffered saline enriched with 0.1% Tween 20 (TBST). Sections were incubated overnight at 4◦C with the specific antibody (monoclonal mouse anti-mouse α-SMA or monoclonal rat anti-mouse F4/80 from Sigma– Aldrich, St. Louis, MO, United States) and Serotec, Kidlington, United Kingdom, respectively) diluted in TBST with 1% BSA. Primary antibody binding was detected after incubating the sections with a horseradish peroxidase conjugated-secondary antibody (polyclonal anti-mouse IgG HRP, R&D System, Minneapolis, MN, United States) and polyclonal anti-rat IgG HRP, Serotec, Kidlington, United Kingdom) for 2 h at 4◦C, followed by a 15 min exposure to enzyme substrate 3-amino-9-ethylcarbazole (AEC). Sections were washed with TBST between all steps and weakly counterstained with hematoxylin for easy identification of tissue structures. In negative controls, primary antibody was omitted and tissues were incubated with antibody diluent only. The images were captured through light microscope (Olympus BX51) coupled to a video camera (Olympus DP72), and analyzed with software Image Pro Plus 6.2 (Media Cybernetics Inc., Rockville, MD, United States). The number of positive pixels was divided by the field area and expressed as pixels/µm<sup>2</sup> .

### Binding Assays

The capacity of LASSBio-897 to bind to A2A receptors was assessed using classical competition assays at equilibrium, in two different conditions. In the first assay, we used 150 µg protein of a rat striatum membrane preparation, rich in A2A receptors (Cunha et al., 1996), and the agonist [3H]-CGS21680 (10 nM) as radioligand, in experimental conditions reported by Luthin and Linden (1995), i.e., incubation during 2 h at 4◦C in a medium containing MgCl<sup>2</sup> 5 mM, EDTA-Na<sup>2</sup> 1 mM, and Tris-HCl 50 mM (pH 7.4). For the second assay, we used 10 µg protein of a membrane preparation of HEK293 cells overexpressing the human A2A receptors (RBHA2AM400UA, PerkinElmer, United States). In this assay, we used the endogenous agonist [ <sup>3</sup>H]-adenosine (25 nM) as the radioligand. The incubation was performed at 25◦C for 1 h in a medium containing EDTA-Na<sup>2</sup> 1 mM, MgCl<sup>2</sup> 10 mM, an adenosine deaminase inhibitor [erythro-9-(2-Hydroxy-3-nonyl) adenine hydrochloride-EHNA, 10 µM], an adenosine uptake inhibitor (nitrobenzylthioinosine-NBI, 3 µM) and Tris-HCl 50 mM (pH 7.4). For both assays, the non-specific binding was estimated using 30 µM NECA and the filters were rapidly washed with cold Tris-HCl 5 mM (pH 7.4), either 3 × 4 mL ([3H]-CGS21680) or 2 × 2 mL ([3H] adenosine). For the assay with [3H]-adenosine, the glass fiber filters (GMF 3, Filtrak, Germany) were previously soaked in 0.5% polyethyleneimine. For exploring the effect of LASSBio-897 on [3H]-ZM241385 dissociation kinetics, rat striatal membranes

were incubated at 25◦C in a medium containing [3H]-ZM241385 0.5 nM, EDTA-Na<sup>2</sup> 1 mM and Tris-HCl 50 mM (pH 7.4). After 45 min, the dissociation of the [3H]-ZM241385-receptor complex was initiated by addition of NECA 30 µM with or without 30 µM LASSBio-897.

### Cell Culture

We used Glosensor cAMP Human Embryonic Kidney 293 (HEK293G) cells (Promega, Madison, WI, United States), which is a cell line that stably expresses the biosensor variant encoded by the pGloSensorTM-20F cAMP plasmid; and Chinese Hamster Ovary cells stably expressing both the human A<sup>1</sup> (CHO-A1), A2A (CHO-A2A), A2B (CHO-A2B), or A<sup>3</sup> (CHO-A3) receptor and a reporter gene, Secreted Placental Alkaline Phosphatase (SPAP), under the transcriptional control of a six CRE promoter (McDonnell et al., 1998). Cells were used until the 30th passage. HEK293G were maintained in Dulbecco's modified Eagle's medium (DMEM) while all CHO cell lines were maintained in DMEM/Nutrient mix F12 (DEMEM/F12). All culture mediums were supplemented with 10% fetal calf serum (FCS) and 2 mM <sup>L</sup>-glutamine and all cell lines were grown at 37◦C in a humidified 5% CO2: 95% air atmosphere.

### GloSensor cAMP Assay

The GloSensorTM cAMP assay is an extremely sensitive technique to detect changes in intracellular concentration of cAMP using a live-cell system in a non-lytic format. The assay uses genetically encoded biosensor variants with cAMP binding domains fused to mutant forms of Photinus pyralis luciferase, that promote large increases in light output. The analysis of kinetic measurements of cAMP in HEK293G cells was performed following the manufacturer's instructions (Promega, Madison, WI, United States). Briefly, cells were grown to 90% confluence in 96-well plates. Then, the media was removed and replaced with 100 µL of 6% v/v dilution of GloSensorTM cAMP reagent stock solution in HBSS, and the cells incubated for 2 h. LASSBio-897 and/or agonists (50 µL each, diluted in HBSS) were then added to each well concomitantly, and the plate was then read at 37◦C in an EnVision Luminometer plate reader (PerkinElmer, Akron, OH, United States) in a kinetic mode, from 10 s to 1 h. The peak light emission in the time course of the reaction was recorded. In the experiments where we used antagonists or inhibitors, they were added 30 min before stimulation.

### CRE-Mediated SPAP Transcription Measurement

CRE-dependent SPAP reporter activity was evaluated as previously described (Baker et al., 2010). Briefly, CHO-A1, CHO-A2A, CHO-A2B, or CHO-A3 cells were incubated with LASSBio-897 and/or adenosine for 6 h at 37◦C in a 5% CO2: 95% atmosphere air. In the case of CHO-A1 and CHO-A3 cells, we added forskolin (3 µM) 30 min after stimulation with adenosine. CRE-dependent SPAP gene transcription was quantified by color change caused by hydrolysis of p-nitrophenol phosphate, using MRX plate reader (Dynatech Labs, Chantilly, VA, United States) at 405 nm absorbance.

### Molecular Docking

The crystal structure of the A2A receptor with PDB code 4EIY (Liu et al., 2012) was chosen to perform molecular docking studies of LASSBio-897, using the genetic algorithm software GOLD 5.2 (CCDC). The number of genetic operations (crossover, migration, mutation) in each docking run used in the searching procedure was set to 100,000. Docking analyzes were performed on the orthosteric site and on the allosteric site. The binding sites were selected to include all amino acid residues located within a 10 Å distance from the cocrystalized ligand (ZM241385) at the orthosteric site and a 10 Å distance from the sodium ion at the allosteric site. The docking runs at the orthosteric site were done allowing complete flexibility only for the ligand. However, even in the rigid protein mode, the program optimizes hydrogen bond geometries by rotating hydroxyl and amino groups of amino acid side chains. At the allosteric site, some amino acid side chains were allowed to stay flexible during the docking runs. This was done because the LASSBio-897 molecule occupies a higher volume than the sodium ion, thus the flexibility of amino acid side chains could enable a better fit of the molecule into this site. The resulting poses were classified according to the GoldScore fitness score function (Jones et al., 1997).

### Statistical Analysis

Results were expressed as mean ± SD. All data were analyzed in blind flashing and evaluated to ensure normal distribution. Statistical analysis was performed with one-way ANOVA followed by the multiple comparison test of Newman–Keuls– Student. In lung mechanical experiments, statistical analysis was performed with two-way ANOVA followed by the multiple comparison test of Bonferroni. The P-values ≤ 0.05 were considered statistically significant. In binding experiments at equilibrium, we used the model of competition for one binding site (top and bottom fixed at 100% and 0%, respectively) for analysis of the specific binding data by non-linear regression analysis (GraphPad Prism 5) and estimation of the IC<sup>50</sup> (concentration of the unlabeled ligand that inhibits the binding of the radioligand by 50%).

### RESULTS

### CGS 21680 Inhibits Silica-Particle Induced AHR

While investigating whether or not the A2A receptor would be a relevant therapeutic target in silicosis, we studied the effect of the interventional intraperitoneal treatment with the selective A2A agonist CGS 21680 (0.5 or 1 mg/kg) on AHR triggered by silica particle intranasal instillation in mice. As illustrated in **Figure 2**, values of airway resistance and lung elastance following aerosolized methacholine (3–27 mg/mL) were elevated 28 days after silica instillation, compared to negative control mice challenged with saline (silica particle vehicle), pointing out a state of airway hyper-reactivity (AHR) in line with previous studies (Tripathi and Pandey, 2010). CGS 21680 treatment completely

reversed silica-induced AHR regarding both airway resistance (**Figure 2A**) and lung elastance changes (**Figure 2B**) at dose of 1 mg/kg, being partially effective at 0.5 mg/kg.

### LASSBio-897 Effects on Lung Function, Granuloma Formation, and Collagen Deposition in the Lung Parenchyma

In order to evaluate the effect of the interventional treatment with LASSBio-897 on pathological changes triggered by silica particle intranasal instillation in mice, we gave the compound orally (1–5 mg/kg) once a day for 7 days. As in the case of the CGS 21680 treatment, the treatment started at day 21 post-silica, when lung inflammatory changes and other dysfunctions were already established as reported (Ferreira et al., 2013; Trentin et al., 2015). As shown in **Figure 3**, LASSBio-897 reversed silica-induced AHR regarding both airway resistance (**Figure 3A**) and lung elastance changes (**Figure 3B**) at doses of 2 and 5 mg/kg given orally, being inactive at 1 mg/kg.

Histological evaluation of lung tissue samples, 28 days after intranasal instillation of silica particles, showed a parenchymal accumulation of inflammatory cells associated with granuloma formation (**Figure 4B**) and increased collagen deposition (**Figure 4F**), findings which were not observed in mice subjected to intranasal instillation of saline (**Figures 4A,E**, respectively). LASSBio-897 treatment of silica-stimulated mice, at oral doses of 5 mg/kg, resulted in a significant reduction in granulomatous and fibrogenic responses, as illustrated by the representative histological sections, shown in **Figures 4C,G**, respectively. Quantitative assessments revealed that silica-induced granuloma (**Figure 4D**) and collagen deposition (**Figure 4H**) were equally inhibited by LASSBio-897 at doses of 2 mg/kg and 5 mg/kg, but they were not modified at the dose of 1 mg/kg.

### LASSBio-897 Effects on the Number of Macrophages, Myofibroblasts, and Silica Particles Present in the Lung Parenchyma

We assessed the effect of LASSBio-897 on the number of macrophages and myofibroblasts in the lung parenchyma of mice exposed to silica, as attested by the immunohistochemistry of lung sections for F4/80 and α-SMA, respectively. Silicotic mice showed an increase in the levels of F4/80<sup>+</sup> cells (**Figure 5B**) and myofibroblasts (**Figure 5F**) accumulated in granuloma areas, compared to samples from negative control mice (**Figures 5A,E**, respectively). Oral treatment of silicotic mice with LASSBio-897 (5 mg/kg) reduced the immune reactivity for F4/80 (**Figure 5C**) and α-SMA (**Figure 5G**), suggesting a substantial inhibitory effect upon the levels of infiltrating macrophages and myofibroblasts. Quantitative data for the impact of LASSBio-897 (1–5 mg/kg) upon F4/80 (**Figure 5D**) and α-SMA (**Figure 5H**) pointed out that immune reactivity for α-SMA were inhibited at the doses of 2 and 5 mg/kg, while F4/80 expression were reduced only at 5 mg/kg, and both changes remaining unaltered at 1 mg/kg.

Furthermore, under light microscopy, in a system equipped with polarizing filters, we visualized silica crystals present in the lung parenchyma sections from silicotic mice (**Figure 5J**). Histological sections from mice stimulated with saline were employed as negative controls (**Figure 5I**). Remarkably, LASSBio-897 (5 mg/kg) notably reduced the amount of silica crystals identified in the lung tissue (**Figure 5K**). **Figure 5L** shows that LASSBio-897, at 2 and 5 mg/kg but not at 1 mg/kg, significantly inhibited the amount of silica dispersed in the lung.

### LASSBio-897 Effects on the Lung Tissue Production of Pro-Inflammatory Mediators

We also examined the effect of LASSBio-897 on lung tissue production of key inflammatory mediators in silicosis. We noted that mice reacted to silica challenge with increased lung tissue levels of TNF-α, IL-6, and MIP-2/CXCL-2, as compared with samples from control mice. The interventional treatment with LASSBio-897 (2 and 5 mg/kg) significantly inhibited the

FIGURE 3 | LASSBio-897 inhibits airway hyper-reactivity (AHR) in silicotic mice. Silica (10 mg) was given i.n. and analyses were performed 28 days later. Oral treatment with LASSBio-897 was performed daily for 7 days, starting 21 days after silica instillation. AHR was induced by provocation with increasing concentrations of methacholine and measured as airway resistance (A) and lung elastance (B) parameters. Silicotic non-treated animals received an equal amount of vehicle (DMSO 0.1%). Data are expressed as the means ± SD from six mice. These results are representative of two independent assays. <sup>+</sup>P < 0.05 compared to saline-provoked group. <sup>∗</sup>P < 0.05 compared to silica-provoked group.

silica-induced augmentation in the levels of the assessed pro-

### LASSBio-897 Is a Ligand of A2A Receptors

Initially, we tested the effect of LASSBio-897 in a classical competition assay using rat striatum membranes. In these

saline-provoked group. <sup>∗</sup>P < 0.05 compared to silica-provoked group.

conditions, saturation experiments with [3H]-CGS21680 revealed a single class of specific binding sites with a Bmax of 1070 ± 90 f mol/mg protein and a K<sup>d</sup> of 207 ± 55 nM. **Figure 6A** shows that LASSBio-897 decreased the specific binding of [3H]-CGS21680 in a concentration-dependent manner, with an IC<sup>50</sup> of 9.3 µM (95% confidence interval (CI): 6.4–15.3 µM). In the experiments reported here, the

inflammatory cytokines (**Table 1**).

Data are expressed as the means ± SD from six mice. Scale bars = 200 µm. These results are representative of two independent assays. <sup>+</sup>P < 0.05 compared to

an equal amount of vehicle (DMSO 0.1%). White arrows indicate silica particles in (J) and (K). Data are expressed as the means ± SD from six mice. Scale bars = 200 µm. These results are representative of two independent assays. <sup>+</sup>P < 0.05 compared to saline-provoked group. <sup>∗</sup>P < 0.05 compared to silica-provoked group.


TABLE 1 | Effect of LASSBio-897 on cytokine/chemokine generation in the lung tissue of silicotic mice.

LASSBio-897 (2 and 5 mg/mouse) was given 21 days after silica instillation (10 mg) and the analyses were performed 24 h post-silica. Values represent the mean ± SD. These results are representative of two independent assays. <sup>+</sup>P < 0.05 vs. saline-challenged group; <sup>∗</sup>P < 0.05 vs. silica-challenged group.

total, non-specific and specific CPM (counts per min) of the control were 374, 213, and 161, respectively (mean of two experiments). As differences exist in binding affinity and potency of some agonists between human and rat (Kull et al., 1999), we decided to test the effect of LASSBio-897 on the binding of the endogenous agonist [3H]-adenosine to human A2A receptors. As shown in **Figure 6B**, a similar result was obtained with LASSBio-897 decreasing the binding of the agonist radioligand with an IC<sup>50</sup> of 11.4 µM (95% CI: 9.0–14.4 µM). In the experiment performed with membranes of cells transfected with the human receptor, the total, non-specific and specific CPM of the control were 990, 151, and 799, respectively. To verify if these effects of LASSBio-897 were due to competition at the orthosteric site and not to a negative allosteric modulation (alternative hypothesis for such results), we used two different approaches using [3H]-ZM-241385 as radioligand and the rat

striatal membrane preparation, for economical and practical reasons. Indeed, this classical selective radioligand for the A2A receptor allows a much better signal than [3H]-CGS21680 or [ <sup>3</sup>H]-adenosine in this preparation (in our conditions, the specific binding corresponds to around 90% of the total binding) and has been used successfully in different protocols in our laboratory (Noel and do Monte, 2017). In the first experiment, we used the most classical, and more sensitive, approach for investigating directly an allosteric effect, the radioligand dissociation assay (Hill et al., 2014), already used by us in previous studies (Lopes et al., 2004). **Figure 6C** shows that LASSBio-897 did not increase the dissociation kinetics of the orthosteric ligand (here [3H]-ZM-241385) as expected for a negative allosteric modulator. The second assay is more indirect but allows addressing a putative allosteric effect that could be detected only with the endogenous orthosteric ligand (probe

FIGURE 6 | LASSBio-897 binds at the orthosteric site of A2A receptors, but has no negative allosteric effect. LASSBio-897 inhibits [3H]-CGS21680 binding in rat striatal membranes (A) and [3H]-adenosine binding in a membrane preparation of human A2A receptors (B) in a concentration-dependent manner. The specific binding of the radioligands are expressed as percent of the control without competing drug. The fitted curves were obtained by non-linear regression analysis using the model of competition for one binding site, as detailed in the methods. (C) Effect of LASSBio-897 (30 µM) on [3H]-ZM241385 dissociation kinetics in rat striatal membranes. After 45 min incubation at 25◦C, the dissociation of the [3H]-ZM241385-receptor complex was initiated by addition of NECA 30 µM with or without 30 µM LASSBio-897. The fitted curves were obtained by non-linear regression using the model of one phase exponential decay. In this experiment, the specific binding of the control corresponds to 95% of the total binding. (D) Effect of LASSBio-897 (10 µM) on the competition curve of adenosine for [3H]-ZM-241385 binding in rat striatal membranes. The incubation was performed at 25◦C during 2 h. Specific binding of [3H]-ZM-241385 is expressed as percent of binding in the absence (control) or presence of 10 µM LASSBio-897. At this concentration, LASSBio-897 inhibits the binding of [3H]-ZM-241385 by only 13% and the specific binding corresponds to 88% of the total binding. Data are from two (A), or one (B–D), experiments performed in triplicate.

dependency). **Figure 6D** shows that 10 µM LASSBio-897 did not modify the affinity of adenosine for its binding to the A2A receptor, measured indirectly from a competition assay with the antagonist [3H]-ZM-241385.

### LASSBio-897 Is an Activator of A2A Receptors

In HEK293G cells, stimulation with 3 µM forskolin, 10 µM NECA, or 5 µM CGS21680 induced a fast production of high amounts of intracellular cAMP (**Figures 7A,C**). In contrast, LASSBio-897 produced a small increase of intracellular cAMP levels, and only at the highest concentration used (50 µM) (**Figures 7B,D**). However, the co-incubation with LASSBio-897 (10 and 50 µM) increased the cAMP production induced by 10 and 30 µM adenosine (**Figure 8A**). Treatment with the A2A receptor selective antagonist SCH 58261 blocked the potentiating effect of LASSBio-897 on adenosine-induced cAMP production in a concentration dependent manner (**Figure 8B**). However, although treatment with A2B receptor selective antagonist PSB 603 drastically decreases adenosine-induced cAMP production, addition of LASSBio-897 slightly increases the cAMP concentration, suggesting that its potentiating effect is not mediated by the A2B receptor (**Figure 8C**).

### LASSBio-897 Does Not Alter Adenosine Metabolism, But Acts Selectively on Adenosine A2A Receptor

In order to assess whether LASSBio-897 potentiating effect upon cAMP production by adenosine could be due to inhibition of either adenosine metabolism or transport, we pre-incubated

HEK293G cells for 30 min with inhibitors of two enzymes, which metabolize adenosine, as well as with an inhibitor of adenosine reuptake transporters. Individually, both LASSBio-897 and adenosine deaminase inhibitor pentostatin increased the effect of 100 µM adenosine on intracellular cAMP production (**Figure 9A**). When used together, pentostatin and LASSBio-897 produced a greater increase of the effect of 100 µM adenosine (**Figure 9A**). Similarly, both LASSBio-897 and adenosine kinase blocker 5-iodotubercidin increased the effect of 100 µM adenosine on cAMP production. The association of 5-iodotubercidin with LASSBio-897 produced an even greater increase of the effect of 100 µM adenosine (**Figure 9B**). Finally, we noted that the inhibitor of adenosine transporters dipyridamole also increased the effect of 100 µM adenosine similarly to LASSBio-897. Once more, co-incubation with both dipyridamole and LASSBio-897 produced a greater increase of the effect of adenosine (**Figure 9C**).

In order to address the selectivity of LASSBio-897 effect for the four subtypes of adenosine receptors, we used an assay based on CRE-mediated gene transcription (Baker et al., 2010). To observe the activation of A<sup>1</sup> and A<sup>3</sup> receptors, we added forskolin in order to raise the basal adenylyl cyclase activity and subsequent CRE-gene transcription activation as shown in **Figures 10A,G**, respectively. In CHO-A1 transfected cells, we observed that adenosine induced a biphasic response in the presence of forskolin. First, there was an inhibitory Gi-mediated suppression of CRE gene transcription, while at higher agonist concentration, a Gs-mediated enhancement of the response to forskolin was noted (**Figure 10B**). LASSBio-897 did not alter the effect of adenosine on A<sup>1</sup> (**Figure 10B**), A2B (**Figures 10E,F**), and A<sup>3</sup> receptors (**Figure 10H**) at any concentration analyzed. In CHO-A2A transfected cells, LASSBio-897 enhanced the Emax of adenosine for its effect on CRE gene transcription (**Figures 10C,D**).

### LASSBio-897 Can Bind to the Orthosteric and Allosteric Sites of A2A Receptor

The molecular docking results showed that LASSBio-897 is predicted to be able to interact favorably both at the orthosteric

vehicle (DMSO 0.1%). Data points are means ± SD of triplicate determinations from a single experiment and are representative of three experiments. SCH, SCH 58261. PSB, PSB 603.

and allosteric sites. **Figure 11A** indicates that LASSBio-897 could replace the co-crystallized ligand ZM-241385 at the orthosteric site without approaching the sodium ion pocket. The main interaction observed in the orthosteric site solution involved Asn253. As expected from its greater volume, the solutions at the allosteric pocket indicate that LASSBio-897 occupies the sodium ion binding site and also part of the orthosteric site (**Figure 11A**). The main interaction observed at this site was with Ser91, a

residue directly implied in the interaction with the sodium ion at the allosteric pocket.

In order to test the hypothesis of LASSBio-897 binding to the allosteric site of A2A receptor, we treated HEK293G cells with amiloride, a negative allosteric modulator of adenosine receptors. Amiloride alone reduced the effect of 100 µM adenosine, and blocked the potentiating effect of LASSBio-897 on adenosineinduced cAMP production (**Figure 11B**).

were concomitantly stimulated with different concentrations of LASSBio-897 and adenosine. Six hours after stimulation, CRE-mediated SPAP transcription was measured by spectrophotometry. The bars represent SPAP secretion from non-stimulated cells. Control group received an equal amount of vehicle (DMSO 0.1%). Data points are means ± SD of triplicate determinations from a single experiment and are representative of three experiments. FSK, forskolin.

### DISCUSSION

Prior studies have highlighted the anti-inflammatory properties of some compounds of the class of N-acylhydrazones, but the mechanism of action remains poorly understood (Fraga and Barreiro, 2006; Kummerle et al., 2012). The present work investigated the therapeutic potential of the N-acylhydrazone LASSBio-897 in experimental silicosis and explored its putative mechanism of action.

Our results revealed that the interventional treatment with LASSBio-897 or the A2A receptor agonist CGS 21680 reversed silica-induced AHR in mice. LASSBio-897 also inhibited the progress of major pathological changes triggered by nasalinstilled silica particles, including granuloma formation and

collagen deposition, in the mouse lung parenchyma. The effect appeared associated with a reduction in the content of F4/80+macrophages, α-SMA<sup>+</sup> myofibroblasts and crystalline silica particles in the lungs, in parallel with a significant decrease in the lung tissue levels of pro-inflammatory cytokines and chemokines. Binding studies suggested that LASSBio-897 has some affinity for the A2A receptor and functional assays indicated that it is a weak activator of cAMP production in HEK293G cells. Also, LASSBio-897 increased the adenosineinduced cAMP production in cells expressing A2A receptor, being inactive in cells transfected with A1R, A2BR, or A3R, through a mechanism unrelated to blockade of adenosine metabolism. Finally, docking studies indicated that LASSBio-897 could bind at both orthosteric and allosteric sites of A2A receptor, which is in line with the ability of amiloride – a negative allosteric modulator of adenosine receptor – to prevent the synergistic effect of LASSBio-897 and adenosine. These findings suggest that LASSBio-897 is indeed a promising therapeutic strategy to control silicosis. It is likely to act at the A2A receptor through a non-canonical mechanism based on the fact that native A2A receptors are probably predominantly present as homodimers (Ferre et al., 2014). Although merely putative, we hypothesize that LASSBio-897 could act as a bitopic ligand (Lane et al., 2014) where binding at the allosteric site of one protomer (considering a dimer composition), would increase the efficacy of adenosine at the other protomer (cAMP assays). On the other hand, binding of LASSBio-897 at the orthosteric site would explain the results of the binding competition assays (decrease of radiolabeled orthosteric ligand binding).

Alveolar macrophages display a pivotal role in the recognition, uptake, and clearance of environmental particulate matters that traffic into the airways (Blank et al., 2007). The pathogenesis of the silicosis is largely attributed to the direct damage by inhaled crystalized silica particles to alternatively activated macrophages localized in the airway wall (Kawasaki, 2015). When this barrier is broken, free silica crystals translocate into the interstitial space and are taken up by M1 macrophages, which eventually sequester the particulate matter in the lung (Huaux, 2007). By releasing pro-inflammatory cytokines and chemokines, the inflammatory macrophages are also involved in inducing granuloma formation, which further overlaps with an extensive fibrotic response throughout the lung interstitial area (Kawasaki, 2015). The characteristic features of human silicosis can be modeled in mice by intranasal instillation of crystalized silica particles (Ferreira et al., 2013, 2016; Trentin et al., 2015). LASSBio-897 (2 and 5 mg/kg), given orally from days 21–27 post-silica, significantly inhibited an extensive area of fibro-granulomatous formation, which reach about 40% of the lung parenchyma in non-treated mice 28 days post-challenge.

Because fibro-granulomatous lesions can take important part of the lung of patients with silicosis, a respiratory deficit is expected to occur in such patients. Actually, in mild silicosis, and particularly in the early phase of the disease, spirometry results are frequently negative. However, in more severe cases, increased airway resistance and residual volume have been reported in

clinical studies (de Mesquita Junior et al., 2006; Leung et al., 2012; Sa et al., 2013). Using invasive whole-body plethysmography in mice exposed to silica particles, we noted that the LASSBio-897 treatment significantly reduced the elevated levels of lung resistance and elastance, also inhibiting the state of airway hyperreactivity to the bronchoconstrictor methacholine as compared to non-treated silicotic controls. Similarly to the granuloma and fibrotic responses, also the bronchial hyper-reactivity noted in animal models of silicosis has been associated in a causative manner to inflammatory mediators generated locally (Tripathi and Pandey, 2010; Ferreira et al., 2013).

In attempt to further clarify the mechanism that underpins the protective effect of LASSBio-897 in this model, we found first that the treatment reduced the levels of inflammatory agents such as TNF-α, IL-6, and MIP-2/CXCL-2. These mediators are largely implicated in the recruitment and activation of macrophages and fibroblasts, which play a central role in the genesis of silica particle-induced tissue injury in the respiratory tract (Driscoll, 2000; Tripathi and Pandey, 2010; Fazzi et al., 2014). This is relevant since the granulomatous inflammation induced by silica particles consists predominantly of macrophages (Hamilton et al., 2008). In addition, myofibroblasts are responsible for the increased collagen production in silica-induced fibrosis (Huaux, 2007). In fact, we noted a significant decrease in the levels of F4/80 and α-SMA immunoreactive cells in lung tissue samples in those mice treated with the active dose of LASSBio-897 (5 mg/kg), which pointed out the decrease in the number of macrophages and myofibroblasts, respectively. Therefore, LASSBio-897 may exert its protective effect by interfering with both inflammatory and fibrotic responses. A further and important aspect of the effect on experimental silicosis is the decrease in the amount of crystals of silica dispersed in the lung interstitial space. It is noteworthy that silica particles can be drained through the lymphatic system into the lymph nodes, particularly when they are free enough to be carried out from the interstitial space (Ferreira et al., 2013). Thus, by reducing the silica-induced fibro-granulomatous response, LASSBio-897 may favor silica particle mobility and elimination from the lung parenchyma via lymphatic draining.

The screening of 10 µM LASSBio-897 at the 98 potential targets of the Cerep's "Diversity Profile" platform (Poitiers, France), showed an inhibitory activity superior to 30% at only the A2A receptor (72%), 5-HT transporter (56%), and NE transporter (42%). As activation of A2A receptor is a potent inductor of anti-inflammatory response in several models of lung diseases (Bonneau et al., 2006), we utilized several in vitro and in silico models for investigating how LASSBio-897 could modulate A2A receptor signaling. First, we showed that LASSBio-897 inhibited the binding of both [3H]-CGS21680, a selective A2A receptor agonist, and [3H]-adenosine, the endogenous agonist, in rat striatum membranes and human A2A receptors, respectively, indicating that LASSBio-897 either competes with the orthosteric ligand of the A2A receptor or exerts a negative allosteric effect by decreasing the affinity of the radioligand. Note that these competition binding assays cannot estimate a putative effect of LASSBio-897 on the efficacy of the orthosteric ligands. The two experimental approaches used for addressing this question were able to discard the second hypothesis.

Then, we tested the ability of LASSBio-897 to modulate cAMP production in vitro. For this purpose, we choose the HEK293G cells since they express the A2A receptors whose activation increases the intracellular levels of cAMP (Ongini et al., 1999). Only at the highest concentration (50 µM), LASSBio-897 induced an increase in cAMP production that was much smaller than that observed with NECA and CGS21680, suggesting that LASSBio-897 could be a weak partial agonist of the A2A receptor. Surprisingly, LASSBio-897 increased the production of cAMP induced by adenosine, when the two drugs were co-incubated. In fact, if LASSBio-897 were only a classical partial agonist, it should decrease, and not increase, the effect of adenosine. In order to test if the activation of A2A could be involved in the curative effect of LASSBio-897 in silica particle-induced lung injury, we evaluate the effect of selective A2A agonist CGS 21680. CGS 21680 decreases silica particle-induced AHR, evidenced by the parameters of airway resistance and lung elastance, in a similar way to that seen with LASSBio-897 treatment, suggesting that the curative effect of LASSBio-897 on silica particle-induced lung injury can be related to the its ability in increase the effect of adenosine on A2A.

To challenge our hypothesis of a synergistic interaction between LASSBio-897 and adenosine at the A2A receptor level, we explored three distinct approaches. Initially, we pre-treated HEK293G cells with the A2A antagonist SCH 58261. This A2A antagonist in both concentrations (0.1 and 1 µM) blocked the synergic effect of LASSBio-897 on adenosine-induced cAMP, suggesting that this compound in fact acts on A2A receptor. Although we noted that SCH 58261 inhibited the synergic effect of LASSBio-897 in the range of low nanomolar concentrations, we evaluated the effect of LASSBio-897 on the A2B receptor, once HEK293 cells also express the A2B receptors (Cooper et al., 1997). We noted that LASSBio-897 maintained its ability to increase the production of cAMP induced by adenosine even in the presence of the A2B antagonist PSB 603, suggesting that LASSBio-897 did not act on this receptor. Then, we pre-treated HEK293G cells with adenosine metabolism or transport blockers. We showed that inhibitors of adenosine deaminase (pentostatin), adenosine kinase (5-iodotubercidin) or adenosine transporters (dipyridamole) increased the synergistic effect of LASSBio-897 on adenosine-induced cAMP, indicating that our compound did not act blocking metabolism or reuptake of adenosine. Together, these data suggest that the effect of LASSBio-897 would not be due to increase of the supply of adenosine to the A2A receptors. Next, we analyzed the selectivity of LASSBio-897 toward the A2A receptors, since HEK293 cells express other receptors able to activate adenylyl cyclase, including the A2B receptors (Cooper et al., 1997). To address this point, we used CHO cells transfected with A1, A2A, A2B, or A<sup>3</sup> receptors. We observed that LASSBio-897 increased the adenosine-induced activation of A2A but not of A1, A2B, and A<sup>3</sup> receptors, indicating that LASSBio-897 selectively modulates the A2A receptors.

In order to test the hypothesis of a positive modulator allosteric effect of LASSBio-897, we performed a molecular docking study using the crystal structure of the A2A receptor

cooperativity across the dimer for the binding of adenosine to the other protomer. AD, adenosine; L-897, LASSBio-897.

with PDB code 4EIY (Liu et al., 2012). The docking analysis showed that LASSBio-897 has potential to bind at both orthosteric and sodium ion allosteric sites of the A2A receptor. Since the pre-treatment with amiloride, a negative allosteric modulator, abolished the synergistic effect of LASSBio-897 on adenosine-induced cAMP, we propose that LASSBio-897 modulates positively the A2A receptor activity by binding at the sodium ion allosteric site. A putative molecular mechanism of action that could integrate all our data could be suggested according to what had been proposed by the Christopoulos's group (Lane et al., 2014) for their data on SB269652 effect on a dimeric D<sup>2</sup> receptor complex. Accordingly, we are suggesting as a working hypothesis that LASSBio-897 could be a bitopic ligand able to bind on one A2A receptor protomer to modulate allosterically the effect of adenosine to the orthosteric site of a second protomer, considering the plausible existence of homodimers of this receptor in our assays. This hypothesis could explain the displacement of agonist ligands in our binding assays (competition for the orthosteric site) and also the positive allosteric modulation in the cAMP assay, considering that this effect is mainly an increase in efficacy of adenosine (**Figure 12**).

The fact that LASSBio-897 is able to increase the adenosineinduced activation of A2A receptor is indeed interesting, since adenosine accumulates extracellularly in inflamed and fibrotic tissues through rapid conversion of adenine nucleotides released from many cells, including mast cells, nerves and endothelial cells, to adenosine (Linden, 2001; Sitkovsky and Lukashev, 2005). Furthermore, extracellular adenosine stores at inflammatory tissue usually act as an organ protection (Linden, 2001). During inflammation, the defensive action of adenosine occurs through the activation of A2A receptors, which culminates in anti-inflammatory and pro-resolution responses, including inhibition of pro-inflammatory cytokines TNF-α and IL-6 and augmentation of anti-inflammatory cytokine IL-10 (Pinhal-Enfield et al., 2003), besides enhancing alternative macrophage activation (Koroskenyi et al., 2011; Csoka et al., 2012). In addition, although the majority of works described the antiinflammatory effect of A2A receptor using orthosteric agonists, it has been shown that allosteric ligands able to modulate positively the A2A receptor reduced LPS-induced mouse model of inflammation (Welihinda and Amento, 2014).

Taken together, our findings show that therapeutic treatment with LASSBio-897 reduced granulomatous inflammation and

fibrosis into the lungs, associated with improvement of lung function in silicotic mice. LASSBio-897 may represent a promising therapeutic strategy to control silicosis, probably by acting at the A2A receptor through a new mechanism involving synergism with the endogenous agonist.

### AUTHOR CONTRIBUTIONS

fphar-08-00778 October 28, 2017 Time: 16:21 # 15

VC: Contributions to design of the work, acquisition and analysis of data, illustration, revision for intellectual content and final approval. TF: Acquisition and analysis of data, illustration, revision for intellectual content and final approval. AdA: Acquisition and analysis of data, illustration and final approval. FN: Acquisition and analysis of data, illustration, critical revision and final approval. RT and CF: Acquisition and analysis of data, illustration, revision for important intellectual content and final approval. CS'A: Acquisition and analysis of data, revision for important intellectual content, supervision and final approval. EB: Revision for important intellectual content, supervision and final approval. PR: Contributions to design of the work, illustration, revision for important intellectual

### REFERENCES


content, supervision and final approval. MM: Design of the work, illustration, drafting of the manuscript, supervision and final approval.

### FUNDING

The work was supported by grants from National Institute of Science and Technology on Drugs and Medicines (INCT-INOFAR) N◦ 465.249/2014-0; Conselho Nacional de Desenvolvimento Científico and Tecnológico (CNPq) and Fundação Carlos Chagas de Amparo à Pesquisa do Estado do Rio de Janeiro (FAPERJ).

### ACKNOWLEDGMENTS

We express our gratitude to Prof. Stephen J. Hill for providing all the facilities of the Institute of Cell Signaling at University of Nottingham, including cell systems and reagents, concerning the in vitro experimental work associated with adenosine receptors in this study.



CHO-K1 cells transfected with the human β2-adrenoceptor. Br. J. Pharmacol. 125, 717–726. doi: 10.1038/sj.bjp.0702139


**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 Carvalho, Ferreira, de Arantes, Noël, Tesch, Sant'Anna, Barreiro, Fraga, Rodrigues e Silva and Martins. 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.

# Volatile Oil from Amomi Fructus Attenuates 5-Fluorouracil-Induced Intestinal Mucositis

Ting Zhang, Shan H. Lu, Qian Bi, Li Liang, Yan F. Wang, Xing X. Yang, Wen Gu\* and Jie Yu\*

College of Pharmaceutical Science, Yunnan University of Traditional Chinese Medicine, Kunming, China

Amomi Fructus has been used to treat digestive diseases in the context of traditional Chinese medicine, so we evaluated the effects of a volatile oil from Amomum villosum (VOA) on intestinal mucositis induced by 5-fluorouracil (5-FU). We measured the effect of VOA and its main active constituent, bornyl acetate (BA), on body weight, food intake, diarrhea, inflammatory cytokines, the mucosal barrier, and gut microbiota. VOA and BA significantly increased the rats' body weight, relieved diarrhea, and reversed histopathological changes in the gut and inflammation. VOA significantly inhibited apoptosis and alleviated the endoenteritis by downregulating p38 MAPK and caspase-3 expression. VOA and BA strengthened the intestinal mucosal barrier by increasing zonula occludin-1 and occludin expression. VOA and BA reduced the amount of pathogenic bacteria and increased the abundance of probiotics. Thus, VOA prevented the development and progression of intestinal mucositis after chemotherapy.

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

#### Reviewed by:

Satish Ramalingam, SRM University, India Soon Yew Tang, University of Pennsylvania, United States

#### \*Correspondence:

Wen Gu guwen1230@qq.com Jie Yu cz.yujie@gmail.com

#### Specialty section:

This article was submitted to Inflammation Pharmacology, a section of the journal Frontiers in Pharmacology

Received: 09 August 2017 Accepted: 17 October 2017 Published: 09 November 2017

#### Citation:

Zhang T, Lu SH, Bi Q, Liang L, Wang YF, Yang XX, Gu W and Yu J (2017) Volatile Oil from Amomi Fructus Attenuates 5-Fluorouracil-Induced Intestinal Mucositis. Front. Pharmacol. 8:786. doi: 10.3389/fphar.2017.00786 Keywords: intestinal mucositis, Amomi Fructus, volatile oil, bornyl acetate, 5-fluorouracil

### INTRODUCTION

Intestinal mucositis, which is characterized by a decrease in villi length and the disruption of crypt cell homeostasis, is a common toxic side effect of the cancer chemotherapeutics 5-fluorouracil (5-FU) and irinotecan. This toxicity causes severe diarrhea and morphological mucosal damage, which limits the safety and clinical application of the drugs. Intestinal mucositis has many facets, including microstructural damage of small intestinal tissue, an injured intestinal mucosal barrier and inflammation. Small intestinal mucosal barrier integrity is frequently disrupted in various acute and chronic intestinal diseases (Turner, 2009), and tight junction proteins of the small intestinal mucosa, such as zonula occludin (ZO) and occludin, are needed to maintain the intestinal epithelial barrier (Suzuki, 2013). In addition, caspase and cysteine protease are needed for the morphological and biochemical changes that occur during apoptosis. Caspase-3, which is required for the activation of various apoptotic-stimulating factors, can target substrates and cause cell disassembly and DNA fragmentation (Flanagan et al., 2016). Damage to the small intestinal mucosal barrier can cause endotoxin-translocation-induced endotoxemia, which exacerbates inflammatory infiltration. Meanwhile, commensal intestinal microbiota may influence all of the phases of mucosal pathogenesis (van Vliet et al., 2010). Thus, treatment with prebiotics and probiotics may alleviate intestinal mucositis (Wada et al., 2010; Justino et al., 2014, 2015; Araújo et al., 2015). Efforts to reduce chemotherapy-induced intestinal damage, with traditional or conventional medicine have not been successful (Sonis, 2004; Cathy, 2010; Lam et al., 2010; Wang et al., 2014, 2015).

Amomi Fructus, the dry and mature fruit of Amomum villosum Lour., A. villosum Lour. var. xanthioides T. L. Wu et Senjen, and A. longiligulare T. L. Wu, has been recorded and used in traditional Chinese medicine as an excellent crude drug for the treatment of digestive system disorders (Commission of Chinese Materia Medica, 1999; Commission of Chinese Pharmacopoeia, 2015). Amomi Fructus is also authorized as a food by the China Food and Drug Administration<sup>1</sup> . Amomi Fructus has been shown to have a significant effect on the recovery of mice intestinal flora balance that was disturbed by antibiotics (Yan et al., 2013). In addition, aqueous extracts may improve and promote intestinal function (Huang and Zhou, 2006). However, whether Amomi Fructus works and its mechanism in the treatment of intestinal mucositis have yet to be elucidated. Therefore, we focused on the potential roles of volatile oil from A. villosum (VOA) and bornyl acetate (BA), the main component of VOA, on the regulation of intestinal microflora, the adjustment of the inflammatory process and oxidative stress, maintaining intestinal permeability, and alleviating the intestinal mucositis process to provide a more theoretical and scientific foundation for its clinic use.

### MATERIALS AND METHODS

### Medicinal Materials

Amomum villosum Lour. (**Figures 1A,B**) was collected in Jingping County, Honghe Prefecture of Yunnan Province, China. The plants were identified as A. villosum Lour. by Jie Yu, an Associate Professor at Yunnan University of Traditional Chinese Medicine. Voucher specimens were deposited in the Herbarium of Pharmacognosy, Yunnan University of Traditional Chinese Medicine. BA (**Figure 1C**), was purchased from Jingzhu Biotechnology Co., Ltd. (Nanjing, China, purity > 98%, GC). 5-FU was obtained from XuDong HaiPu Pharmaceutical Co., Ltd. (Shanghai, China). Live combined Bifidobacterium and Lactobacillus tablets (CBL) were purchased from Inner Mongolia Shuangqi Pharmaceutical Co., Ltd. (China).

### Volatile Oil from A. villosum

Powder (100 g) of A. villosum fruit was soaked with 800 mL distilled water for 12 h (4◦C). VOA, gathered by steam distillation, was stored at 4◦C after drying with Na2SO4.

### GC-MS Characterization of VOA

Gas chromatography-mass spectrometry (GC-MS) was used to characterize VOA (HP6890GC/5973MS, Agilent). Chromatographic separation was achieved on an HP-5MS (30 mm × 0.25 mm, 0.25 µm) fused-silica capillary column. The injector temperature was 250◦C and the column pressure was 100 kPa. Helium was used as the carrier gas (1.0 mL/min) and the injection volume was 0.1 µL (split ratio of 50:1). The temperature was 80◦C for 2 min and increased to 280◦C for 20 min (3◦C/min).

Mass spectrometry conditions were as follows: ion source temperature, 230◦C; electron energy, 70 eV; interface temperature, 250◦C; quadrupole temperature, 150◦C; and mass scan range, 30–500 amu. Wiley7n.l mass spectral database was used to compare data with a standard spectrum to identify the components of every peak.

### Animals

Fifty-four, 8-week-old male Spraque–Dawley rats (Da Shuo Biotech Co., Ltd., Chengdu, China, Certificate of Quality No: 0016254) were kept in a temperature-controlled room with free access to water and food, and they fasted for 2 h before all of the treatments. The study was approved by the Institutional Ethical Committee on Animal Care and Experimentation of Yunnan University of Traditional Chinese Medicine (R-0620150028). All reasonable efforts were made to minimize animal suffering.

### Rats Model of Intestinal Mucositis Induced by 5-FU

After adaptive feeding for 3 days, the rats were randomized to one of nine groups (N = 6/group): normal controls (CON); 5-FU-induced intestinal mucositis model (MOD); live combined Bifidobacterium and Lactobacillus tablets (CBL, 450 mg/kg) positive controls; VOA low, medium, and high groups (VOA.L, 8 mg/kg; VOA.M, 16 mg/kg; and VOA.H, 32 mg/kg); BA low, medium, and high groups (BA.L, 2 mg/kg; BA.M, 4 mg/kg; and BA.H, 8 mg/kg). All of the rats, except for the normal controls, received 5-FU (35 mg/kg, daily, i.p.) for 5 days. Researchers used personal protective gear to prevent exposure to 5-FU and CBL. At the end of the study, the drugs were destroyed by the Experimental Center of Yunnan University of Traditional Chinese Medicine.

<sup>1</sup>http://www.nhfpc.gov.cn/ewebeditor/uploadfile/2013/07/20130712155225821. doc

The dose of A. villosum was 3–6 g/kg in clinical medium and the BA concentration was required to be not less than 1% in A. villosum according to the Commission of Chinese Pharmacopoeia (2015). The yield of VOA was 3.9% from A. villosum in our study and BA was calculated as 1%, which was the minimum content of BA regulated by the Commission of Chinese Pharmacopoeia (2015). The clinical dose for humans was calculated as 4.5 g/kg in this study. These factors were considered in our dosage conversation, and we then converted the human dosage into the rat dose according to the body surface area. We concluded that the dose of VOA was 16 mg/kg and BA was 4 mg/kg in rats, which was considered to be the medium dose. Finally, the low, medium, and high dose of each administration group was calculated by 1:2:4. VOA and BA were dissolved in distilled water with 0.1% Tween-80 by an ultrasonic mixer at 16◦C and then stored at 4◦C. Before chemotherapy (4 h), the rats received CBL, VOA, and BA (via gavage) once daily for 12 days. **Figure 2** depicts the treatment schedule.

Whether or not the rats had diarrhea, and their weight and food intake were recorded daily. Rats were killed 12 days after treatment using 10% chloral hydrate anesthetization. Blood samples and small intestinal tissue were collected.

### Morphology and Histopathology Observation of Small Intestinal Tissue

Segments of the jejunum were collected, fixed, and stained with hematoxylin and eosin to measure the villus height and crypt depth. Three tissues from different rats in each group and three images (10×) per section were analyzed for morphological studies.

### IL-6, ROS, TNF-α, NF-κB, and MPO in Blood Samples and Small Intestinal Tissue

Blood samples were collected from the retro-orbital venous plexus and hepatic portal vein at 2 h after treatment in the morning at the end of the study. Serum was centrifuged at 12,000 × g for 15 min and analyzed immediately. Serum interleukin-6 (IL-6) and reactive oxygen species (ROS) were measured with ELISA kits purchased from Cusabio Biotech Co., Ltd. (China).

At the end of the treatment, the rats were killed via 10% chloral hydrate (0.3 mL/100 g, ip). Small intestine tissue samples were excised, weighed, and washed in 0.9% saline. Tissues (100 mg) were rinsed with PBS and homogenized in 1 mL of PBS and then stored overnight at −20◦C. Two freezethaw cycles were performed to break the cell membranes, and homogenates were centrifuged for 10 min at 6,500 × g, 4◦C. The supernatant was collected for biochemical analysis. ROS, tumor necrosis factor-α (TNF-α), nuclear factor of kappa b (NF-κB), and myeloperoxidase MPO in the intestinal homogenate were measured with ELISA kits purchased from Cusabio Biotech Co., Ltd. (China).

### Hepatic Portal Vein Lipopolysaccharide (LPS)

Hepatic portal vein blood samples were centrifuged at 6,500 × g for 10 min at 4◦C after being collected under anesthesia. Lipopolysaccharide (LPS) was measured using a tachypleus amebocyte lysate test purchased from Chinese Horseshoe Crab Reagent Manufacturers Co., Ltd. (Xiamen, China).

### Flow Cytometric Measurement of Occludin, ZO-1, and Caspase-3

A single-cell suspension was prepared by cutting and disrupting one representative intestinal sample from each group through a 70-µm filter membrane. Cells were diluted with staining buffer (1 × 10<sup>6</sup> cells/mL) after blockage with 3% FBS. For analysis, the cells were incubated with an anti-occludin antibody (EPR8208, Abcam, United States), anti-ZO-1 antibody (21773-1-AP, Proteintech, United States) and an anti-caspase antibody (25546-1-AP, Proteintech, United States) for 2 h, and then they were incubated with fluorescein isothiocyanate (FITC, SA00003-2) in the dark for 1 h. Occludin, ZO-1, and caspase-3 expression was measured using flow cytometry (FACSCalibur, Becton, Dickinson, and Company, United States).

### Western Blot for MAPK

Intestinal tissues (100 mg) were homogenized using 1 mL PIRA lysis buffer, and centrifuged (12,000 × g, 10 min, 4◦C). Then, samples were kept on ice for 20 min. The supernatant was collected for Western blotting, and the extracted proteins were quantified using the bicinchoninic acid (BCA) method. Next, 50 µg of total protein was added to each sample. Proteins were resolved with 10% sodium dodecyl sulfate (SDS)-polyacrylamide gels (PAGE) and transferred to polyvinylidene fluoride (PVDF) membranes. Transferred membranes were blocked with 5% albumin bovine V (BSA) to inhibit non-specific proteins. Then, the primary antibodies (9212S, 1:1.000 dilution; Cell Signaling, Danver, MA, United States) were added, with β-actin (20536-1-AP, 1:1.000 dilution; Proteintech Group, United States) used as an internal reference. After incubating with primary antibodies, the membranes were washed with TBS/Tween-20 (TBST) three times and then incubated with the secondary antibodies (SA00001-2, 1:10.000 dilution; Proteintech Group, United States). Each protein band was visualized with an enhanced chemiluminescence (ECL)

detection system (Proteintech Group, United States) and quantified with Quantity One Analysis Software (Bio-Rad Laboratories, Inc.).

### 16S rDNA Gene Sequencing of Rat Feces

Rat feces samples were collected on the last day and placed in a sterile centrifuge tube. They were then evenly ground in liquid nitrogen and preserved at −80◦C for future inspection. All of the fecal samples from the treatment groups were blended, and transferred to a mortar and ground to a fine powder. DNA was extracted according to the instructions of the stool DNA kit (Omegea Bio-tek, United States).

To determine bacterial diversity and composition in the feces samples, we used the protocol described by Caporaso's group (Caporaso et al., 2010). PCR amplifications were conducted with a 515f/806r primer set that amplifies the V4 region of the 16S rDNA gene. This primer set has few biases and yields accurate phylogenetic and taxonomic information. The reverse primer contains a 6-bp error-correcting barcode unique to each sample. DNA was amplified as described in the literature (Magoc and ˇ Salzberg, 2011). Sequencing was conducted on an Illumina MiSeq platform (Novogene, Beijing, China).

### Statistical Analysis

Data are means + SD. One-way analysis of variance (ANOVA) was used to compare multiple groups (p < 0.05, < 0.01, and < 0.001). Graphics were created with Origin 6.1 (MicroCal Software, United States). A 95% confidence interval (CI) was used as a threshold to identify potential outliers in all of the samples, and clustering was analyzed with TMEV Clustering (Mev Development Team).

### RESULTS

### A. villosum Lour. Characterization

A total of 65 compounds were determined, and 58 compounds were successfully identified, which represented over 99% of the total oil composition. As shown in **Table 1**, BA (54.54%), camphor (17.92%), camphene (6.76%), limonene (5.25%), borneol (4.07%), myrcene (1.97%), α-pinene (1.50%), β-caryophyllene (0.85%), β-pinene (0.80%), and α-pepperene


(0.54%) were detected in VOA. BA was considered as the representative compound of VOA.

### General Condition of Experimental Animals

Initial animal weight and food intake were similar among the groups, and the controls gained weight over time (**Figure 3A**). Rats treated with 5-FU had decreased weight from the fourth to the eighth day and the weight of these animals was the lowest until the end of the experiment. Treatment with CBL, VOA, and BA prevented (p < 0.01) significant weight loss in the MOD group. During the study, the average MOD food intake was 10.81 g/day, which was roughly half of the normal group (**Figure 3B**). CBL, VOA, and BA treatments reversed the appetite reduction in the MOD group. Compared with the MOD group, the food intake of the CBL, VOA, and BA groups was increased by about 50%.

### VOA Improved Morphology and Microstructures of Small Intestinal Tissue

Fecal appearance and shape indicated diarrhea in the MOD group, and all of the treatments relieved 5-FU-induced watery diarrhea (**Figure 4A**). Compared with the controls, the MOD

group had histopathological changes in the jejunum, such as mucosa with shortened villi with vacuolated cells, crypt necrosis, and intense inflammatory cell infiltration. Treatment with VOA, BA, and CBL for 12 days reversed these changes in limited fashion (**Figure 4B**).

The administration of 5-FU induced significant decreases in villus height, increased crypt depth, and decreased villus: crypt ratio (**Figure 4C**). **Figure 4C** shows that CBL, VOA, and BA treatment reversed 5-FU-induced reductions in villus height. Similarly, a smaller villus: crypt ratio was observed in the MOD group, and this was reversed with the drug treatments, especially with the VOA.H treatment (p < 0.01).

### VOA Prevented Serum and Intestinal Inflammation

After an injection of 5-FU, we observed a significant increase (p < 0.01) in the serum IL-6 concentration (3.07 ± 0.79 pg/mL) compared with the controls (1.54 ± 0.39 pg/mL), suggesting that the inflammation was elevated in the MOD rats. Treatment with VOA and BA significantly reduced the IL-6 levels by 26.71%, 40.06%, 38.76%, 30.94%, 45.60%, and 42.34% (**Figure 5A**). In the meantime, we found a significantly elevated ROS concentration in the sera of the MOD group (2.48 ± 0.41 U/mL) compared with the control group (1.93 ± 0.18 U/mL); this indicated that the model rats induced by 5-FU were in an inflammatory status. The increase was significantly inhibited after treatment with CBL (1.98 ± 0.31 U/mL) (**Figure 5B**). In brief, VOA.M, VOA.H, and BA suppressed the 5-FU-mediated increase in IL-6, and VOA.H suppressed the ROS increase compared with the MOD group.

As shown in **Figures 5C,D**, the concentrations of ROS and TNF-α in the small intestinal tissue of the intestinal mucositis rats was increased significantly (to 20.42 ± 1.10 U/mL and 184.7442 ± 25.71 pg/mL, respectively). Similarly, MPO and NF-κB concentrations were also increased by 67.44% and 51.21%, respectively, in the intestinal mucositis rats. All of the treatments could effectively reduce (p < 0.001) these inflammatory factor levels. **Figure 5C** shows that treatment

with BA significantly reduced ROS concentrations by 24.06%, 44.62%, and 53.18%, while **Figure 5D** shows that treatment with BA significantly reduced TNF-α concentrations by 69.43%, 72.02%, and 73.77%, respectively. VOA treatment reduced MPO (55.63%, 60.03%, and 49.46%, respectively) and NF-κB concentrations (44.68%, 46.74%, and 43.84%, respectively) (**Figures 5E,F**). Thus, VOA and BA prevented inflammation in the small intestinal tissue after 5-FU chemotherapy.

### VOA Downregulated p38 MAPK and Caspase-3 Protein Expression

The concentrations of p38 MAPK in the small intestine were detected by Western blotting. Our results revealed

that p38 MAPK expression in the MOD group was significantly higher than the CON (control) group (p < 0.01), while the therapeutic drugs could successfully inhibit the expression of p38 MAPK. Treatment with VOA and BA reduced (p < 0.01) the p38 MAPK levels by 37.29%, 39.83%, 60.17%, 12.71%, 23.73%, and 41.53%, respectively (**Figure 6A**).

Caspase-3 protein in the small intestinal tissue was measured with flow cytometry. **Figure 6B** shows that 5-FU significantly increased the expression of caspase-3 compared with the control. In contrast, both VOA and BA could effectively reduce (p < 0.01 and p < 0.001) the expression of caspase-3 in a dose-dependent manner. These data suggested that they could significantly inhibit cell apoptosis and alleviate the development of endoenteritis.

## VOA Strengthened the Intestinal Mucosal Barrier and Inhibited Enterogenous Endotoxin

The 5-FU treatment increased LPS in the MOD group more than CON group (p < 0.001; **Figure 7A**). VOA (p < 0.01) and BA (p < 0.05) showed a satisfactory effect with respect to the inhibition of LPS rise. This inhibition effect may be related to its beneficial effects on gut microbiota equilibrium.

**Figures 7B,C** show that the expression of ZO-1 and occludin was significantly lower (p < 0.001) in MOD rats compared to the CON rats, and this agreed with a previous study (Song et al., 2013). VOA and BA increased ZO-1 and occludin protein expression to normal levels. VOA and BA improved the intestinal mucosal barrier by increasing the expression of ZO-1 and

occludin protein and blocking endotoxin translocation-induced endotoxemia.

### VOA Regulated Intestinal Microbial Balance and Altered Its Structure and Composition

The structure and composition of gut microbiota were dramatically altered by 5-FU injection. Results of the microbial classification at the level of family showed that the relative abundance of Bacteroidaceae, Helicobacteraceae, Enterobacteriaceae, Porphyromonadaceae, and Streptococcaceae in the CON group was 1.19%, 0.32%, 0.02%, 0.25%, and 0.02%, respectively. However, their relative abundance in the MOD group increased to 10.40%, 1.54%, 0.28%, 1.09%, and 0.07%, respectively. The relative abundance of Helicobacteraceae, Enterobacteriaceae, Porphyromonadaceae, and Streptococcaceae was significantly reduced after treatment with BA, and Bacteroidaceae was reduced by VOA.M, VOA.H, and BA. Meanwhile, the relative abundance of Lactobacillaceae, which are usually considered probiotics, decreased from 4.69% in the CON group to 3.15% in the MOD group. VOA and BA treatment

reincreased the abundance of Lactobacillaceae to 5.35–13.45% and 14.03–18.63%, respectively (**Figure 8A**).

Results of the microbial classification at the level of genus showed that the relative abundance of the Bacteroides (Bacteroidaceae), Ruminococcus (Ruminococ caceae), Helicobacter (Helicobacteraceae), Escherichia (Enterobacteriaceae), and Parabacteroides (Porphyromo nadaceae) in the CON group was 1.19%, 1.48%, 0.03%, 0.01%, and 0.25%, respectively. However, their relative abundance in the MOD group increased to 10.40%, 2.14%, 1.24%, 0.28%, and 1.09%, respectively. The relative abundance of Ruminococcus, Helicobacter, Escherichia, and Parabacteroides was significantly reduced after treatment with BA, and Bacteroides was reduced by VOA.M, VOA.H, and BA. The relative abundance of Lactobacillus (Lactobacillaceae) in the CON group was 4.69%, and it decreased to 3.15% in the MOD group. After treatments with VOA and BA, the abundance of Lactobacillus increased to 5.35–13.45% and 14.03–18.63%, respectively (**Figure 8B**).

In addition, according to the species annotations and the abundance information at the level of genus, the heat map of the most abundant 35 genus was plotted and cluster analysis was conducted (**Figure 9**). They could be divided into four categories: BA.L/BA.M/BA.H; CON/CBL; MOD/VOA.H/VOA.L; and VOA.M/BA, and we found that BA.M and CBL were the closest. We concluded that BA increased the abundance of probiotics, such as Lactobacillus and Bifidobacterium, while VOA increased the abundance of Roseburia and Coprococcus. We suggested that BA and VOA might regulate intestinal microecology in two ways not just one way.

Pearson correlation and cluster analyses were performed with clean operational taxonomic unit (OTU) data and we identified 52 key variables, which were significantly altered after the VOA and BA treatments. These were correlated to alternations in ROS, MPO, LPS, and caspase-3 (**Figure 10**). Among them, Bacteroidaceae (four OTUs, from Bacteroidetes) were positively correlated with ROS, MPO, and caspase-3. S24-7 (eight OTUs, from Bacteroidetes), Ruminococcaceae (five OTUs, from Firmicutes), and Desulfovibrionaceae (one OTU, from Proteobacteria) were positively correlated with LPS. In addition, a few bacteria from Bacteroidetes (three OTUs) and Firmicutes (one OTU) were negatively correlated with LPS (**Figures 10A,C**).

Clustering analysis of the OTUs (**Figure 10B**) showed that among the 52 key OTUs, most of them in the MOD rats were increased when compared to the CON group. Using the Euclidean distance metric, the MOD group was at a long distance from the CON group. However, all of the treatment groups could shorten the distance to the CON group and the VOA and BA groups were clustered into two trees, suggesting that VOA and BA had a significant effect on the regulation of the 52 key OTUs. Thus, gut microbiota in the MOD rats were significantly altered compared to CON rats, and CLB, VOA, and BA treatment could partially recover the gut microbiota equilibrium.

### DISCUSSION AND CONCLUSION

The VOA and BA treatments prevented the occurrence of diarrhea and reversed weight loss and diminished food intake, all symptoms of 5-FU-induced intestinal mucositis. In addition, VOA and BA treatments significantly improved the histopathological changes in rat intestinal mucositis induced by 5-FU, perhaps due to a reduction in inflammatory ROS, IL-6, TNF-α, and NF-κB, decreased MPO, decreased p38 MAPK and caspase-3 proteins, and improved intestinal mucosal barrier function. VOA and BA also contributed to the regulation of the intestinal microbiota balance.

Intestinal injury may induce p53/PUMA-mediated apoptotic, brush-border hydrolase activity changes, blunted villus height, crypt deepening, and increased crypt cell apoptosis with decreased proliferation (Zhan et al., 2014). Chemotherapy drugs often indiscriminately damage DNA, tumor cells, and normal cells, causing increased ROS and reactive nitrogen species (RNS). They may also activate downstream signaling pathways to induce local inflammatory responses and cause damage to intestinal epithelial cells (Sonis, 2004). ROS, which are crucial mediators of downstream biological events, are then produced. VOA appeared to block 5-FU-induced increases in ROS in blood and the small intestines, and these data agreed with a report by Arifa's group

(Arifa et al., 2014), who observed that inflammasome activation was dependent on ROS and that anti-neoplastic drugs induced mucositis via inflammatory factors in mice.

DNA damage, non-DNA damage, and ROS activate several transduction pathways that then activate transcription factors, such as NF-κB (Sonis, 2002). NF-κB activates downstream genes, including LPS and pro-inflammatory cytokines, such as TNF-α, IL-6, and IL-1 (Rinkenbaugh and Baldwin, 2016). In the present study, VOA and BA significantly reduced the expression of NF-κB and reduced pro-inflammatory cytokines (IL-6 and TNF-α) that had been increased by 5-FU; this suggested that VOA and BA play protective roles in the inflammatory state.

p38 MAPK, an important member of the MAPK family, participates in cell proliferation, apoptosis, and differentiation (Horie et al., 2012; Chen J.Y. et al., 2014; Lee et al., 2014). VOA and BA treatment inhibited MAPK and capase-3 signaling and suppressed MPO activity, decreased apoptosis, and protected

the integrity of the intestinal mucosal barrier. BA had a dosedependent relationship with the regulation of ROS, TNF-α, MPO, and related proteins (p38 MAPK and caspase-3). VOA had a dose-dependent relationship with respect to the regulation of p38 MAPK and caspase-3. This may be due to different the compositions of VOA and BA.

Lipopolysaccharide, a cell component of Gram-negative bacteria, is delivered to the liver via the portal vein in endotoxemia (Lin et al., 2015). Endotoxin leads to chronic low-grade inflammation in patients with intestinal mucositis. LPS is thought to disrupt the epithelial barrier integrity by downregulating the tight junction proteins (Sheth et al., 2007; Han et al., 2016). We measured the tight junction, ZO-1, and occludin proteins in the small intestinal tissue and LPS in the hepatic portal veins. There is substantial evidence suggesting that tight junctions play pivotal roles in the pathophysiology of chemotherapy drug-induced gut toxicity, so we hypothesized that VOA could improve 5-FU-induced small intestinal mucositis,

which may be related to changes in tight junction protein expression and reduced LPS, and it is perhaps mediated by the MAPK and NF-κB pathways.

We also measured the influence of commensal intestinal microbiota, such as Escherichia, Bacteroides, Helicobacter, Desulfovibrio, Ruminococcus, Parabacteroides, and Clostridium. Cassmann's group (Cassmann et al., 2016) found that dogs with chronic enteropathies harbored more mucosal bacteria from Bacteroides, Escherichia, and Enterobacteriaceae. Studies show that Helicobacter pylori is an important factor in gastroduodenal diseases as H. pylori infection leads to chronic gastritis in children and adults (Mansour-Ghanaei et al., 2010). Asonum's group reported that H. pylori is a major cause of transdifferentiation into intestinal metaplasia that causes gastric cancer (Asonuma et al., 2009). Desulfovibrio was significantly increased in acute and chronic ulcerative colitis at multiple levels within the colon (Rowan et al., 2010). Ruminococcus is a symbiotic anaerobic bacteria present in the gastrointestinal tract and its overgrowth occurs in inflammatory bowel disease (IBD). In patients with irritable bowel syndrome, there were fewer Bifidobacterium and more Ruminococcus and Bacteroides abundances than healthy (Shukla et al., 2015). Dziarski's group (Dziarski et al., 2016) found that Parabacteroides and Bacteroides promoted colitis. Clostridium perfringens is a Gram-positive anaerobic pathogen that is usually associated with skin and soft tissue infections, gastrointestinal infections, and occasionally bacteremia (Antony et al., 2009).

In our study, BA and VOA significantly reduced Escherichia, Bacteroides, Helicobacter, Desulfovibrio, Ruminococcus, Parabacteroides, and Clostridium, which are pathogenic bacteria. BOA increased Lactobacillus and Bifidobacterium, the latter of which has been shown to decrease NF-κB activation (Beg, 2004), leading to decreased endotoxin and plasma IL-6 (O'Hara et al., 2006). Lactobacillus facilitates the maintenance of the intestinal membrane integrity during Shigella dysenteriae 1 infection in rats (Moorthy et al., 2009). VOA increased Roseburia and Coprococcus compared with the MOD rats, and study suggest that Roseburia, Coprococcus, and Ruminococcus were significantly reduced, whereas pathogens Escherichia and Enterococcus were prevalent in patients with IBD (Chen L.P. et al., 2014). This study support our findings.

Furthermore, we identified 52 key variables that were significantly altered after VOA and BA treatments, and clustering analysis suggested that the VOA and BA treatments partially

### REFERENCES


recovered gut microbiota equilibria. In addition, BA or VOA dose effects were clustered together, while BA or VOA were in a wide range; this might be due to the fact that VOA is a mixture and BA is a monomeric compound. To the best of our knowledge, this is the first study that evaluates the effects of VOA on the intestinal epithelial barrier, inflammation, and the intestinal microbial imbalance induced by 5-FU. We hope these data will inform future novel probiotic-based treatments for intestinal toxicity associated with anticancer therapy. It is the theoretically possible that VOA or BA could interfere with the anticancer effect of 5-FU, and thus reduce the side effects of 5-FU. However, other study indicated that 5-FU retained its anti-cancer efficacy when used in combination with other medicinal products in different cancer models. The combined effects of sinomenine and 5-FU on esophageal carcinoma were superior to those of the individual compounds, and the drug combination did not increase the side effects of chemotherapy (Wang et al., 2013). Definitely, the benefit-risk profile of the combined treatment of 5-FU with VOA/BA is also worth studying. We would like to find these answers in our future research.

### AUTHOR CONTRIBUTIONS

TZ and SL conducted the study, analyzed the data, and prepared the manuscript. QB, LL, and YW collected and interpreted the data. XY, WG, and JY designed the study, provided the drugs and research facilities, and revised the manuscript.

### FUNDING

This research was supported by the National Natural Science Foundation of China (81460623, 81660596, and 81660684), the Natural Science Foundation of Yunnan Province (2014FA035, 2014FZ083, and 2016FD050), the Southern Medicine Collaborative and Innovation Center (30270100500), and the Young and Middle-Aged Academic and Technological Leaders of Yunnan (2015HB053).

### ACKNOWLEDGMENT

We would like to thank Novogene for the technical assistance with 16S rDNA sequencing and analysis.

IL-18 in mice. Am. J. Pathol. 184, 2023–2034. doi: 10.1016/j.ajpath.2014. 03.012



responses by Bifidobacterium infantis and Lactobacillus salivarius. Immunology 118, 202–215. doi: 10.1111/j.1365-2567.2006.02358.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 Zhang, Lu, Bi, Liang, Wang, Yang, Gu and Yu. 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.

# Carnosic Acid Alleviates BDL-Induced Liver Fibrosis through miR-29b-3p-Mediated Inhibition of the High-Mobility Group Box 1/Toll-Like Receptor 4 Signaling Pathway in Rats

### Edited by:

David Sacerdoti, Università degli Studi di Padova, Italy

#### Reviewed by:

Jiiang-Huei Jeng, National Taiwan University, Taiwan Mirko Pesce, Università degli Studi "G. d'Annunzio" Chieti – Pescara, Italy Shailendra Pratap Singh, New York Medical College, United States

> \*Correspondence: Jihong Yao yaojihong65@hotmail.com

#### Specialty section:

This article was submitted to Inflammation Pharmacology, a section of the journal Frontiers in Pharmacology

Received: 06 July 2017 Accepted: 21 December 2017 Published: 19 January 2018

#### Citation:

Zhang S, Wang Z, Zhu J, Xu T, Zhao Y, Zhao H, Tang F, Li Z, Zhou J, Gao D, Tian X and Yao J (2018) Carnosic Acid Alleviates BDL-Induced Liver Fibrosis through miR-29b-3p-Mediated Inhibition of the High-Mobility Group Box 1/Toll-Like Receptor 4 Signaling Pathway in Rats. Front. Pharmacol. 8:976. doi: 10.3389/fphar.2017.00976 Shuai Zhang<sup>1</sup> , Zhecheng Wang<sup>1</sup> , Jie Zhu<sup>1</sup> , Ting Xu<sup>1</sup> , Yan Zhao<sup>1</sup> , Huanyu Zhao<sup>1</sup> , Fan Tang<sup>1</sup> , Zhenlu Li<sup>2</sup> , Junjun Zhou<sup>1</sup> , Dongyan Gao<sup>1</sup> , Xiaofeng Tian<sup>2</sup> and Jihong Yao<sup>1</sup> \*

<sup>1</sup> Department of Pharmacology, Dalian Medical University, Dalian, China, <sup>2</sup> Department of General Surgery, Second Affiliated Hospital of Dalian Medical University, Dalian, China

Fibrosis reflects a progression to liver cancer or cirrhosis of the liver. Recent studies have shown that high-mobility group box-1 (HMGB1) plays a major role in hepatic injury and fibrosis. Carnosic acid (CA), a compound extracted from rosemary, has been reported to alleviate alcoholic and non-alcoholic fatty liver injury. CA can also alleviate renal fibrosis. We hypothesized that CA might exert anti-liver fibrosis properties through an HMGB1-related pathway, and the results of the present study showed that CA treatment significantly protected against hepatic fibrosis in a bile duct ligation (BDL) rat model. CA reduced the liver expression of α-smooth muscle actin (α-SMA) and collagen 1 (Col-1). Importantly, we found that CA ameliorated the increase in HMGB1 and Toll-like receptor 4 (TLR4) caused by BDL, and inhibited NF-κB p65 nuclear translocation in fibrotic livers. In vitro, CA inhibited LX2 cell activation by inhibiting HMGB1/TLR4 signaling pathway. Furthermore, miR-29b-3p decreased HMGB1 expression, and a dual-luciferase assay validated these results. Moreover, CA down-regulated HMGB1 and inhibited LX2 cell activation, and these effects were significantly counteracted by antago-miR-29b-3p, indicating that the CA-mediated inhibition of HMGB1 expression might be miR-29b-3p dependent. Collectively, the results demonstrate that a miR-29b-3p/HMGB1/TLR4/NFκB signaling pathway, which can be modulated by CA, is important in liver fibrosis, and indicate that CA might be a prospective therapeutic drug for liver fibrosis.

Keywords: carnosic acid, bile duct ligation, high-mobility group box-1, miR-29b-3p, liver fibrosis

**Abbreviations:** α-SMA, α-smooth muscle actin; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BDL, bile duct ligation; CA, carnosic acid; Col-1, collagen 1; HMGB1, high-mobility group box 1; MMP9, matrix metalloprotein 9; NFκB, nuclear factor kappa B; TGF-β1, transforming growth factor-beta 1; TGF-β R1, transforming growth factor-beta receptor type 1; TLR4, toll-like receptor 4.

### INTRODUCTION

fphar-08-00976 January 17, 2018 Time: 16:30 # 2

Liver fibrosis, a symptom of the progression of chronic liver diseases, is characterized by excessive extracellular matrix (ECM) deposition, distorted hepatic architecture and damaged normal function, which is a main reason for the development of fibrosis into cirrhosis or even hepatocellular carcinoma (Hernandez-Gea and Friedman, 2011; Schuppan and Kim, 2013). In the recent years, the mechanisms underlying the pathogenesis of liver fibrosis have been increasingly studied. Previous studies have reported that the activation of hepatic stellate cells (HSCs) plays an important role in the progression of liver fibrosis (Friedman, 2008; Mallat and Lotersztajn, 2013). HSCs become activated and differentiate into fibroblasts when liver injury occurs; specifically, these cells lose their epithelial characteristics and acquire the characteristics of mesenchymal cells, which show increased expression of α-smooth muscle actin (α-SMA) and Col-1 (Zhou et al., 2007). Therefore, controlling the activation of HSCs is essential to liver fibrosis.

High-mobility group box-1 (HMGB1) as a nuclear non-histone chromosomal protein that can bind to the minor groove of DNA, participates in DNA repair and replication and energy homeostasis (Bianchi and Manfredi, 2007; Kang et al., 2009). Several recent experimental reports have proven that HMGB1 is markedly increased in fibrotic liver diseases (Tu et al., 2012; Seo et al., 2013; Wang et al., 2013; Li X. et al., 2016). HMGB1 is an important inflammatory response mediator (Scaffidi et al., 2002), and HMGB1/toll-like receptors (TLRs) have been shown to play critical roles in liver inflammation and liver fibrosis (Mencin et al., 2009; Berzsenyi et al., 2011; Tu et al., 2012). Following BDL, TLR4-deficient mice exhibits significantly reduced inflammation and hepatic fibrosis, indicating that TLR4 is essential in liver fibrosis (Zhu et al., 2012; Hoshino et al., 2016). Moreover, HMGB1 also promotes the release of pro-inflammatory mediators by acting on its target receptors, leading to nuclear translocation of transcription factors such as NF-κB (Li X. et al., 2016). Therefore, the HMGB1/TLR4/NF-κB signaling pathway plays a pivotal role in liver fibrosis, and modulation of this signaling pathway is an appealing strategy for the inhibition of liver fibrosis.

Rosmarinus officinalis L (Lamiaceae) is an herbal plant that is extensively used by the food industry due to its beneficial health properties (Raškovic et al., 2014 ´ ). CA, phenolic compound that is extracted from the leaf of rosemary (Raškovic et al., 2014 ´ ), exhibits many pharmacological activities, including antisteatosis, antioxidant, antiapoptosis and antitumor activities (Jordan et al., 2012; Park and Mun, 2013; Petiwala et al., 2014; Shan et al., 2015). CA can stimulate SIRT1 activity, and subsequently mediates anti-apoptosis by deacetylating downstream factors, including p66Shc and CHREBP (Raškovic et al., 2014 ´ ; Yan et al., 2014; Shan et al., 2015; Gao et al., 2016). CA can alleviate alcoholic and non-alcoholic fatty liver (Raškovic et al., 2014 ´ ; Shan et al., 2015; Gao et al., 2016) and hepatic ischemia reperfusion injury in rats (Yan et al., 2014). Moreover, CA has been reported to alleviate renal fibrosis in rats by mediating the NOX signaling pathway (Jung et al., 2016). However, its effect on hepatic fibrosis and mechanism of action remain unclear. Alcoholic and non-alcoholic fatty liver disease, particularly in the late stages, are more likely to develop into the progression of liver fibrosis. We hypothesized that CA might have anti-liver fibrosis properties.

MicroRNA (miRNAs), which constitute a type of posttranscriptional regulators, are highly conserved, small, non-coding RNAs that bind to the 3<sup>0</sup> untranslated region (30 -UTR) complementary sequences of many target mRNAs. miRNAs are important modulators of pathophysiological processes (Neilson and Sharp, 2008; Pirola et al., 2015), are expressed in a developmental stage-specific and tissue-specific manner and specifically reduce mRNA stability (Neilson and Sharp, 2008). In recent years, an increasing number of polyphenols extracted from herbal medicines have been reported to exhibit pharmacological activity by acting on miRNA. Salvianolic acid B inhibits the Hh signaling pathway and reduces liver fibrosis by promoting the expression of miR-152 (Yu et al., 2015). CA exerts antitumor activity by down-regulating miR-15b (Gonzalez-Vallinas et al., 2014), and increases miR-34a to exert an anti-apoptotic effect (Shan et al., 2015).

Based on the miRNA database and our preliminary experiment, we found that miR-29b-3p might play a key role in control of HMGB1 expression in BDL-induced liver fibrosis. The main objectives of this study were as follows: (1) to elucidate the role of the antifibrosis effect of CA in protecting against BDL; (2) to test whether the activity of CA against liver fibrosis is associated with the signaling mechanisms of HMGB1/TLR4/NF-κB pathway in BDL; and (3) to investigate whether CA inhibits HMGB1/TLR4/NF-κB by promoting the expression of miR-29b-3p.

### MATERIALS AND METHODS

### Reagents

The purity of CA, which was purchased from Shanghai Winherb Medical Science Co., Ltd. (Shanghai, China), was 98%. CA was extracted from rosemary through ethanol extraction, and the coarse extracts were dissolved in macroporous adsorption resin, eluted with different concentrations of alcohol, and concentrated under reduced pressure.

The olive oil used was Aceite De Oliva Virgen Extra of BELLINA and was purchased from Joybuy.com (Beijing, China).

### Experimental Animals

Adult male Sprague-Dawley rats weighing between 180 and 220 g were obtained from the Experimental Animal Center of Dalian Medical University (Dalian, China). CA was dissolved in olive oil. The rats were maintained in a temperature-controlled chamber (22 ± 2 ◦C) with a 12-h light/dark cycle and a relative humidity of 40–60%. The rats had free access to food and water. Forty experimental rats were divided into five groups with eight rats per group: (A) sham-operated; (B) sham+CA (60 mg/kg/day); (C) BDL; (D) BDL-CA (30 mg/kg/day); and (E) BDL-CA (60 mg/kg/day). The abdominal cavities of the rats in Groups A and B were opened, and the common bile duct was isolated but not ligated. The bile ducts of the rats in Groups C, D, and E were ligated (Aubé et al., 2007). The

CA concentration administered intragastrically to the rats in Group D was 30 mg/kg/day. The rats in Groups B and E were given CA intragastrically at 60 mg/kg/day, whereas those of Groups A and C were intragastrically given the same amount of solvent olive oil. Twenty-four hours after surgery, the rats in Groups B, D, and E were subjected to the CA treatment for 21 days. After 3 weeks of treatment, the rats were sacrificed by anesthesia after an entire night of fasting. Serum and liver samples were collected for further experiments. The study was subject to approval by the institutional animal care committee of Dalian Medical University. All procedures in this study were performed according to institutional guidelines and the Guide for the Care and Use of Laboratory Animals.

### Cell Culture and Transfection

The human LX2 hepatic cell line was purchased from China Cell Culture Center (Shanghai, China). The cells were cultured in 1640 medium (GIBCO BRL, United States) that containing 10% fetal bovine serum (FBS, GIBCO BRL, United States) in an incubator with 5% CO<sup>2</sup> at 37◦C. Following the manufacturer's instructions, the cells were treated with 20 µM CA dissolved in DMSO and diluted with DMEM for 12 h before the extraction of total protein or RNA. Transfected experiments were performed using 2 µg pcDNA3.1/HMGB1, 50 nM mimic-miR-29b-3p, 50 nM mimic-miR-300 or 50 nM antagomiR-29b-3p (GenePharma) and Lipofectamine 3000 (Invitrogen, United States) according to the manufacturer's instructions. pcDNA3.1 (GenePharma) and a random RNA duplex (GenePharma) was used as negative control. After transfection for 24 h, the mimic group was harvested, the pcDNA3.1/HMGB1 and the antagomir group was cultured with or without 20 µM CA for another 12 h. LX2 cells were harvested post-transfection and processed for total RNA and protein extraction. The CA concentration and exposure time were determined based on the previous reports (Shan et al., 2015) combined with the cytotoxicity of CA on LX2 cells and L02 cells (**Supplementary Figure S1**). The cytotoxicty of transfection or plasmid with/without CA on LX2 cells were also assayed (**Supplementary Figure S2**). The supplementary results were obtained by the method described in the part of Supplementary Materials and Methods.

### Serum Levels of Total Bilirubin (Tbil), Alanine Aminotransferase (ALT), and Aspartate Aminotransferase (AST)

Blood samples were collected from the abdominal aorta and centrifuged at 3000 rpm for 15 min to obtain serum. The manufacturer's recommended protocols (Nanjing Jiancheng Corp, China) were used for measurements of the serum Tbil, ALT and AST levels.

### Liver Histological Examination

Liver tissue samples were embedded in paraffin and cut into 5-µm-thick sections for hematoxylin and eosin (H&E) staining, Masson staining and immunohistochemistry (IHC) staining (Tao et al., 2017). The sections were examined by light microscopy.

### Quantitative RT-PCR

Total RNA from livers and cells was isolated using the TRIzol reagent (TaKaRa, China) according to the manufacturer's protocols, and reverse transcribed using a TaqMan miRNA Reverse Transcription Kit. The RNA was quantified using an Applied Biosystems 7300 System (Applied Biosystems, United States) using Quantitative RT-PCR (qPCR) with TaqMan miRNA assay Kit (GenePharma Corp, China). The quantity and purity of total RNA samples were tested by UV spectroscopy (Thermo Fisher Scientific, United States). The miRNA expression level was normalized to endogenous expression of RNA U6.

### Western Blotting

Total protein was extracted from liver tissues, and cells were lysed with RIPA buffer, PMSF (Beyotime, China) and Cocktail (Biotool, China). Nuclear extracts were isolated with the Protein Ext Mammalian Nuclear and Cytoplasmic Protein Extraction Kit (TransGen Biotech). The protein from each sample was selected by electrophoresis in 8–12% SDS-PAGE gels (Bio-Rad, United States). These strips were cultured overnight with the designated primary antibodies: HMGB1 (Cell Signaling Technology, United States), TLR4 (Proteintech, China), NF-κB p65 (Proteintech, China), α-SMA (Proteintech, China), Col-1 (Abcam, British), TGF-β1 (ABclonal, China), MMP9 (Wanleibio, China), TGF-β R1 (Wanleibio, China) and GAPDH (Proteintech, China). After incubation with corresponding secondary antibodies at 37◦C, bands were exposed and developed after the addition of enhanced chemiluminescenceplus reagents (Advansta, United States). Western blot (WB) images were analyzed with a Gel-Pro Analyzer (Version 5.0; Media Cybernetics, Rockville, MD, United States).

### Dual-Luciferase Reporter

Dual-luciferase reporter plasmids of miR-29b-3p-HMGB1 were purchased from GenePharma Corp. (GenePharma, China). The plasmid and the miR-29b-3p mimic or the miR-29b-3p negative mimic were co-transfected into LX2 cells. Twenty-four hours post-transfection, the firefly and Renilla luciferase activities were measured with a Double-Luciferase Reporter Assay Kit (TransGen Biotec, China), and the firefly luciferase activity was normalized to the Renilla luciferase activity.

### Statistical Analyses

A two-tailed unpaired Student's t-test or one-way analysis was used for comparison among the groups, and for one-way ANOVA, Tukey's method was used for statistical analysis. (version 5.0; GraphPad Prism Software, United States). Differences with P-value < 0.05 were considered significant.

### RESULTS

### Effects of CA on Liver Injury in BDL Rats

To confirm the effects of CA on rat liver injury caused by BDL, we performed H&E to determine the degree of liver injury. As indicated by H&E staining, BDL-induced

the formation of regenerative nodules and prominent hepatic necrosis in rat liver tissues, and these effects were ameliorated by CA (**Figure 1A**). The dynamic alterations in the liver architecture of BDL rats after 3 weeks of CA treatment were reflected by quantitative biochemical assessments of liver damage. As shown in **Figures 1B–D**), the serum ALT, AST, and Tbil levels clearly increased in response to BDL compared with those of the sham-operated group (P < 0.01). CA treatment significantly decreased the ALT and AST levels in a dose-dependent manner (P < 0.01). However, the Tbil levels did not change after CA supplementation. These data confirmed that CA protects rats against BDL-induced liver injury.

### CA Ameliorates BDL-Induced Liver Fibrosis in Rats

Hepatic fibrosis is characterized by excessive ECM deposition, particularly type 1 collagens and α-SMA (Hernandez-Gea and Friedman, 2011). The effect of CA against BDL-induced hepatic fibrosis was evaluated through Masson staining, which is a classical histopathological technique used for observing collagen. In BDL rats, extensive accumulation of collagen was observed, and this accumulation was characterized by hyperplasia of the lattice fibers and collagenous fibers in the portal area without ward extension (**Figure 2A**). The expression of α-SMA in liver was revealed by immunohistochemistry staining (**Figure 2B**). It showed that CA supplementation markedly decreased α-SMA expression in the liver. Then we investigated whether CA could ameliorate BDL-induced fibrogenic gene expression by western blot. As is shown, CA significantly reduced the protein expression of α-SMA and Col-1 (**Figure 2C**). These results suggested that CA can alleviate BDL-induced liver fibrosis.

### CA-Mediated Protection against BDL-Induced Liver Fibrosis Involves the Down-Regulation of HMGB1, TLR4, and NF-κB

The HMGB1/TLR4/NF-κB signaling pathway induces the proliferation, migration and pro-fibrotic effects of HSCs and enhances the related collagen expression and pro-fibrotic cytokine production in liver fibrosis (Berzsenyi et al., 2011; Zhu et al., 2012; Hoshino et al., 2016). We measured changes in the protein expression of HMGB1 and TLR4 and the nuclear translocation of NF-κB in the liver through western blot. As shown in **Figure 3A**, the HMGB1, TLR4, and nuclear NF-κB p65 expression levels were significantly increased in the BDL group compared with those in the sham-operated group, and this up-regulation was abrogated by CA in a dose-dependent manner. We also observed the expression of the key protein HMGB1 by immunohistochemistry (**Figure 3B**), and the results were consistent with the western blot finding. These findings suggested that CA-mediated protection against BDL-induced liver fibrosis might involve the down-regulation of HMGB1, TLR4 and NF-κB.

## CA Inhibits LX2 Cell Activation by Inhibiting HMGB1/TLR4/NF-κB

To determine whether the CA-mediated reduction in the activation of LX2 cells was related to the inhibition of HMGB1 expression, we transfected LX2 cells treated with CA with an HMGB1 over-expression plasmid, and the protein expression levels of α-SMA, HMGB1, TLR4 and NF-κB p65 were detected by western blot. As shown in **Figure 4**, CA decreased the expression of α-SMA, HMGB1, TLR4 and nuclear NF-κB p65 protein in LX2 cells, and this decrease was abrogated by HMGB1 overexpression.

### Regulatory Effect of miR-29b-3p on HMGB1 Expression

miRNAs regulate gene expression by binding to the 3<sup>0</sup> -UTR of target gene mRNAs and control approximately 60% of mammalian genes (Neilson and Sharp, 2008). Therefore, we hypothesized that miRNAs might be associated with the regulation of HMGB1 in BDL-induced liver fibrosis. To investigate the miRNA expression profiles in cholestatic liver, we referred to the literature associated with the BDL model to determine the down-regulated miRNAs (Yang et al., 2012, 2015; Shifeng et al., 2013; Marzioni et al., 2014; Meng et al., 2014; Xiao et al., 2015; Wang et al., 2015) (**Table 1**). Among the 25 downregulated miRNAs, miR-300 and miR-29b-3p were predicted to bind to the 3<sup>0</sup> -UTR of HMGB1 mRNA. Therefore, the expression of the two miRNAs was further analyzed by qPCR. The qPCR results indicated that the miR-300 and miR-29b-3p levels were significantly lower (P < 0.01) in the BDL rats than in the controls (**Figure 5A**). Based on the miRNA database and the results of our experiment, we determined the effect of mimic-miR-300 and mimic-miR-29b-3p on the expression of HMGB1 in LX2 cells. As shown in **Figures 5B,C**, mimic-miR-29b-3p significantly decreased HMGB1 protein expression, whereas mimic-miR-300 had no effect on HMGB1 protein expression in LX2 cells.

### CA Reduces HMGB1 Expression and LX2 Activation by Enhancing miR-29b-3p

To investigate the effects of miR-29b-3p over-expression on the activation of HSCs, α-SMA expression in LX2 cells was determined by western blot. The results revealed that the level of α-SMA was decreased by treatment with mimic-miR-29b-3p (**Figure 6A**), indicating that miR-29b-3p can inhibit the activation of LX2 cells. Given that HMGB1 was predicted to be a putative target of miR-29b-3p (**Figure 6B**), the protein levels of HMGB1 were found to be decreased by mimic-miR-29b-3p (**Figure 5B**). We then generated an HMGB1 3<sup>0</sup> -UTR luciferase reporter containing the miR-29b-3p-binding sites (HMGB1 wild-type 3<sup>0</sup> -UTR) or mutated sites (HMGB1 Mut 3 0 -UTR). The construct was cotransfected into LX2 cells with the miR-29b-3p mimic or the miRNA negative control (con-mimic). The experimental results showed that the miR-29b-3p mimic significantly reduced the luciferase activity driven by the wildtype 3<sup>0</sup> -UTR of HMGB1 compared with the con-mimic in LX2 cells. Moreover, the luciferase activities of the mutated HMGB1

FIGURE 1 | Effects of CA on liver injury in BDL-induced rat models. (A) Histological changes in the liver of BDL rats visualized through hematoxylin and eosin (H&E) staining (x100). The hepatic necrosis and foci has been marked with arrows. The experimental groups subjected to H&E staining were as follows: sham; sham + CA (60 mg/kg); BDL; BDL + CA (30 mg/kg); and BDL + CA (60 mg/kg). (B) Serum ALT levels (n = 8); (C) serum aspartate aminotransferase (AST) levels (n = 8); (D) serum total bilirubin (TBil) levels (n = 8). The data are presented as the means ± SD. ∗∗P < 0.01 versus the sham group; ##P < 0.01 versus the BDL group.

3 0 -UTR and the empty vector were not inhibited by the miR-29b-3p mimic (**Figure 6C**). These results confirmed that HMGB1 is a direct target of miR-29b-3p.

For further studies on the effects of CA on HMGB1 in cells, we detected the expression of miR-29b-3p by qPCR. As shown in **Figures 5A**, **6D**, the in vivo, miR-29b-3p expression was significantly increased by CA treatment, and this increase was positively associated with the protective effect of CA in BDL rats. In vitro experiments revealed the same tendency. We subsequently transfected LX2 cells with the miR-29b-3p antagomir in the presence or absence of CA treatment and assessed the expression levels of HMGB1, TLR4, and α-SMA and the translocation of p65 to the nucleus. Significantly lower levels of HMGB1, TLR4, nuclear NF-κB p65 and α-SMA protein were observed in the CA treated compared with the control, and these proteins were increased by the miR-29b-3p antagomir (**Figure 6E**). Therefore, we concluded that CA might target miR-29b-3p to inhibit HMGB1 expression in BDL-induced


TABLE 1 | Down-regulation of putative HMGB1-targeting miRNA in BDL-induced rat model, which through the literature and TargetScan human database (http://www.targetscan.org/vert\_70/).

liver fibrosis. The results showed that CA decreased HMGB1 expression and LX2 activation at least partially in an miR-29b-3p-dependent manner.

### CA-Mediated Protection against BDL-Induced Liver Fibrosis Involves the Down-Regulation of MMP9, TGF-β1, and Its Receptor

Matrix remodeling and the TGF signaling pathway are also important in liver fibrosis (Hu et al., 2009; Li X.M. et al., 2016; Xu et al., 2016; Wu et al., 2017; Yu et al., 2017). We investigated whether CA treatment would alter the expression of MMP9 (a class of enzymes that involved in the degradation of the ECM), as well as TGF-β1 and its receptor, TGF-β R1, in BDL-induced liver fibrosis. The expressions of MMP9, TGF-β1, TGF-β R1 were abnormally increased in the BDL groups, and CA reversed these increases (**Figures 7A,B**). Furthermore, we found the same outcomes in LX2 cells (**Figures 7C,D**), indicating that CA-induced inhibition of hepatic fibrosis might be associated with the down-regulation of MMP9, TGF-β1, and the TGF-β1 receptor.

### DISCUSSION

Carnosic acid, a phenolic compound extracted from Rosmarinus officinalis L, is a well-known antiadipogenic and antioxidant agent (Petiwala et al., 2014; Raškovic et al., 2014 ´ ). In recent years, CA has been reported to regulate cell proliferation and differentiation, maintain normal cell function and exhibit antiapoptotic properties (Wang et al., 2011, 2012; Kar et al., 2012). In addition, CA reportedly ameliorates liver injury (Gao et al., 2016) and inflammation (Tang et al., 2016) and inhibits kidney fibrosis by inhibiting the NOX signaling pathway (Jung et al., 2016). However, the mechanisms of hepatic fibrosis regulation by CA are unclear. Here, we generated a BDL-induced liver fibrosis model exhibiting liver fibrosis and injury and we found that CA has favorable characteristics for the treatment of BDL-induced liver fibrosis, as indicated by the improved liver pathology, decreased expression of α-SMA and Col-1. Furthermore, we illuminated the molecular mechanisms through which CA protects against BDL-induced liver fibrosis.

It has been reported that chronic hepatic inflammation could result in hepatic fibrosis and cirrhosis (Schuppan and Kim, 2013). Accumulating evidence indicates that HMGB1 is closely involved in fibrotic diseases, including lung fibrosis, cystic fibrosis, liver fibrosis and pulmonary fibrosis (He et al., 2007; Hamada et al., 2008; Rowe et al., 2008; Tu et al., 2012), whereas the inhibition of HMGB1 signaling can act against experimental models of fibrotic disorders (Ge et al., 2011; Entezari et al., 2012). Moreover, HMGB1 can bind to cell surface receptors, such as RAGE, TLR2, and TLR4, to exert its effects (Mencin et al., 2009; Berzsenyi et al., 2011). In particular, TLR4 plays a crucial role

co-transfected with reporter constructs containing HMGB1 3'UTRs with (HMGB1-UTRwt) or without (HMGB1-UTRmt) miR-29b-3p-binding sites and the miR-29b-3p mimic or control (n = 3). The data are presented as the means ± SD. ∗∗P < 0.01 versus the control group. (D) The level of miR-29b-3p in LX2 cells after CA treatment for 12 h was assayed with qPCR (n = 3). The data are presented as the means ± SD. ∗∗P < 0.01 versus the control group. (E) The expression level of α-SMA, HMGB1, TLR4, and nuclear NF-κB p65 protein in LX2 cells were determined by western blot (n = 3). For functional analyses, LX2 cells were transfected with antago-miR-29b-3p or antago-miR-29b-3p control. After 24 h, the cells were exposed to 20 µM CA for 12 h. The data are presented as the means ± SD. ∗∗P < 0.01 versus the control group; #P < 0.05 versus the CA group.

in hepatic inflammation and liver fibrosis (Zhu et al., 2012; Wang et al., 2013; Hoshino et al., 2016). We found that the CA-mediated protection against BDL involves HMGB1/TLR4 down-regulation and the nuclear translocation of NF-κB. Furthermore, CA decreased the expression of α-SMA, HMGB1, and TLR4 protein and nuclear NF-κB p65 in LX2 cells; however, the decreases were abrogated by HMGB1 over-expression. Therefore, CA-induced protection against liver fibrosis is associated with the HMGB1/TLR4/NF-κB pathway.

In recent years, miRNAs have been identified as regulators of many biological processes (Neilson and Sharp, 2008). The silencing and over-expression of miRNAs might be involved in the progression of specific diseases, including fibrosis-related diseases (Sekiya et al., 2011; Yu et al., 2015, 2016). It has been reported that some miRNAs were down-regulated in BDL-induced liver fibrosis (Yang et al., 2012, 2015; Shifeng et al., 2013; Marzioni et al., 2014; Meng et al., 2014; Wang et al., 2015; Xiao et al., 2015). We predicted that miR-300 and miR-29b-3p bind to the 3<sup>0</sup> -UTR of HMGB1 mRNA and found that the expression of miR-300 and miR-29b-3p in the liver was consistent with that obtained in a previous report. We subsequently discovered that the protein level of HMGB1 decreased only following treatment with the miR-29b-3p mimic. In addition, the expression of α-SMA in LX2 cells was inhibited by mimic-miR-29b-3p. Luciferase assays also confirmed that HMGB1 is a target of miR-29b-3p. These data suggest that miR-29b-3p might play a key role in the control of HMGB1 expression.

The animal experiments revealed that CA treatment could significantly alleviate BDL-induced liver fibrosis in rats and reverse the decrease in miR-29b-3p. Therefore, we hypothesized that CA-induced protection against BDL-induced liver fibrosis occurred through miR-29b-3p up-regulation. We then explored this possibility in vitro and found that miR-29b-3p was up-regulated in LX2 cells after CA treatment. In addition, decreased HMGB1 expression was observed in the mimic-miR-29b-3p group compared with that in the control group. Importantly, the miR-29b-3p antagomir could reverse the CA-mediated inhibition of the HMGB1/TLR4/NF-κB pathway and α-SMA levels. Therefore, CA increases miR-29b-3p to inhibit the HMGB1/TLR4/NFκB pathway, thereby attenuating BDL-induced liver fibrosis.

Effects of drugs on disease is achieved through a variety of channels, while Matrix remodeling and the TGF signaling pathway have essential roles in fibrosis (Hu et al., 2009; Li X.M. et al., 2016; Xu et al., 2016; Wu et al., 2017; Yu et al., 2017). We found that CA provides protection against BDL-induced liver fibrosis by inhibiting MMP9 as well as TGF-β1 and its receptor, TGF-β R1. We will further investigate whether CA inhibits fibrosis relative to matrix remodeling and the TGF signaling pathway. We hope to provide more ideas for future research of CA and liver fibrosis.

In summary, our data indicate that the miR-29b-3p/HMGB1/TLR4/NF-κB signaling pathway might be essential for liver fibrosis and that CA could be a promising therapeutic agent in liver fibrosis by modulating the miR-29b-3p/HMGB1/TLR4 signaling pathway. The antifibrotic functions of CA require further clinical investigation for the treatment of patients with chronic liver diseases.

### AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: SZ, ZW, ZL, and JY. Performed the experiments: SZ, ZW, JZ, TX, YZ, HZ, and FT. Analyzed the data: SZ, ZW, JjZ, and DG. Wrote and provided suggestion regarding the manuscript: SZ, XT, and JY. Funding support: DG and JY. All authors reviewed the manuscript.

### ACKNOWLEDGMENTS

fphar-08-00976 January 17, 2018 Time: 16:30 # 10

This work received the support of grants from the Chinese National Natural Science Foundation (81473266 and 81503126).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar. 2017.00976/full#supplementary-material

FIGURE S1 | Detection of LX2 and L02 cell viability after CA treatment by CCK8 assay. (A) Impact of a 12-h treatment with different concentrations of CA, including 0, 5, 10, 20, and 40 µM to the survival rate of LX2 cells. (B) Effect of CA

### REFERENCES


(20 µM) on the survival rate of LX2 cells at different time including 0, 3, 6, 12, and 24 h. (C) Impact of treatment with CA (20 µM) for 12 h on the survival rate of L02 cells. The data are presented as the means ± SD (n = 8). <sup>∗</sup>P < 0.05 versus the control group; ∗∗P < 0.01 versus the control group.

FIGURE S2 | The cytotoxicity of transfection or plasmid with/without CA in LX2 cells was assessed through a CCK8 assay. (A) Survival rate of LX2 cells transfected with con-29b-3p/antagomir-29b-3p with/without CA. The cells were divided into five groups: control, con-29b-3p, con-29b-3p + CA, antagomir-29b-3p, and antagomir-29b-3p + CA. (B) Survival rate of LX2 cells transfected with pcDNA3.1 or pcDNA3.1/HHMGB1 with/without CA. The cells were divided into five groups; control, pcDNA3.1-control, pcDNA3.1-control + CA (20 µM), pcDNA3.1-HMGB1, and pcDNA3.1-HMGB1 + CA. The data are presented as the means ± SD (n = 8). <sup>∗</sup>P < 0.05 versus the control group; ∗∗P < 0.01 versus the control group; #P < 0.05 versus the antagomir-29b-3p or pcDNA3.1-HMGB1 group.

in vitro antioxidant and antimicrobial activities of Rosmarinus officinalis (L.) methanolic extracts. J. Agric. Food Chem. 60, 9603–9608. doi: 10.1021/jf302881t


essential oil and its hepatoprotective potential. BMC Complement. Altern. Med. 14:225. doi: 10.1186/1472-6882-14-225


**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 Zhang, Wang, Zhu, Xu, Zhao, Zhao, Tang, Li, Zhou, Gao, Tian and Yao. 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.

# Age-, Gender-, and in Vivo Different Doses of Isoproterenol Modify in Vitro Aortic Vasoreactivity and Circulating VCAM-1

Betzabé Nieto-Lima<sup>1</sup> , Agustina Cano-Martínez <sup>1</sup> \*, María E. Rubio-Ruiz <sup>1</sup> , Israel Pérez-Torres <sup>2</sup> and Verónica Guarner-Lans <sup>1</sup> \*

<sup>1</sup> Department of Physiology, Instituto Nacional de Cardiología "Ignacio Chávez", Mexico City, Mexico, <sup>2</sup> Department of Pathology, Instituto Nacional de Cardiología "Ignacio Chávez", Mexico City, Mexico

Different human-like cardiomyopathies associated to β-adrenergic stimulation are

#### Edited by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

#### Reviewed by:

Owen Woodman, RMIT University, Australia Gautham Yepuri, University of Fribourg, Switzerland

#### \*Correspondence:

Agustina Cano-Martínez cmamx2002@yahoo.com.mx Verónica Guarner-Lans gualanv@yahoo.com

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 08 August 2017 Accepted: 09 January 2018 Published: 24 January 2018

#### Citation:

Nieto-Lima B, Cano-Martínez A, Rubio-Ruiz ME, Pérez-Torres I and Guarner-Lans V (2018) Age-, Gender-, and in Vivo Different Doses of Isoproterenol Modify in Vitro Aortic Vasoreactivity and Circulating VCAM-1. Front. Physiol. 9:20. doi: 10.3389/fphys.2018.00020 experimentally modeled in animals through variations in dose, route, and duration of administration of different cardiotoxic drugs. However, associated changes in the vasculature and their relation to systemic inflammation, and the influence of cardiovascular diseases risk factors (gender and age) upon them are seldom analyzed. Here we studied the effect of age and gender on the vasoreactivity of aortas from mice subjected to in vivo repeated β-adrenergic stimulation with different doses of isoproterenol (ISO) in association with circulating inflammatory cytokines. Young (2 months) and old (18 months) male and female mice received 0 (control), 5, 40, 80 or 160 µg/g/d of ISO (7 days, s.c.). IL-1α, IL-4 and vascular cell adhesion molecule-1 (VCAM-1) were quantified in plasma. In vitro, norepinephrine-induced vasoconstriction and acetylcholine-induced relaxation were measured in aortas. No differences in contraction, relaxation, IL-1α, and IL-4 were found between control young males and females. Age decreased contraction in males and relaxation was lower in females and abolished in males. VCAM-1 was higher in young males than in females and increased in old mice. Vasoconstriction in ISO-treated mice results as a bell-shaped curve on contraction in young and old males, with lower values in the latter. In females, ISO-160 increased contraction in young females but decreased it in old females. Vasorelaxation was reduced in ISO-treated young males and females. ISO-80 and 160 reduced vasorelaxation in old females, and intermediate doses relaxed aortas from old males. VCAM-1 was higher in young and old males with ISO-80 and 160; while VCAM-1 was higher only with ISO-160 in old females. Our results demonstrate that repeated βadrenergic stimulation modifies vascular reactivity depending on gender, age, and dose. Females were less sensitive to alterations in vasoreactivity, and young females required a higher amount of the adrenergic stimuli than old females to show vascular alterations. Changes were independent of IL-1α and IL-4. VCAM-1 only changed in old females stimulated with ISO 160. Our results highlight the relevance of considering and comparing in the same study females and aged organisms to improve the accuracy of applications to clinical studies.

Keywords: gender, age, inflammatory cytokine, VCAM-1, isoproterenol, aortic vasoreactivity, NE-induced contraction, Ach-induced vasorelaxation

## INTRODUCTION

β-adrenoceptors (β-AR) are essential regulators of the cardiovascular homeostasis. They are located in the heart, but also in vascular smooth muscle cells, where they mediate vasodilating effects of endogenous catecholamines (Chruscinski et al., 2001). Their overstimulation with catecholamines is associated with heart damage and eventually heart failure (Grimm et al., 1998; Osadchii, 2007; Shao et al., 2013). Heart damage and failure decrease cardiac output and induce changes in vascular function leading to increased systemic resistance to compensate for the decreased output, thus helping to maintain perfusion to vital organs (Ledoux et al., 2003). In this paper, we study the effect of repeated daily β-adrenergic overstimulation with isoproterenol (ISO), a non-selective β-adrenergic agonist on aortic vascular reactivity.

β-adrenergic stimulation also increases circulating inflammatory cells (Mills et al., 2002; Barnes et al., 2015), elevating inflammatory mediators including vascular cell adhesion molecule-1 (VCAM-1) and tumor necrosis factor alpha (Chen et al., 2005; Han et al., 2016). Systemic inflammatory cytokines such as interleukin-1 alpha (IL-1α) are also increased when there is damage to organs including the cardiac tissue (Sprague and Khalil, 2009). In addition, the damaged heart is infiltrated with eosinophils that secrete interleukin-4 (IL-4) (Diny et al., 2017). Changes in circulating inflammatory mediators and cell adhesion molecules influence vascular function (Sprague and Khalil, 2009). Particularly VCAM-1 promotes the recruitment of inflammatory cells and other inflammatory cytokines (Cook-Mills et al., 2011). This inflammatory environment in the vascular tissue can trigger remodeling (Ganss et al., 2002; Meloche et al., 2017), cause stiffness of the aorta (Tomiyama et al., 2017) and produce vascular dysfunction, thus modifying the vascular compensatory function.

Experimentally, different pathological cardiac phenotypes result from the stimulation with ISO depending on the dose, route or duration of the administration. These phenotypes resemble different human-like cardiomyopathies including myocardial infarction (Hohimer et al., 2005; George et al., 2010), heart failure (Grimm et al., 1998), cardiac hypertrophy (Ma et al., 2011) and cardiomyopathies induced by stress such as takotsubo syndrome (Shao et al., 2013). However, changes in vascular function after the stimulation of the β-AR are seldom explored, and circulatory markers of inflammation are not usually measured (Davel et al., 2008; Han et al., 2016).

Furthermore, factors such as gender and age influence the incidence of cardiovascular diseases. Aging increases 10-fold the risk of cardiovascular morbidity between ages 50 and 80 (Ghebre et al., 2016). Prevalence of coronary heart disease is higher in men in all age ranges until after 75 years of age (Mosca et al., 2011). However, in women, relative mortality risk linked with cardiac hypertrophy which is the most potent cardiovascular risk factor after age (Levy et al., 1990) is higher (Deo et al., 2011). Aging is also accompanied by increases in inflammation and oxidative stress (El Assar et al., 2013; Wu et al., 2014), by decreases in the contractile responses of vascular smooth muscle and only in certain animal species by elevations in blood pressure (Harvey et al., 2015). The ability of β-AR to respond to catecholamine stimulation declines with age (Sato et al., 1995; Ferrara et al., 2014). Intact blood vessels from female murine models produce or release more endothelial-derived releasing factors such as nitric oxide and less endothelial-derived contracting factors (Kauser and Rubanyi, 1995). Also, the impaired endotheliumdependent aortic relaxation in old male mice is due to enhanced superoxide production via NADPH oxidase, while the relative preservation of endothelial function in female-old aortas may be due to enhanced superoxide scavenging (Takenouchi et al., 2009). Although gender and age play an important role in determining the incidence of cardiovascular diseases (Bhupathy et al., 2010), female and old organisms are less frequently considered as targets for the study of diseases, toxicity, drugs, and therapies.

Here we studied the differences in in vitro norepinephrine (NE)-induced contraction and acetylcholine (Ach)-induced vasorelaxation in young and old male and female mice, previously injected in vivo with different repeated doses of ISO. This study is important since sensitivity to heart damage caused by the stimulation with different doses of catecholamines that act on the β-adrenergic system is modulated by risk factors such as age and gender. In addition, there is little information on the relation of β-adrenergic stimulation, vasculature function and inflammation which potentiates damage, and therefore, our aim was also to measure the circulating inflammatory mediators IL-1α, IL-4, and VCAM-1 which are known to influence vascular function.

### MATERIALS AND METHODS

### Animals and in Vivo Treatment with Isoproterenol

All procedures followed the guidelines established by the Federal Regulation for Experimentation and Animal Care (SAGARPA, NOM-062-ZOO-1999, México) and the experimental protocol was registered and approved by our institution; protocol INCICH-10-695. Young (2 months old) and old (18 months old) male and female Balb/c mice were used. Animals were kept with food and water ad libitum and under a 12:12 light:dark cycle. Before treatment began, animals were inspected to ensure their good health. For each age and gender, five groups of four mice each were formed. Each group received subcutaneously one of the following doses of ISO for seven days: 5, 40, 80 and 160 µg/g/d. These doses were chosen considering reported minimal and maximal ranges that induce heart damage but not the death of the organisms (Wallner et al., 2016). The administration for 7 days was chosen since we reported that it produces damaging effects on the heart (Nieto-Lima et al., 2013, 2016). One group of each age and gender received saline solution as a vehicle. 24 h after the last administration, mice were weighed, sacrificed and blood was collected by heart puncture in syringes with EDTA. Plasma was obtained by centrifuging the blood at 2500 g for 10 min at 4◦C. Ventricles and aortas were removed. Ventricles were washed in phosphate buffer solution and weighed. The ventricular weight (VW) to body weight (BW) ratio (VW/BW) was used as a macroscopic evidence of heart damage. Aortas were immediately placed in oxygenated normal tyrode solution (containing in mM: 140 NaCl, 5 KCl, 1 CaCl2, 1 MgCl2, 5 HEPES, and 5.5 glucose; pH 7.4) and used to determine in vitro contraction to NE and relaxation to Ach.

### In Vitro NE-Induced Vasoconstriction and Ach-Induced Vasorelaxation in Aortas from Vehicle- and Isoproterenol-Treated Young and Old Male and Female Mice

Aortas from 4 mice of each group were obtained and carefully cleaned from connective and adipose tissue, taking care not to damage the endothelium. 1 or 2 segments from every aorta were used totaling 6–8 vascular reactivity assays. Tension measurements were made as previously described (Rubio-Ruiz et al., 2014). Briefly, a 1.5 g resting tension was applied to aortic rings (segments of about 3–4 mm long). This tension has been tested previously and found to be optimal under our experimental conditions. The aortas were allowed to rest for 60 min, with the replacement of the tyrode solution every 20 min. As in most studies of vascular reactivity (Kamata et al., 1989; Baños et al., 1997; Ponnoth et al., 2008; Rubio-Ruiz et al., 2014), the aortas were stimulated twice with NE (1 µmol/L). Endothelial integrity was tested by Ach induced relaxation (10 µmol/L) (Furchgott et al., 1984) in pre-contracted aortas with NE (1 µmol/L). Vasorelaxation was determined by cumulative concentration-response curves to Ach (10−4–10−<sup>9</sup> M) on NE- (1 µmol/L) precontracted aortic rings. The half maximal response to Ach (pEC50), expressed as –log10 of the molar concentration of EC50, and the maximum relaxation response (Emax) were calculated. To have an approach on the participation of nitric oxide (NO) on the vasorelaxation, we measured nitrates (NO<sup>−</sup> 3 ) and nitrites (NO<sup>−</sup> 2 ), as previously reported (Pérez-Torres et al., 2014). NO is a relaxing factor synthesized not only from Larginine by NO synthases (NOSs) but also from its inert metabolites, the nitrites and nitrates. NO<sup>−</sup> <sup>3</sup> was reduced to NO<sup>−</sup> 2 by nitrate reductase enzyme reaction. Ten µl of serum were added to 5 µl nitrate reductase (0.020 units) and 30 µl of buffer (0.14 M KHPO4, pH 7.35) and incubated for 30 min at 37◦C. At the end of the incubation period, 50 µl of sulfanilamide 1% and 50 µl of N-naphthyl-ethyldiamine 0.1% were added, and the total volume was adjusted to 1 ml. The calibration curve was obtained with a solution of KNO<sup>2</sup> ranging 5–0.078 M. The absorbance was measured at 540 nm.

### Inflammatory Profile Quantification

Plasma circulating levels of IL-1α (RRA00, R&D Systems), IL-4 (R4000, R&D Systems), and VCAM-1 (E-EL-R1061, Elabscience) were determined by enzyme-linked immunosorbent assay (ELISA) following the manufacturer's instructions using 10 µl of serum from all experimental animal groups (n = 4 per group).

### Statistical Analysis

Statistical analysis of vascular reactivity assays, pEC<sup>50</sup> and Emax were performed by two-way ANOVA, followed by Student-Newman-Keuls or Dunn tests, using the Sigma Stat program (Jandel Scientific). When comparing control values between gender and age, Student's t-test was used with the same program. Statistical analysis of plasma interleukins was made using Student's t-test with the Sigma Stat program (Jandel Scientific). Statistical analysis of body weight was performed by two-way ANOVA, followed by Sidak test, and VW/BW was performed by one-way ANOVA, followed by Dunnet test, using Prisma software. Results are expressed as the mean ± standard error of the mean (SEM). Differences were considered statistically significant when p < 0.05.

### RESULTS

### In Vitro NE-Induced Vasoconstriction and Ach-Induced Vasorelaxation of Aortas from Control Young and Old Male and Female Mice

The contraction induced by NE was slightly stronger in aortas from males than from females of 2 months of age (0.21 ± 0.02 g vs. 0.18 ± 0.03 g, respectively) but the difference was not statistically significant. In aged mice, contraction significantly decreased in male aortas, while it remained constant in females (old males: 0.06 ± 6.8<sup>∗</sup> 10−<sup>3</sup> g vs. old females: 0.17 ± 0.01 g, respectively).

Aortic rings exhibited a concentration-dependent vasorelaxation in response to Ach. The maximal relaxation was similar in aortas from male and female mice of 2 months of age (Emax = 76.7 ± 0.9% vs. 71.6 ± 0.4%, respectively) (**Table 1**). Vasorelaxation significantly decreased in aged female aortas (Emax = 71.6 ± 0.4% vs. 50.0 ± 2.2%); and in male aortas, vasorelaxation was abolished (**Figures 1A,B**). The EC<sup>50</sup> was not altered in any group (**Table 1**). There was a clear tendency of NO<sup>−</sup> 3 and NO<sup>−</sup> 2 to decrease with age which was not statistically significant in both genders (males: young 20.53 ± 4.07 nM/ml vs. old 12.82 ± 4.22; females: young 10.22 ± 4.86 vs. old 7.25 ± 2.13 nM/ml).

### In Vitro NE-Induced Vasoconstriction and Ach-Induced Vasorelaxation of Aortas from Isoproterenol-Treated Young and Old Male and Female Mice

In vivo treatment with different doses of ISO produced a bell-shaped curve on the NE-induced contraction in aortas from young male mice (**Figure 2A**); while in young females, only the highest dose of ISO significantly increased vascular contraction (**Figure 2B**). The contraction in old male mice showed the same tendency as aortas from young male mice but with a reduced contraction force (**Figure 2A**). In contrast, in aged female mice, only the in vivo treatment with the highest dose of ISO significantly decreased vascular contraction when compared to its control and young females for the same dose (**Figure 2B**).

As shown in **Figures 3A,B**, ISO treatment significantly decreased vascular relaxation of aortic rings from young male and female mice in a non-dose-dependent manner; however, the decrease in the maximum relaxation response was more



Values are means ± SEM, n = 6–8 vasoreactivity assays.

Aortas were pre-constricted with NE 1µM. ISO doses were 5, 40, 80, and 160 µg/g/d for 7 days. pEC<sup>50</sup> (–log<sup>10</sup> de EC50), sensitivity to Ach; Emax , maximum relaxation response; ISO, isoproterenol; ND, not detected. \*p < 0.05 vs. vehicle same group; &p < 0.05 vs. male same age and dose; #p < 0.05 vs. young same sex and dose determined by Student's t-test (controls) and two-way ANOVA and post hoc tests.

pronounced in males (50 vs. 30%, respectively; **Table 1**). ISO treatment in young male mice did not affect the sensitivity to Ach, while in young females, the dose of 5 µg/g/d significantly diminished the response to Ach, as observed from the increased value of pEC<sup>50</sup> (**Table 1**). In vivo treatment with 5, 40, and 80 µg/g/d of ISO relaxed the aortas from old male mice, and there was no relaxation with the highest dose (**Figure 3C** and **Table 1**). In old female mice, ISO significantly reduced the relaxation (in approximately 50%) with all of the doses (**Figure 3D** and **Table 1**). There were no significant changes in NO<sup>−</sup> 3 and NO<sup>−</sup> 2 concentration with any of the doses of ISO used in young or aged males and females.

### Plasma Concentration of Il-1α and Il-4 and Circulating VCAM-1 in Control and Isoproterenol-Treated Mice

Changes in plasma concentrations of circulating VCAM-1, IL-1α, and IL-4 are shown in **Figure 4**. Plasma concentrations of IL-1α and IL-4 were not different between vehicle-treated males and females and between vehicle-treated young and old mice. VCAM-1 was significantly higher in control males than females (44.5 ± 3.9 vs. 33.5 ± 9.1 pg/mL), and levels were increased with age, although they did not reach statistical significance. Old male mice had the highest levels of VCAM-1 when compared to the rest of the groups (by approximately 4.13–fold levels to the rest of the groups).

In vivo treatment with ISO did not modify plasma levels of IL-1α and IL-4 in any of the groups. Circulating VCAM-1 was considerably increased in young males with the doses of 80 and 160 µg/g/d (3.39- and 3.26-fold vs. its corresponding vehicle), without changes in old males. Meanwhile, circulating VCAM-1

FIGURE 1 | Ach-induced relaxation in NE-precontracted aortic rings from 2 months-old (continuous line) and 18 months-old (discontinuous line) male (A) and female (B) mice. Values are means ± SEM; n = 6–8 vasoreactivity assays. #p < 0.01 old males vs. young males; \*p < 0.05 old females vs. young females; &p < 0.01 old females vs. old males determined by Student's t-test.

was not modified by ISO in young females, but it was significantly increased with dose of 160 µg/g/d in old females (2.58-fold vs. control).

### Effect of in Vivo Treatment with Isoproterenol on Body Weight and VW/BW

Changes in ventricular weight (VW) to body weight (BW) ratio (VW/BW) were quantified to corroborate that the doses used induced heart damage. Therefore, the observed changes in vasoreactivity can be related to cardiac damage. **Table 2** shows changes in body weight (BW) and ventricular weight to body weight index (VW/BW) in young and old male and female mice treated with vehicle or ISO. BW of vehicle-treated male mice was significantly higher than that of females for both ages (young: 26.65 ± 0.55 g vs. 21.95 ± 0.52 g; old: 30.86 ± 0.71 g vs. 26.93 ± 0.66 g, respectively). Old mice had a higher weight than young mice in both genders (p < 0.05). VW/BW was higher in vehicle-treated males than in females in both ages (young: 4.7 ± 0.06 vs. 4.4 ± 0.08, old: 5.0 ± 0.07 vs. 4.1 ± 0.12, respectively) but only reached statistical significance in vehicletreated old mice. Treatment with ISO did not modify the gender and age-associated BW, but it significantly increased VW/BW in all groups.

months-old (open bars) male (A) and female (B) aortas from vehicle- and ISO-treated mice. Values are means ± SEM; n = 6–8 vasoreactivity assays. ISO doses were 5, 40, 80, and 160 µg/g/d for 7 days. ISO, isoproterenol. \*p < 0.05 vs. vehicle same group; &p < 0.05 vs. male same age and dose; #p < 0.05 vs. young same sex and dose determined by two-way ANOVA and post hoc tests.

### DISCUSSION

Overstimulation of the β-adrenergic system with daily doses of ISO causes heart damage inducing inflammation of the tissue and having consequences on the vasculature. In this paper, we studied the effects of repeated β-adrenergic overstimulation with different doses of ISO on vascular reactivity.

We found that repeated β-adrenergic stimulation with different doses of ISO modifies vascular reactivity (VR) in an age-, gender-, and dose-dependent pattern being young males the most susceptible to vascular alterations; whereas old mice tended to keep their homeostasis, but with different strategies according to the gender. Even though changes in vascular tone may be associated with systemic inflammation induced by the repeated β-adrenergic stimulation, in our study, we found no changes in the circulating levels of the inflammatory mediators IL-1α and IL-4 and only VCAM-1 seemed to be associated with vascular alterations in old females stimulated with the highest ISO dose. However, we cannot rule out that other biochemical and cellular markers of inflammation and local inflammation markers could be modified.

### Effect of Gender and Age on Vascular Reactivity and Inflammation

Gender and age modulate the incidence and progression of vascular and heart-related diseases. In the present study, we found no differences in VR between vehicle-treated young males and females. This response is in agreement with previous reports (Júnior and Cordellini, 2007; Loria et al., 2014).

Vasoconstriction usually declines with age due to decreases in cell density, aortic compliance and aortic contractility and increases in aortic diameter, the thickness of the medial layer of the aortic wall and collagen volume fraction (Wheeler et al., 2015). The decrease in endothelin-1 secretion could also decrease contraction (Leblanc et al., 2013) during aging. Our results corroborate that vascular contraction decreases with age but only in males. Similar results in aged animals have been previously reported (Gurdal et al., 1995; Shipley and Muller-Delp, 2005; Blough et al., 2007; Wheeler et al., 2015).

Gender differences in contraction during aging may be related to sex hormones. Estrogens improve endothelial function (Usselman et al., 2016), and inflammation (Zhu et al., 2013) throughout life. They lower levels of oxidative stress (Zhu et al., 2013; Angeloni et al., 2017). This decrease may translate into less damage to the structure and function of the vasculature and the heart. Testosterone may also be involved, since postmenopausal women with chronic heart failure are benefited by testosterone supplementation and hypertensive men have lower serum testosterone levels than normotensive men of the same age (Lopez-Ruiz et al., 2008).

Endothelium-dependent relaxation mediated by Ach has been reported to fall during rat maturation (Soltis and Newman, 1992). Our study corroborates that relaxation decreases with age and this response is more pronounced in males. One of the mediators of vasorelaxation is NO, and plasma nitrite serves as a reservoir of NO bioactivity (Cao et al., 2009; Waltz et al., 2015). Here we found a tendency of reduced levels of NO<sup>−</sup> 3 and NO<sup>−</sup> 2 during aging in males and females. However, we cannot exclude the participation of endothelium in relaxation because nitrates and nitrates were measured in plasma, which does not allow us to ensure that its source is exclusively the endothelium.

Besides cardiovascular modifications, aging also influences the inflammatory system. In the present study, circulating IL-1α and IL-4 were not different among individuals from different gender and age, but young vehicle-treated males had higher levels of circulating VCAM-1. At the vascular level, previous studies indicate that circulating pro-inflammatory factors may activate endothelial cells to promote an atherogenic phenotype, which may result in endothelial dysfunction, ventricular dysfunction and heart failure (Marchais et al., 1993; Belz, 1995; Ganss et al., 2002; Safar et al., 2012; Meloche et al., 2017; Tomiyama et al., 2017). Endothelium-independent regulation might also participate since the involvement of the endothelium was

not evidenced in our experiments. In accordance with the participation of VCAM-1 in the alteration of the heart and vascular function, increased levels of this molecule have been correlated with the development of cardiovascular diseases in healthy middle-aged men during a 6.6 years follow-up study (Schmidt et al., 2009). Even more, VCAM-1 has been proposed as a reference molecule for diagnosis of viral myocarditis (Gao et al., 2013) and is used to identify subjects at risk for events related to heart failure (Savic-Radojevic et al., 2013). Therefore, higher levels of VCAM-1 may contribute to a faster or sooner decline in male mice health, at least when compared to females. In line with this, we found that old males showed the highest value of VW/BW and VCAM-1 and an altered VR (reduced vasoconstriction and non-existent vasorelaxation with Ach). In contrast, old females only showed a slight reduction in vasorelaxation without significant changes in VW/BW, vasoconstriction, and VCAM-1.

### Effect of Isoproterenol on VR and Inflammation in Young and Aged Male and Female Mice

In the present work, the repeated stimulation of the β-AR system with different doses of ISO that caused heart damage also triggered different responses in VR and inflammation according to the gender and age of the organisms.

We found that, in response to β-adrenergic stimulation, young male aortas showed a bell-shaped contraction curve with increases in 3 of 4 doses of ISO and this response was decreased with age. In young female aortas, only the highest dose of isoproterenol produced a significant increase in contraction. In old females, a decrease in contraction was registered only with the highest dose of ISO. Gender differences in the susceptibility to β-adrenergic stimulation have been previously reported, being males more susceptible (Page et al., 2008; Elmes et al., 2009; Michel et al., 2017). In agreement with this, our results show that in young females the dose of catecholamines needed to trigger a vasoreactivity response is higher than in males. Thus, young females are more resistant to vascular responses triggered by β-adrenergic stimulation. Gender-related differences in VR have been previously reported in murine models of diabetes and hypertension (Robert et al., 2005; Takenouchi et al., 2010). These studies also showed that diseases modify the vasorelaxing response in females. There are also changes in VR of aortas of rats during pregnancy that are dependent on the gestational age (Jain et al., 1998); providing evidence of the effect of gender on VR under pathological conditions.

The β2-AR/Giα signaling pathway (Davel et al., 2014), and endothelial nitric synthase uncoupling mediated by oxidative stress (Davel et al., 2014), might participate in the increases in aortic contraction in young males by elevating ROS, generation and impairing NO bioavailability (Davel et al., 2006, 2008). Opposite to males, estrogens in females may protect them from alterations in contraction in most of the doses of ISO used. Estrogens reduce oxidative stress and improve endothelial

TABLE 2 | Body weight and ventricular weight/body weight ratio from young (2 months) and old (18 months) male and female mice treated with isoproterenol.


Values are means ± SEM, n = 4 mice. ISO doses were 5, 40, 80 and 160 µg/g/d for 7 days. BW, body weight; VW, ventricular weight; ISO, isoproterenol. #p < 0.05 vs. vehicle same group; \*p < 0.05 vs. male same age and dose; &p < 0.05 vs. young same sex and dose determined by one- (VW/BW ratio) and two-way ANOVA (BW) and post hoc tests.

function (Baylis, 2009). Under pro-inflammatory conditions, estrogens inhibit vasoconstrictor prostanoid production in endothelial cells and activity in intact arteries through G proteincoupled estrogen receptor (GPER) (Meyer et al., 2015).

In old males, a reduced aortic response is expected since the ability of β-AR to respond to the stimulation by catecholamines is declined. In females, the selective blockage of β2-AR increases the contraction to NE (Al-Gburi et al., 2017). Another interpretation might come from the activation of systems such as the renin-angiotensin system which is also influenced by age and gender (Costa et al., 2016).

In our study, ISO-administration was accompanied by a decrease in relaxation in young males and females at all doses. However, we did not find significant changes in the NO<sup>−</sup> 3 and NO<sup>−</sup> 2 levels with the ISO treatment with age nor with gender. It has been reported that decreases in β-adrenoceptor density (Kiuchi et al., 1993; Gaballa et al., 2001) play an important role in the β-adrenoceptor-mediated reduction in vasorelaxation in animal models of heart failure (Mathew et al., 1993; Nasa et al., 1996; McGoldrick et al., 2007). Autooxidation of catecholamines produces hydrogen peroxide and superoxide anion, and exogenous hydrogen peroxide

which might cause contractions in aortas from normotensive rats (Rodríguez-Martínez et al., 1998). Even more, superoxide anion degrades NO, an agent that increases the relaxation and suppresses the contraction of the vessels, leading to the formation of peroxynitrite, an oxidant capable of causing tissue damage (Bouloumié et al., 1997). Similar to autooxidation of catecholamines, β-AR overstimulation may induce degradation of NO through endothelial nitric oxide synthase uncoupling (by decreasing the concentration of biopterin), oxidative stress and inflammation (Davel et al., 2008; Xu et al., 2016). Protection by estrogens may attenuate oxidative stress and inflammation (Viña et al., 2013; Lamas et al., 2015), reducing the degradation of NO, and thus attenuating decreases in vasorelaxation. The latter agrees with the lower decrease in relaxation that we observed in young females compared to young males, although this data did not reach statistical significance. Furthermore, human female macrovessels express more dilatory β1- and β3-adrenoreceptors than male vessels. This gender-specific difference is attenuated with aging (Al-Gburi et al., 2017).

In aortas from old female mice, the highest doses of ISO significantly decreased the maximum response to Ach. Surprisingly, the treatment with ISO at 5, 40, and 80 µg/g/d induced vasorelaxation in aortas from old male mice. We do not know the exact reason for this effect, but we hypothesize that ISO could be mediating the vasorelaxant effect by the stimulation of β2-AR that activate the ATP-sensitive potassium channels in vascular smooth muscle (Fauaz et al., 2000).

The combination of the changes in relaxation and contraction that we observed suggests that young females are more resistant than young males to vascular responses triggered by β-adrenergic stimulation. This is reflected by the decreased relaxation with all the doses without changes in contraction with the exception of the 160 dose in females, while in males, relaxation decreased and contraction increased. This last combination is more likely to lead to the occlusion of the vessels. Opposite to young mice, old mice treated with ISO tended to keep the homeostasis of the vascular function with most of the doses of ISO, but probably through different mechanisms. For instance, in our study, contraction and relaxation increased with the majority of the doses in old males; while in old females there were no changes in vasorelaxation and vasoconstriction when they were treated with different doses of isoproterenol. These results highlight the relevance of including males and females young and old when trying to test new treatments.

Even if inflammation and changes in the vascular function have been associated with several conditions that involve the βadrenergic overactivation (Zeiher et al., 1994; Hadi et al., 2005; Esler et al., 2006; Stapleton et al., 2008; Hong, 2010; Webb et al., 2011; Sandoo et al., 2013; Moxon et al., 2014), our results show no correlation between the inflammatory mediators (IL-1α, IL-4) and VR. Similar results with a no-significant relationship between the chronotropic 25 dose (CD25) of ISO and interleukin 6 or soluble tumor necrosis factor receptor-1 have been previously reported (Euteneuer et al., 2012). However, we found that the levels of VCAM-1 were higher in males than in females and that they increased with age. VCAM-1 is a soluble biomarker of endothelial activation, which promotes a response of the macroand microvasculature to an injury involving several phenotypic changes of the luminal surface of endothelial cells (Jefferson et al., 2013). These changes are necessary for efficient leukocyte adherence and diapedesis to injured tissues. Aging is associated with an elevated level of a chronic endothelial activation (Belliere et al., 2015) and thus, VCAM-1 is expected to be high in this condition. Importantly, an increase in the levels in VCAM-1 and the lack of detection of other inflammatory cytokines have been reported in the coronary circulation of mice treated with β2 adrenergic blockers. This suggests that β-adrenergic stimulation could regulate VCAM-1 independently of other cytokines (Chen et al., 2005). However, further research on this issue is still needed. In general, our results support the idea that there are differences in male and female responses at different stages of life in association to the repeated overstimulation of the β-AR system.

In conclusion, our results showed that repeated β-adrenergic stimulation with different doses of catecholamines (ISO) modifies VR in an age-, gender- and dose-dependent manner being young males more susceptible to damage. Old mice tend to keep their homeostasis, but with different strategies according to the gender. These alterations in VR accompany the heart damage which was evidenced by the increases in VW/BW. A vicious cycle could be established with a tendency to threaten life unless interrupted. Although inflammation and vascular tone are associated with β-adrenergic stimulation, this association might not necessarily involve the participation of the inflammatory cytokines IL-1α and IL-4. Only circulating VCAM-1 levels showed changes in old females stimulated with the highest ISO dose. Therefore, other cellular and humoral participants of the inflammatory response should also be assayed in future studies. Importantly, these results highlight the relevance of considering the gender and age of organisms in studies aiming to determine mechanisms that lead to the incidence of diseases.

### AUTHOR CONTRIBUTIONS

AC-M and BN-L: designed the study, performed the treatments, obtained the heart and aortic tissues and plasma samples, drafted the interpreted results and wrote and reviewed the manuscript; MR-R: performed the assay of vascular reactivity and drafted the interpreted results; IP-T: performed the assay to determine the inflammatory and nitrite profile; VG-L: was responsible reviewing the manuscript.

### FUNDING

This work was supported by Grants from CONACyT-169736, INCICH-10-695, INCICH-12-758 to Agustina Cano-Martínez.

### ACKNOWLEDGMENTS

We thank Florencio Hernández for his help in the animal care.

### REFERENCES


function and morphology in rheumatoid arthritis: a six-year prospective study. Arthr. Res. Ther. 15:R203. doi: 10.1186/ar4396


**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 Nieto-Lima, Cano-Martínez, Rubio-Ruiz, Pérez-Torres and Guarner-Lans. 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 Modulatory Properties of Chronic Antidepressant Drugs Treatment on the Brain Chemokine – Chemokine Receptor Network: A Molecular Study in an Animal Model of Depression

Ewa Trojan, Joanna Slusarczyk, Katarzyna Chamera, Katarzyna Kotarska, ´ Katarzyna Głombik, Marta Kubera and Agnieszka Basta-Kaim\*

Department of Experimental Neuroendocrinology, Institute of Pharmacology, Polish Academy of Sciences, Kraków, Poland

#### Edited by:

Enrique Hernandez-Lemus, National Institute of Genomic Medicine, Mexico

### Reviewed by:

Dumitru A. Iacobas, New York Medical College, United States Jyothi Thyagabhavan Mony, University of Pittsburgh, United States

> \*Correspondence: Agnieszka Basta-Kaim basta@if-pan.krakow.pl

#### Specialty section:

This article was submitted to Inflammation Pharmacology, a section of the journal Frontiers in Pharmacology

Received: 16 May 2017 Accepted: 16 October 2017 Published: 01 November 2017

#### Citation:

Trojan E, Slusarczyk J, Chamera K, ´ Kotarska K, Głombik K, Kubera M and Basta-Kaim A (2017) The Modulatory Properties of Chronic Antidepressant Drugs Treatment on the Brain Chemokine – Chemokine Receptor Network: A Molecular Study in an Animal Model of Depression. Front. Pharmacol. 8:779. doi: 10.3389/fphar.2017.00779 An increasing number of studies indicate that the chemokine system may be the third major communication system of the brain. Therefore, the role of the chemokine system in the development of brain disorders, including depression, has been recently proposed. However, little is known about the impact of the administration of various antidepressant drugs on the brain chemokine – chemokine receptor axis. In the present study, we used an animal model of depression based on the prenatal stress procedure. We determined whether chronic treatment with tianeptine, venlafaxine, or fluoxetine influenced the evoked by prenatal stress procedure changes in the mRNA and protein levels of the homeostatic chemokines, CXCL12 (SDF-1α), CX3CL1 (fractalkine) and their receptors, in the hippocampus and frontal cortex. Moreover, the impact of mentioned antidepressants on the TGF-β, a molecular pathway related to fractalkine receptor (CX3CR1), was explored. We found that prenatal stress caused anxiety and depressive-like disturbances in adult offspring rats, which were normalized by chronic antidepressant treatment. Furthermore, we showed the stress-evoked CXCL12 upregulation while CXCR4 downregulation in hippocampus and frontal cortex. CXCR7 expression was enhanced in frontal cortex but not hippocampus. Furthermore, the levels of CX3CL1 and CX3CR1 were diminished by prenatal stress in the both examined brain areas. The mentioned changes were normalized with various potency by chronic administration of tested antidepressants. All drugs in hippocampus, while tianeptine and venlafaxine in frontal cortex normalized the CXCL12 level in prenatally stressed offspring. Moreover, in hippocampus only fluoxetine enhanced CXCR4 level, while fluoxetine and tianeptine diminished CXCR7 level in frontal cortex. Additionally, the diminished by prenatal stress levels of CX3CL1 and CX3CR1 in the both examined brain areas were normalized by chronic tianeptine and partially fluoxetine administration. Tianeptine modulate also brain TGF-β signaling in the prenatal stress-induced animal model of depression. Our results provide new evidence that not only prenatal stress-induced

**295**

behavioral disturbances but also changes of CXCL12 and their receptor and at less extend in CX3CL1–CX3CR1 expression may be normalized by chronic antidepressant drug treatment. In particular, the effect on the CXCL12 and their CXCR4 and CXCR7 receptors requires additional studies to elucidate the possible biological consequences.

Keywords: CXCL12, CX3CL1, chemokine receptors, hippocampus, frontal cortex, prenatal stress, antidepressant drugs, TGFβ/Smad pathway

### INTRODUCTION

fphar-08-00779 October 30, 2017 Time: 18:11 # 2

Depression is a common and serious mental disorder that affects nearly 350 million people worldwide (Murray et al., 2013). Despite the significant social burden that stems from this disease, there are still significant gaps in our scientific understanding of the biological basis and progression of this illness. Currently, psychiatric disorders, including depression, are believed to have a multifactorial origin that involves molecular, cellular, structural and functional dysfunctions in various brain areas, which makes the task of understanding the background of depression based on one hypothesis impracticable (Krishnan and Nestler, 2008). Due to the complexity of depression, unsurprisingly, pharmacotherapy directed toward one particular mechanism is often marginally effective. This issue explains why current antidepressant drug treatment is usually effective in only 50% of patients, and clinical data show that patients respond to this medication only after weeks or months of chronic treatment (Masi and Brovedani, 2011). Moreover, there is major socioeconomic pressure to find new, attractive targets for developing more effective strategies (Papakostas and Ionescu, 2015).

A growing body of evidence indicated that beyond the reciprocal influence of genetic and environmental factors, depressive symptoms are frequently associated with inflammatory processes (Dowlati et al., 2010; Kubera et al., 2011; Maes, 2011; Liu et al., 2012; Ogłodek et al., 2014). Among immune mediators, chemokines (chemotactic cytokines) are the main proteins responsible for the regulation of inflammatory processes in the brain. Chemokines are a group of small (8–14 kDa) polypeptides, which mediate their biological effects via interactions with 7-transmembrane G protein-coupled receptors. Their basic role in the periphery is the recruitment of leukocytes to maintain the function of the immune system (Gerard and Rollins, 2001; Moser et al., 2004). However, in the brain, data indicate that these proteins are also involved in the modulation of nervous system functions and in the restoration and maintenance of brain homeostasis (Stuart et al., 2015). Through the activation of diverse signaling pathways, chemokines regulate migration (Yin et al., 2013; Bardina et al., 2015; Merino et al., 2015), proliferation of neuronal stem/progenitor cells (Peng et al., 2007; Patel et al., 2012), control axon elongation (Pujol et al., 2005), synaptic pruning processes (Paolicelli et al., 2011) and blood–brain barrier (BBB) permeability. In addition to their role in neuromodulation, studies have shown the significance of chemokines in regulating neuroendocrine functions and mediating the activity of specific neurotransmitter and neuropeptide systems (Heinisch and Kirby, 2009; Di Castro et al., 2016). Recently, researchers have found that some chemokines are key in maintaining the interaction between neuronal and glial cells in both the developing and adult brain (Gómez-Gonzalo et al., 2017; Zhang et al., 2017). When the diverse activity of chemokines in brain was taken into account, chemokines were divided into categories based on their biological activity.

CX3CL1 (fractalkine) and CXCL12/SDF-1α (stromal cellderived factor-1α) are among the homeostatic chemokines that are constitutively expressed in the brain. CX3CL1 shows higher expression in the brain than in the periphery. Complementary expression of CX3CL1 mainly on neurons and CX3CR1 on microglia establishes a unique communication system between these cells, where CX3CL1–CX3CR1 signaling is responsible for control of microglial activation (Lyons et al., 2009). Importantly, our previous data demonstrated the antiinflammatory properties of CX3CL1 as well as its important role in the regulation of behavioral disturbances in an animal model of depression (Slusarczyk et al., 2016 ´ ). Another constitutively expressed chemokine in the brain is CXCL12 – a CXC-chemokine, which modulates the immune response, but its impact on neuronal development and plasticity and involvement in anxiety disorders mediation should be taken into account (Yang et al., 2016). In addition to the classical CXCR4 receptor, CXCL12 acts also via atypical CXCR7, which may operate as a β-arrestin-biased receptor.

Considering the chemokine system as the third major communication system of the brain, this study examine changes in expression of chemokine and chemokine receptors in animal model of depression. Moreover, we evaluated the impact of chronic administration of various antidepressant drugs: tianeptine – an atypical antidepressant, which enhances re-uptake of serotonin; venlafaxine – a serotonin – norepinephrine reuptake inhibitor (SNRI); and fluoxetine – a selective serotonin reuptake inhibitor (SSRI), on the gene expression and protein levels of CX3CL1 and CXCL12 and their receptors CX3CR1, CXCR4, CXCR7 in the hippocampus and the frontal cortex of adult rats in an animal model of depression. Taking into account previous data (Chen et al., 2002) that indicated a role of transforming growth factor β (TGF-β) signaling in the modulation of some chemokine receptor levels, particularly CX3CR1, and in search of the underlying mechanism of the impact of antidepressant drug treatment on chemokine receptor dysfunction in depression, we also focused on the activation of the canonical intracellular pathways linked to TGFβ and its receptors TGFβr1 and TGFβr2, such as phosphorylated Smad2/3, as well as Smad4 and Smad7 levels in the hippocampus and the frontal cortex of adult rats in an animal model of depression.

In the present study, we used a universally recognized animal model of depression based on a prenatal stress procedure (Morley-Fletcher et al., 2003). In this model, behavioral changes and abnormalities in the function of the neuroendocrine system were observed, which were normalized by chronic antidepressant treatment (Szymanska et al., 2009 ´ ; Budziszewska et al., 2010). Our previous study found that prenatal stress impairs the activity of the immune system not only in the periphery but also in the hippocampus and the frontal cortex, which are important brain areas in the pathogenesis of depression, leading to prolonged microglial activation and malfunction in the chemokine-chemokine receptor axis in adult offspring (Szczesny et al., 2014; Slusarczyk et al., 2016 ´ ).

### MATERIALS AND METHODS

### Animals

Sprague-Dawley rats (200–250 g upon arrival) that were obtained from Charles River (Sulzfeld, Germany) were maintained under standard conditions (at room temperature of 23◦C, 12/12 h light/dark cycle, lights on at 06:00 am), with food and water available ad libitum. Two weeks after arrival, vaginal smears were obtained daily from the female rats to determine the phase of the oestrous cycle. On the pro-oestrus day, the females were placed with males for 12 h and subsequently checked for the presence of sperm in vaginal smears. Pregnant females were randomly assigned to control and stress groups (n = 10 in each group). All experimental protocols were approved by the Committee for Laboratory Animal Welfare and Ethics of the Institute of Pharmacology, Polish Academy of Sciences, Cracow and met the criteria of the International Council for Laboratory Animals and Guide for the Care and Use of Laboratory Animals Consent procedure: 1037/2013.

### Stress Procedure

The prenatal stress procedure was performed as previously described (Morley-Fletcher et al., 2003). Briefly, pregnant females were subjected to three stress sessions daily, beginning on the 14th day of pregnancy and continuing until delivery. At 9:00 am, 12:00 pm and 5:00 pm, the rats were placed in plastic cylinders (7 cm × 12 cm) and exposed to bright light (150 W, 1800–2000 lx). Control, pregnant females were left undisturbed in their home cages. Male offspring were selected from 21-day-old litters for the experiment. They were housed in groups of five animals per cage (one or two animals from each litter) under standard conditions. At 3 months of age, the offspring of the control and stressed mothers underwent the first behavioral verification in the forced swim test (**Figure 1**).

### Forced Swim Test (FST)

The forced swim test (FST, Porsolt test) was conducted according to a previously described method (Detke et al., 1995). Briefly, the animals were individually subjected to two trials during which they were forced to swim in a cylinder (50 cm high, 18 cm in diameter) filled with water (23◦C) to a height of 35 cm. There was a 24-h interval between the first (pre-test) and second (test) trial. The first trial lasted 15 min, and the second trial lasted 5 min. The total durations of immobility, mobility (swimming) and climbing were measured throughout the second trial (Porsolt et al., 1978; Detke et al., 1995).

### Antidepressant Drug Administration

After the behavioral verification, the control and prenatally stressed offspring were divided into eight experimental groups (CONTROL+VEH, CONTROL+FLU, CONTROL+VEN, CONTROL+TIA, STRESS+VEH, STRESS+FLU, STRE SS+VEN, STRESS+TIA; six animals per group) and were treated with antidepressant drugs for 21 days. Fluoxetine (Eli Lilly, Indianapolis, IN, United States), venlafaxine (Sequoia Research, United Kingdom), and tianeptine (Tocris Bioscience, United Kingdom) were intraperitoneally injected once per day between 9:00 am and 10:00 am at a dose of 10 mg/kg, which was diluted in 0.9% saline. The CONTROL+VEH and STRESS+VEH groups received 0.9% saline (Polpharma, Poland). For pharmacological verification of the animal model of depression, animals underwent the elevated plus-maze and again forced swim procedures on the last days of chronic antidepressant drug treatment (according to the schedule illustrated in **Figure 1**).

### Elevated Plus-Maze Test

The elevated plus-maze test was performed as previously described (Pellow et al., 1985). The maze was elevated to a height of 50 cm above the floor and illuminated from below by a dim light (15 W). To allow the animals to adapt to the experimental conditions, they were placed in the experimental room for 1 h before the test. Each subject was individually placed in the central area of the maze facing the closed arm and observed for 5 min. The results are presented as the average number of entries into the open arms and the time in seconds (s) spent in the open arms. An entry was recorded when the animal entered the arm with all four limbs. The behavioral study was not blinded.

### Tissue Collection

Twenty-four hours after the last injection of antidepressant drugs, the animals were sacrificed under non-stress conditions by rapid decapitation. The frontal cortices (FCx) and hippocampi (Hp) were dissected onto ice-cold glass plates, and the tissues were frozen on dry ice and stored at −80◦C (for ELISA and Western blot assays) or at −20◦C in RNALater <sup>R</sup> solution (Applied Biosystems, United States) prior to total RNA extraction.

### Tissue Preparation and Determination of Protein Concentration

All tissue samples were placed in 2-ml Eppendorf <sup>R</sup> tubes filled with an appropriate buffer and homogenized using a Tissue Lyser II (Qiagen, Inc., Valencia, CA, United States). The samples were aliquoted and stored at −20 to −80◦C until use to avoid freeze– thaw cycles. In all experiments, the protein concentrations of the analyzed samples were determined using a BCA Protein Assay Kit (Sigma–Aldrich, St. Louis, MO, United States). Protein concentrations were measured at a wavelength of 562 nm using

a Tecan Infinite 200 Pro spectrophotometer (Tecan, Mannedorf, Germany) in triplicates for each sample using bovine serum albumin as a standard.

### Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR)

Total RNA was isolated using a RNeasy Mini Kit (Qiagen, Hilden, Germany). The samples were homogenized in an appropriate volume of the lysis buffer supplied with the kit, and isolation of total RNA was performed with strict adherence to the manufacturer's instructions. RNA concentrations were measured using a NanoDrop ND-1000 Spectrophotometer (Thermo Scientific, Wilmington, DE, United States).

Identical amounts of RNA (1 µg) were reverse – transcribed into cDNA using a commercial RT-PCR kit (Applied Biosystems, Foster City, CA, United States). All reactions were run under the conditions recommended by the manufacturer (total reaction volume 20 µl). The cDNAs were subsequently amplified using PCR Master Mix (Applied Biosystems, Foster City, CA, United States) and TaqMan probes and primers for the following genes: CX3CL1 (Rn00593186\_m1), CX3CR1 (Rn00591798\_m1), CXCL12 (Rn00573260\_m1), CXCR4 (Rn01483207\_m1), and CXCR7 (Rn00584358\_ m1). Amplification was performed using a 20 µl mixture containing PCR Master Mix, the cDNA used as the PCR template, TaqMan forward and reverse primers and 250 nM of a hydrolysis probe labeled at the 5<sup>0</sup> -end with the fluorescent reporter FAM and at the 3<sup>0</sup> -end with a quenching dye. The thermal cycling conditions were 2 min at 50◦C and 10 min at 95◦C, followed by 40 cycles at 95◦C for 15 s and 60◦C for 1 min. The threshold value (Ct) for each sample was set during the exponential phase of the PCR, and the 11Ct method was used for data analysis. The expression levels were normalized to the Ct for beta-2 microglobulin (b2m) (Rn00560865\_m1) as a reference gene.

### Enzyme-Linked Immunosorbent Assay (ELISA)

The tissue samples were homogenized in PBS buffer (Sigma–Aldrich, St. Louis, MO, United States) containing protease inhibitor cocktail (Sigma–Aldrich, St. Louis, MO, United States), phosphatase inhibitor cocktail (Sigma–Aldrich, St. Louis, MO, United States), 1 mM sodium orthovanadate (Sigma–Aldrich, St. Louis, MO, United States), and 1 mM phenylmethanesulfonyl fluoride (Sigma–Aldrich, St. Louis, MO, United States). The lysates were shaken in an ice bath for 30 min and cleared by centrifugation (14,000 rpm, 4◦C, 30 min).

The levels of CX3CL1 (fractalkine, Cloud Clone Corporation, Houston, TX, United Stats), CX3CR1 (fractalkine receptor, Cusabio, Washington, DC, United States), SDF-1 (Cloud Clone Corporation, Houston, TX, United States), CXCL12/SDF-1 receptor 4 (CXCR4, MyBiosource, San Diego, CA, United States), CXCL12/SDF-1 receptor 7 (CXCR7, MyBiosource, San Diego, CA, United States) and transforming growth factor β (TGFβ, Cloud Clone Corporation, Houston, TX, United States) in the cortical and hippocampal homogenates were measured using a commercially available enzyme-linked immunosorbent assay (ELISA) kit. Briefly, standards or samples (50 or 100 µl) were pipetted into 96-well plates coated with rat CX3CL1, CX3CR1, CXCL12/SDF-1, CXCR4, and CXCR7 antibodies and incubated. After the plates were extensively washed, HRPconjugated streptavidin was pipetted into the wells, and the samples were incubated. The wells were washed and 3,3<sup>0</sup> ,5,5<sup>0</sup> tetramethylbenzidine (TMB) was added. In this assay, the color develops in proportion to the concentration of the measured protein. The reactions were terminated by the addition of a stop solution. The absorbance was determined using a Tecan Infinite 200 Pro (Tecan, Mannedorf, Germany) system set to the appropriate wavelength (nm). The detection limits were as follows: CX3CL1, 0.054 ng/ml; CX3CR1, 23.5 pg/ml; CXCL12/SDF-1, 6.5 pg/ml; CXCR4, 1 pg/ml; CXCR7, 0.1 ng/ml; and TGFβ, 5.5 pg/ml. Inter-assay precision was as follows: CX3CL1 < 12%, CX3CR1 < 10%, CXCL12/SDF-1 < 12%, CXCR4 < 10%, CXCR7 < 10%, and TGFβ < 12%. Intra-assay precision was as follows: CX3CL1 < 10%, CX3CR1 < 8%, CXCL12/SDF-1 < 10%, CXCR4 < 8%, CXCR7 < 8%, and TGFβ < 10%. Positive controls for each assay were provided by the manufacturers.

### Western Blot

The tissue samples were homogenized in 2% SDS buffer (Sigma–Aldrich, St. Louis, MO, United States). Samples containing equal amounts of protein were heated at 95◦C for 5 min in 4x Laemmli sample buffer (Bio-Rad, Hercules, CA, United States). Next, proteins were separated by SDS-PAGE (4–20% gel) under constant voltage (200 V) and were electrophoretically transferred to PVDF membranes (Trans-Blot Turbo; Bio-Rad, Hercules, CA, United States). The blots were blocked in 5% blocking buffer (5% bovine serum albumin) for 1 h at room temperature (RT) and incubated overnight at 4◦C with the following primary antibodies that had been diluted in a

SignalBoost Immunoreaction Enhancer Kit (Millipore, Warsaw, Poland): anti- Smad2/3 (3102, Cell Signaling, Danvers, MA, United States), anti-phospho-Smad2/3 (8828, Cell Signaling, Danvers, MA, United States), anti-Smad4 (9515, Cell Signaling, Danvers, MA, United States), anti-Smad7 (TA322546, OriGene, Rockville, MD, United States), anti-TGFβr2 (TA347477, OriGene, Rockville, MD, United States), and anti-TGFβr1 (AP01457PU-N, Acris Antibodies, Herfold, Germany). The blots were then incubated at RT with one of the following peroxidaseconjugated secondary antibodies: goat anti-rabbit IgG HRP (PI 1000, Vector Laboratories, Peterborough, United Kingdom) or horse anti-mouse IgG HRP (PI-2000, Vector Laboratories, Peterborough, United Kingdom) for 1–2 h. The immune complexes were detected using Pierce <sup>R</sup> ECL Western Blotting Substrate (Thermo Fisher, Pierce Biotechnology, Carlsbad, CA, United States) and visualized using a Fujifilm LAS-1000 System (Fuji Film, Tokyo, Japan). The blots were washed two times for 5 min each in TBS; stripped using stripping buffer containing 100 µl of Tris-HCl (pH = 6.7), 2% SDS and 700 µl of 2-mercaptoethanol (all from Sigma–Aldrich, St. Louis, MO, United States); washed two additional times for 5 min each in TBS; blocked; and reprobed with an antibody against GAPDH (MAB374, Millipore, Warsaw, Poland) as an internal loading control at a dilution of 1:5000 in a SignalBoost Immunoreaction Enhancer Kit. All membranes were stripped twice. The relative levels of immunoreactivity were densitometrically quantified using Fujifilm Multi Gauge software (Fuji Film, Tokyo, Japan).

### Statistical Analysis

The outcomes of the behavioral studies are presented as the means ± SEM. The data obtained in the ELISA study are presented as weight units (pg or ng) per milligram of protein ± SEM; those for RT-PCR are presented as the average fold ± SEM, and for Western blot analysis, the results are presented as percentage of the control ± SEM. The normality of variable distribution and homogeneity of variances were checked by Shapiro–Wilk test and Levene's test, respectively. The significance of the differences between the means was evaluated by one- or two-way analysis of variance (ANOVA), with Duncan's post hoc test if appropriate. When the assumptions of ANOVA were not fulfilled, Kruskal–Wallis ANOVA (by ranks) for multiple comparisons was used. A value of p < 0.05 was considered statistically significant. All of the statistical analyses were performed using Statistica software, version 10.0 (Statsoft, Tulsa, OK, United States).

### RESULTS

### The Impact of Prenatal Stress and Chronic Antidepressant Drug Administration on the Immobility, Swimming, and Climbing Times of the Forced Swim Test

Consistent with previous reports (Budziszewska et al., 2010; Głombik et al., 2016), the prenatal stress procedure significantly prolonged immobility time in the FST (F1,<sup>57</sup> = 100.65; p < 0.05; **Table 1**). Moreover, when compared to control animals, adult rats exposed to the prenatal stress procedure exhibited a reduction in swimming (F1,<sup>57</sup> = 100.66; p < 0.05; **Table 1**), as well as climbing time (F1,<sup>57</sup> = 54,56; p < 0.05; **Table 1**), indicating depressive-like behavior.

Next, to determine whether chronic tianeptine, venlafaxine, or fluoxetine administration affected the behavioral changes evoked by prenatal stress, we assessed the rats during the FST again. As we previously demonstrated, enhanced immobility time (p < 0.05), shortened swimming (p < 0.05), and climbing (p < 0.05) were detected in prenatally stressed offspring compared with control offspring. Furthermore, we revealed a significant effect of drugs (F3,<sup>39</sup> = 13.02; p < 0.05; **Figure 2A**) on the immobility time. Post hoc comparisons revealed that tianeptine (p < 0.05), venlafaxine (p < 0.05), and fluoxetine (p < 0.05) shortened immobility time in prenatally stressed offspring. We also observed a significant effect of the drugs (F3,<sup>39</sup> = 13.02; p < 0.05; **Figure 2B**) on swimming time. Moreover, post hoc comparisons revealed that tianeptine (p < 0.05), venlafaxine (p < 0.05), and fluoxetine (p < 0.05) extended the swimming time in stressed offspring. Regarding climbing time (**Figure 2C**), we observed that only tianeptine (F3,<sup>38</sup> = 7.94; p < 0.05) prolonged the climbing time in prenatally stressed rats.

### The Impact of Prenatal Stress and Chronic Antidepressant Drug Administration on Anxiety-Like Behavior in the Elevated Plus-Maze Test

To assess anxiety-like behavior in adult rats, we performed the elevated plus-maze test. As we previously demonstrated (Głombik et al., 2017), the prenatal stress procedure leads to a significant reduction in the number of entries into the open arms (F1,<sup>34</sup> = 32.46; p < 0.05; **Figure 2D**) of the maze and a significant decrease in the time spent in them (F1,<sup>34</sup> = 88.57; p < 0.05; **Figure 2E**). However, we did not observe differences in the number of entries into the closed arms and the time spent in them (data not shown).

Post hoc comparisons showed that tianeptine (p < 0.05) and venlafaxine (p < 0.05) enhanced the number of entries into the open arms of the maze (p < 0.05) and the time spent in them (p < 0.05).

TABLE 1 | The effect of prenatal stress on the times for immobility, swimming, and climbing in the forced swim test.


The results are presented as the mean ± SEM, <sup>∗</sup>p < 0.05 in comparison to control group, n = 24–26 for each group. Statistics: one-way ANOVA.

means ± SEMs, with n = 5–6 for each group. <sup>∗</sup>p ≤ 0.05 vs. control Veh group; #p ≤ 0.05 vs. prenatally stressed Veh group. ANOVA (two-way), followed by Duncan's test or Kruskal–Wallis ANOVA (by ranks) for multiple comparisons.

### The Impact of Prenatal Stress and Chronic Antidepressant Drug Administration on the mRNA Expression and Protein Levels of CX3CL1 and Its Receptor CX3CR1 in the Hippocampus and the Frontal Cortex of Adult Offspring Rats

In the present study, the analyses of tissue samples revealed that the prenatal stress procedure did not affect the gene expression of CX3CL1 (**Table 2A**) in the hippocampus. However, in the frontal cortex of prenatally stressed offspring, the mRNA expression of CX3CL1 was significantly diminished (F1,<sup>39</sup> = 5.97; p < 0.05). Chronic administration of tianeptine (p < 0.05) and fluoxetine (p < 0.05) normalized the prenatal stress-induced changes in CX3CL1 expression in the frontal cortex (**Table 2B**). On the other hand, ELISA revealed significantly diminished protein levels of CX3CL1 in both the hippocampus (F1,<sup>36</sup> = 9.67; p < 0.05; **Figure 3A**) and the frontal cortex (F1,<sup>36</sup> = 7.66; p < 0.05; **Figure 3B**). Only chronic administration of tianeptine


TABLE 2 | The effect of prenatal stress (PS) and antidepressant drugs treatment [tianeptine (Tia), venlafaxine (Ven), or fluoxetine (Flu)] on the mRNA expression of chemokine and chemokine receptors in the hippocampus (A) and frontal cortex (B) of adult rats.

Results are expressed as average fold ± SEM. The number of animals in each group: n = 5–6. <sup>∗</sup>p ≤ 0.05 vs. control Veh group; #p ≤ 0.05 vs. prenatally stressed Veh group Statistics: two-way ANOVA followed by Duncan test. When the assumptions of ANOVA were not fulfilled, Kruskal–Wallis ANOVA (by ranks) for multiple comparisons was used. p < 0.05. Bolded values are statistically significant.

FIGURE 3 | The effect of prenatal stress and antidepressant drugs treatment [tianeptine (Tia), venlafaxine (Ven), or fluoxetine (Flu)] on CX3CL1 level (ng/mg of protein) in the hippocampus (A) and frontal cortex (B) and on the level of its receptor – CX3CR1 (pg/mg of protein) in these brain areas (C,D). The data are presented as the means ± SEM, with n = 5–6 for each group. <sup>∗</sup>p < 0.05 vs. control Veh group; #p < 0.05 vs. prenatally stressed Veh group. ANOVA (two-way), followed by Duncan's test.

ANOVA (two-way), followed by Duncan's test.

normalized the changes in CX3CL1 levels caused by prenatal stress in both brain areas (p < 0.05; **Figures 3A,B**).

In the next set of experiments, analyses of hippocampal (F1,<sup>33</sup> = 19.28; p < 0.05) and cortico-frontal (F1,<sup>38</sup> = 4.92; p < 0.05) samples revealed a significant down-regulation of the mRNA expression of CX3CR1 in prenatally stressed rats compared with control animals. Chronic treatment with tianeptine normalized the gene expression changes in both brain areas (p < 0.05; **Tables 2A,B**). Furthermore, we demonstrated that the prenatal stress procedure-induced diminished protein levels of CX3CR1 in the hippocampus (F1,<sup>35</sup> = 0.59; p < 0.05; **Figure 3C**) and the frontal cortex (F1,<sup>39</sup> = 18.20; p < 0.05; **Figure 3D**) were normalized by tianeptine treatment (p < 0.05). Moreover, fluoxetine (p < 0.05) attenuated the changes in

CX3CR1 protein levels caused by prenatal stress in the hippocampus (**Figure 3C**).

### The Impact of Prenatal Stress and Chronic Antidepressant Drug Administration on the mRNA Expression and Protein Levels of CXCL12 and Its Receptors: CXCR4 and CXCR7 in the Hippocampus and the Frontal Cortex of Adult Offspring Rats

The results of ANOVA showed a significant increase in CXCL12 expression in the hippocampus (F1,<sup>33</sup> = 35.26; p < 0.05) of prenatally stressed offspring (**Table 2A**). Chronic administration of tianeptine (p < 0.05) normalized these changes. As shown in **Figure 4**, we observed statistically significant up-regulation of CXCL12 levels in the hippocampus (F1,<sup>38</sup> = 41.15; p < 0.05; **Figure 4A**) and the frontal cortex (F1,<sup>37</sup> = 40.67; p < 0.05; **Figure 4B**). Chronic administration of tianeptine (p < 0.05), venlafaxine (p < 0.05), and fluoxetine (p < 0.05) in the hippocampus, while tianeptine (p < 0.05) and venlafaxine in the frontal cortex (p < 0.05) normalized the changes in CXCL12 levels caused by prenatal stress (**Figures 4A,B**).

In the next set of experiments, we reported a significant decrease in the gene expression levels of CXCR4 in the hippocampus (F1,<sup>39</sup> = 28.59; p < 0.05) and the frontal cortex (F1,<sup>38</sup> = 20.39; p < 0.05; **Tables 2A,B**). Chronic administration of tianeptine (p < 0.05) and venlafaxine (p < 0.05) normalized these changes only in the hippocampus. Our experiments showed that the prenatal stress procedure diminished the CXCR4 concentration in the hippocampus (F1,<sup>38</sup> = 2.42; p < 0.05; **Figure 4C**) and the frontal cortex (F1,<sup>39</sup> = 0.17; p < 0.05; **Figure 4D**). We observed the impact of venlafaxine (p < 0.05) in the hippocampus and the effect of fluoxetine (p < 0.05) in the frontal cortex on the CXCR4 concentration in control animals. Among the tested antidepressants, post hoc comparison found that only fluoxetine treatment enhanced CXCR4 levels in the hippocampus of prenatally stressed offspring (p < 0.05; **Figure 4C**).

Since CXCL12 also exerts biological activity via CXCR7, we therefore examined the gene expression and protein levels of CXCR7 in both the hippocampus and the frontal cortex. A significant increase in CXCR7 expression levels in the hippocampus (F1,<sup>38</sup> = 68.25; p < 0.05) and the frontal cortex (F1,<sup>39</sup> = 45.81; p < 0.05) of prenatally stressed offspring was observed, which was attenuated only by chronic tianeptine treatment (p < 0.05; **Tables 2A,B**).

Furthermore, our study demonstrated that neither prenatal stress nor chronic antidepressant administration affected CXCR7 concentrations in the hippocampus (**Figure 4E**). In contrast, prenatally stressed rats exhibited significantly increased levels of CXCR7 in the frontal cortex (F1,<sup>47</sup> = 1.08; p < 0.05), while chronic tianeptine (p < 0.05) or fluoxetine (p < 0.05) administration normalized this parameter (**Figure 4F**).

### The Impact of Prenatal Stress and Chronic Antidepressant Drug Administration on the Protein Levels of TGFβ and Both Receptors, TGFβr1 and TGFβr2, in the Hippocampus and the Frontal Cortex of Adult Offspring Rats

As reported previously, the malfunction in the TGFβ pathway may play an important role in the regulation of the brain CX3CL1–CX3CR1 axis (Chen et al., 2002). Therefore, we examined the effect of the prenatal stress procedure and chronic antidepressant drug administration on TGFβ, TGFβr1, and TGFβr2 levels in the hippocampus and the frontal cortex. Regarding the differences between control and prenatally stressed rats, a significant decrease in TGFβ protein levels in the hippocampus (F1,<sup>37</sup> = 13.68; p < 0.05) and the frontal cortex (F1,<sup>37</sup> = 25.56; p < 0.05) was observed (**Figures 5A,D**). Moreover, we demonstrated that only tianeptine (p < 0.05) ameliorated prenatal stress-induced reductions in TGFβ levels in the hippocampus and frontal cortex (**Figures 5A,D**), while venlafaxine (p < 0.05) administration only in the frontal cortex (**Figure 5D**).

Next, we showed that neither prenatal stress nor antidepressant treatment affected the levels of TGFβr1 or TGFβr2 in the hippocampus of adult offspring (**Figures 5B,C**). In contrast, in **Figures 5E,F**, we demonstrated significant down-regulation of TGFβr1 (F1,<sup>30</sup> = 5.75; p < 0.05) and TGFβr2 (F1,<sup>26</sup> = 2.20; p < 0.05) in the frontal cortex of adult offspring after prenatal stress. Tianeptine (p < 0.05) treatment normalized changes in both TGFβr1 (**Figure 5E**) and TGFβr2 concentrations (**Figure 5F**), while venlafaxine (p < 0.05) only normalized changes in TGFβr2. Additionally, our study found that venlafaxine and fluoxetine diminished TGFβr2 levels in control rats (**Figure 5F**) (Supplementary Data Sheet 1).

### The Impact of Prenatal Stress and Chronic Antidepressant Drug Administration on the Canonical TGFβ Receptors Pathways in the Hippocampus and the Frontal Cortex of Adult Offspring Rats

Since data demonstrated that Smad signaling is essential for TGFβ gene responses, we measured the impact of the prenatal stress procedure and antidepressant administration on the level of the phosphorylated active form of Smad2/3 (pSmad2/3/Smad2/3) as well as protein levels of Smad4 and Smad7 (**Figure 6**). We found that prenatal stress did not affect pSmad2/3/Smad2/3 and Smad4 levels in the hippocampus of adult prenatally stressed animals (**Figures 6A,B**). However, the level of Smad7 in this structure in prenatally stressed offspring was significantly up-regulated (F1,<sup>28</sup> = 5.42; p < 0.05; **Figure 6C**) and normalized by chronic tianeptine (p < 0.05) or venlafaxine (p < 0.05) administration (**Figure 6C**). In the frontal cortex, we did not observe the impact of either prenatal stress or antidepressant drug treatment on pSmad2/3/Smad2/3

(**Figure 6D**) and Smad4 (**Figure 6E**) levels. However, in control offspring, the down-regulation of Smad4 was noted following treatment with each antidepressant. Importantly, as in the hippocampus, the level of Smad7 in the frontal cortex of prenatally stressed offspring was up-regulated (F1,<sup>35</sup> = 10.96; p < 0.05; **Figure 6F**) (Supplementary Data Sheet 1).

### DISCUSSION

Prenatal stress is a well-recognized animal model of depression. The usefulness of this procedure as a model of depression has been verified by reports of compliance with the requirements for construct, face and predictive validity (Morley-Fletcher et al.,

2003). Our study confirmed that behavioral deficits are present in the offspring of rat dams that were stressed during the last week of pregnancy (Trojan et al., 2016; Głombik et al., 2017). Moreover, prolonged immobility and diminished swimming time were normalized by chronic tianeptine, fluoxetine, or venlafaxine treatment. Additionally, tianeptine administration attenuated the deficits in climbing behavior caused by stress. Tianeptine and venlafaxine attenuated anxiety-like behavior evoked by prenatal stress, as assessed by an increase in the number of entries and in the time spent in the open arms of the maze.

Our study demonstrated that chronic treatment with antidepressants, in addition to the beneficial impact on the behavioral alterations, attenuated in drug-dependent manner the malfunction evoked by prenatal stress in the chemokine – chemokine receptor network in the hippocampus and the frontal cortex of adult offspring. Considering that CXCL12,

CX3CL1 and their receptors play a crucial role in brain homeostasis and the course of pathological conditions, e.g., neuroinflammation, disturbances in these molecules expression, have been proposed as a potential background of brain disorders, including depression.

In the present study, we clearly demonstrated that prenatal stress increases CXCL12 levels in the hippocampus and the frontal cortex of adult offspring. Importantly, elevated CXCL12 levels in the hippocampus were normalized by chronic administration of all tested antidepressants, while levels in the frontal cortex by tianeptine and venlafaxine treatment.

Data show that CXCL12 is expressed under homeostatic conditions, whereas its expression is strongly up-regulated during inflammation, hypoxia or ischemia. However, recent data showed that CXCL12 may also have an anti-inflammatory property that mediates immune cell recruitment, which leads to limited inflammation. Thus far, only some experimental data demonstrated CXCL12 changes in the brain in stressrelated model (Slusarczyk et al., 2015 ´ ). Interestingly, a potential GABA-mediated inhibitory role of secreted CXCL12 on the serotoninergic system, which may lead to depressive behavior, has been observed (Réaux-Le Goazigo et al., 2013). In line with this finding, an increase in the levels of CXCL12 and RANTES has been observed in patients with major depression, and suppression of these molecules has been found following treatment with antidepressant drugs (e.g., fluoxetine) (Shen et al., 2010). Notably, enhanced brain CXCL12 levels may also potentiate TNF-α release that not only causes production of glutamate, which directly affects astrocytes, but also leads to activation of receptors on microglia. Consequently, enhanced TNF-α production can lead to neurotoxicity (Calì et al., 2008, 2010). Therefore, in our study, the normalizing impact of antidepressants on CXCL12 levels in the studied brain areas may have a broad, indirect anti-inflammatory value.

CXCL12 is well-known to mediate biological function in brain through the two following receptors: classical CXCR4 and non-classical CXCR7 (Banisadr et al., 2002, 2003). We observed that prenatal stress diminished the levels of CXCR4 in both examined areas. However, chronic fluoxetine treatment attenuated prenatal stress-induced changes only in the hippocampus. Since CXCR4 modulates neurotransmitter release and hormone secretion from neuroendocrine cells (Bezzi et al., 2001) and is crucial in the modulation of brain inflammation (Schönemeier et al., 2008), the observed malfunction of the CXCR12–CXCR4 axis in the present study may interfere with these processes. The diminished expression levels of CXCR4 may also affect neuronal viability of rat brain cortical neurons, as well as several metabolic parameters (Merino et al., 2016). Interestingly, the results of Felszeghy et al. (2004) showed that dexamethasone down-regulates CXCR4 receptor expression. Therefore, we may postulate that enhanced levels of endogenous glucocorticoids in prenatally stressed rats (Szymanska et al., 2009 ´ ) may be responsible for decreased CXCR4 brain expression. To date, the mechanisms responsible for the impact of antidepressants on the CXCL12–CXCR4 brain axis have not been studied in any animal model of depression,

so further understanding of the fluoxetine signaling pathways downstream of CXCR4 will be crucial.

In contrast to CXCR4 in the present study, we observed enhanced CXCR7 receptor levels, which were normalized by tianeptine or fluoxetine treatment, in the frontal cortex of prenatally stressed rats. CXCR7 does not activate Gα<sup>i</sup> signaling but acts as a β-arrestin-biased receptor (Levoye et al., 2009; Rajagopal et al., 2010). The broad expression of CXCR7 receptor in the brain may suggest CXCR4-independent function of CXCL12 (Banisadr et al., 2016). Therefore, in the present study, prenatal stress procedure-induced down-regulation of CXCR4 and up-regulation of CXCR7 receptors may lead to disparate effects of CXCL12. Our study is the first showing the ability of chronic fluoxetine treatment to normalize the brain CXCL12- CXCR4-CXCR7 axis in an animal model of depression.

The present study demonstrated that among antidepressants tested only chronic tianeptine administration normalized the diminished protein levels of CX3CL1 and CX3CR1 in both the hippocampus and the frontal cortex of adult offspring. Thus far, data concerning the impact of antidepressant drugs on CX3CL1–CX3CR1 brain signaling in animal models of depression are scarce and ambiguous. Treatment with imipramine and agomelatine did not reverse the changes in the dorsal hippocampus evoked by chronic mild stress, while chronic imipramine or lurasidone treatment diminished CX3CR1 mRNA expression in these structures (Rossetti et al., 2016). Recently, the Alboni et al. (2016) group demonstrated that the impact of chronic fluoxetine administration on the CX3CL1–CX3CR1 axis in the hippocampus depends on the quality of the living environment. In stressful conditions, CX3CL1 gene expression was reduced in fluoxetine-treated mice, whereas in enriched conditions, no changes were found. In the case of venlafaxine, its chronic treatment ameliorated depressive-like behavior and restored microglial morphology in control (wide type) animals, whereas resistance to stress-induced depressive-like behavior and changes in microglial morphology after venlafaxine treatment in CX3CR1 deficient mice was recently noted (Hellwig et al., 2016). In the present study, however, venlafaxine had no effect on CX3CR1 and CX3CL1 expression in both brain regions – though its anxiolytic and anti-depressive properties were observed. Therefore, further studies are needed to unequivocally identify the involvement of CXCL1–CXCR1 axis in behavioral effects of this drug.

On the other hand, in our previous functional study, we demonstrated that 7 days after icv CX3CL1 treatment, adult prenatally stressed rats showed an increased swimming and climbing time and a decreased immobility time. Furthermore, CX3CL1 administration reduced stress-induced anxiety-like behavior (Slusarczyk et al., 2016 ´ ). Taking into account this observation, we can speculate that, in contrast to venlafaxine, beneficial effect of tianeptine on behavioral disturbances observed in adult offspring may be at least partially related with normalization of the CX3CL1–CX3CR1 axis. In the context of our results, one particularly relevant observation is that proper CX3CL1 and CX3CR1 concentrations are required for maintaining microglia in a "resting state" through modulation of their activity and pro-inflammatory factor release (Cardona

et al., 2006; Corona et al., 2010; Frick et al., 2013). Thus, the normalization of the brain CX3CL1–CX3CR1 axis by tianeptine may be the result of its suppressive effect on microglial overactivation and pro-inflammatory factor release (Slusarczyk et al., ´ 2015), which was demonstrated in our previous study in prenatally stressed offspring. The limitation of our study is the fact that we performed our research only in male offspring, whereas recent data showed sex differences in microglial density, morphology and activation in hippocampus and frontal cortex (Schwarz et al., 2012). Moreover, the sex-dependent expression of CX3CL1–CX3CR1 in brain in basal condition was demonstrated (Bollinger et al., 2016), which might determine not only the susceptibility to stress, the profile of behavioral disturbances but maybe also tianeptine action.

Several studies indicate that TGFβ is an interesting suppressor of microglial cell activation (Suzumura et al., 1993; Lodge and Sriram, 1996; Chen et al., 2002). In this study, we demonstrated diminished concentrations of TGFβ in the hippocampus and the frontal cortex of prenatally stressed rats. Tianeptine treatment enhanced TGFβ levels in both examined areas, while venlafaxine increased levels in the frontal cortex. In line with these results, rats subjected to a chronic mild stress (model of depression) exhibit significant decreases in TGFβ in the brain (You et al., 2011), while depressive patients show a reduction in TGFβ serum levels (Sutcigil et al., 2007; Musil et al., 2011). Interestingly, TGFβ levels showed a significant negative correlation with the Hamilton Depression Rating Scale (HDRS) (Myint et al., 2005), while TGFβ levels increased in depressed patients after 6 weeks of treatment with fluoxetine, venlafaxine, or paroxetine (Lee and Kim, 2006). Because TGFβ signaling is mainly localized to microglia (Kiefer et al., 1995), it is considered the alternative regulator of their activation (Spittau et al., 2013). Specifically, TGFβ increased CX3CR1 and reduced IL-1β expression in activated microglia, leading to resolution of microglial activation and a return to baseline behavior after LPS challenge (Henry et al., 2009; Wynne et al., 2010). Moreover, in aged mice, TGFβ inhibition resulted in an inflammatory phenotype of microglia, including enhanced IL-1β and lower CX3CR1 expression levels. Additionally, the mRNA levels of both TGFβ and CX3CR1 dynamically change in a corresponding time-dependent manner, indicating a relationship between these two protein systems. Because our study found that tianeptine up-regulated TGFβ release, this signaling pathway should be taken into account as a conceivable mechanism of the action of tianeptine on CX3CR1 concentrations in the hippocampus and the frontal cortex of prenatally stressed rats.

In the present study, we examined the canonical TGFβ signaling cascade, which initiates TGF-β binding to the type II transmembrane receptor serine/threonine kinases (TGFβr2) that in turn assemble with, phosphorylate and activate the type I receptor (TGFβr1; ALK5). We observed that prenatal stress diminished TGFβr2 and TGFβr1 levels in the frontal cortex. We also reported that chronic tianeptine treatment normalized the stress-induced malfunction in both TGFβ receptors, while venlafaxine only affected TGFβr2. Activated TGFβr1 may phosphorylate the downstream effectors Smad2 and Smad3, which then associate with Smad4 (Edlund et al., 2003; Bakkebø et al., 2010). Additionally, alternative non-Smad pathways, including ERK1/2, JNK, p38 MAPK and the tyrosine kinase Src or PI3K, can be activated (Moustakas and Heldin, 2005). In our study, neither prenatal stress nor antidepressant treatment affected Smad2/3 phosphorylation, while all tested antidepressants diminished Smad4 levels in control offspring. In contrast to our study, Dow et al. (2005) indicated that chronic antidepressant treatment increases TGFβ-mediated phosphorylation of Smad2 (pSmad2). The lack of impact of antidepressants on pSmad2 in our study may be due to the fact that Smad2 and 3 are also phosphorylated and activated by several other ligands and receptor complexes of the TGFβ family or directly by other kinases (Moustakas and Heldin, 2009). Moreover, other pathways may induce inhibitory Smads (I-Smads), consequently suppressing TGFβ receptor signaling. Among these signaling molecules, TNF-α and IL-1β induce Smad7 (Bitzer et al., 2000; Mitchell et al., 2014). Interestingly, our results showed that prenatal stress up-regulated Smad7 levels in the hippocampus and the frontal cortex and thus exerted an inhibitory effect on TGFβ signaling. Moreover, chronic tianeptine or venlafaxine treatment diminished prenatal stress-induced up-regulation of Smad7 in the hippocampus. In parallel, Smad7 interferes with TGFβ signaling by binding to TGFβr1 to prevent Smad2/3 phosphorylation and activation or recruitment of the protein phosphatase or the ubiquitin ligases to the receptor, leading to either receptor dephosphorylation or proteasomal degradation (Koinuma et al., 2003; Kamiya et al., 2010).

Taken together, the outcomes of the present study show that prenatal stress leads to anxiety and depressive-like disturbances in adult animals. Moreover, for the first time, we demonstrated that the evoked by prenatal stress procedure dysfunctions of constitutively expressed brain chemokines, CXCL12, and their receptors and at less extend CX3CL1 and CX3CR1, in hippocampus and frontal cortex of adult offspring, were attenuated by chronic antidepressant drug treatment. This drugdependent action seems to be bi-directional and manifests as inhibition of pro-inflammatory CXCL12 expression and partially stimulation of anti-inflammatory CX3CL1 and TGFβ release.

### AUTHOR CONTRIBUTIONS

AB-K, ET, and JS were responsible for the conception and design ´ of the study. ET, JS, KG, and KK performed behavioral analyses. ´ ET, JS, KC, and KK were responsible for biochemical analyses ´ of the samples. ET was responsible for the interpretation of the data. ET and JS drafted the article. MK helped to write the final ´ version of the manuscript. All authors revised the paper critically for important intellectual content and gave final approval of the version to be published.

### FUNDING

This work was supported by grant no. 2013/09/B/NZ7/04096, National Science Centre, Poland and partially by the statutory

founds of the Department of Experimental Neuroendocrinology Polish Academy of Sciences. ET and KC are recipients of scholarships from the KNOW, sponsored by the Ministry of Science and Higher Education, Poland. Publication charge was supported by KNOW funds MNiSW-DS-6002-4693-26/WA/12.

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar. 2017.00779/full#supplementary-material



possible role of suppressors of cytokine signaling proteins. J. Neuroimmunol. 276, 37–46. doi: 10.1016/j.jneuroim.2014.08.001


dependent upon hypoxia. CNS Neurosci. Ther. 19, 145–153. doi: 10.1111/cns. 12049


**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 Trojan, Slusarczyk, Chamera, Kotarska, Głombik, Kubera and ´ Basta-Kaim. 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.

# Much More than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain

Luiz H. A. Cavalcante-Silva<sup>1</sup> , Éssia de Almeida Lima<sup>2</sup> , Deyse C. M. Carvalho<sup>3</sup> , José M. de Sales-Neto<sup>1</sup> , Anne K. de Abreu Alves <sup>2</sup> , José G. F. M. Galvão<sup>1</sup> , Juliane S. de França da Silva<sup>1</sup> and Sandra Rodrigues-Mascarenhas 1, 2, 3 \*

<sup>1</sup> Programa de Pós-Graduação em Produtos Naturais e Sintéticos Bioativos, Laboratório de Imunobiotecnologia, Centro de Ciências da Saúde, Universidade Federal da Paraíba, João Pessoa, Brazil, <sup>2</sup> Programa de Pós-Graduação em Biotecnologia, Laboratório de Imunobiotecnologia, Centro de Biotecnologia, Universidade Federal da Paraíba, João Pessoa, Brazil, <sup>3</sup> Programa Multicêntrico de Pós-graduação em Ciências Fisiológicas, Laboratório de Imunobiotecnologia, Centro de Biotecnologia, Universidade Federal da Paraíba, João Pessoa, Brazil

#### Edited by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

#### Reviewed by:

Bruno Vogt, University of Bern, Switzerland Timothy J. Moss, Ritchie Centre, Australia

\*Correspondence: Sandra Rodrigues-Mascarenhas sandra@cbiotec.ufpb.br

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 06 September 2017 Accepted: 24 October 2017 Published: 10 November 2017

### Citation:

Cavalcante-Silva LHA, Lima ÉA, Carvalho DCM, Sales-Neto JM, Alves AKA, Galvão JGFM, Silva JSF and Rodrigues-Mascarenhas S (2017) Much More than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain. Front. Physiol. 8:895. doi: 10.3389/fphys.2017.00895 Since the discovery of ouabain as a cardiotonic steroid hormone present in higher mammals, research about it has progressed rapidly and several of its physiological and pharmacological effects have been described. Ouabain can behave as a stress hormone and adrenal cortex is its main source. Direct effects of ouabain are originated due to the binding to its receptor, the Na+/K+-ATPase, on target cells. This interaction can promote Na<sup>+</sup> transport blockade or even activation of signaling transduction pathways (e.g., EGFR/Src-Ras-ERK pathway activation), independent of ion transport. Besides the well-known effect of ouabain on the cardiovascular system and blood pressure control, compelling evidence indicates that ouabain regulates a number of immune functions. Inflammation is a tightly coordinated immunological function that is also affected by ouabain. Indeed, this hormone can modulate many inflammatory events such as cell migration, vascular permeability, and cytokine production. Moreover, ouabain also interferes on neuroinflammation. However, it is not clear how ouabain controls these events. In this brief review, we summarize the updates of ouabain effect on several aspects of peripheral and central inflammation, bringing new insights into ouabain functions on the immune system.

Keywords: ouabain, immune system, peripheral inflammation, cell migration, cytokines, neuroinflammation

### INTRODUCTION

Although the cardiotonic steroid ouabain was originally identified as a plant secondary metabolite (e.g., from Strophantus gratus and Acokanthera ouabaio), it was later described as an endogenous mammalian substance (Hamlyn et al., 1991) such as other cardiotonic steroids (e.g., marinobufagenin and digoxin) (Bagrov et al., 2009). Ouabain was found in bovine adrenal gland (Laredo et al., 1994; Schneider et al., 1998), adrenal gland tumors (Blanco and Wallace, 2013), bovine hypothalamus (Tymiak et al., 1993), bovine hypophysis (Schoner, 2002), and human plasma (Hamlyn et al., 1991; Ferrandi et al., 1997). It is noteworthy that ouabain isolated from mammalian tissues and body fluids is structurally, biochemically, and immunologically indistinguishable to ouabain isolated from plants (Schoner, 2002).

Ouabain synthesis seems to occur in the zona glomerulosa and fasciculata of the adrenal gland cortex (Masugi et al., 1988; Laredo et al., 1995), using hydroxycholesterol, pregnenolone, and progesterone (Hamlyn et al., 1998; Schoner and Scheiner-Bobis, 2007) as precursors. Thereafter, ouabain is released into the circulation after stimulation by adrenocorticotropic hormone (Lewis et al., 2014), epinephrine (Schoner and Scheiner-Bobis, 2005), angiotensin II (Laredo et al., 1997), and α1-adrenergic receptor agonists (Schoner, 2002; Schoner and Scheiner-Bobis, 2007). In rats, the physiological role of ouabain is associated to vasculature tone control and natriuresis (Nesher et al., 2009).

Ouabain levels are increased in different conditions such as chronic renal insufficiency (Stella et al., 2008), chronic salt intake (Blanco and Wallace, 2013), congestive heart failure (Manunta et al., 2009, 2010), hypertension (Hauck and Frishman, 2012), pregnancy (Dvela-Levitt et al., 2015), and primary hyperaldosteronism (Rossi et al., 1995). This steroid is also associated with stress conditions, such as physical exercise (Antolovic et al., 2000; Bauer et al., 2005). In addition, high levels of ouabain are correlated with cortisol concentration (Berendes et al., 2003), which reinforces its role as a stress hormone.

Besides its role as a Na+/K+-ATPase (sodium pump) inhibitor (Lingrel, 2010), which is associated with cardiovascular effects (Hamlyn and Blaustein, 2013; Blaustein et al., 2016), ouabain, at low concentrations, triggers Na+/K+-ATPasemediated signaling pathways (Xie and Askari, 2002; Xie and Cai, 2003). These relayed signals cascades include Src kinase, MAPK, and NF-κB activation, reactive oxygen species release and others (Saunders and Scheiner-Bobis, 2004; Aperia, 2007). Ouabain induces several biological regulatory effects, including cell proliferation, hypertrophy, apoptosis (Bagrov et al., 2009) resulting in different functional outcomes. Additionally, it has been demonstrated that ouabain modulates various immune system functions (Rodrigues-Mascarenhas et al., 2009), including inflammation. In this mini-review, we present the relationship between ouabain and inflammatory process.

### OUABAIN AND IMMUNE SYSTEM

The immune system is a highly specialized network of lymphoid organs, cells, humoral factors, and cytokines, which acts in order to maintain homeostasis (Parkin and Cohen, 2001). The relationship between ouabain and immune system was first studied when Quastel and Kaplan (1968) demonstrated that this steroid inhibits lymphocytes proliferation induced by the mitogen phytohaemagglutinin. This effect was later confirmed by several other reports that used different stimuli (e.g., anti-CD3 and IL-2) (Jensen et al., 1977; Dornand et al., 1986; Redondo et al., 1986; Olej et al., 1994; Brodie et al., 1995; Szamel et al., 1995). This phenomenon could be related to CD25 (Pires et al., 1997) and IL-2 reduced expression (Dornand et al., 1986; Szamel et al., 1995) induced by ouabain, since both molecules are required for lymphocyte proliferation. Moreover, ouabain reduces regulatory T cells absolute number in mice (Silva et al., 2015). It is noteworthy that CD25 is highly expressed and fundamental to regulatory T cells survival (Setiady et al., 2010). Besides that, it was also reported that ouabain induces cell death in stimulated lymphocytes (Olej et al., 1998; Esteves et al., 2005; Panayiotidis et al., 2010).

In thymocytes, T lymphocyte precursors, ouabain is able to modulate different events such as intracellular calcium concentration increase (Echevarria-Lima et al., 2003). It was also observed that this effect is related to CD69 increased expression, a molecule associated with cell activation, induced by ouabain (Rodrigues-Mascarenhas et al., 2003). Additionally, ouabain induces in vitro intracellular free radicals accumulation and thymocytes death (Smolyaninova et al., 2013). In vivo, ouabain synergizes with hydrocortisone increasing T lymphocyte precursors death by apoptosis (Rodrigues-Mascarenhas et al., 2006), which reinforces its role as a stress-related hormone. Moreover, ouabain reduced NFAT expression and P-p38 levels, after concanavalin A stimulation (Rodrigues-Mascarenhas et al., 2008, 2009). This later data support the fact that ouabain modulates cell signaling (Xie and Askari, 2002).

Ouabain is also able to modulate in vivo B lymphocytes dynamics, decreasing mature B cells in the bone marrow, spleen and peripheral blood (de Paiva et al., 2011), although IgG and IgM levels were not affected by ouabain. On the other hand, there was an increase in B lymphocytes of mesenteric lymph node, probably by CD62L reduced and CXCR5 increased expression (da Silva et al., 2015).

Despite ouabain effects on B and T lymphocytes, natural killer (NK) cells seem to be resistant to ouabain. In fact, ouabain did not affect NK cell cytotoxic activity, in neither the absence nor presence of stimulatory agents (de Moraes et al., 1989). However, ouabain inhibits lymphokine-activated killer (LAK) cell generation induced by IL-2 (Olej et al., 1994).

Many lymphocytes functions rely on antigen presenting cells (APCs), in which dendritic cells (DC) have a highlighted role together with macrophages (Steinman, 2012). The influence of ouabain on DC was also described. Nascimento et al. (2014) demonstrated that ouabain modulates dendritic cells markers and IL-12 production during activation by TNF-α. Additionally, ouabain affects monocyte/macrophage activation (Sowa and Przewłocki, 1997; Teixeira and Rumjanek, 2014). Indeed, ouabain reduces CD14 expression, a molecule involved in foreign antigens recognition, in human monocytes (Valente et al., 2009; Teixeira and Rumjanek, 2014). Moreover, ouabain inhibits a proinflammatory monocyte subset (mCD14+CD16+) appearance in vitro, which may indicate that this steroid also modulates the inflammatory response.

### OUABAIN AS A MODULATOR OF INFLAMMATION

Inflammation is an immunological complex response that can be triggered by pathogen- and damage-associated molecular patterns and is able to restore tissue homeostasis (Medzhitov, 2010). Acute inflammatory process is mainly characterized by vascular (e.g., vasodilation and vascular permeability) and cellular (e.g., leukocytes migration) alterations, resulting in five cardinal (clinical) signs: redness, swelling, heat, pain, and disturbance of function. Uncontrolled or unresolved inflammation can lead to homeostatic imbalance and chronic diseases, including cardiovascular and neurodegenerative diseases (Scrivo et al., 2011). Besides immune system role in inflammation, other systems, such as endocrine and nervous system, can also regulate this physiological response (Padro and Sanders, 2014; Procaccini et al., 2014). In fact, many hormones are known to affect inflammation, such as glucocorticoids (Cain and Cidlowski, 2017), and ghrelin, a pituitary-derived hormone (Baatar et al., 2011). In the following topics, ouabain role in the inflammatory process will be discussed.

### Ouabain and Peripheral Inflammation

Acute peripheral inflammation initiates after inflammatory signals recognition (e.g., infection and tissue injury) by resident cells, such as mast cells and macrophages. This recognition promotes mediators release (e.g., vasoactive amines and prostaglandins), which stimulates rapid effects on the vasculature, including vasodilation and fluid extravasation (i.e., increased vascular permeability; Medzhitov, 2008). One of the first reports associating ouabain and inflammation revealed that this steroid suppresses vascular permeability in the sheep skin and pleural cavity induced by the irritant agent turpentine (Lancaster and Vegad, 1967). Later, our group demonstrated that ouabain given intraperitoneally decreases zymosan-induced plasma extravasation in mice peritoneal cavity (Leite et al., 2015) and reduces the mouse paw edema stimulated by several phlogistic agents (de Vasconcelos et al., 2011). However, Gonçalves-de-Albuquerque et al. (2014) showed that intratracheal administration of ouabain induces lung edema formation in mice. It is important to consider that ouabain effect on lung edema must be related to Na+/K+-ATPase inhibition in alveolar cells (Gonçalves-de-Albuquerque et al., 2014), while ouabain effects demonstrated by our group may be associated with cell signaling mechanisms in immune cell (e.g., P-p38 and NF-κB activity inhibition; Mascarenhas et al., 2014; Leite et al., 2015).

Vasodilation and vascular permeability are events tune regulated by vasoactive amines. Histamine, which plays a critical role among these vasoactive molecules, is released by perivascular mast cells together with other mediators (e.g., newly synthesized cytokines and tryptases) during inflammation (Nathan, 2002). Different ouabain effects on histamine secretion by mast cells have been described. Okazaki et al. (1976) reported that ouabain inhibits antigen-induced histamine release on guinea-pig mast cells. On the other hand, several studies revealed that ouabain increased histamine secretion induced by different agents on rat mast cells (Frossard et al., 1983; Amellal et al., 1984; Knudsen et al., 1992; Lago et al., 2001), while no ouabain effect on human mast cells (Senol et al., 2007) and basophils (Magro, 1977) were reported. Different protocols and species variation in the ouabain sensitivity of Na+/K+-ATPase (Abeywardena et al., 1984; Herzig and Mohr, 1984; Wang et al., 2001) could explain this discrepant ouabain effects on mast cell degranulation.

Upon initiation of acute inflammation, circulating leukocytes are able to recognize molecules (e.g., selectins, integrins, and chemokines) on activated vascular endothelium and, after rolling and adhesion steps, they cross blood vessel barrier and reach inflamed tissue (Ley et al., 2007; Vestweber, 2015; Kourtzelis et al., 2017). Neutrophils are the first cells recruited to the injured site, followed by other inflammatory cells such as monocytes (Kolaczkowska and Kubes, 2013; Wang and Arase, 2014). These polymorphonuclear leukocytes are important not only to eliminate microorganisms but they also play a key role in inflammation resolution (Mayadas et al., 2014; Sugimoto et al., 2016). Moreover, neutrophils role in chronic inflammation has been described and they are pointed as a target to emerging therapeutic strategies (Soehnlein et al., 2017). Considering this, blocking neutrophil recruitment appears to be a crucial way to avoid inflammation maintenance.

Ward and Becker (1970) initially described ouabain inhibitory effect on rabbit neutrophil migration toward bacterial chemotactic factors in vitro. In agreement with this study, our group revealed that ouabain pretreatment reduces mice neutrophil migration induced by zymosan, a component of the cell wall of yeast Saccharomyces cerevisiae, (Leite et al., 2015) and by Leishmania amazonensis (Jacob et al., 2013). These data provide clear evidence that ouabain inhibits neutrophil recruitment induced by microbial agents. This ouabain effect was also demonstrated in peritoneal inflammation induced by mitogen concanavalin A (de Vasconcelos et al., 2011). Furthermore, in airway allergic inflammation model, ouabain has an anti-migratory effect (Galvão et al., 2017). Ray and Samanta (1997) have also demonstrated that ouabain impairs in vitro human neutrophil migration, by interfering with IL-8 receptor recycling. Additionally, other studies have demonstrated that ouabain also decreases lung cancer cells migration (Liu et al., 2013), possibly by reducing the expression of molecules related to cell adhesion (e.g., integrins and ICAM) (Takada et al., 2009; Ninsontia and Chanvorachote, 2014) and cell migration (e.g., Src, Akt, and FAK) (Pongrakhananon et al., 2013; Shin et al., 2015).

However, the inhibitory effect of ouabain on cell migration seems to depend on the presence of a previous inflammatory stimulus, since ouabain itself given by inhalation (Feng et al., 2011) or intratracheally (Gonçalves-de-Albuquerque et al., 2014) causes acute lung inflammation with increased neutrophil migration. This proinflammatory effect was followed by LTB<sup>4</sup> and PGE<sup>2</sup> high levels, both lipid mediators associated with cell migration. Moreover, it has been demonstrated that ouabain at high concentrations induces VCAM-1 expression (an adhesion molecule) in murine endothelial cells (Bereta et al., 1995). Na+/K+-ATPase inhibition may be, at least partially, responsible for this ouabain effect (Lacroix-Lamandé et al., 2012; Gonçalves-de-Albuquerque et al., 2014), but Na+/K+-ATPase-dependent activation of signaling cascades (e.g., ERK and p38 MAPK) cannot be ruled out (Bereta et al., 1995; Feng et al., 2011). Indeed, Leu et al. (1973) demonstrated that ouabain stimulates guinea-pig alveolar and peritoneal macrophages migration independent of the sodium pump.

A different pattern of cytokines is present since the inflammation onset until resolution phase. These soluble proteins are secreted by a variety of cells (e.g., immune e non-immune cells) and allow intercellular communication,

mediating and regulating inflammatory process. An imbalance in proinflammatory (e.g., TNF-α) and anti-inflammatory (e.g., IL-10) cytokine production could entail inflammatory disorders (Tayal and Kalra, 2008; Sugimoto et al., 2016). Monocytes/macrophages are immune cells that act as a key source of cytokines because of their plasticity ability (i.e., change their pattern of cytokines and functions when exposed to different signals; Gordon and Taylor, 2005; Mantovani et al., 2014). It has been described that ouabain itself can stimulate human monocytes to secrete cytokines such as IL-1α, IL-1β, IL-6, and TNF-α (Foey et al., 1997; Matsumori et al., 1997, 2000; Teixeira and Rumjanek, 2014). Some different results are related to IL-6 and TNF-α production, which could be associated with individual variability of human donors. Indeed, critically ill patients with high levels of ouabain had higher serum concentrations of these proinflammatory cytokines and other inflammatory markers, such as C-reactive peptide and serum amyloid A, when compared to patients with low ouabain concentrations (Berendes et al., 2003). Interestingly, ouabain enhances interleukin-10 levels in human monocytes (Teixeira and Rumjanek, 2014). Recently, Kobayashi et al. (2017) demonstrated that ouabain effect on IL-1β release, in both macrophages and cardiac tissue, is related to NLRP3 inflammasome activation, which in turn is mediated through K <sup>+</sup> efflux. It is noteworthy that in this study the authors used high doses of ouabain both in vitro and in vivo. This contrasts with other studies that show a reduction of a different pattern of cytokines, including IL-1β (Leite et al., 2015) with lower ouabain doses in presence of inflammatory stimulus (Jacob et al., 2013; Galvão et al., 2017).

Despite the well-established proinflammatory activities of some cytokines, such as TNF-α and IL-6, some studies have demonstrated their anti-inflammatory role (Liu et al., 1998; Zakharova and Ziegler, 2005; Masli and Turpie, 2008; Scheller et al., 2013). In this regard, use low concentrations of ouabain as cytokine immunoregulator could be useful in different clinical conditions. In fact, ouabain at low doses reverses sepsis-induced immunoparalysis by increase TNF-α, IFN-γ, and GM-CSF levels and improve mice survival (Dan et al., 2014).

Pain is another cardinal sign of inflammation, which is also modulated by ouabain. de Vasconcelos et al. (2011) demonstrated that intraperitoneal administration of ouabain reduces nociceptive behavior in mice model of inflammatory pain (i.e., acetic acid induced writhing test). This steroid also induces supraspinal antinociceptive activity, related to opioid mechanisms, since naloxone, an opioid antagonist, inhibits its effect. In addition, other studies revealed that ouabain intracerebroventricular (i.c.v.) (Calcutt et al., 1971) and intratechal (i.t.) (Zeng et al., 1999) injections at relative high doses (micrograms) produce central antinociception and potentiate morphine and clonidine central antinociceptive effect, mainly by enhancement of cholinergic transmission at the spinal cord level (Zeng et al., 1999, 2007). In contrast, it has also been shown that low doses (nanograms) of ouabain (i.c.v.) antagonize opioid receptor agonists (Masocha et al., 2003, 2016; Gonzalez et al., 2012) and that ouabain (i.t.) itself did not cause antinociception (Horvath et al., 2003). This pain modulation by ouabain, which depends on the dose and administration route used, suggests that it can modulate events in the central nervous system such as neuroinflammation.

### Ouabain and Neuroinflammation

Some studies have described ouabain role in the central nervous system (CNS). However, the effects of this steroid on neuroinflammation can be controversial. In a study with rat hippocampus, ouabain anti-inflammatory effect against neuroinflammation induced by LPS was observed. Acute intraperitoneal pre-treatment with this steroid reduced iNOS and IL-1β mRNA levels. In addition, ouabain also reduced p65 subunit NF-κB translocation and IκB degradation, both mechanisms important to inflammatory process (Kinoshita et al., 2014). However, when ouabain is administrated by intrahippocampal route in a concentration that does not inhibit Na+/K+-ATPase, it induces activation of NF-κB and increases iNOS and TNF-α mRNA levels (Kawamoto et al., 2012). As well as in peripheral inflammation, the antiinflammatory effect of ouabain on hippocampus was only observed after an inflammatory stimulus. This could explain the different effects regarding NF-κB activation. Additionally, another study demonstrated the ability of ouabain to restore the lipid composition of rat hippocampal membranes in neuroinflammation induced by LPS (Garcia et al., 2015).

In rat cerebellar cell culture, ouabain at high concentrations induced NF-κB activation and consequent TNF-α and IL-1β cytokines increase through NMDA-Src-Ras pathway in absence of inflammatory stimulus (de Sá Lima et al., 2013). However, ouabain decreases IL-1β release in LPS-stimulated astrocytes (Forshammar et al., 2011). In spite of that, ouabain did not modulate IL-1β release in LPS-stimulated microglia, while increased TNF-α release at low concentration (Forshammar et al., 2013). Therefore, ouabain role in cytokine production at CNS level depends on cell type and concentration used. Besides that, ouabain effects on CNS could be associated with its role in

### REFERENCES


bipolar and depressive disorders (Goldstein et al., 2006; Tonin et al., 2014). The interaction between neuroinflammation and cardiac steroids is more substantially detailed by Orellana et al. (2016).

### CONCLUSIONS AND PERSPECTIVES

In summary, compelling evidence indicates that ouabain has a pro- and anti-inflammatory effects (**Figure 1**), which mainly depends on its concentration and functional state of cells (i.e., absence or presence of inflammatory stimulus), corroborating other ouabain effects on the immune system. However, to the best of our knowledge, studies relating ouabain and chronic inflammation are missing. In addition, more details about ouabain mechanism of action are necessary. Lastly, ouabain effects on the inflammatory process could be better explored in order to establish possible strategies for pharmacological treatment of immune dysregulation/inflammatory diseases.

### AUTHOR CONTRIBUTIONS

Conceived and designed the manuscript: SR-M and LC-S. Wrote the manuscript: LC-S, AAA, DC, ÉAL, JG, JMS-N, and JFS. Final version: SR-M.

### ACKNOWLEDGMENTS

LC-S and JG are supported by Ph.D. fellowship from "Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)." DC, ÉAL, JMS-N, and JFS are supported by fellowship from "Coordenação de Aperfeiçoamento de Pessoal de Nível Superior" (CAPES).

of beta-blockade and angiotensin-converting enzyme inhibition. Hypertension 45, 1024–1028. doi: 10.1161/01.HYP.0000165024. 47728.f7


is modulated by in vivo treatment with steroid ouabain. Immunobiology 221, 368–376. doi: 10.1016/j.imbio.2015.09.020


protein phosphorylation and intracellular pH. Cell Signal. 13, 515–524. doi: 10.1016/S0898-6568(01)00169-3


of the immune system by ouabain. Ann. N. Y. Acad. Sci. 1153, 153–163. doi: 10.1111/j.1749-6632.2008.03969.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 Cavalcante-Silva, Lima, Carvalho, Sales-Neto, Alves, Galvão, Silva and Rodrigues-Mascarenhas. 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.

# Corrigendum: Much More than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain

Luiz H. A. Cavalcante-Silva<sup>1</sup> , Éssia de Almeida Lima<sup>2</sup> , Deyse C. M. Carvalho<sup>3</sup> , José M. de Sales-Neto<sup>1</sup> , Anne K. de Abreu Alves <sup>2</sup> , José G. F. M. Galvão<sup>1</sup> , Juliane S. de França da Silva<sup>1</sup> and Sandra Rodrigues-Mascarenhas 1, 2, 3 \*

#### Edited and reviewed by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

\*Correspondence: Sandra Rodrigues-Mascarenhas sandra@cbiotec.ufpb.br

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 22 December 2017 Accepted: 03 January 2018 Published: 12 January 2018

#### Citation:

Cavalcante-Silva LHA, Lima ÉA, Carvalho DCM, Sales-Neto JM, Alves AKA, Galvão JGFM, Silva JSF and Rodrigues-Mascarenhas S (2018) Corrigendum: Much More than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain. Front. Physiol. 9:1. doi: 10.3389/fphys.2018.00001

<sup>1</sup> Programa de Pós-Graduação em Produtos Naturais e Sintéticos Bioativos, Laboratório de Imunobiotecnologia, Centro de Ciências da Saúde, Universidade Federal da Paraíba, João Pessoa, Brazil, <sup>2</sup> Programa de Pós-Graduação em Biotecnologia, Laboratório de Imunobiotecnologia, Centro de Biotecnologia, Universidade Federal da Paraíba, João Pessoa, Brazil, <sup>3</sup> Programa Multicêntrico de Pós-graduação em Ciências Fisiológicas, Laboratório de Imunobiotecnologia, Centro de Biotecnologia, Universidade Federal da Paraíba, João Pessoa, Brazil

Keywords: ouabain, immune system, peripheral inflammation, cell migration, cytokines, neuroinflammation

#### **A corrigendum on**

### **Much More than a Cardiotonic Steroid: Modulation of Inflammation by Ouabain**

by Cavalcante-Silva, L. H. A., Lima, É. A., Carvalho, D. C. M., Sales-Neto, J. M., Alves, A. K. A., Galvão, J. G. F. M., et al. (2017). Front. Physiol. 8:895. doi: 10.3389/fphys.2017.00895

In the original article, we neglected to include the funder CAPES/PROCAD-2013, grant number 2951/2014. The authors apologize for this error and state that this does not change the scientific conclusions of the article in any way.

**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 Cavalcante-Silva, Lima, Carvalho, de Sales-Neto, Alves, Galvão, Silva and Rodrigues-Mascarenhas. 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.

# Key Inflammatory Processes in Human NASH Are Reflected in Ldlr−/−.Leiden Mice: A Translational Gene Profiling Study

Martine C. Morrison<sup>1</sup> , Robert Kleemann1,2, Arianne van Koppen<sup>1</sup> , Roeland Hanemaaijer <sup>1</sup> and Lars Verschuren<sup>3</sup> \*

<sup>1</sup> Department of Metabolic Health Research, The Netherlands Organization for Applied Scientific Research (TNO), Leiden, Netherlands, <sup>2</sup> Department of Vascular Surgery, Leiden University Medical Center, Leiden, Netherlands, <sup>3</sup> Department of Microbiology and Systems Biology, The Netherlands Organization for Applied Scientific Research (TNO), Leiden, Netherlands

#### Edited by:

Carlos Rosales, Universidad Nacional Autónoma de México, Mexico

#### Reviewed by:

Ramani Soundararajan, University of South Florida, United States Savneet Kaur, Institute of Liver and Biliary Sciences, India

> \*Correspondence: Lars Verschuren lars.verschuren@tno.nl

#### Specialty section:

This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology

Received: 11 October 2017 Accepted: 08 February 2018 Published: 23 February 2018

#### Citation:

Morrison MC, Kleemann R, van Koppen A, Hanemaaijer R and Verschuren L (2018) Key Inflammatory Processes in Human NASH Are Reflected in Ldlr−/−.Leiden Mice: A Translational Gene Profiling Study. Front. Physiol. 9:132. doi: 10.3389/fphys.2018.00132 Introduction: It is generally accepted that metabolic inflammation in the liver is an important driver of disease progression in NASH and associated matrix remodeling/fibrosis. However, the exact molecular inflammatory mechanisms are poorly defined in human studies. Investigation of key pathogenic mechanisms requires the use of pre-clinical models, for instance for time-resolved studies. Such models must reflect molecular disease processes of importance in patients. Herein we characterized inflammation in NASH patients on the molecular level by transcriptomics and investigated whether key human disease pathways can be recapitulated experimentally in Ldlr−/−.Leiden mice, an established pre-clinical model of NASH.

Methods: Human molecular inflammatory processes were defined using a publicly available NASH gene expression profiling dataset (GSE48452) allowing the comparison of biopsy-confirmed NASH patients with normal controls. Gene profiling data from high-fat diet (HFD)-fed Ldlr−/−.Leiden mice (GSE109345) were used for assessment of the translational value of these mice.

Results: In human NASH livers, we observed regulation of 65 canonical pathways of which the majority was involved in inflammation (32%), lipid metabolism (16%), and extracellular matrix/remodeling (12%). A similar distribution of pathways across these categories, inflammation (36%), lipid metabolism (24%) and extracellular matrix/remodeling (8%) was observed in HFD-fed Ldlr−/−.Leiden mice. Detailed evaluation of these pathways revealed that a substantial proportion (11 out of 13) of human NASH inflammatory pathways was recapitulated in Ldlr−/−.Leiden mice. Furthermore, the activation state of identified master regulators of inflammation (i.e., specific transcription factors, cytokines, and growth factors) in human NASH was largely reflected in Ldlr−/−.Leiden mice, further substantiating its translational value.

Conclusion: Human NASH is characterized by upregulation of specific inflammatory processes (e.g., "Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes," "PI3K signaling in B Lymphocytes") and master regulators (e.g., TNF, CSF2, TGFB1). The majority of these processes and regulators are modulated in the

**320**

same direction in Ldlr−/−.Leiden mice fed HFD with a human-like macronutrient composition, thus demonstrating that specific experimental conditions recapitulate human disease on the molecular level of disease pathways and upstream/master regulators.

Keywords: liver, NASH, inflammation, molecular, gene expression, translational, mouse, human

### INTRODUCTION

Non-alcoholic fatty liver disease (NAFLD) is one of the most important causes of chronic liver disease worldwide (Wong et al., 2015; Younossi et al., 2017). The prevalence of NAFLD is rising in close association with the increasing prevalence of obesity, insulin resistance, and dyslipidemia, all of which are risk factors for NAFLD (Siddiqui et al., 2015; Chang et al., 2016; Dongiovanni et al., 2016; Katsiki et al., 2016). NALFD encompasses a spectrum of liver disease: ranging from the relatively benign hepatic steatosis, which is characterized by the accumulation of lipids in the liver, to non-alcoholic steatohepatitis (NASH), the progressive form of NAFLD.

NASH is characterized by the presence of hepatocellular damage and inflammation (Rinella, 2015), which in concert can drive the development of liver fibrosis (Fujii and Kawada, 2012), the strongest predictor of NAFLD-related mortality (Angulo et al., 2015; Ekstedt et al., 2015). The hepatic inflammatory response in NASH is poorly characterized on the molecular level and thought to originate from a combination of various chronic pro-inflammatory triggers (Tilg and Moschen, 2010). In addition to direct lipotoxicity resulting from the build-up of pro-inflammatory lipid species in the liver (such as free fatty acids, diglycerides, ceramides, and free cholesterol) (Alkhouri et al., 2009), the inability of hepatocytes to cope with an increased metabolic load is thought to lead to ER stress, metabolic dysfunction and production of reactive oxygen species, which in turn can contribute to exacerbation of the hepatic inflammatory response (Takaki et al., 2014). On top of that, it is thought that extrahepatic pro-inflammatory signals from the adipose tissue and the gut can drive hepatic inflammation in NASH (Tilg and Moschen, 2010; Takaki et al., 2014).

The exact molecular events in liver that contribute to this chronic inflammatory condition are not well-characterized, and much remains unknown about the nature of these molecular inflammatory processes in NASH. While a number of studies has explored genome-wide hepatic gene expression in NASH patients (Younossi et al., 2005; Zhang et al., 2012; Moylan et al., 2014; Arendt et al., 2015; Teufel et al., 2016) none of these has focused on unraveling the inflammatory response. Since studies based on human liver biopsy material generally do not allow in-depth mechanistic studies or timeresolved investigation of disease progression, pre-clinical disease models are key to the development of a deeper mechanistical understanding of pathogenesis and are required for testing of new therapeutic interventions (Hebbard and George, 2011). To study mechanisms of disease development in pre-clinical models, it is critical that the model employed is reflective of human disease processes.

A wide variety of animal models for NASH is available, each with their own specific advantages and disadvantages (Ibrahim et al., 2016; Jacobs et al., 2016). The translational value of these experimental models is often judged on basis of histopathological features but not on the molecular pathophysiological level, i.e., whether they recapitulate the disease pathways that are evoked in NASH patients. Hence, the translational value of most models is currently under debate, the more so because many models are not sufficiently validated. A recent study compared different animal models with the full spectrum of NAFLD patients using gene profiling and concluded that none of the investigated animal models mimics the complete spectrum of molecular processes involved in humans (Teufel et al., 2016). However, it has been shown that diet-inducible models showed at least some similarities in the development of disease. Therefore this study and other reports (Hebbard and George, 2011; Mulder et al., 2016) suggest that diet-induced models may best reflect sub-processes of human disease phenotypes and underlying pathogenesis.

In the current study we characterized specifically the subprocess of inflammation in human NASH and evaluated the translational value of the high-fat-diet (HFD)-fed Ldlr−/−.Leiden mouse with regards to the hepatic inflammatory response in NASH. This NASH model was not included in the comparison of hepatic gene expression in murine and human NASH described above (Teufel et al., 2016). Ldlr−/−.Leiden mice display phenotypical and histopathological characteristics of NASH patients when fed a human like-HFD without requiring amino acid and choline deficiency or supraphysiologic levels of cholesterol in the diet (Liang et al., 2014; Morrison et al., 2016; van Koppen et al., 2018). More specifically, they develop NASH in the context of an obese phenotype with hypercholesterolemia, hypertriglyceridemia, and hyperinsulinemia as observed in many patients (Anderson and Borlak, 2008; Loomba and Sanyal, 2013). Histopathologically, these mice show presence of micro- and macrovesicular steatosis in the liver, hepatocellular hypertrophy with hepatocellular disintegration and sporadic ballooning, lobular inflammation (mixed-cell inflammatory infiltrates) and marked hepatic fibrosis that progresses with prolonged HFD treatment. In a recent comparison of NASH-induced regulation of hepatic gene expression (using an unbiased approach), HFD-fed Ldlr−/−.Leiden mice were found to recapitulate many of the gene expression changes observed in human NASH (van Koppen et al., 2018).The current study focuses specifically on the inflammatory component of NASH, first defining the main molecular inflammatory processes present in human NASH using available published information and subsequently exploring the representation of these processes in HFD-fed Ldlr−/−.Leiden mice.

### MATERIALS AND METHODS

### Human Hepatic Gene Expression Dataset

For the investigation of molecular inflammatory processes in human NASH, samples were selected from a published dataset accessible at the NCBI Gene Expression Omnibus (GEO) database, accession GSE48452. This dataset originates from a study on DNA methylation patterns in NASH patients (Ahrens et al., 2013) that included morbidly obese patients with biopsy-confirmed NAFLD pre- and post-bariatric surgery, as well as healthy controls. In this study, gene expression levels were measured using Affymetrix Human Gene 1.1 ST array (Affymetrix Inc., Santa Clara, California, USA). For the purpose of the present study, we selected the pre-bariatric surgery samples from NASH patients (n = 17) and compared them with the healthy controls (n = 12) (see sample list in Supplemental Table 1). The probe-level, background-subtracted, expression values were used as input for lumi package (Du et al., 2008) of the R/Bioconductor (http://www.bioconductor.org; http:// www.r-project.org) to perform quality control and quantile normalization. Differentially expressed probes were identified using the limma package of R/Bioconductor (Wettenhall and Smyth, 2004), calculated values of P < 0.01 were used as threshold for significance.

### Murine Hepatic Gene Expression Dataset

Murine gene expression data was obtained from a previously published study (van Koppen et al., 2018) and the dataset is accessible at the NCBI GEO database via accession number GSE109345. Detailed methods of tissue collection and processing as well as next generation sequencing (NGS) and statistical analysis are described in van Koppen et al. (2018). Briefly, n = 15 Ldlr−/−.Leiden mice (94% C57BL/6J background, 6% 129S1/SvImJ background) were fed an energy-dense high-fat diet (HFD; 45 kcal% fat from lard, 17 kcal% sucrose) for 24 weeks to induce NASH (steatosis, inflammation, early fibrosis) in the context of obesity, hyperlipidemia and hyperinsulinemia. Hepatic gene expression dataset of HFD-treated mice was generated and data were expressed relative to age-matched chowfed controls (n = 15).

### Data- and Statistical Analysis

Differentially expressed probes (from human dataset) and counts (from mouse dataset) confined with the statistical cut-off were used as input for pathway analysis through Ingenuity Pathway Analysis (IPA) suite (www.ingenuity.com, accessed 2017) as described previously (Morrison et al., 2015; Mulder et al., 2016; van Koppen et al., 2018). An upstream regulator analysis, which is part of IPA, predicts the activation state of a protein, enzyme or transcription factor based on the expression pattern of the genes downstream of this factor. The z-score indicates the predicted activation state of a transcription factor or key regulator: z ≤ −2 indicates relevant inhibition (shown in green), z ≥ 2 indicates relevant activation (shown in red). Pathway-based overlap analysis was performed by using Venny 2.1 (Oliveros, 2007) and heatmaps were generated using a web-based tool (Babicki et al., 2016). The genes that belong to significantly regulated inflammatory canonical pathways were visualized using Neo4J (Neo4j, Inc, San Mateo, CA, USA) a graph database with query-based calculations.

## RESULTS

### Key Inflammatory Processes and Regulators in NASH Liver Biopsies

To define the major inflammatory processes that are modulated during NASH development, gene expression profiles in liver biopsies from NASH patients were compared to those of healthy control livers using a published human dataset (GSE48452). In total 12 healthy controls and 17 patients with biopsy-proven NASH were analyzed. At a statistical cut-off of P < 0.01, 972 genes were differentially expressed between the two groups, i.e., 519 genes were upregulated and 453 genes were downregulated in NASH livers relative to healthy controls. All these NASHassociated genes were used as input for pathway analysis, which integrates the expression of a multitude of genes in predefined biological and disease-associated pathways thus allowing summation of multiple (small) gene expression changes to provide information on the effect on entire pathways rather than single genes.

The top 20 most significantly enriched pathways are visualized in **Figure 1A**; this top 20 includes canonical pathways involved in various (patho)physiological processes such as lipid metabolism, inflammation and hepatic fibrosis. When taking all significantly enriched canonical pathways [−log(P-value) > 2] into account, the majority of pathways appear to be related to inflammation (32%) while other pathways are related to lipid metabolism (16%) or extracellular matrix remodeling (12%; **Figure 1B**).

A total of 13 pathways was related to inflammation, among which "Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes," "PI3K signaling in B Lymphocytes" and "Leukocyte Extravasation Signaling" (**Figure 2A**). Subsequent refined analysis of the genes that are part of these pathways showed that the expression of the majority of the individual genes is significantly upregulated in NASH patients relative to controls. The consistency of this upregulation is exemplified for the pathway "Leukocyte Extravasation Signaling" (**Figure 2B**).

We subsequently analyzed inflammatory processes in NASH in more detail by performing an upstream regulator analysis. This analysis predicts the activation state (indicated by a positive/negative z-score for activation/inhibition) of upstream regulators (e.g., transcription factors and cytokines) in NASH livers relative to healthy controls. **Table 1** shows the inflammation-related upstream regulators that were significantly activated (in red) or inhibited (in green) in NASH patients. We observed an activation of classical regulators of inflammation such as the cytokines TNF (Tumor necrosis factor alpha), CSF2 (Granulocyte-macrophage colony-stimulating factor) and TGFB1 (Transforming growth factor beta 1) as well as factors that can indirectly affect inflammatory processes, such as ESR2 (Estrogen receptor 2), and PLG (Plasminogen). In addition, a known inhibitor of inflammation AHR (Aryl hydrocarbon

compared to normal controls. Values are expressed as –log(p-value). (B) Circle chart which classifies canonical pathways into more general biological processes and illustrates the proportion of pathways in each of these categories.

receptor) (Li et al., 2011) was predicted to be inactive in NASH relative to healthy controls.

### Key Inflammatory Processes and Regulators in Murine NASH Development

To investigate whether similar inflammatory processes were evoked in murine NASH, we studied hepatic gene expression profiles of HFD-fed Ldlr−/−.Leiden mice, which display phenotypical and histopathological characteristics of NASH patients (Liang et al., 2014; Morrison et al., 2016). We compared hepatic gene expression profiles of 24-week HFD-fed mice (at the stage of hepatic steatosis, inflammation and early fibrosis) to those of age-matched healthy controls (chow-fed Ldlr−/−.Leiden mice) using data from a recently published study (van Koppen et al., 2018, GSE 109345). In total, 2,680 genes were differentially expressed between the two groups (1,639 upregulated, 1,046

genes.

downregulated; P < 0.001). Subsequent pathway analysis revealed that, as seen in human NASH, the majority of regulated pathways is related to inflammation (36%) while many of the other most significantly regulated pathways are involved in lipid metabolism (24%) and extracellular matrix remodeling (8%) (**Figure 3A**). A total of 27 pathways were related to inflammation, for instance "Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes," "IL-8 signaling," "Production of Nitric Oxide and Reactive Oxygen Species in Macrophages" and "Leukocyte Extravasation Signaling" (**Figure 3B**). As observed in human NASH, the expression of most of the individual genes that make up these inflammatory pathways were upregulated in HFD-fed mice relative to healthy controls, as is illustrated for the "Leukocyte Extravasation Signaling" pathway (**Figure 3C**). To determine which inflammation-related upstream regulators are involved in the development of murine NASH, an upstream regulator analysis was performed to identify significantly activated or inhibited regulators in HFD-fed Ldlr−/−.Leiden mice relative to their age-matched healthy controls (shown in **Table 2**). This analysis revealed significant activation of many classical chemokines (e.g., CCL2, CCL5, CXCL2, CXCL3), cytokines (e.g., TNF, TGFB1, IL1B, CSF2) and transcriptional



The z-score indicates the predicted activation state of a transcription factor or key regulator: z ≤ −2 indicates relevant inhibition (shown in green), z ≥ 2 indicates relevant activation (shown in red). The p-value indicates significant enrichment of the genes downstream of a regulator (p < 0.01 was considered statistically significant).

regulators of inflammation (e.g., NF-KB, STAT4, JUN) as well as factors with indirect links to inflammation (e.g., FGF2, FOXO1, PLG). Several anti-inflammatory factors were found to be significantly inactivated in NASH livers including AHR and others (e.g., IL10RA, IL1RN) which is consistent with the observations in humans.

### Overlap in Inflammatory Processes between Human and Experimental NASH

Next, we investigated in more detail whether these NASHassociated molecular inflammatory processes are similar in human and murine NASH, and to what extent the individual genes are similarly regulated. To evaluate the overlap of canonical inflammatory pathways between human and murine NASH, pathways were integrated using an overlay Venndiagram. Thirteen inflammatory pathways were significantly regulated in human NASH and 27 inflammatory pathways were significantly regulated in murine NASH. Interestingly, a large part-−11 out of 13 pathways—of the human inflammatory processes were also regulated in HFD-fed Ldlr−/−.Leiden mice (**Figure 4A**), including several macrophage-related pathways such as "Production of Nitric Oxide and Reactive Oxygen Species in Macrophages," "Macropinocytosis Signaling," and "Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes" but also pathways related to other inflammatory cell types, such as "Dendritic Cell Maturation," "B Cell Receptor Signaling," and "Natural Killer Cell Signaling" (**Table 3**). Next, we investigated the differentially expressed genes underlying the shared pathways in human NASH and identified which of these were also regulated in the HFD-fed Ldlr−/−.Leiden mouse (**Figure 4B**). We then evaluated the conformity in direction of these genes that were differentially regulated in both human and murine NASH (i.e., upregulation or downregulation relative to control) (**Figure 4C**) and found that 83% of the shared genes (in either human or murine NASH) changed in the same direction in human and mouse NASH.

Finally, we investigated the overlap of upstream regulators between human and mouse NASH using an overlay Venndiagram (**Figure 5A**). A total of 25 inflammation-related upstream regulators was significantly regulated in human NASH, and 76 inflammation-related upstream regulators were significantly regulated in HFD-fed mice. Of the 25 inflammationrelated factors that were regulated in human NASH, 18 were also affected in HFD-fed Ldlr−/−.Leiden mice, including the pathways downstream of the cytokines TNF, TGFB1 and CSF2 which were activated in both human and murine NASH. Analysis of the direction of regulation, i.e., predicted activation or predicted inhibition of the upstream regulator, showed that the majority of the overlapping upstream regulators was regulated in the same direction in Ldlr−/−.Leiden mice as in NASH patients (**Figure 5B**) supporting the translational value of the experimental model conditions.

### DISCUSSION

This study defines the major molecular inflammatory pathways and key regulators in human NASH using gene profiling data from liver biopsies, and provides evidence that the major molecular responses in humans can be replicated experimentally in a diet-inducible NASH model, the HFD-fed Ldlr−/−.Leiden mouse.

Studies on human gene expression in NASH are scarce, typically include only a small number of patients, and datasets are not always made publicly available (Moylan et al., 2014; Arendt et al., 2015; Teufel et al., 2016; Lefebvre et al., 2017). Typically, the patients used therein are drawn from a patient population with known heterogeneity (Machado and Diehl, 2016; Younossi et al., 2016) for instance due to variations in disease etiology, ethnicity, gender, lifestyle and dietary habits. Herein we used an open access gene profiling dataset (Ahrens et al., 2013; Teufel et al., 2016) and, despite these limitations, such as a limited number of samples, were able to identify a number of significantly regulated inflammatory pathways and upstream regulators, indicating that these may be key inflammatory processes that

biological processes and illustrates the proportion of pathways in each of these categories. (B) Visualization of all inflammation-related canonical pathways in HFD-fed Ldlr−/−.Leiden mice. (C) Visualization of the expression change in HFD-fed Ldlr−/−.Leiden mice as compared to chow controls for the canonical pathway "Leukocyte Extravasation Signaling." Red color indicates significant upregulated genes and green color indicates significant downregulated genes.




(Continued)


The z-score indicates the predicted activation state of a transcription factor or key regulator: z ≤−2 indicates relevant inhibition (shown in green), z ≥ 2 indicates relevant activation (shown in red). The p-value indicates significant enrichment of the genes downstream of a regulator (p < 0.001 was considered statistically significant).

are common in many NASH patients. Many of the identified pathways describe processes to do with inflammatory cells, such as "Leukocyte Extravasation Signaling" and "Chemokine Signaling" which are required for the infiltration of immune cells into the liver, one of the diagnostic criteria for NASH (Kleiner et al., 2005). Several were related to the activity of macrophages (e.g., "Production of Nitric Oxide and Reactive Oxygen Species in Macrophages," "Macropinocytosis Signaling," and "Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes"), a cell type that is believed to play an important role in the inflammatory response in NASH (Itoh et al., 2013; Alisi et al., 2017) and is thus considered a promising target for treatment (Tacke, 2017). Immune cell infiltration in general, and macrophage responses in specific, can be the result of various pro-inflammatory stimuli (for instance cholesterol or gut-derived LPS; Heymann and Tacke, 2016), and may therefore represent a downstream phenomenon in NASH that can be the result of various disease-inducing pathways, providing a potential

FIGURE 4 | Representation of human key molecular pathways and associated genes in HFD-fed Ldlr−/−.Leiden mice. (A) Venn diagram to visualize the overlap in canonical pathways between human NASH biopsies (red circle) and Ldlr−/−.Leiden mice (blue circle). (B) Network visualization of overlapping canonical pathways (gray nodes, A: Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes, B: PI3K Signaling in B Lymphocytes, C: Leukocyte Extravasation Signaling, D: IL-8 Signaling, E: Natural Killer Cell Signaling, F: Macropinocytosis Signaling, G: B Cell Receptor Signaling, H: Role of Macrophages Fibroblasts and Endothelial Cells in Rheumatoid Arthritis, I: Production of Nitric Oxide and Reactive Oxygen Species in Macrophages; J: Dendritic Cell Maturation; K: iCOS-iCOSL Signaling in T Helper Cells) and the associated significantly expressed genes in human NASH (blue and yellow nodes). The blue nodes represent genes that were also regulated in the Ldlr−/−.Leiden mouse, the yellow nodes represent genes that were not significantly regulated in the Ldlr−/−.Leiden mouse. (C) Heatmap visualization of genes underlying the common pathways that are regulated in both human NASH biopsies and HFD-fed Ldlr−/−.Leiden mice relative to their respective controls. Red color indicates upregulated genes and green color indicates downregulated genes. The rightmost column shows the genes that do not share their direction of regulation between human and mouse in red.

explanation for its involvement across the heterogenous patient population.

To provide meaningful information on disease development, disease mechanisms, or effects of new therapeutic interventions, it is critical that a pre-clinical model not only reflects histopathological features of human disease, but also recapitulates human (patho)physiological processes on the molecular level. We found that the vast majority of inflammatory pathways and upstream regulators that were significantly modulated in human NASH were also significantly modulated in the Ldlr−/−.Leiden mouse (11 out of 13 pathways and 18 out of 25 upstream regulators), indicating that this model recapitulates the key inflammatory processes of human NASH.

Inbred mouse strains such as the Ldlr−/−.Leiden mouse are genetically homogenous and allow study of NASH development under standardized experimental conditions as reported (Morrison et al., 2016; Schoemaker et al., 2017). Furthermore, mice liver samples are collected at a relatively uniform stage of disease development depending on the time of high-caloric diet feeding (Liang et al., 2014; Arnoldussen et al., 2017), while human samples show more varying histopathology within a disease stage. In reflection of this high degree of homogeneity, substantially more genes were differentially expressed relative to healthy control in the Ldlr−/−.Leiden mice than in the human samples, and we also found a larger number of pathways and upstream regulators significantly modulated in mice (with a

### TABLE 3 | Overlap analysis of significantly regulated pathways in human and murine NASH.


Significance of enrichment for canonical pathways is indicated by –log(p-value).

higher level of significance). For instance, we observed regulation of several T-cell related pathways in HFD-fed Ldlr−/−.Leiden mice that were not significantly regulated in the human NASH samples (e.g., "Th1 and Th2 Activation Pathway," "T Helper Cell Differentiation," "T Cell Receptor Signaling"). Although much remains unknown about the role of T cells in the development and progression of NASH, alterations in this immune cell population have been reported for NASH patients (Inzaugarat et al., 2011), specifically in more advanced disease stages (i.e., fibrosis score >F2; Gadd et al., 2014). The lack of regulation of T cell-related pathways observed in the current study may be reflective of the relatively low presence of fibrosis in this patient cohort (mostly F1). In general, pathways that were found to be significantly enriched in mice but not in patients may constitute pathways that are more variably enriched in human disease (e.g., in a subset of NASH patients but not in others—resulting in an average expression level that does not pass the threshold for enrichment). Conversely, there were 2 pathways that were regulated in human NASH but not in the Ldlr−/−.Leiden mouse; "LPS-stimulated MAPK Signaling" and "Chemokine Signaling." In both pathways it is specifically the MAPK expression which is distinctive in human compared to mouse liver tissue. Since mice

are housed under pathogen-free conditions (SPF) it is amenable that activation of MAPK by LPS is more likely to be observed in the human situation.

Besides phenotypically readily apparent subtypes of NASH patients, such as lean vs. obese patients (Kumar and Mohan, 2017) or diabetic vs. non-diabetic patients (Puchakayala et al., 2015), recent efforts on subtyping NAFLD/NASH patients on the basis of their serum metabolome (Alonso et al., 2017; Iruarrizaga-Lejarreta et al., 2017) have revealed that patients may also be classified on the basis of their molecular disease patterns and have provided insight into molecular pathways that may be impaired in some but not other patients (i.e., synthesis of S-adenosylmethionine). A similar analysis in the Ldlr−/−.Leiden mouse revealed that on the metabolome level this model reflects a substantial proportion of NAFLD/NASH patients (Morrison et al., 2017) but the underlying disease mechanisms remain unclear. Further exploration on the gene expression level may shine light on different disease etiologies and the molecular pathways involved in various diseaseinducing mechanisms, as well as their divergent representation in the NASH patient population. However, such studies require large cross-sectional gene expression datasets of wellcharacterized and uniformly graded biopsies of different stages that allow classification of patients into molecular subtypes.

Given this large diversity of disease-inducing mechanisms in human NASH, it is unlikely that one single preclinical model for NASH will reflect the entire spectrum of underlying molecular responses observed in patients. However, individual models may recapitulate specific aspects of the spectrum of molecular responses seen in humans and can be of value to study that particular mechanism (Morrison et al., 2015; Iruarrizaga-Lejarreta et al., 2017; Zimmer et al., 2017). At the same time, one must always be aware of the limitations of the model employed. The Ldlr−/−.Leiden model for instance, while mimicking many aspects of human NASH (e.g., pathophysiology, histology, underlying molecular processes; Liang et al., 2014; Morrison et al., 2016; Mulder et al., 2016; van Koppen et al., 2018), is unsuitable to study interventions that require a functioning LDL receptor (Zimmer et al., 2017). Although many different (diet-inducible) experimental models of NASH have been reported (reviewed elsewhere Takahashi et al., 2012), the vast majority of these models has only been characterized on the pathophysiological and histological level and thus their value in the study of human disease processes remains unclear and requires further investigation.

Altogether, this study defines, to our knowledge for the first time, the key molecular inflammatory responses in biopsies of NASH patients and demonstrates that these are reflected in HFD-fed Ldlr−/−.Leiden mice. Comparative gene profiling approaches may help to better estimate the translational value of preclinical models for the NASH population in general, or specific subgroups of NASH patients in particular. Models that are validated on the molecular level against human disease pathways and key regulators of inflammation, constitute important tools to evaluate new therapeutics that target these pathways.

### AUTHOR CONTRIBUTIONS

MM, RK, AvK, RH, and LV: Study concept and design; MM, and LV: Writing the manuscript; MM, and LV: Analysis and interpretation of data; MM, RK, AvK, RH, and LV: Critical revision of manuscript.

### REFERENCES


### FUNDING

The work was funded by the TNO research program "Predictive Health Technologies" and supported by a grant from ZonMW (grant#114025001).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2018.00132/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 © 2018 Morrison, Kleemann, van Koppen, Hanemaaijer and Verschuren. 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.

digital media

of impactful research

article's readership