# NON-CODING RNAS AND HUMAN DISEASES

EDITED BY : Yujing Li, Ge Shan, Zhao-Qian Teng and Thomas S. Wingo PUBLISHED IN : Frontiers in Genetics

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-832-1 DOI 10.3389/978-2-88963-832-1

#### 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

# NON-CODING RNAS AND HUMAN DISEASES

Topic Editors: Yujing Li, Emory University, United States Ge Shan, University of Science and Technology of China, China Zhao-Qian Teng, Institute of Zoology, Chinese Academy of Sciences (CAS), China Thomas S. Wingo, Emory University, United States

Citation: Li, Y., Shan, G., Teng, Z.-Q., Wingo, T. S., eds. (2020). Non-Coding RNAs and Human Diseases. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-832-1

# Table of Contents

*06 Editorial: Non-Coding RNAs and Human Diseases* Yujing Li, Ge Shan, Zhao-Qian Teng and Thomas S. Wingo *09 Long Noncoding RNA Can Be a Probable Mechanism and a Novel Target for Diagnosis and Therapy in Fragile X Syndrome* Ge Huang, He Zhu, Shuying Wu, Manhua Cui and Tianmin Xu *20 CeRNA Expression Profiling Identifies KIT-Related circRNA-miRNA-mRNA Networks in Gastrointestinal Stromal Tumour* Ning Jia, Hanxing Tong, Yong Zhang, Hiroshi Katayama, Yuan Wang, Weiqi Lu, Sumei Zhang and Jin Wang *32 The Role of Non-Coding RNAs in Neurodevelopmental Disorders* Shuang-Feng Zhang, Jun Gao and Chang-Mei Liu *42 Aberrant Expression of Pseudogene-Derived lncRNAs as an Alternative Mechanism of Cancer Gene Regulation in Lung Adenocarcinoma* Greg L. Stewart, Katey S. S. Enfield, Adam P. Sage, Victor D. Martinez, Brenda C. Minatel, Michelle E. Pewarchuk, Erin A. Marshall and Wan L. Lam *55 LncRNAs GIHCG and SPINT1-AS1 Are Crucial Factors for Pan-Cancer Cells Sensitivity to Lapatinib* Zhen Xiang, Shuzheng Song, Zhenggang Zhu, Wenhong Sun, Jaron E. Gifts, Sam Sun, Qiushi Shauna Li, Yingyan Yu and Keqin Kathy Li *73 Molecular Mechanisms in Clear Cell Renal Cell Carcinoma: Role of miRNAs and Hypermethylated miRNA Genes in Crucial Oncogenic Pathways and Processes* Eleonora A. Braga, Marina V. Fridman, Vitaly I. Loginov, Alexey A. Dmitriev and Sergey G. Morozov *94 Transcriptomics Analysis of Circular RNAs Differentially Expressed in Apoptotic HeLa Cells* Bilge Yaylak, Ipek Erdogan and Bunyamin Akgul *104 Analyses of a Panel of Transcripts Identified From a Small Sample Size and Construction of RNA Networks in Hepatocellular Carcinoma* Zhiyong Sheng, Xiaolin Wang, Geliang Xu, Ge Shan and Liang Chen *118 Long Non-coding RNA in Neuronal Development and Neurological Disorders* Ling Li, Yingliang Zhuang, Xingsen Zhao and Xuekun Li *130 MicroRNAs and Androgen Receptor: Emerging Players in Breast Cancer* Erika Bandini and Francesca Fanini *139 Non-coding RNA in Fragile X Syndrome and Converging Mechanisms Shared by Related Disorders* Yafang Zhou, Yacen Hu, Qiying Sun and Nina Xie *150 MiR-378a-5p Regulates Proliferation and Migration in Vascular Smooth Muscle Cell by Targeting CDK1*

> Shaoyan Liu, Yanyan Yang, Shaoyan Jiang, Hong Xu, Ningning Tang, Amara Lobo, Rui Zhang, Song Liu, Tao Yu and Hui Xin


Elvezia Maria Paraboschi, Giulia Cardamone, Giulia Soldà, Stefano Duga and Rosanna Asselta


Adenilson Pereira, Fabiano Moreira, Tatiana Vinasco-Sandoval, Adenard Cunha, Amanda Vidal, André M. Ribeiro-dos-Santos, Pablo Pinto, Leandro Magalhães, Mônica Assumpção, Samia Demachki, Sidney Santos, Paulo Assumpção and Ândrea Ribeiro-dos-Santos

*327 The Epigenetics of Alzheimer's Disease: Factors and Therapeutic Implications*

Xiaolei Liu, Bin Jiao and Lu Shen

*337 miR-205-5p Mediated Downregulation of PTEN Contributes to Cisplatin Resistance in C13K Human Ovarian Cancer Cells* Xiaoyan Shi, Lan Xiao, Xiaolu Mao, Jinrong He, Yu Ding, Jin Huang, Caixia Peng and Zihui Xu

# Editorial: Non-Coding RNAs and Human Diseases

Yujing Li <sup>1</sup> \*, Ge Shan<sup>2</sup> , Zhao-Qian Teng<sup>3</sup> and Thomas S. Wingo1,4

*<sup>1</sup> Department of Human Genetics, Emory University, Atlanta, GA, United States, <sup>2</sup> CAS Key Laboratory of Innate Immunity and Chronic Disease, School of Life Sciences, University of Science and Technology of China, Hefei, China, <sup>3</sup> State Key Laboratory of Stem Cell and Reproductive Biology, Institute of Zoology (CAS), Beijing, China, <sup>4</sup> Department of Neurology, Emory University, Atlanta, GA, United States*

Keywords: miRNA, ncRNAs, lncRNA, circRNA, piRNA, cancer biology, neurologcial disorders

**Editorial on the Research Topic**

#### **Non-Coding RNAs and Human Diseases**

Non-coding RNA (ncRNA) are functional RNA molecules that are not translatable into proteins (Djebali et al., 2012; Lonsdale et al., 2013; Forrest et al., 2014). Initially, ncRNAs referred to tRNAs and rRNAs (Brown et al., 1992; St Laurent et al., 2015). Recent technical advances have led to the discovery and characterization of many new classes of ncRNAs (Hüttenhofer and Vogel, 2006). These new ncRNAs species include snRNAs, snoRNAs, miRNAs, siRNAs, piRNAs, exRNAs, long non-coding RNA (lncRNAs), scaRNAs, and circRNAs (He and Hannon, 2004; Gu et al., 2007; Esteller, 2011; Redzic et al., 2014; Wu and Yang, 2015). While not all of their functions are known, many of the ncRNA species appear to play essential roles regulating transcription and translation of genes and transcription of ncRNAs themselves. Thus, there is little surprise that ncRNAs are identified as playing important roles in normal physiologic processes, complex human traits, and human diseases (Diederichs et al., 2016; Li et al., 2018; Fernandes et al., 2019; Vijayan and Reddy, 2020). This special issue focused on the ncRNA, particularly circRNAs, lncRNAs, miRNAs, and their role in human disease. The aim of this issue is to provide a broad overview of current research on the diverse work being done to elucidate the role of ncRNAs in disease. A major theme that emerged was the potential role of miRNAs as prognostic markers or biomarkers of disease.

NcRNAs AND CANCER

ncRNAs play vital roles in tumorigenesis and tumor progression that is incompletely understood. Several investigators addressed the role of circRNA-miRNA-mRNA networks in different cancers, including hepatocellular carcinoma (HCC) (Sheng et al.), gastrointestinal stromal tumors (Jia et al.), and cervical cancer (Liu C. et al.,). In a perspective, Molin et al. address the significance of circRNAs in MLL arranged acute leukemia (MLLre) recombinome. In lung adenocarcinoma, (Stewart et al.,) performed a large-scale analysis of lncRNAs and find evidence for deregulated pseudogene-derived lncRNAs associated with cancer survival. microRNAs are among the better known ncRNAs. Here, Pereira et al., identify miRNAs associated with the development of gastric cancer to novel targets and potential early-stage indicators. The role of miRNAs to identify HCC progression was also explored using fectal-derived miRNAs by Wang et al.. The role of miRNA and mRNA expression in endometrial cancer by Xu et al. using the The Cancer Genome Atlas (TCGA) identified mRNAs and miRNAs associated with patient survival suggesting a potential role for miRNAs in predicting clinical outcomes. The role of polymorphisms in miRNA and their contributions to cancer risk was explored by Choupani et al. by investigating the association of polymorphisms mir-196a-2 rs11614913 and mir-149 rs2292832 with multiple cancers (e.g., gynecological cancers, ovarian, breast, and HCC) in their updated meta-analysis. Finally, five review articles highlight the recent advances

#### Edited by:

*William Cho, Queen Elizabeth Hospital (QEH), Hong Kong*

#### Reviewed by:

*Mohammadreza Hajjari, Shahid Chamran University of Ahvaz, Iran*

#### \*Correspondence:

*Yujing Li yli29@emory.edu*

#### Specialty section:

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

Received: *03 February 2020* Accepted: *29 April 2020* Published: *25 May 2020*

#### Citation:

*Li Y, Shan G, Teng Z-Q and Wingo TS (2020) Editorial: Non-Coding RNAs and Human Diseases. Front. Genet. 11:523. doi: 10.3389/fgene.2020.00523*

**6**

in the understanding of lncRNAs, circRNAs, and miRNAs in cancer initiation and progression, provide insight into their potential as biomarkers and therapeutic targets (Dong et al.; Plousiou and Vannini; Bandini and Fanini; Khan et al.).

To date, ncRNAs, particularly lncRNAs and miRNAs, are implicated in resistance or sensitivity to chemotherapy. Work by Xiang et al. analyzed the Cancer Cell Line Encyclopedia database and identified 44 of ncRNAs differentially expressed and significantly related to resistance or sensitivity of therapy for the advanced or metastatic breast cancer using Lapatinib, a small molecule inhibitor of HER1 and HER2 receptors. Shi et al. assayed miRNAs to identify the cisplatin-resistance in C13K human ovarian cancer cell and its cisplatin-sensitive OV2008 parental cells, and they found miR-205-5p led to PTEN downregulation and the subsequent enhancement of its downstream target p-AKT significantly contributes to cisplatin resistance in C13K Cells. Finally, Wan et al. addressed the cardiotoxicity from compound doxorubicin (DOX), a broad-spectrum anti-tumor drug. Their study found that the severe heart failure incurred by DOX based chemotherapy attributed to the enhanced expression of p21.

### NcRNAs AND LIVER AND CARDIOVASCULAR DISEASES

Non-alcoholic fatty liver disease (NAFLD) is a prevalent chronic liver condition that is associated with liver failure and HCC, Huang Z. et al. identified miRNAs that regulate the level of CYP3A4, an important drug-metabolizing enzyme associated with the progression of NAFLD, and they found that miR-200a-3p and miR-150-5p appear to directly regulate CYP3A4 and are involved in free fatty acid (FFA)-induced steatosis, implicating them in the NAFLD pathogenesis. The role of miRNAs was also examined in coronary artery disease (CAD), a prevalent human disease by Liu S. et al. who identified the miR-378a-5p-CDK1 axis as important in the proliferation and migration of vascular smooth muscle cells that can cause development of atherosclerosis and treatment failure for CAD.

#### REFERENCES


#### NcRNAs AND NEUROLOGIC DISEASE

The role of lncRNAs to potentially address the wide-range of clinical severity and age-at-onset for spinocerebellar ataxia type 3, a rare neurodegenerative disease identified six lncRNAs that were initially identified in blood and then tested in cerebellum Li T. et al. In multiple sclerosis (MS), the most common chronic neurologic disease in young adults, identified circRNAs MSassociated genes and showed evidence that top MS-GWAS results are enriched for blocks with circRNAs thereby suggesting a potential novel role for circRNAs in MS pathogenesis (Paraboschi et al.). Review articles focused on recent progress of ncRNAs in many neurologic diseases including Alzheimer's Disease (Liu X. et al.), fragile X syndrome (Huang G. et al.; Zhou et al.), and neurodevelopmental disorders (Li L. et al.; Zhang et al.).

As the research and review articles of this issue highlight, ncRNAs are a diverse group of RNA species that are likely important contributors to human disease. They provide a unique window into disease pathogenesis and ability to target networks of genes and cellular processes. Future work to understand ncRNAs and their role in human illness will undoubtedly be aided by incorporating ncRNAs with transcriptomic and proteomic data. Such studies will be able to provide a more complete model of disease pathogenesis. We hope that future work will also examine the potential role of ncRNAs in prospectively collected samples from clinical trials or population-based samples to evaluate their suitability as prognostic indicators and biomarkers of disease.

### AUTHOR CONTRIBUTIONS

YL, Z-QT, GS, and TW drafted the editorial. YL and TW revised the editorial with contributions from all authors. All authors approved the final version.

#### FUNDING

This work was supported by NIH grants RF1 AG057470, U01 AG061357, P50 AG025688, R56 AG062256, R56 AG060757, and R01 AG056533 (to TW).


Wu, N., and Yang, B. B. (2015). The biological functions of non-coding RNAs: from a line to a circle. Discoveries 3:e48. doi: 10.15190/d. 2015.40

**Conflict of Interest:** 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 © 2020 Li, Shan, Teng and Wingo. 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.

# Long Noncoding RNA Can Be a Probable Mechanism and a Novel Target for Diagnosis and Therapy in Fragile X Syndrome

*Ge Huang, He Zhu, Shuying Wu, Manhua Cui and Tianmin Xu\**

*The Second Hospital of Jilin University, Changchun, China*

Fragile X syndrome (FXS) is the most common congenital hereditary disease of low intelligence after Down syndrome. Its main pathogenic gene is fragile X mental retardation 1 (FMR1) gene associated with intellectual disability, autism, and fragile X-related primary ovarian insufficiency (FXPOI) and fragile X-associated tremor/ataxia syndrome (FXTAS). FMR1 gene transcription leads to the absence of fragile X mental retardation protein (FMRP). How to relieve or cure disorders associated with FXS has also become a clinically disturbing problem. Previous studies have recently shown that long noncoding RNAs (lncRNAs) contribute to the pathogenesis. And it has been identified that several lncRNAs including FMR4, FMR5, and FMR6 contribute to developing FXPOI/FXTAS, originating from the FMR1 gene locus. FMR4 is a product of RNA polymerase II and can regulate the expression of relevant genes during differentiation of human neural precursor cells. FMR5 is a sense-oriented transcript while FMR6 is an antisense lncRNA produced by the 3′ UTR of FMR1. FMR6 is likely to contribute to developing FXPOI, and it overlaps exons 15–17 of FMR1 as well as two microRNA binding sites. Additionally, BC1 can bind FMRP to form an inhibitory complex and lncRNA TUG1 also can control axonal development by directly interacting with FMRP through modulating SnoN–Ccd1 pathway. Therefore, these lncRNAs provide pharmaceutical targets and novel biomarkers. This review will: (1) describe the clinical manifestations and traditional pathogenesis of FXS and FXTAS/FXPOI; (2) summarize what is known about the role of lncRNAs in the pathogenesis of FXS and FXTAS/FXPOI; and (3) provide an outlook of potential effects and future directions of lncRNAs in FXS and FXTAS/FXPOI researches.

Keywords: long noncoding RNA, fragile X syndrome, fragile X-related primary ovarian insufficiency, fragile X-associated tremor/ataxia syndrome, FMR4, FMR6, BC1, TUG1

### INTRODUCTION

Fragile X syndrome (FXS) is a congenital hereditary disease associated with low intelligence, and is second common only to Down syndrome. The main pathogenic gene of the fragile X syndrome (FXS) is the fragile X mental retardation 1 (FMR1) gene located at Xq27.3, and it was first cloned in 1991 by Verkerk et al. (1991). Between Xq27 and Xq28, the chromosomes

*Edited by:* 

*Ge Shan, University of Science and Technology of China, China*

#### *Reviewed by:*

*Marianna Aprile, Italian National Research Council (CNR), Italy Sandeep Kumar, Emory University, United States*

> *\*Correspondence: Tianmin Xu xutianmin@126.com*

#### *Specialty section:*

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

*Received: 14 December 2018 Accepted: 30 April 2019 Published: 22 May 2019*

#### *Citation:*

*Huang G, Zhu H, Wu S, Cui M and Xu T (2019) Long Noncoding RNA Can Be a Probable Mechanism and a Novel Target for Diagnosis and Therapy in Fragile X Syndrome. Front. Genet. 10:446. doi: 10.3389/fgene.2019.00446*

**9**

are abnormally concentrated during meiosis, forming an extremely fragile filamentous site. Therefore, this syndrome is regarded as a fragile X chromosome syndrome. As well, the 5′ end untranslated region (UTR) of the gene has a highly conserved CpG island with a length of 250 bp, which includes a trinucleotide repeat expansion (CGG)n whose sequence is abnormally amplified and methylated in individuals with fragile X chromosome syndrome (Bardoni and Mandel, 2002). Based on the CGG trinucleotide repeat numbers, the sequence is divided into full mutation, permutation, gray zone, and normality (Allingham-Hawkins et al., 1999; Bardoni and Mandel, 2002). In addition, full mutation, an expansion beyond 200 repeats, is associated with typical clinical symptoms of FXS which is responsible for intellectual disability, autism, and so on. Besides, permutation which corresponds to 55–200 repeats may contribute to development of fragile X-related primary ovarian insufficiency (FXPOI) and the fragile X-associated tremor/ataxia syndrome (FXTAS). A size of 45–54 is known as a gray zone, while normality corresponds to 5–44 repeats (Wittenberger et al., 2007; Ciaccio et al., 2017).

There is a lot of evidence that all proteins originate from only 1.2% of the genome, although 40% or higher of the human genome is transcribed into RNA (Carninci et al., 2005; Cheng et al., 2005; Consortium et al., 2007), known as non-protein-coding RNA (ncRNA) and enriched in the brain (Djebali et al., 2012). As well, ncRNAs are classified into different categories based on their functions and sizes. For instance, ncRNAs are differentiated into short and long noncoding RNAs (lncRNAs). Also, we introduced the long noncoding RNAs (lncRNAs) in this article, having a wide range of functions. They are also a kind of ncRNA with a length of more than 200 nucleotides while short noncoding RNAs certainly have less than 200 nucleotides. Short noncoding RNAs include microRNAs (miRNAs), small-interfering RNA (siRNA), and piwi-interacting RNA (piwi-RNA) and small nuclear RNAs (snRNA) (Zhou et al., 2019). For example, miRNAs are noncoding single-stranded RNAs consisting of 18–25 nucleotides in length (Akbari Kordkheyli et al., 2019). Previous studies have recently indicated that lncRNAs contribute to the pathogenesis of both the full mutation and premutation carriers, especially the nervous disorders. Still, there is no summary of function and mechanism of lncRNAs in FXS patients and premutation carriers, while the article sums up and explains the link between FXS and lncRNAs in detail.

### FUNCTIONS AND MECHANISMS OF LONG NONCODING RNAs

NcRNAs play a significant role in human diseases. It is also demonstrable that ncRNAs have multiple functions. For example, pi-RNAs can repress translation (Fire et al., 1998; Carmell et al., 2007; Houwing et al., 2007), miRNAs can inhibit translation, while siRNAs can lead to silencing of a wide range of genetic targets and degrade mRNA (Fire et al., 1998; Dana et al., 2017). And a lot of evidence implicates that numerous protein-coding genes such as FMR1 have antisense transcripts. In some instances, antisense transcription manipulation may result in sense transcription inhibition or make sense transcription more stable. These phenomena are known as discordant regulation and concordant regulation, respectively (Katayama et al., 2005). However, it is still not certain whether the exact regulation mechanisms of sense transcriptions are regulated by antisense partners, which may be various and complicated; thus, there is a need for more and further studies. Previous researches have revealed that the expression pattern of ncRNAs has changed in many patients suffering from human disorders, such as cardiovascular disease, neuronal dysfunction, and cancer (Haemmig and Feinberg, 2017; Nicolas, 2017; Cai et al., 2019; Shao et al., 2019; Zhang et al., 2019). They all indicate that ncRNAs can be functionally related to human diseases. Therefore, ncRNAs are probably taken for potential drug targets (Khalil et al., 2008; Nicolas, 2017). Short noncoding RNAs have been extensively explored and reviewed that they tend to affect the gene expression through the interference with translation or posttranscriptional mechanisms (Rother and Meister, 2011; Ojha et al., 2019). However, we only learned in recent years that lncRNAs can contribute to regulating cellular functions and/or physiology as regulatory factors. As the function of the lncRNAs gradually surfaces, lncRNAs have attracted more and more attention as potential biomarkers and/or drug targets (Kung et al., 2013; Tang et al., 2014; Zhu et al., 2018; Zou et al., 2018).

In particular, lncRNAs can originate from both sense and antisense chains of genes that can encode proteins. As well, they are likely to be a transcript of promoter, intron, and 3′ end region. More specifically, lncRNAs are grouped into five categories according to the nearest protein-coding genes, which are sense, antisense, bidirectional, intronic, and intergenic lncRNAs. Besides, protein-coding genes are defined as the sense DNA. Transcriptions of sense lncRNAs incline to the occurrence of the sense DNA strand, which is different in the antisense lncRNAs. However, they all overlap one exon or more. Bidirectional lncRNAs are transcribed from the promoter in two directions, and their length is usually several hundred base pairs (bps) (Sarfi et al., 2019). Intronic lncRNAs do not overlap any exons and are the transcripts that originate from introns in any directions. Meanwhile, intergenic lncRNAs are stand-alone, meaning that they can exist in the sequence space without contain protein-coding genes, also known as large intergenic (or intervening) ncRNAs (Kung et al., 2013; Kumar and Goyal, 2017). At the same time, lncRNAs can interact with DNA, RNA, and protein, and other biological macromolecules and perform many biological functions such as regulation of the activity of transcription and the epigenetic landscape of their original locus (Wang and Chang, 2011; Huarte, 2015; Klinge, 2018). In addition, lncRNAs usually play a crucial role in cis- or trans-regulation of gene expression at their original sites or other locus in the genome. They also perform scaffolding function and remodeling chromatin through recruiting epigenetic complexes and ribonucleotide nucleotide proteins. The functions of other lncRNAs are executed *via* targeting mRNAs or regulating the post-transcriptional mechanism of genes. The subcellular locations of lncRNAs have a significant effect on their functional properties. The lncRNAs which are located in the nucleus can regulate genetic transcription and perform epigenetic modification by binding DNA to generate RNA–DNA triplex complex. As well, cytoplasmic lncRNAs can affect the stability of mRNAs and act on posttranscriptional regulation (Mercer and Mattick, 2013). For example, lncRNAs which are located in the nucleus can contribute to both RNA processing and protein modifications *via* interacting with RNA-binding proteins (RBPs) (He et al., 2019). In addition, they may contribute to up-regulated expression of mRNAs by acting as miRNA sponges and reduce the regulatory effects of miRNAs (Kinney and Pradhan, 2013; Militello et al., 2017). Also, they may be involved in protein synthesis or interact with other proteins and RNAs to influence cellular signaling cascades (Kinney and Pradhan, 2013; Sanchez Calle et al., 2018). LncRNAs can also generate a secondary and/or tertiary structure that provides multiple binding sites for proteins and other regulatory RNAs (Liu et al., 2017). For example, lncRNAs may bind DNA-binding proteins and prevent DNA-binding proteins from attaching to related transcription factors. A typical example is that some lncRNAs can stop DNMT1 from binding its targeted DNA. Thus, the methylation of the targeted DNA has been affected to some extent. As a result, transcriptional activation of the gene is influenced (Hung et al., 2011). As well, lncRNAs can play an important role in DNA damage response and cellular division (Hajjari et al., 2014). No matter the biological mechanism of lncRNAs, there is enough evidence that lncRNAs are involved in the numerous normal and abnormal cell functions.

#### THE CLINICAL MANIFESTATIONS AND TRADITIONAL PATHOGENESIS OF FRAGILE X SYNDROME

In essence, men are more likely to get FXS than women. This is because the X chromosome is linked to recessive inheritance. In fact, about 80% of male patients have intellectual disabilities. According to IQ, the severity is classified as mild, moderate, and severe mental disability, with corresponding IQ values at 40–54, 50–70, and less than 20, respectively. Many patients with full mutation are disabled moderately (Greco et al., 2002). They tend to suffer from midface hypoplasia, language barrier, autism, and macro-orchidism (Verkerk et al., 1991) while women are considered to be the carriers of FXS. More specifically, about 70% of women have normal intelligence as carriers, and female FXS patients only tend to have mild mental retardation. Female premutation is likely to suffer from FXPOI, leading to a reproductive decline.

The full mutation is responsible for FXS associated with inherited intellectual and developmental disability. With full mutation, the promoters and CpG islands of FMR1 gene are highly methylated. Meanwhile, the associated histone proteins are hyperacetylated and chromatin aggregated. Then, the silencing of FMR1 gene transcription is attributed to the absence of fragile X mental retardation protein (FMRP), a protein product encoded by the causative gene (Fu et al., 1991; Ali et al., 2017). The expression of FMRP begins at the early stage of development and lasts a lifetime. The expression of FMRP widely exists within all mammalian tissues. However, it is particularly abundant in the testis and brain. In the brain, FMRP exists mainly in the cytoplasm of neurons, including soma, dendrites, and synapses. FMRP, a kind of mRNA-binding protein, can associate with ribosomes and be involved in the aggregation of mRNA as well as regulation of the transcription efficiency of targeted genes. It also participates in protein synthesis of axons and dendrites. Therefore, the absence of FMRP might lead to the abnormal translation of mRNA and the abnormal structure and function of synapsis, which would affect the function of the nervous system (Pieretti et al., 1991; Sutcliffe et al., 1992; Darnell and Klann, 2013). The deduction supports that the deficient expression of FMRP is responsible for intellectual disability, thus patients exhibit a series of clinical manifestations (Weiler et al., 1997; Brown et al., 1998; Antar et al., 2005). Meanwhile, FMRP can potentially play an important role in the nucleus. Some existing researches have reported that FMRP may also be involved in mediating the DNA-damage response pathway through binding to methylated H3K79 chromatin (Liu et al., 2012).

### THE CLINICAL MANIFESTATIONS AND TRADITIONAL PATHOGENESIS OF FRAGILE X-ASSOCIATED TREMOR/ ATAXIA SYNDROME AND FRAGILE X-RELATED PRIMARY OVARIAN INSUFFICIENCY

#### The Clinical Manifestations of Fragile X-Associated Tremor/Ataxia Syndrome and Fragile X-Related Primary Ovarian Insufficiency

Premature ovarian insufficiency (POI) is the cessation of ovarian function before 40 years of age. POI refers to the loss of germination and hormone function before normal physiological menopause as a result of the exhaustion of ovarian follicles (Hoek et al., 1997). As well, the risk of menstrual dysfunction, diminished ovarian reserve, and infertility is increased due to POI. In contrast, the age of menopause is decreased for POI patients. Laboratory tests have revealed that hypoestrogenism and elevated gonadotropin serum levels, which are characterized by low estradiol (E2) levels (<20 pg/ml), increased gonadotropin levels (follicle-stimulating hormone ([FSH] > 20 IU/l), low anti-Müllerian hormone (AMH) levels – <0.5 ng/ml (<1 ng/ml), and low inhibin B levels (Bellipanni et al., 2001). FSH levels may vary from cycle to cycle. Meanwhile, AMH is considered the best marker of ovarian reserve. As well, estrogen deficiency leads to the first symptoms: excessive sweating, tension, diminished libido, hot flushes, weakness, and mucous membrane dryness. In addition, substantial chronic hypoestrogenism may cause bone injuries and a higher risk of bone fracture. Even for younger women with POF, it is possible to have a consequent decrease in bone mineral density. Therefore, densitometry testing is necessary. Deficient estrogen levels are associated with metabolic disorders, thus can lead to cardiovascular diseases such as hypercholesterolaemia, atherosclerosis, and urogenital atrophy – infections and vaginal dryness (Fink et al., 2018). However, lower fertility or even infertility is one of the most troubling POF-associated problems for young women. However, for a majority of the women suffering from premature ovarian insufficiency (POI), etiology is still completely unknown. The risk of POI for premutation carriers with CGG repeat on one allele has been found to be high (up to 35%), while only 1% of women in the general population suffer from POI. As well, noncarriers might experience menopause 5 years later than premutation carriers, since these carriers' ovarian function is obviously destroyed (Bretherick et al., 2005; Bodega et al., 2006; Mailick et al., 2014). Recent studies have demonstrated that overt premature ovarian insufficiency is correlated with premutation repeat lengths. Also, there is a linear relationship between the size of CGG trinucleotide repeats and the risk of POI/ovarian phenotype. With an increase in the size of repeats, the risk of POI elevates and reaches a plateau, but decreases beyond 80–100 repeats (Sullivan et al., 2005; Ennis et al., 2006; Tejada et al., 2008). Allen and colleagues realized that the average age of menopause tends to decline in all premutation carriers, which is apparently the medium-sized repeats with the lowest menopause age. It means that mediumsized repeats have lower odds ratio for fertility and an increased rate of dizygotic twinning compared with normal individuals while the rate of spontaneous abortion does not increase. This indicates that it puts no damage on the quality of oocyte. Both low and high repeats tended to suffer the same experience from insufficient ovarian reserve, although not as serious as medium-sized carriers (Allen et al., 2007).

FXTAS is a neurodegenerative disorder characterized by progressive intention tremor, gait ataxia, psychiatric symptoms, parkinsonism, cognitive decline, and autonomic disorders, which typically occurs after 50 years of age (Hagerman et al., 2001; Jacquemont et al., 2003; Kong et al., 2017). Main principal neuropathological features of FXTAS include Purkinje cell loss, brain atrophy, and ubiquitin-positive intranuclear inclusions which are mostly present in single, large (~2–5 μm of diameter), and spherical aggregates (Weiler et al., 1997; Greco et al., 2002, 2007; Jacquemont et al., 2003). These aggregates exist in different areas of the brain, especially in the hippocampus, and they are also found in the brain's Purkinje cells. Additionally, rare intranuclear inclusions, which are ubiquitin-positive and contain various chaperones, are also detected in the tissues outside of the central nervous system (Iwahashi et al., 2006). In particular, FXTAS is more common in males than in females (Rodriguez-Revenga et al., 2009). Previous studies have found that FMR1 premutation can contribute to FXTAS development (Garcia-Arocena and Hagerman, 2010). As well, female permutation carriers do not suffer from FXTAS in most cases due to the defense mechanism that the pathogenic mutant, allele (located at X-chromosome) may become inactivated randomly. In addition, it affects an estimated 46% of males and 17% of females. Compared to carriers of the CGG premutation allele with control (<30) CGG repeat size, the FMR1 mRNA level of the former has increased 2 to 8-folds *via* brain and blood analysis (Tassone et al., 2000, 2004). Tassone and colleagues attributed the higher FMR1 mRNA level to an increased transcriptional activity of the FMR1 gene (Tassone et al., 2007), which may be due to epigenetic modifications in the result of CGG repeat expansion itself (Todd et al., 2010; Usdin and Kumari, 2015).

#### The Traditional Pathogenesis of Fragile X-Associated Tremor/Ataxia Syndrome and Fragile X-Related Primary Ovarian Insufficiency

However, little is known about the mechanisms that contribute to the development of FXPOI and FXTAS. These individuals with an allele of an expansion beyond 200 CGC trinucleotide repeats lead to the incomplete absence of FMRP, but they still perform their function normally. Thus, insufficient FMRP does not submit to be the culprit of developing FXPOI/FXTAS (Sherman et al., 2014). Kenneson put forward that FMR1 RNAs' transcription of the permutation carriers was positively relevant with the size of CGG trinucleotide repeats while the FMRP translation was negatively correlated with the size of CGG trinucleotide repeats (Kenneson et al., 2001). The excessive FMR1 mRNAs in premutation carriers contribute to several proteins' dysregulation and deposition. Along with FMR1 mRNAs, these proteins are present in several parts of the body in the form of cell inclusions, which is located in CNS, peripheral nervous system (particularly autonomic ganglia), pituitary, and Leydig cells (Greco et al., 2006; Gokden et al., 2009). Therefore, partial FMRP deficiency and/or RNA toxicity may be involved in the pathogenic mechanism of FXPOI/ FXTAS. In addition, abnormal translation of the CGG repeats can result in the production of polyalanine (FMRpolyA). The polyglycine (FMRpolyA) and other polypeptides containing proteins are neurotoxic (Hall and Berry-Kravis, 2018). FMRpolyG proteins can be detected in the brains and other tissues of individuals, but FMRpolyA could only be demonstrated in transfected cells with FXTAS. The phenomenon illustrated that FMRpolyA may contribute to the development of FXTAS (Hukema et al., 2015).

The theory suggests that dynamic intranuclear long rCGG RNA, translated by the gene with CGG trinucleotide repeats, can contribute to making normal cells dead and nonfunctional *via* tracking and binding to a wide range of RNA-binding proteins (RBP). The intranuclear long rCGG RNA is sequestersspecific and its sequestration can make the viability of normal cells decreased (Sellier et al., 2010). Also, the phenomenon is confirmed in another study, suggesting the fully mutated carriers or those with 4,199 methylated repeated alleles lead to silencing the expression of FMR1 gene, but did not suffer from FXPOI or FXTAS. That leads to the conclusion that the silence of disease-causing gene (FMR1) and protein products (FMRP) is not the main culprit. On the contrary, FMR1 transcript levels increase with premutation carriers. In addition, Elizur et al. have confirmed the fact that FMR1 mRNAs of both male and female premutation carriers are up-regulated, and FMR1 mRNAs play a significant role in FXTAS and FXPOI (Elizur et al., 2014).

### CATEGORIES OF PATHOGENIC LONG NONCODING RNAs ORIGINATING FROM FMR1 GENE

Previous studies have indicated that lncRNAs contribute to the pathogenesis of both permutation and full mutation carriers, especially nervous disorders. In addition, it is reported that the expression of lncRNAs is different, and several relevant lncRNAs originate from the FMR1 gene locus in both FXS patients and premutation carriers. This suggests that they may be markers to diagnose or evaluate relevant disorders (**Figure 1**; Ladd et al., 2007; Khalil et al., 2008). And the discovered lncRNAs originating from FMR1, including FMR4, FMR5, and FMR6 as shown in **Figure 1**. Additionally, BC1 RNA and lncRNA TUG1 which originate from other genes also contribute to making permutation carriers sick.

#### FMR4 Can Affect Cell Proliferation or Differentiation

FMR4, an untranslated primate-specific lncRNA (2.4 kb), is transcribed upstream of FMR1 in the antisense direction. It is widely expressed in body development. In particular, FMR4 has been widely detected during our growing up years. Also, the expression of FMR4 exists in some adult tissues such as brain, small intestine, spleen, colon, liver, and placenta except in the ovaries, prostate, pancreas, or testes (Hinds et al., 1993). During embryonic and/or fetal development, FMR4 is likely to express in these orangs and tissues (ovaries, prostate, pancreas, or testes). Similarly, the expression of FMR4 is detected highly in kidney and heart of fetus. Considering that a number of people suffering from FXS have cardiac dysfunctions, such as prolapse of mitral valve and aortic root dilation, FMR4 which expresses highly in heart may have a functional role to play in the relevant pathogenic mechanism (Sreeram et al., 1989).

FMR4, a product of RNA polymerase II, can be detected in normal people as well as in premutation carriers, but not in FXS with full mutation. It also has similar half-life to FMR1 mRNA. Similar to FMR1 mRNA level, the expression of FMR4 is up-regulated in premutation carriers and silenced in brain tissue of full mutation carriers (FXS). Some previous researches have shown that overexpression and knockdown of FMR4 can alter the expression of these genes, which has an effect on cellular proliferation or differentiation. As a chromatographic transcript, FMR4 can induce the changes of transcriptional levels by directly aiming at mRNAs splicing, editing, or stability, and by binding histone-modifying enzymes to develop complexes. However, it is not certain whether these observed results are affected by epigenetic changes, RNA-protein interactions, or downstream effects. The study also confirmed the methyl-CpGbinding domain protein 4 (MBD4), which is a negatively responsive gene of FMR4. MBD4 is a member of MBD nucleoprotein family, and it has two domains: one is methyl-CpG-binding domain which can specifically bind methylated CpG, and the other one is a DNA glycosylase domain which is associated with the catalytic activity. Thus, MBD4 is crucial in DNA mismatch repair, inhibition of transcription, and the regulation of apoptosis (Yakovlev et al., 2017). In addition, the study revealed that FMR4 very likely shares a bidirectional promoter with FMR1. The expression of FMR4 is developmentally regulated and shows negative relation to the expression of both FMR1 and MBD4 in human neural precursor cells, which are in differentiation. Therefore, all evidence implicates that as a kind of LncRNAs and FMR4 can regulate the function of relevant genes which can regulate gene, and the transcript may act in cellular development (Peschansky et al., 2015). However, Peschansky and colleagues also suggested that the genes regulated by FMR4 enrich and are involved in cell proliferation and neural development. S-phase marker assays further exhibited that FMR4 may up-regulated cell proliferation, rather than differentiation of human neural precursor cells (hNPCs). They confirmed the theory by using transfection, qPCR, and subcellular fractionation, and other technologies. The experimental group is that HEK293T cells are transfected with pcDNA3.1-FMR4, while the control group is that cells are transfected with empty pcDNA3.1 control vector or silencer negative control siRNA aiming at FMR4. And FMR4 is highly expressed in the experimental group but not in the control group. They found that the silencing and overexpression of FMR4 can lead to the genome-wide change in histone methylation. The situation also

exists in other mRNAs which have been confirmed to act in developmental or neurophysiological roles. FMR4 works mainly by forming scaffolds for the recruitment of histone-modifying complexes or other proteins to affect the stability, splicing, or editing of relevant mRNAs (**Figure 2**; Peschansky et al., 2016).

The phenomenon was also found in the research of Ahmad M and colleagues. As mentioned above, there are no overlaps between the FMR4 and FMR1 because FMR4 lies in the upstream of FMR1 with 2.4 kb in the antisense direction. Thus, they used relevant siRNAs to knockout FMR1 and found that there has been no change in the expression of FMR4. Thus, they reported that expression of FMR4 cannot be affected by FMR1. These observations have indicated that FMR4 is not directly derived from regulatory transcript for FMR1. But siRNA knockdown of FMR4 can cause an increase of cell apoptosis and the changes of cell cycle. In addition, they simultaneously found that the overexpression of FMR4 resulted in an increased proliferation *in vitro*. Thus, FMR4 has a significant effect on cell proliferation *in vitro*. The evidence implicated that FMR4 can act at anti-apoptosis in HeLa cells and HEK293T and suggests that studying genomic locus can make us discover unknown functions of the gene. They also speculate that changes in the expression of FMR4 may affect the clinical manifestations of FXS or associated disorders (Khalil et al., 2008).

Another transcript, ASFMR1, originates from the CGG expanded repeats in the 5′ UTR of FMR1, and was reported by Ladd and colleagues. The expression of ASFMRI is silenced in full carriers and increased in premutation carriers, with the changes being similar to FMR4. Ladd and colleagues also evaluated that FMR4 is likely to be nested in the 3′ UTR of ASFMR1 due to its one splice variant overlapping that of FMR4. Besides, ASFMR1 is widely expressed in human tissues with relatively high expression in brain. The ASFMR1 transcript is transported to the cytoplasm and contains a potential proline-rich ORF, indicating that ASFMR1 has a conserved cellular function and can potentially be associated with FXS and FXTAS (Ladd et al., 2007).

#### FMR5 Is a Kind of Sense-Oriented Long Noncoding RNA

FMR5 is a sense-oriented lncRNA and its transcription begins around 1 kb upstream from the FMR1 transcription start site (TSS) which overlaps with the FMR1 promoter. Meanwhile, FMR5 and FMR6 are discovered by using a new technology called "Deep-RACE." This technology can combine next-generation sequencing with rapid amplification of cDNA ends (RACE). The expression of FMR5 reportedly appeared in some brain tissue with full mutation, permutation carriers, and normal individuals (Pastori et al., 2014). There are similar expression levels of FMR5 in brain tissue with normal people, full mutation, and permutation carriers, which shows that FMR5 transcription is not related to chromatin methylated modifications. Kumari and Usdin also indicated that the transcription of low-riched transcripts such as FMR5 may be repressed by the essence of negative histone marks in the FMR1 locus. This is consistent with the discovery that trimethylation of histone H4 at lysine 20 (H4K20me3) as a negative chromatin mark as well as trimethylation of histone H3 at lysine 9 (H3K9me3) are related to exon 1 of the FMR1 gene, such as CGG expanded repeats, but not connected with the promoter region (Kumari and Usdin, 2010). At the same time, lower levels of three positive chromatin marks, including H3K4 dimethylation (H3K4me2), H3 acetylation (H3Ac), and H4 acetylation (H4Ac) combine with the FMR1 promoter in full mutation carriers (Gheldof et al., 2006). Kumari and Usdin suggested negative histone modifications on silenced FMR1 may contribute to developing FXS because these modifications enrich FX alleles, and the intrinsic and local repeats may lead to the silence of FMR1 (Kumari and Usdin, 2010).

#### FMR6 Regulates Translational Efficiency and/or Stability of FMR1

The expression of FMR5 and that of FMR6 are dependent on completely different patterns. FMR6, a spliced lncRNA, is

target gene. (B) FMR4 can play role in DNA mismatch repair, transcriptional inhibition, and apoptosis regulation as a negatively regulating factor of MBD4. (C) FMR4 can induce the changes of transcriptional levels by directly aiming at mRNAs splicing, editing, or stability, or by forming scaffolds for the recruitment of functional protein.

transcribed by the 3′ UTR of FMR1 overlapping exons 15–17 in the antisense direction, and FMR6 may combine with the FMR1 mRNA because it is complementary to the 3′ region of FMR1. A study found that selective siRNA aiming to knockdown the nonoverlapping regions of the β-secretase-1 antisense transcript (BACE1-AS) made the expression of β-secretase-1 (BACE1) mRNA and protein decreased, which indicated that BACE1-AS can regulate BACE1 mRNA and BACE1 protein expression subsequently (Faghihi et al., 2008). And it follows that FMR6 may also regulate the splicing and stability of FMR1 mRNA and the expression level of FMRP. As well, FMR6 overlaps two microRNA binding sites, including miR-19a and miR-19b located in the 3′ UTR of FMR1, and the lncRNA may regulate translational efficiency or stability of FMR1 mRNA by the another way of combining with microRNAs (Pastori et al., 2014).

The expression of FMR6 is down-regulated in brain tissue from premutation carriers and fragile X patients (Pastori et al., 2014). Meanwhile, Shai E. Elizur and colleagues carried out a study to evaluate whether the accumulation of lncRNAs contributes to developing FXPOI. Their research suggested that FMR6 is expressed in ovarian granular cells from both premutation carriers and fragile X patients similar to FMR1 mRNA, and there is a marked nonlinearity between the FMR6 level of ovarian granular cells and the size of CGG repeats. Females in the medium-range CGG repeats (80–120) obviously keep up with higher FMR6 levels of granulosa cells. In addition, the transcription level of FMR6 is negatively associated with the number of oocytes detected. These findings indicate that in ovary granulosa cells of females with FXPOI, the accumulation of both FMR6 and FMR1 except FMR4, may result in ovarian dysfunction. Although the FMR1 premutation can result in primary ovarian insufficiency, the relationship is nonlinear but still not exact between ovarian reserve and CGG repeat numbers. We can speculate FXPOI is the result of an increased accumulation of FMR1 and FMR6 lncRNAs in ovarian granular cells in the intermediate range (80–120 CGG repeat). Also, more well-studied observations are required to explore the exact mechanism by which FMR1 mRNA and FMR6 contribute to FXPOI, and estimate whether these research findings might also be extended to the normal-range CGG repeats (Elizur et al., 2016).

#### THE FMRP-BC1-mRNA INHIBITORY COMPLEX

Presynaptic localization of FMRP has been confirmed in the CNS. FMRP can regulate the development of both axon and dendrite. Meanwhile, methylation of FMR1 gene two-way promoter can regulate the expression of lncRNAs. In addition, FMRP, a kind of protein involved in regulating the efficiency of translation and transporting messenger ribonucleoprotein (mRNP), can combine with the dendritic brain cytoplasmic RNA 1 (ncRNA BC1) to form the FMRP-BC1 complex. This complex can inhibit the translation of a certain subset of FMRP-targeted mRNAs in neurons.

BC1 ncRNAs have been regarded as negative translation regulators. Knockdown of BC1 RNAs' expression can lead to remarkably increased neuronal excitability and epilepsy (Zhong et al., 2009). The BC1 RNA can also play a role as an adaptor molecule to connect several mRNAs with FMRP (Zalfa et al., 2005). BC200, BC1 analog in primates, can have more advantages compared with BC1 because its RNA distribution is also able to indicate the localization of dendrites and neuron-specific expression (Tiedge et al., 1993). Also, the FMRP have exhibited enough binding sites as a regulatory and transport factor, such as two evidently characterized KH domains which can bind RNAs, and the N and C termini which have affinity for RNA. However, only the RGG box of FMRP can only bind RNA with special sequence and/or structure. Similarly, G quartet is a kind of RNA with rich G. It can directly bind FMRP, recognize and connect with the above four domains (Darnell et al., 2001; Schaeffer et al., 2001). Also, a new RNA-binding motif originating from the N terminus (NT) of FMRP has been identified to be capable of binding to BC200 specifically and directly. In addition, the FMRP-BC1/BC200 complex can cover up the signal which can regulate the FMRP cycle in and out of the nucleus. Thus, the transportation of mRNP is affected and the relevant proteins cannot be transported into the nucleus, while the breakdown of the complex permits proteins to reenter the nucleus freely (Zalfa et al., 2005). Studies *in vitro* have indicated that BC1 can play its functional inhibition role by binding both PABA and eIF4A (the translational initiation factor) (Wang et al., 2002). Meanwhile, Lacoux C and colleagues proposed that the interaction between BC1 and FMRP may be regulated by 2′-O-methylation. They demonstrated that BC1 RNAs in neurons are 2′-O-methylated differentially and have an effect in binding FMRP to the complex (**Figure 3**; Lacoux et al., 2012).

### LONG NONCODING RNA TUG1 REGULATE AXONAL DEVELOPMENT *VIA* MODULATING SnoN-CCD1 PATHWAY

Current evidence confirms that lncRNA TUG1 has an obvious and close relationship with FMRP for patients suffering from cancer. Guo and colleagues pointed out that the transcription of FMRP is successfully suppressed in neurons transfected by Fmr1shRNA. They found out that the length and complexity of the dendrites were all reduced in the neurons whose FMRP expression was deficient compared with the control group whose FMRP expression was normal. As such, the data can confirm that FMRP plays a crucial role in axonal development. However, there was no significant change in the length and complexity of the dendrites in TUG1-deficient neurons transfected by TUG1 shRNA. The researchers found that down-regulated TUG1 expression slightly led to developing axon better in neurons and up-regulated TUG1 expression results in significantly shortening the axonal length. Meanwhile, FMRP deficiency led to overexpression of TUG1 and knockdown of TUG1 expression can repair the defects of axonal development in FMRP-deficient neurons. It indicates that TUG1 may interact with FMRP to

specifically regulate axonal development of neurons. At the same time, they found that TUG1 can regulate axonal development by interfering with the SnoN-Ccd1 pathway, which is known to be involved in the development of axons. In addition, the reduced length of axon due to TUG1 up-regulation and FMRP deficiency can be rescued by the overexpression of Ccd1. However, making FMR1 silenced and TUG1 overexpressed does not alter the whole protein expression level of SnoN. It demonstrates that the interaction between TUG1 and FMRP regulates SnoN activity that would not be dependent on the ubiquitin-proteasome system. The system has been proved to be capable of activating SnoN pathway because the ubiquitin ligase Cdh1/APC can accelerate SnoN ubiquitination and subsequent degradation, and can consequently inhibit axonal development (Konishi et al., 2004; Guo et al., 2018).

It has been known that lncRNAs can affect gene expression level *via* binding to specific transcriptional factors and inhibit or enhance the activity of these specific transcriptional factors. Similarly, TUG1 can also inhibit the transcriptional activity of SnoN by binding with SnoN, which would lead to a decreased expression of Ccd1, a well-known SnoN-targeted gene. Therefore, these evidence provide another potential mechanism of how lncRNA TUG1 can regulate the transcriptional activity of SnoN. In other words, lncRNAs can also combine with particular transcriptional factors to repress or promote the activities of these transcriptional factors, thus regulate the expression of relevant genes, just as TUG1 can bind with SnoN to inhibit its transcriptional activity. This will result in the down-regulated expression of Ccd1, known as SnoN-targeted gene (Geisler and Coller, 2013). A large number of previous studies have proved that lncRNA TUG1 is closely related to FMRP for patients suffering from cancer. However, whether the relationship between lncRNA TUG1 and FMRP for patients suffering from FXS and FXPOI/FXTAS is the same or not, there is a need for more in-depth studies to verify it. If the answer is yes, then it could open new avenues of research and treatment for FXS and FXPOI/FXTAS.

### CONCLUSION AND PERSPECTIVE

The expression of FMR6, FMR5, FMR4, and FMR1 is different in patients' brain tissues with FXS and FXTAS/FXPOI, which could probably be as a result of the differences in CGG trinucleotide repeat numbers, DNA methylation degree, or histone modification degree. FMR4, a product of RNA polymerase II, indicated increased expression in premutation carriers and silenced expression in full mutation carriers. It can also regulate the expression of relevant genes during differentiation of human neural precursor cells (Kumari and Usdin, 2010). Meanwhile, both FMR5 and FMR6 are new transcripts from the FMR1 gene locus. The expression of FMR5 appears in some human brain regions in both FXS patients and premutation carriers, while the expression of FMR6 is silenced in premutation and full mutation carriers. For FXS and FXTAS/FXPOI patients, it is feasible that the levels of these transcripts (FMR6, FMR5, FMR4 and FMR1) can correspond to different clinical manifestations and results. Therefore, these lncRNAs may be taken for markers to diagnose and evaluate FXS and FXTAS/FXPOI. And there is a need for additional studies to evaluate whether any undiscovered functional properties of each transcript may result in clinical phenotypes of FXS and FXTAS/FXPOI patients (Pastori et al., 2014). Most importantly, several studies have suggested that FMRP also has a direct interaction with BC1, which affects its functional regulation and transportation. As well, lncRNA TUG1 can bind it to decrease its stability. In addition, lncRNA TUG1 can regulate axonal development by combining with SnoN and mediating SnoN-Ccd1 pathway. Also, there is a need to deeply explore an alternative potential mechanism of modulating the transcriptional activity of SnoN by lncRNA TUG1 and the function of lncRNA TUG1 in patients suffering from FXS and FXPOI/FXTAS (Fabian and Sonenberg, 2012). LncRNAs can be considered a new biomarker for human disease. It has been applied to new diagnostic or prognostic markers in bodily fluid samples, such as urine samples of patients suffering from cancer to detect the lncRNA prostate cancer antigen 3. The lncRNA can improve diagnosis of prostate cancer (Reis and Verjovski-Almeida, 2012). Similarly, to determine whether the lncRNAs' expression levels are related to clinical manifestations of fragile X Syndrome, it is necessary to detect these transcripts in a large number of patients. In particular, blood samples provide the most practical choice for both experimental and potential prognostic purposes. By understanding the mystery of these lncRNAs, the diagnosis and treatment of FXS, along with its associated disorders, may be more accurate and effective in the future.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

GH and HZ wrote the manuscript, carried out image analysis, and contributed to this work equally. SW drew the figures. SW and MC co-supervised GH and HZ, participated in discussions, and commented on the manuscript. TX conceived the idea, directed, and critically reviewed the manuscript.

#### FUNDING

This study was supported by grants from the National Natural Science Foundation of China (81302242), Jilin Province Development and Reform Commission funds (20180201032YY), and Education Department of Jilin Province (JJKH20170840KJ).


*Am. J. Med. Genet.* 94, 232–236. doi: 10.1002/1096-8628(20000918)94:3<232::AID-AJMG9>3.0.CO;2-H


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

*Copyright © 2019 Huang, Zhu, Wu, Cui and Xu. 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.*

# CeRNA Expression Profiling Identifies KIT-Related circRNAmiRNA-mRNA Networks in Gastrointestinal Stromal Tumour

*Ning Jia1,2†, Hanxing Tong3†, Yong Zhang3†, Hiroshi Katayama4, Yuan Wang1, Weiqi Lu3, Sumei Zhang1\* and Jin Wang2\**

*1 Laboratory of Molecular Biology and Department of Biochemistry, Anhui Medical University, Hefei, China, 2 Shanghai Public Health Clinical Center, Fudan University, Shanghai, China, 3 Department of General Surgery, Zhongshan Hospital, Fudan University, Shanghai, China, 4 Department of Molecular Oncology, Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences, Okayama, Japan*

## *Edited by:*

*Yujing Li, Emory University, United States*

#### *Reviewed by:*

*Graziella Curtale, The Scripps Research Institute, United States Feng Wang, Emory University School of Medicine, United States*

#### *\*Correspondence:*

*Sumei Zhang 379236778@qq.com Jin Wang wjncityu@yahoo.com*

*†These authors have contributed equally to this work*

#### *Specialty section:*

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

*Received: 13 December 2018 Accepted: 09 August 2019 Published: 10 September 2019*

#### *Citation:*

*Jia N, Tong H, Zhang Y, Katayama H, Wang Y, Lu W, Zhang S and Wang J (2019) CeRNA Expression Profiling Identifies KIT-Related circRNA-miRNA-mRNA Networks in Gastrointestinal Stromal Tumour. Front. Genet. 10:825. doi: 10.3389/fgene.2019.00825*

Gastrointestinal stromal tumours (GISTs) are the most common human sarcomas and are typically located in the stomach or small intestine. Although circular RNAs (circRNAs) reportedly play vital roles in tumour oncogenesis and progression, the molecular basis of the aggressive tumour biology of these circRNAs in GISTs remains unclear. In this study, we applied SBC ceRNA microarrays to screen for tumour-specific circRNA profiles in GISTs and identified that a total of 5,770 circRNAs and 1,815 mRNAs were differentially expressed in GISTs. Three significantly differential circRNAs (circ\_0069765, circ\_0084097, and circ\_0079471) and their host genes (KIT, PLAT, and ETV1) were also verified in 68 pairs of GISTs and adjacent normal gastrointestinal tissues by qRT-PCR. A GIST-specific circRNA-miRNA-mRNA regulatory network analysis demonstrated that the specific KITrelated regulatory networks involved the three circRNAs, the circRNA host genes and three miRNAs (miR-142-5p, miR-144-3p and miR-485-3p), which may be key regulators of GISTs that could serve as molecular biomarkers and potential therapeutic targets for this malignant disease.

Keywords: circRNAs, KIT, PLAT, ETV1, regulatory networks analysis

### INTRODUCTION

As one of the most common non-epithelial neoplasms, gastrointestinal stromal tumours (GISTs) are located in the stomach (55.6%), small intestine (31.8%), colon and rectum (6.0%), and oesophagus and abdominal cavity (6.2%) and have various clinical features ranging from asymptomatic to nonspecific abdominal discomfort, pain, haemorrhage and tumour abdominal (Joensuu et al., 2012); the incidence of GISTs is 10-15 cases per million per year in 19 countries (Soreide et al., 2016). It is not necessary for GIST patients to exhibit liver metastasis or intraperitoneal dissemination to perform an assessment of the tumour risk. However, clinicopathological factors, including the tumour size, mitotic count and anatomical location, were associated with cancer patient survival (Fletcher et al., 2002; Markku Miettinen and Jerzy Lasota, 2006; Joensuu, 2008), and complete surgical resection remains the primary treatment method for localized GISTs (Ho and Blanke, 2011). GISTs can be characterized by the expression of CD117 or PDGFRA protein in neoplastic cells, and a gain-of-function mutation may exist in

**20**

the type III receptor tyrosine kinase gene (c-KIT or PDGFR-α) at the genetic level (Hirota et al., 1998; Heinrich et al., 2003b). KIT is a receptor tyrosine kinase that is upregulated in the interstitial cells of Cajal, which are pacemakers responsible for digestive movement (Chi et al., 2010). KIT mutations frequently occur in exons 9, 11, 13 and 17 (Heinrich et al., 2003a; Corless et al., 2004) and play a vital role in GIST pathogenesis (Mazur and Clark, 1983; Hirota et al., 1998). In addition, a PDGFR-α mutation, which is present in 1/3 of KIT-negative GIST patients, mainly affects exons 12, 14 and 18 and can induce tyrosine kinase autophosphorylation, activate signalling molecules containing SH2 domains, and initiate various cancer-related pathways (Wozniak et al., 2012).

Additionally, deregulated circular RNAs (circRNAs) were investigated in acute myeloid leukaemia, breast cancer, gastric cancer, and prostate cancer (Patop and Kadener, 2018; Xia et al., 2018) and found to be involved in a variety of tumour-specific progression events, such as proliferation, invasion and metastasis (Li F et al., 2015; Li J et al., 2015; Wilusz, 2017; Yang et al., 2018; Patop and Kadener, 2018). These deregulated circRNAs exhibit cell- or tissue-specific expression, exist in a steady state on tissues, in the cellular nucleus and cytosol, on extracellular exosomes, and in body fluid and may serve as potential biomarkers of cancer (Gao and Zhao, 2018). Several deregulated circRNAs have been reported to contribute to promoting cell proliferation, such as circPVT1 in gastric tumours, circABCB10 in breast tumours and circBANP in colon tumours (Patop and Kadener, 2018). The downregulation of circITCH was also observed in bladder carcinoma, oesophageal squamous cell carcinoma, lung cancer, colon and rectal cancer and hepatocellular carcinoma (Patop and Kadener, 2018). circRNAs, which have a head-totail connected ring structure of exon or intron cyclization, are generated from pre-mRNAs (Wilusz, 2018) and play a sponge role by absorbing microRNAs for binding to the miRNAs of target genes, which could be indirectly influenced by circRNAs forming competing endogenous RNA (ceRNA) networks with circRNAs (Kim et al., 2009). The overexpression of circITCH passively modulated diverse tumour cellular processes by binding miR-17 via specific miRNA-binding sites, which had evident effects on the aggressive biological behaviours mediated by the circITCH/miR-17, miR-224/p21, and PTEN axis (Yang et al., 2018). We previously revealed that the differentially expressed circRNAs (circ\_0062019 and circ\_0057558) and the host gene SLC19A1 of circ\_0062019 could be used as potential novel biomarkers of prostate cancer (Xia et al., 2018). However, to note, no altered circRNAs have been reported in GISTs, and we still lack adequate in-depth knowledge about the biological function of circRNAs in GISTs.

In this study, we first analysed the ceRNA expression profile in gastrointestinal stromal tumour using high-throughput circRNA gene microarray and verified the differential circRNAs in GISTs by qRT-PCR. Our exploration of the circRNA-miRNA-mRNA network could help by adding a new dimension to the study of the molecular mechanisms of GISTs and provide new directions for GIST diagnosis and treatment.

### MATERIALS AND METHODS

#### Patients and Specimens

This study included patients with GIST who underwent partial or complete resection between Sept 2012 and Oct 2017 at Shanghai Public Health Clinical Center, Fudan University, China. The study was approved by the Medical Ethics Commission of Shanghai Public Health Clinical Center. All patients had understood all aspects of the informed consent and signed the informed consent forms before undergoing surgeries. During the operation, 68 pairs of GIST and adjacent normal gastrointestinal tissue samples were collected from obvious lesions and the corresponding gastric or intestinal tissues, which were 1–3 centimetre distant from the tumour edge and contained no obvious cancer cells. After removal from the body, the fresh samples were rapidly intensively chilled in liquid nitrogen within 5 min of excision to avoid degradation. Then, the frozen specimens were stored in a −80°C refrigerator. All enrolled patients were diagnosed for the first time through a pathological examination before undergoing surgical resection. The definitive diagnosis of all cases required tissue biopsy, which relied on endoscopic ultrasound-guided fine-needle aspiration. The tumour histological grading were based on malignancy risk stratification of the gastrointestinal stromal cell tumours by tumour size, mitotic count, and location (Markku Miettinen and Jerzy Lasota, 2006).

### Cell Line, Plasmid and Cell Transfection

The human gastrointestinal stromal tumour cell lines GIST-T1 and GIST-882 were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). The GIST-T1 cells were cultured in Mcoy5A's medium, and the GIST-882 cells were cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% (v/v) foetal bovine serum (FBS) (HyClone, Logan, UT, USA) under the culture conditions of 37°C and 5% CO2. A circ\_0084097 and an NC control pLCDH-ciR empty vector were synthesized by Geneseed Biotech Co. Ltd. (Guangzhou, China) and transfected into the GIST-T1 cells by using Lipofectamine 2000 reagent (Life Technologies Corporation, Carlsbad, CA, USA) following the manufacturer's protocol. The transfection efficiency was assessed using qRT-PCR.

#### RNA Purification and SBC ceRNA Microarrays

The total RNA was isolated and purified with TRIzol reagent (Invitrogen, Carlsbad, CA, USA) and a TIANGEN total RNA Isolation Kit (TIANGEN, Beijing, China) according to the manufacturer's protocol. The isolated RNAs were stored at −80°C. The RNA was qualified, and the RNA integrity number was determined by an Agilent 2100 bioanalyser, while the RNA concentration was analysed using a NanoDrop-2000 spectrophotometer (NanoDrop, USA). For the ceRNA microarray, the included RNA samples were obtained from 3 pairs of GIST and adjacent normal gastrointestinal tissue samples. cRNA was synthesized and amplified with an Agilent Low Input Quick Amp WT Labeling Kit (Santa Clara, CA, US) and can be labelled by cyanine 3-labelled CTP with T7 RNA polymerase. The labelled cRNA was purified by an RNeasy mini kit (Qiagen, USA) and loaded onto SBC Human (4\*180K) ceRNA microarrays including 88,371 circRNAs and 18,853 mRNAs (Shanghai Biotechnology corporation, Shanghai, China). The signals were scanned by an Agilent G2565CA Microarray Scanner. The raw data were obtained by Agilent Feature Extraction (v10.7). After normalization of the raw data with R software, the differentially expressed mRNAs and circRNAs were filtrated according to the fold change and Student t-test. The normalized signal value is the value calculated by log2. All ceRNAs with a fold change (FC) ≥ ± 2, a p-value < 0.05 and intensity > 7.0 were included for further statistical analysis. The complete ceRNA array datasets were deposited in the Gene Expression Omnibus (GEO) database under accession number GSE131481.

#### Regulatory Network and Pathway Analysis of the Differential mRNAs and the Host Genes of the Differential circRNAs in GISTs

To further investigate the functions of these differential mRNAs in GISTs, the functions of the differential genes were annotated with GO and KEGG pathway analyses (Xia et al., 2018). CircInteractome (https://circinteractome.nia.nih.gov/) was used to predict the putative miRNAs of the three circRNAs and the potential circRNA/miRNA interaction (Dudekula et al., 2016). Targetscan7.2 (https://circinteractome.org./ vert\_72) was used to predict the targeted miRNAs of the three host genes. We overlapped the two predicted results. Finally, we selected the top miRNAs with the highest context scores (score >85) to establish a circRNA-miRNA-host gene network, which was illustrated by Cytoscape3.5.

#### Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR) Analysis of the Differentially Expressed circRNAs and Their Host Genes in GISTs

In total, 3 circRNAs were chosen for experimental validation by qRT-PCR. As an exoribonuclease, RNase R can only act on RNA from its 3' to 5' end but does not degrade circRNA (Suzuki et al., 2006). Therefore, to distinguish the expression between the linear mRNA and circRNA, total RNAs were incubated for 20 min at 37°C with or without RNase R (Epicentre Technologies, Madison, WI), and the resulting RNAs were purified using an RNAsimple Total RNA Kit (Tiangen, Beijing, China) and transcribed into cDNA. The cDNAs were synthesized with reverse transcriptase using a PrimeScriptTM RT reagent Kit with gDNA Eraser (TaKaRa). The PCR comprised 50 ng cDNA, 10 μl of 2 x PCR Master mix (SYBR Premix Ex TaqTM II kit) (TaKaRa), 0.8 μl primer forward (10 μM), 0.8 μl primer reverse (10 μM), and 0.4 μl of ROX reference Dye and was performed on an ABI ViiA 7 (Applied Biosystems, DE, USA) as follows: denaturation at 95°C for 10

min, amplification at 95°C for 15 s over 40 cycles, followed by annealing and extension at 60°C for 1 min. The results of the relative expression levels were obtained by calculating the raw data using the 2-ΔΔCt method. 18S rRNA served as an internal control for the normalization. The numbers of exons and exact sequences of circ\_0084097 produced from PLAT, circ\_0069765 from KIT, and circ\_0079471 from ETV1 were validated by Sanger sequencing. All the primers for circ\_0084097, circ\_0069765, and circ\_0079471 were designed by Shanghai Biotechnology corporation and shown in **Tables S1** and **S2**.

#### Statistical Analysis

To compare the GIST and adjacent normal gastrointestinal tissue samples, the significance of the relative quantification validation was conducted by Student t-test for the paired analysis. All tests were 2-sided, and p < 0.05 was regarded as statistical significance. The data were analysed with Statistical Program for Social Sciences (SPSS) 16.0 software (SPSS, Chicago, IL, USA).

### RESULTS

#### Differentially Expressed mRNAs and circRNAs in GISTs

The characteristics of the GIST patient population and the clinical details of the three samples from the GIST patients chosen for the SBC ceRNA arrays are shown in **Table S3**. The ceRNA arrays were performed to investigate the differentially expressed mRNAs and circRNAs in GISTs. Volcano plots were used to present the significant differences in the extracted data between the GIST and adjacent normal gastrointestinal tissue samples and show the expressed difference in mRNAs (**Figure S1A**) and circRNAs (**Figure S1B**) between the GIST and adjacent tissues. Based on the differences in their expression levels, hierarchical clustering showed the differentially expressed mRNA (**Figure 1A**) and circRNA (**Figure 1B**) expression profile among 3 pairs of GIST and adjacent normal gastrointestinal tissue samples. In total, 1,815 mRNAs (839 upregulated mRNAs and 976 downregulated mRNAs) (**Table 1**) and 5,770 circRNAs (3,122 upregulated circRNAs and 2,648 downregulated circRNAs) (**Table 2**) were differentially expressed between the GIST and adjacent normal gastrointestinal tissue samples (p < 0.05 and FC ≥ ± 2). After screening the differentially expressed mRNAs by retrieving the GEO database (GSE112) and utilizing GEO2R in analysing the array data (**Table S4**), Venn diagrams were generated to show the 387 common differentially expressed genes (DEGs) selected in our array and GEO dataset GSE112 (**Figure 1C**). Finally, 95 DEGs were also identified as the host genes of DEcircRNAs in GISTs. In total, 54 circRNA host genes were upregulated, and 41 DEcircRNA host genes were downregulated in the GIST tumour tissues from these three GISTs patients, which was consistent with the expression level of the circRNAs (p < 0.05 and FC ≥ ± 2) (**Figure 1D** and **Table 3**).

FIGURE 1 | Heatmaps and Venn Diagrams showing the differential mRNAs, circRNAs and their host genes in GISTs. Heat maps of the differentially expressed mRNAs (A) and circRNAs (B). Venn Diagrams showing that the common 387 mRNAs (C) were from differential mRNAs in the GEO dataset (GSE112) and our ceRNA array, and 95 common mRNAs were from overlapped 387 genes and differently expressed circRNA host genes in our ceRNA array (D).

### Functional Pathway Analysis of Differential mRNAs and circRNA Host Genes in GISTs

Subsequently, a Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis of these differentially expressed mRNAs and the host genes of the differential circRNAs in GISTs was performed to determine the top 5 pathways of the differential mRNAs, which included Biosynthesis of unsaturated fatty acids, Vitamin B6 metabolism, Notch signalling pathway, Dilated cardiomyopathy, ABC transporters and Hypertrophic cardiomyopathy (HCM) (**Figure 2A**); several circRNA host genes were enriched in the pathways of One carbon pool by folate, D-Glutamine and D-glutamate metabolism, ECMreceptor interaction, Adherens junction, and Nicotinate and nicotinamide metabolism (**Figure 2B**). Moreover, several common pathways involved the differentially expressed mRNAs and host genes of differential circRNAs in GISTs, including vascular smooth muscle contraction, Notch signalling pathway, nicotinate and nicotinamide metabolism, N-Glycan biosynthesis, hypertrophic cardiomyopathy (HCM), focal adhesion ECM-receptor interaction, Dilated cardiomyopathy, axon guidance and Arrhythmogenic right ventricular cardiomyopathy (ARVC).

#### KIT-Related circRNA-miRNA-mRNA Regulatory Network Analysis in GISTs

Among ceRNA expression profiling in GISTs, we found three circRNAs (circ\_0069765, cir\_0084097, and circ\_0079471) and their host genes (KIT, PLAT, and ETV1) were up-regulated in GISTs. The molecular analysis of KIT becomes one of the two gold standards of diagnosis in GISTs. Mutation in the KIT gene is the key oncogenic drivers in the majority of GISTs (Wu et al., 2019), which is also potentiated by a positive feedback circuit that involves the ETS transcription factor ETV1 (Duensing, 2015; Wu et al., 2019). Besides, PLAT (Tissue-Type Plasminogen Activator) as a node with VEGFC, PGF and CHD7 in the functional networks was also verified to be significantly enriched in blood vessel development involved in the tissue specificity of GISTs (Ma et al., 2018), which pushed us to analyze KIT related circRNA-miRNA-mRNA regulatory network in GISTs. Thus, three circRNAs derived from above


#### TABLE 2 | Partially differentially expressed circRNAs in GIST.



parental genes were selected for further investigation although there were some top change circRNAs in **Table 2**. circ\_0069765, which is located on chr4 q12 (chr4:55569889-55603446), is derived from a non-coding regulatory region of KIT (**Figure S2A**). circ\_0079471, which is located on chr7 p21.2 (chr7:13949257-13975521), is a regulatory circRNA within a long non-coding region of ETV1 (**Figure S2B**). However, circ\_0084097 stems from a non-coding regulatory region contained a promoter blank adjacent to the promoter region of PLAT, which is located on chr8 p12 (chr8:42046451-42050729) (**Figure S2C**). Based on the miRNA site prediction, we predicted the targeted miRNAs of the three differential circRNAs in circular RNA Interactomem (https://circinteractome.nia.nih. gov/) (Dudekula et al., 2016). To obtain insight into reciprocal circRNA, miRNA and mRNA regulation, we constructed a regulatory circRNA-miRNA-mRNA network using Cytoscape software and clarified the interaction among the three circRNAs (circ\_0069765, circ\_0084097, and circ\_0079471), their host genes (KIT, PLAT, and ETV1) and seven predicted miRNAs (miR-144-3p, miR-1246, miR-485-3p, miR-142-3p, miR-142-5p, miR-326 and miR-324-5p), which is shown in **Figure 3**. In the figure, the upregulated circRNAs and their host genes are marked in red, and the downregulated miRNAs that had been reported in previous studies investigating cancer tissues are marked in green. Evidently, miR-144-3p, and miR-485-3p are common target miRNAs of all three host genes (KIT, PLAT, and ETV1), and miR-142-5p is a targeted miRNA of KIT and PLAT. We also found that miR-1246 was predicted as the common targets of both circ\_0069765 and circ\_0084097 and their host genes (KIT and PLAT), and miR-326 was predicted as the common targets of both circ\_0069765 and circ\_0079471. Thus, the specific regulatory networks including the three circRNAs (circ\_0069765, cir\_0084097, and circ\_0079471), their host genes (KIT, PLAT, and ETV1) and the three miRNAs (miR-142-5p, miR-144-3p and miR-485-3p) may be key regulators in GISTs.



#### Differential circRNAs (circ\_0069765, circ\_0079471 and circ\_0084097) and Their Host Genes Were Verified in GISTs by qRT-PCR

The genomic structure shows that circ\_0069765 contains six exons from the KIT gene (**Figure S2A**), circ\_0079471 contains four exons from ETV1 gene (**Figure S2B**), and circ\_0084097 contains three exons from PLAT gene (**Figure S2C**). All the "head-to-tail" splicing sites of the three circRNAs are presented in **Figure S2**. The distinct products of these three circRNAs were amplified using outward-facing primers and confirmed by Sanger sequencing (**Figures S3A–C**). We found that circ\_0069765, circ\_0079471 and circ\_0084097 were resistant to RNase R, compared to the linear mRNAs (Data not shown). Next, we detected the expression level

of circ\_0069765, circ\_0079471, circ\_0084097 and their corresponding host genes by real-time PCR (qRT-PCR) analyses. The relative expression of the three circRNAs (circ\_0069765, circ\_0079471 and circ\_0084097) was evidently upregulated in the GIST tissues compared with that in the adjacent noncancerous tissues (p < 0.001); in addition, the three host genes, i.e., KIT, PLAT and ETV1, were upregulated (p < 0.001) (**Figure 4**). The qRT-PCR analyses revealed that 44 of 66 (66.67%) tumours had increased circ\_0069765 (4.68 fold); 60 of 65 (92.30%) tumours had increased host gene KIT mRNA (1404.20-fold) expression; 63 of 68 (92.65%) tumours had increased cir\_0084097 expression (156.86-fold); 61 of 68 (89.71%) tumours had increased host gene PLAT mRNA (462.43-fold) expression; 59 of 68 (86.76%) tumours had increased circ\_0079471 (118.10-fold) expression; and 62 of

FIGURE 2 | Functional pathway analysis of targeted genes of predicted miRNAs and competitive and endogenous regulatory network. GO analysis of targeted genes (A), and KEGG analysis of targeted genes (B).

66 (93.94%) tumours had increased host gene ETV1 mRNA (678.60-fold) expression. These findings were consistent with the tissue microarray data and showed the significant upregulation tendency of the three circRNAs and three host genes. Finally, we identified that markable positive correlations were present between PLAT and three verified circRNAs (p < 0.05) (**Table S5**). We also noted a non-negative correlation between two circRNAs and ETV1 (**Table S5**, \*p < 0.05). Interestingly, an obvious correlation was observed not only between the genes ETV1 and PLAT (p < 0.001) but also between the circRNAs circ\_0069765 and circ\_0079471 and between circ\_0079471 and circ\_0084097 (p < 0.05) (**Table S5**). To clarify the characteristics of these differential circRNAs and their host genes in GIST cancer, a Pearson correlation analysis was applied to analyse the correlation between these circRNAs/their host genes and the corresponding clinical parameters. As shown in **Table 4**, circ\_0084097 and its host gene PLAT are negatively correlated with metastasis of tumours significantly related to the stomach (p < 0.05). PLAT was also negatively correlated with the tumour diameter (p < 0.05) (**Table 4**), indicating that circ\_0084097 and PLAT may be related to the early stage of stomach stromal tumour.

FIGURE 4 | qRT-PCR analysis of the gene expression levels of the three differentially expressed circRNAs and their host genes in GISTs. (A) KIT; (B) PLAT; (C) ETV1; (D) circ\_0069765; (E) circ\_0084097; and circ\_0079471(F).


*Bold values denote statistical significance at the p < 0.05 level.*

### DISCUSSION

In this study, the ceRNA expression profile showed that the mRNA and circRNA expression profile in the gastric stromal tumour tissues was distinguished from that in matched tissues adjacent to the tumour and found that a total of 3,122 circRNAs were significantly

upregulated and 2,648 were significantly downregulated in the tumour tissues. More importantly, 95 differentially expressed genes had been filtered by overlapping circRNA host genes and significant mRNAs of GSE112. We found several common pathways involving the differential mRNAs and the host genes of differential circRNAs in GISTs, including vascular smooth muscle contraction, Notch signalling pathway, nicotinate and nicotinamide metabolism, N-Glycan biosynthesis, Hypertrophic cardiomyopathy (HCM), Focal adhesion, ECM-receptor interaction, Dilated cardiomyopathy, Axon guidance, and Arrhythmogenic right ventricular cardiomyopathy (ARVC) (**Figure 2**). Three molecular inhibitors of the Wnt signalling pathway have been reported to be tumour suppressors in various in vitro and in vivo GIST models harbouring a KIT mutation. The Wnt antagonist DKK4 was apparently downregulated in advanced human GISTs (Zeng et al., 2017). The Notch signalling pathway has also been reported to be a tumour suppressor in GIST cells harbouring a KIT mutation. The downstream target of notch (dominant-negative Hes1) was apparently upregulated in GIST patients with longer relapse-free survival (Yang et al., 2018). In addition, the focal adhesion signalling pathway played a critical role in the proliferation of both imatinib-sensitive and resistant GIST cells (Zeng et al., 2017). We demonstrated that the Wnt, Notch and Focal adhesion signalling pathways are associated with GIST cell proliferation.

Notably, 95 genes were not only differentially expressed linear RNAs but also maternal genes that generated various differentially expressed circular RNAs in our study. circ\_0069765, circ\_0079471 and circ\_0084097 were selected for the validation of the array results,

and we detected the expression of these circRNAs in 68 pairs of tissue samples and showed that the three circRNAs were significantly upregulated in tumour tissues, while their host genes KIT, PLAT and ETV1 had a similar rising trend in expression. Furthermore, the expression levels of these three circRNAs and their host genes were also checked in GIST cell lines. We only found that circ\_0069765 was significantly upregulated in the GIST-T1 and GIST-882 cells and that circ\_0079471 and its host gene ETV1 were overexpressed in the GIST-T1 cells compared to the normal stomach stromal tissue by a qRT-PCR analysis (all p < 0.05) (**Figure S4**).

Web tools for miRNA target-site prediction for circRNA that have a sequence-based recognition system come with the context scores which have the advantage of being predictive for all types of interactions. There is not standard score for selecting top miRNAs. We selected top miRNAs with the high context score (score > 85) for the three differential circRNAs to establish a circRNA-miRNA-host gene network in GIST (**Table S6**) and found that miR-142-3p, miR-142-5p, miR-149, miR-324-5p, miR-326, miR-485-3p and miR-1246 might interact with circ\_0069765, circ\_0079471 and circ\_0084097. Interestingly, miR-1246, miR-142-5p, and miR-324-5p were downregulated in the GIST cells (GIST-882 and GIST-T1)

FIGURE 5 | Gene expression levels of miR-1246, miR-142-5p, and miR-324-5p in GIST-T1 and GIST-882 cells (A–C) (A) miR-1246; (B) miR-142-5p; (C) miR-324-5p and their expression in GIST-T1 with circ\_0079471 by overexpression of circ\_0084097 (D–H) (D) circ\_0084097; (E) circ\_0079471; (F) miR-1246; (G) miR-142-5p; (H) miR-324-5p were analysed by qRT-PCR.

compared to the normal stomach stromal tissue in the qRT-PCR analysis (**Figures 5A**–**C**). In the analysis of the function of the ceRNAs and their interaction, we confirmed that these three miRNAs were also repressed and that circ\_0079471 was upregulated in GIST-T1 cells by overexpression of circ\_0084097 (**Figures 5D**–**H**), which was consistent with our circRNA-miRNA regulatory network analysis in GISTs (**Figure 4**). Thus, these miRNAs may be linked to several host genes, including KIT, PLAT, and ETV1. In GISTs, a KIT proximal domain mutation, especially in exon 11, can induce ligandindependent kinase phosphorylation and activate downstream signal transduction pathway, including AKT, MAPK and STAT (Corless et al., 2011). The molecular targeted agent, Imatinib, blocks KIT / PDGFRA signalling by binding the ATP-binding pocket required for phosphorylation and activation of the receptor. The application of imatinib had changed from a single drug model to a combination with surgical treatment, which was essential to complete surgical resection, alleviate the disease, prolong survival and improve the quality of life, especially among postoperative patients (Huang et al., 2016). Unfortunately, initially sensitive tumours acquired imatinib resistance due to a KIT secondary mutation. Sunitinib and regorafenib are two additional multikinase inhibitors approved as second- and third-line therapies, respectively, and are available for the treatment of imatinib-resistance GIST (Demetri et al., 2006; Demetri et al., 2013). It has been found that non-small cell lung cancer tumourigenesis was suppressed by the overexpression of miR142-5p, which also regulated tumour cell PD-L1 expression and enhanced anti-tumour immunity in pancreatic cancer (Jia et al., 2017; Wang et al., 2017a). The downregulation of miR-142-5p was significantly associated with the recurrence and poor prognosis of gastric cancer (GC) and promoted tumour metastasis by regulating CYR61 expression (Yan et al., 2019). miR-144-3p was significantly downregulated in hepatocellular carcinoma, glioblastoma, multiple myeloma and pancreatic cancer and inhibited proliferation, migration and tumour metastasis by targeting SGK3 (Wu et al., 2017), FZD7 (Cheng et al., 2017), c-Met (Zhao et al., 2017) and FOSB (Liu et al., 2018). The repression of miR-485-3p was also found in breast cancer. The overexpression of miR-485-3p can inhibit mitochondrial respiration and breast cancer cell metastasis by inhibiting PGC-1α expression (Lou et al., 2016). Low serum levels of miR-485-3p were related to poor survival in patients with glioblastoma (Wang et al., 2017b). The miR-324-5p-mediated suppression of NF-κB activation was reported to be responsible for inhibition breast cancer cell invasion and migration (Song et al., 2015). The expression of miR-1246 was downregulated in lung cancer cell lines and cervical cancer tissue, was negatively correlated with the clinical stage and inhibited cell invasion and the EMT by targeting CXCR4 (Yang et al., 2015; Xu et al., 2018). miR-149 was downregulated in ovarian cancer, colorectal cancer and lung cancer. The overexpression of miR-149 increased the drug sensitivity of cancer cells and inhibited the EMT through the FOXM1/cyclin D1/MMP2 axis (Ke et al., 2013; Xu et al., 2015; Sun et al., 2018). Thus, the decreased expression and functional inhibition of these miRNAs in cancer further

support our hypothesis that circ\_0069765, circ\_0079471 and circ\_0084097 function to regulate the more comprehensive circRNAs-miRNAs-genes network.

In summary, the present research revealed the ceRNA expression profiles in GISTs and identified that three circRNAs (circ\_0069765, circ\_0079471 and circ\_0084097) and three host genes (KIT, ETV1 and PLAT) were upregulated in GISTs using qRT-PCR. We further demonstrated that the specific regulatory networks including three circRNAs (circ\_0069765, cir\_0084097, and circ\_0079471), their host genes (KIT, PLAT, and ETV1) and three miRNAs (miR-142-5p, miR-144-3p and miR-485-3p) may be key regulators in GISTs and are likely involved in tumour oncogenesis and progression. In future investigations, it is worth considering the verification of the molecular mechanism of these specific circRNAs to regulate GIST occurrence and development. A greater understanding of the mechanisms of the involvement of specific circRNAs in GIST tumour malignancy is necessary for the identification of possible therapeutic targets.

#### DATA AVAILABILITY

The datasets generated for this study can be found at NCBI using accession number GSE131481 (https://www.ncbi.nlm.nih.gov/ geo/query/acc.cgi?acc=GSE131481).

#### ETHICS STATEMENT

The study included patients with GIST who underwent partial or complete resection at Shanghai Public Health Clinical Center, Fudan University, China between Sept 2016 and Oct 2017. The study was approved by the Medical Ethics Commission of Shanghai Public Health Clinical Center.

#### AUTHOR CONTRIBUTIONS

JW and HT contributed to the conception; NJ, HT, YZ, HK, YW, WL, SZ, and JW analyzed the data; NJ and JW wrote the manuscript; and JW revised the manuscript.

#### FUNDING

This research was supported by a grant from the National Natural Science Foundation of China (81672383), the National Special Research Program of China for Important Infectious Diseases (2018ZX10302103-003). The grant (KY-GW-2017-09) (HT) was from Shanghai Public Health Clinical Center, Shanghai, China.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene.2019.00825/ full#supplementary-material

## REFERENCES


forkhead box transcription factor FOXM1. *Cell Physiol. Biochem.* 35, 499–515. doi: 10.1159/000369715


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

*Copyright © 2019 Jia, Tong, Zhang, Katayama, Wang, Lu, Zhang 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) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The Role of Non-Coding RNAs in Neurodevelopmental Disorders

*Shuang-Feng Zhang1,2,3, Jun Gao4 and Chang-Mei Liu1,2,5\**

1 State Key Laboratory of Stem Cell and Reproductive Biology, Institute of Zoology, Chinese Academy of Sciences, Beijing, China, 2 Savaid Medical School, University of Chinese Academy of Sciences, Beijing, China, 3 School of Life Sciences, University of Science and Technology of China, Hefei, China, 4 Department of Neurosurgery, Peking Union Medical College Hospital, Chinese Academy of Medicine Sciences & Peking Union Medical College, Beijing, China, 5 Institute for Stem Cell and Regeneration, Chinese Academy of Sciences, Beijing, China

Non-coding RNAs, a group of ribonucleic acids that are ubiquitous in the body and do not encode proteins, emerge as important regulatory factors in almost all biological processes in the brain. Extensive studies have suggested the involvement of non-coding RNAs in brain development and neurodevelopmental disorders, and dysregulation of non-coding RNAs is associated with abnormal brain development and the etiology of neurodevelopmental disorders. Here we provide an overview of the roles and working mechanisms of non-coding RNAs, and discuss potential clinical applications of noncoding RNAs as diagnostic and prognostic markers and as therapeutic targets in neurodevelopmental disorders.

#### Edited by:

Yadong Zheng, Lanzhou Institute of Veterinary Research (CAAS), China

#### Reviewed by:

Saijilafu, First Affiliated Hospital of Soochow University, China Shuguang Yang, Institute of Basic Medical Sciences, China

> \*Correspondence: Chang-Mei Liu, liuchm@ioz.ac.cn

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 09 December 2018 Accepted: 25 September 2019 Published: 20 November 2019

#### Citation:

Zhang S-F, Gao J and Liu C-M (2019) The Role of Non-Coding RNAs in Neurodevelopmental Disorders. Front. Genet. 10:1033. doi: 10.3389/fgene.2019.01033

Keywords: non-coding RNA, neurodevelopmental disorder, miRNA, piRNA, snoRNA, lncRNA

## INTRODUCTION

Non-coding RNAs are RNA molecules that are not translated into proteins. Recent advances in genomic sequencing technologies and functional assays enable a more in-depth understanding of their characteristics (Mattick, 2011; Djebali et al., 2012; Obiols-Guardia and Guil, 2017). The transcription process of non-coding RNAs is precisely orchestrated in time and space (Okazaki Y., et al., 2002; Djebali et al., 2012). Different developmental stages or tissue types have distinct transcriptional landscapes (Carninci et al., 2005; Kapranov et al., 2007). The central nervous system is a sophisticated and precise system which is responsible for guiding our daily activities such as sports, learning, emotion and language. Rapidly growing evidence indicated that non-coding RNAs play indispensable roles in brain development, function, and the etiology of neurodevelopmental diseases.

Here, we review the diversity and biogenesis processes of non-coding RNAs, and summarize their versatile roles in neurodevelopmental disorders. We also discuss potential clinical applications of non-coding RNAs as diagnostic and prognostic markers and as therapeutic targets in neurodevelopmental disorders.

## CHARACTERISTICS OF NON-CODING RNAS

Abundant and functionally important types of non-coding RNAs include ribosomal RNAs (rRNAs) and transfer RNAs (tRNAs), as well as regulatory non-coding RNAs which mainly consist of microRNA (miRNA), PIWI-interacting RNA (piRNA), small nucleolar RNA (snoRNA), small interfering RNAs (siRNAs), long non-coding RNA (lncRNA), and Circular RNAs (CircRNAs).

1 **32** Non-coding RNAs play a critical role in epigenetics regulation of gene expression in addition to their roles at the transcriptional and post-transcriptional level (David and Bartel1, 2004; Alexander et al., 2010).

miRNAs are small single-stranded molecules (20–24 nt) that have seed sequences complementary to sequences on target mRNAs transcripts through the 3′UTR, leading to silencing of the target gene. The miRNA gene is transcribed by RNA polymerases II and III to generate a primary microRNA precursor molecule (pri-miRNA). The pri-miRNA then undergoes nuclear cleavage by Drosha/DGCR8 to form a precursor microRNA (pre-miRNA). The pre-miRNA is transported from the nucleus into the cytoplasm by Exportin 5, and then processed by Dicer/TRBP into a miRNA duplex which is unwound by a helicase. The mature miRNA is incorporated into the RNA-induced silencing complex (Risch et al.) which mediates down-regulation of gene expression by either translational repression or mRNA degradation (David and Bartel1; 2004; Esteller, 2011; Rajman and Schratt, 2017) (**Figure 1**).

piRNAs are small non-coding RNAs (24-31 nt) that can silence transposons and regulate gene expression by directing PIWI proteins of Argonaute family to specific genomic loci (Fabio Mohn and Brennecke, 2015) (**Figure 1**). piRNAs biogenesis is divided into primary and secondary pathways (Zamore et al., 2018). In primary pathway, primary piRNAs are transcribed from genomic loci called piRNA clusters (Brennecke et al., 2007; Li et al., 2013). Primary piRNAs are spliced by endonuclease into tail-to-head phased precursor piRNA that are catalyzed by the mitochondrial protein Zucchini/PLD6 (Ipsaro et al., 2012; Nishimasu et al., 2012; Fabio Mohn and Brennecke, 2015). Each pre-piRNA begins with a 5′ monophosphate, a prerequisite for loading RNA into nearly all Argonaute proteins (Schirle et al., 2014; Wang et al., 2014). Once the PIWI protein captures prepiRNA, the 3′ terminal is trimmed by a single-stranded RNA exonuclease called Trimmer/PNLDC1 to the appropriate size (Kawaoka et al., 2011; Tang et al., 2016; Ding et al., 2017).

transcription by targeting specific loci. (E) CircRNAs are able to repress miRNA expressions.

Finally, the small RNA methyltransferase Hen1/HENMT1 adds a 2-O-methyl moiety to the 3′ ends of the mature piRNAs (Horwich et al., 2007; Lim et al., 2015). The secondary piRNA biogenesis pathway, also known as the "ping-pong" cycle, is a piRNA-directed piRNA synthesis pathway that produces a piRNA *via* interaction with PIWI proteins (Brennecke et al., 2007; Gunawardane et al., 2007; Wang et al., 2014) (**Figure 1**). So far, functions of piRNA are mainly concentrated in regulation of genomic stability *via* silencing transposon (**Figure 1**).

snoRNAs are 60–300 nt nucleotide long, metabolically stable RNAs, which are usually concentrated in Cajal bodies or nucleoli (Ding et al., 2008). snoRNAs are produced by the transcription of RDR2. Compared with miRNA, snoRNA transcription events occur only in the nucleus (Vaucheret, 2006). Initial, transcripts of snoRNA enter the cytoplasm for processing and modifying and subsequently return to the nucleus. snoRNA can catalyze sequencespecific 2′-O-methylation and pseudouridine acidification of ribosomal RNA (rRNA) by forming protein complexes with splicing function (Kiss-László et al., 1996; Jingwei Ni et al., 1997). A new study has found that snoRNAs are indispensable for processing and stability of lncRNA (Xing and Chen, 2018).

lncRNA are a family of long-chain non-coding RNA that are usually longer than 200 nt, which regulate various developmental and physiological processes (Hu et al., 2016; Zhou et al., 2016; Lekka and Hall, 2018). Almost all lncRNAs transcripts do not contain open reading frames (Wilhelm et al., 2014), which are produced by RNA polymerase II, followed by capping and forming 3′polyadenylate tails (**Figure 1**) (Lekka and Hall, 2018). lncRNAs can interact with other epigenetic regulators to direct histone-modified enzymes or DNA-methylated enzymes to the specific gene loci and modulate gene expression (Tripathi et al., 2010; Gong and Maquat, 2011). For example, BDNF-AS can recruit EZH2 and PRC2 complex to the promoter region of BDNF to down-regulate the expression of BDNF (Modarresi et al.). In addition, lncRNAs might act as enhancers to activate

gene transcription (Orom et al., 2010), and target miRNAs to silence their inhibitory functions (Cesana et al., 2011; Lai et al., 2013). lncRNAs can also interact with proteins to modulate gene expression in different levels or direct their appropriate spatial subcellular localization (Chu et al., 2015; Lekka and Hall, 2018).

Circular RNAs (CircRNAs) are recently emerged as a new class of endogenous noncoding RNAs (ncRNAs) which might regulate gene expression (Memczak et al., 2013; You et al., 2015). These RNAs are usually processed into loops after transcription(Piwecka et al., 2017). Large-scale sequencing and analysis results have demonstrated that thousands of circular RNAs are present in mammalian and nematodes (Memczak et al., 2013; Liang Chen 2015). The published data also have indicated that CircRNAs are highly abundant in mammalian brain compared to other analyzed tissues (Memczak et al., 2013). What's more, the majority of detected mouse CircRNAs are also expressed as CircRNAs in human brain, which suggests that CircRNAs are very conserved between species (Liang Chen 2015). Identically, transcription of circular RNA is orchestrated by developmental stage and tissue specificity(Hansen et al., 2013; Liang Chen 2015; Piwecka et al., 2017), which indicates that CircRNAs might serve as regulatory RNAs, especially in brain (**Figure 1**). Loss of a mammalian circular RNA locus causes miRNA deregulation and affects brain function (Hansen et al., 2013; Piwecka et al., 2017). For example, cerebellar degeneration-related protein 1 transcript (CDR1as) contains more than 70 selectively conserved miRNA target sites. CDR1as strongly suppresses miR-7 activity and results in increased levels of miR-7 targets (Hansen et al., 2013; Piwecka et al., 2017). Functionally, CDR1as and its interaction with miRNAs are important for sensorimotor gating and synaptic transmission (Piwecka et al., 2017). Using high resolution *in situ* hybridization technology, the researchers found that visualized CircRNAs punctate into the dendrites of neurons, and many CircRNAs change their abundance abruptly at a time corresponding to synaptogenesis. Together, these data indicate that CircRNAs play important roles in regulating synaptic function (You et al., 2015).

### NON-CODING RNA IN NEURODEVELOPMENTAL DISORDERS

Human brain displays the richest repertoire of ncRNA species, and where several different ncRNA molecules are known to be involved in crucial steps for neurodevelopment (Mehler and Mattick, 2006; Taft et al., 2010; Obiols-Guardia and Guil, 2017). Abnormal expression of non-coding RNA has been linked with pathologies of several neurodevelopmental diseases, including Autism spectrum disorders (ASD), Fragile X syndrome (FXS), Down syndrome (DS), Rett syndrome, and Prader-Willi Angelman syndrome (**Table 1**).

#### TABLE 1 | Non-coding RNAs are related to neurodevelopmental disorders.


### Autism Spectrum Disorders

ASD is a developmental disorder that affects communication and behavior, which is characterized by repetitive patterns of behavior, interests, or activities, problems in social interactions, and psychological problems in children (Balachandar et al., 2016). Children with ASD have co-occurring language problems, intellectual disabilities, and epilepsy at higher rates than the general population. While the exact cause of ASD has remained somewhat of a mystery, dozens of genes have been identified to potentially contribute to disease susceptibility (Edward et al., 1985; Bailey et al., 1995; Risch et al., 1999; Klei et al., 2012; Kang, 2014). For example, polymorphisms in the FMR1 gene have been reported to be associated with autism (Reddy, 2005), however, no consistent association between FMR1 polymorphisms and autism has been demonstrated. Therefore, ASD is probably not caused by one single genetic factor. Recent studies suggest that epigenetic mechanisms, such as non-coding RNAs, may play a major role in the pathogenesis of ASD (Constantin, 2017).

Abnormal expression levels of miRNAs, including miR-132, miR-23a, miR-93, miR-106b, miR-146b and miRNA-148b, were observed in the serum, lymphoblastoid cells, or cerebellar cortex of autistic patients (Abu-Elneel et al., 2008; Talebizadeh et al., 2008; Tewarit Sarachana1 2010; Mundalil Vasu et al., 2014) (**Figure 2**). Many autism susceptibility genes are predictive targets of these differently expressed miRNAs, which further strengthens the causal relationship between miRNAs and autism (Abu-Elneel et al., 2008; Talebizadeh et al., 2008; Tewarit Sarachana1, 2010; Constantin, 2017).

A profiling study of circulating serum miRNAs in children with ASD reveals that differentially expressed miRNAs in serum may be involved in the molecular pathway of ASD (Kichukova et al., 2017), and serum miR-197-5p, miR-328-3p, miR-424-5p, miR-619-5p, miR-500a-5p, miR-313-5a, miR-365a-3p and miR-664a-3p may be served as potential biomarkers of ASD (Kichukova et al., 2017) (**Table 1**). Moreover, the interaction between miRNAs and autism risk genes in the cerebellum was observed in animal models of ASD. For example, miR-188 is down-regulated in autism patient (Trindade et al., 2013) and is up-regulated in response to long-term potentiation (Lee et al., 2012). MiR-188 promotes dendritic spine formation by blocking the expression of neuropilin-2, a well-known candidate gene of ASD (Lee et al., 2012) (**Figure 2**).

More than 200 lncRNAs are differentially expressed in the prefrontal cortex and cerebellum of ASD patients (Rennert, 2013). The autism loci containing RAY1/ST7 (suppression of tumorigenicity 7) encode at least four non-coding genes (ST7OT1-4, ST7AS1-4) which are located in the sense or antisense chains that potentially regulate RAY1/ST7. Several rare mutations of RAY1/ST7 or ST7OT1-3 genes have been detected in autistic patients (Vincent et al., 2002; van de et al., 2013) (**Figure 2**).

Genome-wide differential expression analysis of blood samples from ASD patients (Carninci et al.) identified 2407 up-regulated and 1522 down-regulated lncRNAs in peripheral blood leukocytes of ASD. The pathway enrichment analysis of these differently expressed lncRNAs revealed that they were mainly involved in synaptic vesicle circulation, long-term inhibition and long-term potentiation of neural pathways (van de et al., 2013). Differential expression of lncRNAs *SHANK2-AS* and *BDNF-AS* was also observed in ASD (van de et al., 2013; Wang et al., 2015; Tang et al., 2017) (**Figure 2**). In neurons, *SHANK2-AS* and *Shank2* can form double-stranded RNA that inhibit the expression of *Shank2*. Overexpression of *SHANK2-AS* reduces the complexity of neurites, and inhibits the proliferation of neuronal stem cells and promotes their apoptosis (Luo et al., 2018).

Moesin is a protein that in human is encoded by the MSN gene, which regulates neuronal structure and immune response. MSN-binding *MSNP1-AS* is highly expressed in postmortem cerebral cortex samples from individuals with ASD. Increased *MSNP1-AS* expression is also observed in individuals carrying the ASD-associated rs4307059 T allele (Kerin et al., 2012) (**Figure 2**). By targeting the *MSNP1-AS* gene promoter, *MSNP1-AS* knockdown disrupts the expression of 318 genes in neuroblastoma neural progenitor cells, many of which are involved in chromatin organization and immune response, indicating multiple transcriptional and translational functions of *MSNP1AS* in ASD-relevant biological processes (DeWitt et al., 2016) (**Table 1**).

## Fragile X Syndrome

FXS is the most common inherited cause of mental disorder and ASD (Lin, 2015; Hecht et al., 2017). FXS is caused by FMR1 (fragile X mental retardation 1) inactivation or dysfunction (Verkerk et al., 1991). FMR1 is required for normal neuronal connectivity and plasticity. The *FMR1* gene contains a CGGrepeat present in the 5′UTR which can be unstable upon transmission to the next generation. FXS patients have a repeat length exceeding 200 CGGs that generally leads to methylation of the repeat and the promoter region, resulting in silencing of *FMR1* gene expression (Fu et al., 1991; Dominique Heitz et al., 1992; Tang et al., 2017).

Several miRNAs have already been proven to involve in the development of FXS (Lin, 2015). During the embryonic stage, miR-302 is specifically expressed by embryonic stem cells, and it blocks the translation of FMR1, which is required to repress differentiation. At blastocyst stage, down-regulation of miR-302 promotes FMRP synthesis and subsequent neuronal development. In the normal neuronal development, FMRP, as an RNA-binding protein, interacts with miR-125 and miR-132 to regulate the signal transduction of metabolic glutamate receptors (mGluR1) and N-methyl-D-aspartate receptors (NMDAR) (Lin, 2015) (**Figure 2**). In addition, let-7c, miR-9, miR-100, miR-124, miR-125a, miR-125b, miR-127, miR-128, miR-132, miR-138, miR-143 and miR-219 might also interact with fragile X mental retardation protein (FMRP) to regulate neuronal development (Lin, 2015) (**Table 1**).

LncRNAs have recently emerged to influence the pathogenesis of FXS (Esteller, 2011; Taft et al., 2010; van de et al., 2013). The FMR1 bidirectional promoter is capable of translating lncRNA *FMR4* or *FMR1-AS1* which is an antisense transcription that overlaps the CGG repeat region (Khalila

and mechanism diagram of Fragile X Syndrome. microRNAs and long non-coding RNAs regulate Fragile X Syndrome process and might be a class of biomarkers of Fragile X Syndrome. (C) Dysregulation of microRNAs related to Down Syndrome in human fetal hippocampus and heart samples. (D) Left, the expression level of several microRNAs up-regulates in Rett Syndrome patients and these microRNAs play important roles in modulating neuronal development. Right, relationship between Mecp2 and Non-coding RNAs. (E) Microdeletion of SnoRNA SNRD (HBI-85) leads to Prader-Willi syndrome like phenotype. Long non-coding RNA Ube3a-ATS represses Ube3a, which gives rise to Angelman syndrome.

2009; Ladd et al., 2007) (**Figure 2**). *FMR4* plays a critical role in regulating cell cycle, proliferation, and apoptosis of human neural precursor cells (Khalila 2009).

LncRNAs FMR5 and FMR6 have recently been linked to FXS (Pastori et al., 2014) (**Figure 2**). FMR5 is a sense lncRNA transcribed upstream of the FMR1 promoter, while FMR6 is an antisense transcript overlapping the 3′-UTR of FMR1. The expression of FMR4, FMR5, and FMR6 is detectable in the majority of patient leukocyte RNA samples, suggesting that it may be reliable biomarkers for FXS (Wahlestedt, 2013).

### Down Syndrome

DS is a neurodevelopmental disorder caused by the presence of all or part of a third copy of chromosome 21 (Asim et al., 2015). This disorder has been characterized with many clinical manifestations, including dementia, defects of the immunity system and congenital heart, and abnormalities of facial growth, gastrointestinal tract, and endocrine system (Malinge et al., 2009). Five miRNAs (miR-99a, let-7c, miR-125b-2, miR-155 and miR-802) have been found to be overexpressed in human fetal hippocampus and heart samples from individuals with DS (Fillat and Altafaj, 2012) (**Figure 2)**. Let-7c and miR-125b have been shown to enhance neuronal aging and degeneration (Chawla et al., 2016). Recent evidence suggests that DS dementia strongly correlates with overexpression of miR-155 on chromosome 21 with concomitant reduction of multiple CNS-functional targets, including BACH1, CoREST1, Cyclin D1, BCL6, BCL10, BIM, and SAPK4 (Tili et al., 2018) (**Figure 2)**.

### Rett Syndrome

RTT is a neurodevelopmental disorder caused by the loss of function of methyl-CpG-binding protein 2 (MeCP2) (Chahrour and Zoghbi, 2007; Obiols-Guardia and Guil, 2017). Because chromosome Y does not exist MeCP2, the disease occurs almost entirely in women (Weng et al., 2011; Obiols-Guardia and Guil, 2017). MeCP2 protein is highly expressed in neurons, acting as a transcriptional repressor and activator, depending on the context (Luikenhuis et al., 2004). Growing evidence suggests that various non-coding RNAs might play important roles in the development of RTT (Obiols-Guardia and Guil, 2017).

Due to direct or indirect deregulation following MeCP2 loss of function, disrupted miRNA expression has been reported in the disease progress of RTT (Urdinguio et al., 2010; Lyu et al., 2016). For instance, miR-184, miR-30a, miR-381, and miR-495 are aberrantly up-regulated in MeCP2 knockout mice (Nomura et al., 2008; Wu et al., 2010) (**Figure 2)**. These miRNAs are known for repressing the expression of important modulators of neuronal development, such as *Bdnf* and *Numbl* (Liu et al., 2010; Wu et al., 2010). MeCP2 also interacts with pri-miRNA processing machines and affects their activity. For example, DGCR8/Drosha complex can be inhibited by MeCP2, thus affecting nuclear miRNA processing and dendritic growth (Cheng et al., 2014). Interestingly, miRNAs can also regulate MeCP2 transcription, such as miR-130a (Zhang et al., 2016), miR-132 (Lyu et al., 2016), miR-200a, and miR-302c (Rodrigues et al., 2016) (**Figure 2)**. Future studies on the link between MeCP2 and the miRNA population will broaden our knowledge of regulatory network affected in RTT and will help develop better therapeutic strategies.

PiRNA expression levels are altered globally in the absence of MeCP2 (Saxena et al., 2012). There are at least 12 hippocampusabundant piRNAs up-regulated with a fold change of over 1.5 in the cerebellum of MeCP2 KO mice. Among them, DQ541777, which is implicated in regulating the size of dendritic spines, is the 5th most abundant piRNA in the cerebellum libraries (Lee et al., 2011; Saxena et al., 2012) **Figure 2**. More specific functions of these dysregulated piRNAs in the pathophysiology of RTT are expected to be discovered in near future.

The aberrant lncRNA transcriptome is also present in the brain of RTT mice. For instance, the AK081227 and AK087060 transcripts are up-regulated in MeCP2-null brains (Petazzi et al., 2013). The overexpression of AK081227 mediated by the Mecp2 loss is associated with the down-regulation of its host coding protein gene *Gabrr2*, a major inhibitory neurotransmitter in the mammalian brain where it acts at GABA receptors, which are ligand-gated chloride channels (Petazzi et al., 2013). The neurotrophic BDNF is known to be aberrantly diminished in RTT individuals (Katz et al., 2016), it can be speculated that the lncRNA BDNF-AS might be an important therapeutic target for treating RTT. Although lncRNAs play an important role in neuronal development, their roles in the pathogenesis of RTT is largely unknown (Mo 2015; Lin et al., 2014; Ng et al., 2013).

### Prader-Willi Syndrome and Angelman Syndrome

Chromosome 15q11-q13 is a region containing a lot of genomic imprinting genes (Kalsner and Chamberlain, 2015). Prader-Willi Syndrome (PWS) and Angelman Syndrome (AS) are two different types of neurodevelopmental disorders which are caused by loss of function or overexpression of at least one imprinted gene at the 15q11-q13 locus (Kalsner and Chamberlain, 2015). PWS is characterized by intellectual disability, irritability, short stature, and low fertility/hypogonadism (Buiting, 2010). NECDIN and small ribonucleoprotein polypeptide N (SNRPN) are functionally related to the pathological features of the disease (Francoise Musctelli, 2000). SNRPN downstream introns contain SNRD116 (HBII-85) snoRNA clusters, and paternal genetic microdeletion of SNRPN clusters might lead to PWS (Sahoo et al., 2008; de Smith et al., 2009) (**Figure 2)**. Actually, mice with deletion of MBII85 snoRNA clusters demonstrated obvious PWS phenotypes (Skryabin et al., 2007), indicating non-coding RNAs can be tightly regulated and may play critical roles in the pathology of PWS.

The pathological features of AS include delayed development, intellectual disability, severe speech impairment, and problems with movement and balance (Buiting, 2010). AS is caused by the deletion and/or mutation of *Ube3a* on the maternal chromosome. While in patients with AS the maternal *Ube3a* allele is inactive, the paternal allele is intact but epigenetically silenced through the *Ube3a-ATS* part of LNCAT (large non-coding antisense transcript) at the Ube3a locus (Runte et al., 2001). Elucidating the mechanisms of how Ube3a-ATS involves in silencing the paternal *Ube3a* may lead to new therapies for AS (Landers et al., 2005; Meng et al., 2012) (**Figure 2)**.

### Conclusions

Non-coding RNAs have emerged as important regulators in the brain development and function. Although the number and functional subclasses of non-coding RNAs has steadily increased, it still likely represents only a small fraction of the total RNA transcriptome underlying the ontogeny and functional complexity of mammalian brain functions in health and disease. Aberrant expression of non-coding RNAs has linked with various neurodevelopmental diseases. The regulatory network of non-coding RNAs in neurodevelopmental disorders is very complicated, and the molecular mechanisms of non-coding RNA causing neurodevelopmental diseases are still largely unknown. However, we still hope that this review raises awareness of the central roles that large non-coding RNAs and their complex regulatory networks in brain development and function.

With more defined molecular function and mechanism, Non-coding RNAs have the great potential to serve as disease biomarkers or drug targets, especially in neurodevelopmental disorders. For example, miR-132, miR-23a, miR-93, miR-106b, miR-146b and miRNA-148b are abnormally expressed in autistic patients (Mahesh Mundalil Vasu1, 2014; Trindade et al., 2013). miR-99a, let-7c, miR-125b-2, miR-155 and miR-802 are found to be overexpressed in human samples from individuals with Down Syndrome(Fillat and Altafaj, 2012). In fragile X syndrome, the expression of FMR4, FMR5, and FMR6 is detectable in the majority of patient leukocyte RNA samples, suggesting that it may be reliable biomarkers for Fragile X Syndrome (Wahlestedt, 2013). The analysis and integration of this information with other datasets will get new clues of non-coding RNAs as reliable biomarkers.

A large number of studies have suggested that non-coding RNAs might be new promising targets for the treatment of neurodevelopmental disorders. However, several challenges remain to be investigated before non-coding RNAs can be routinely used in outbreak investigation and clinical practice. Firstly, more efficient non-coding RNAs delivery systems are needed to develop. Secondly, the biology of non-coding RNAs, such as their structural motifs, stability, degradation, and gene regulatory network, needs further investigations. Finally, preclinical research and clinical trials are required to determine the safe dose and therapeutic potentials of noncoding RNAs.

### REFERENCES


#### In addition, how non-coding RNAs operate in CNS at the molecular, cellular and more hierarchical neural network levels still remains elusive. Therefore, it is important to discover molecules based on further elucidating more pathways of non-coding RNA roles in the CNS and how non-coding RNA dysfunction leading to neurodevelopmental disorders. Based on more understanding of non-coding RNAs in the CNS, the researchers will probably develop new diagnostic and therapeutic approaches for neurodevelopmental disorders.

### AUTHOR CONTRIBUTIONS

C-ML directed the manuscript preparation. S-FZ, JG, and C-ML wrote the manuscript.

### ACKNOWLEDGMENTS

This work was supported by grants from the National Key Research and Development Program of China Project (Grant No. 2016YFA0101402), the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant No. XDA16010302), the National Science Foundation of China (No. 91753140, 81771224, 31571043).

Lifespan in Drosophila. *PloS Genet.* 12, e1006247. doi: 10.1371/journal. pgen.1006247


microRNAs in a mouse model of Rett syndrome. *Proc. Natl. Acad. Sci. U.S.A.* 107, 18161–18166. doi: 10.1073/pnas.1005595107


Zhou, S., Ding, F., and Gu, X. (2016). Non-coding RNAs as Emerging Regulators of Neural Injury Responses and Regeneration. *Neurosci. Bull.* 32, 253–264. doi: 10.1007/s12264-016-0028-7

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

*Copyright © 2019 Zhang, Gao and Liu. 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.*

# Aberrant Expression of Pseudogene-Derived lncRNAs as an Alternative Mechanism of Cancer Gene Regulation in Lung Adenocarcinoma

Greg L. Stewart<sup>1</sup> \*, Katey S. S. Enfield1,2, Adam P. Sage<sup>1</sup> , Victor D. Martinez<sup>1</sup> , Brenda C. Minatel<sup>1</sup> , Michelle E. Pewarchuk<sup>1</sup> , Erin A. Marshall<sup>1</sup> and Wan L. Lam<sup>1</sup>

<sup>1</sup> BC Cancer Research Centre, Vancouver, BC, Canada, <sup>2</sup> The Francis Crick Institute, London, United Kingdom

Edited by: Yujing Li, Emory University, United States

#### Reviewed by:

Sergio Verjovski-Almeida, University of São Paulo, Brazil Nithyananda Thorenoor, Pennsylvania State University, United States

> \*Correspondence: Greg L. Stewart gstewart@bccrc.ca

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 16 November 2018 Accepted: 11 February 2019 Published: 06 March 2019

#### Citation:

Stewart GL, Enfield KSS, Sage AP, Martinez VD, Minatel BC, Pewarchuk ME, Marshall EA and Lam WL (2019) Aberrant Expression of Pseudogene- Derived lncRNAs as an Alternative Mechanism of Cancer Gene Regulation in Lung Adenocarcinoma. Front. Genet. 10:138. doi: 10.3389/fgene.2019.00138 Transcriptome sequencing has led to the widespread identification of long non-coding RNAs (lncRNAs). Subsequently, these genes have been shown to hold functional importance in human cellular biology, which can be exploited by tumors to drive the hallmarks of cancer. Due to the complex tertiary structure and unknown binding motifs of lncRNAs, there is a growing disparity between the number of lncRNAs identified and those that have been functionally characterized. As such, lncRNAs deregulated in cancer may represent critical components of cancer pathways that could serve as novel therapeutic intervention points. Pseudogenes are non-coding DNA sequences that are defunct relatives of their protein-coding parent genes but retain high sequence similarity. Interestingly, certain lncRNAs expressed from pseudogene loci have been shown to regulate the protein-coding parent genes of these pseudogenes in trans particularly because of this sequence complementarity. We hypothesize that this phenomenon occurs more broadly than previously realized, and that aberrant expression of lncRNAs overlapping pseudogene loci provides an alternative mechanism of cancer gene deregulation. Using RNA-sequencing data from two cohorts of lung adenocarcinoma, each paired with patient-matched non-malignant lung samples, we discovered 104 deregulated pseudogene-derived lncRNAs. Remarkably, many of these deregulated lncRNAs (i) were expressed from the loci of pseudogenes related to known cancer genes, (ii) had expression that significantly correlated with protein-coding parent gene expression, and (iii) had lncRNA protein-coding parent gene expression that was significantly associated with survival. Here, we uncover evidence to suggest the lncRNApseudogene-protein-coding gene axis as a prominent mechanism of cancer gene regulation in lung adenocarcinoma, and highlights the clinical utility of exploring the non-coding regions of the cancer transcriptome.

Keywords: long non-coding RNAs, pseudogenes, lung cancer, gene regulation, non-coding RNAs

## INTRODUCTION

fgene-10-00138 March 4, 2019 Time: 19:43 # 2

Lung cancer is an enormous health burden, representing the most common cause of cancer death worldwide. While advances in imaging technology have improved early detection, the outcome for lung cancer patients remains poor. This is largely attributed to limited therapeutic options and short-term response due to tumor heterogeneity. Despite extensive sequencing efforts to characterize exomes and protein-coding transcriptomes of lung tumors, less than half of lung adenocarcinoma (LUAD) patients harbor clinically targetable mutated driver genes, highlighting the need to explore alternative mechanisms of cancer gene deregulation (Chan and Hughes, 2015).

Long non-coding RNAs (lncRNAs, > 200 nt) have emerged as important players in cell biology. LncRNAs have been observed to function through a wide variety of regulatory mechanisms, targeting DNA, proteins, and other RNA species; specifically with defined roles in RNA degradation or stabilization, protein translocation and complex formation, and recruitment of complexes to transcriptional loci (Lee et al., 1999; Wang et al., 2011; Kung et al., 2013; Simon et al., 2013). This broad functional repertoire has been shown to be exploited by many cancer types, including LUAD, to drive various hallmarks of cancer (Bhan et al., 2017; Peng et al., 2017). One of the first cancerassociated lncRNA transcripts, MALAT1 (Metastasis Associated Lung Adenocarcinoma Transcript 1) was discovered in lung cancer. MALAT1 is overexpressed in metastatic lung tumors and functions in trans through transcriptional regulation of multiple genes involved in cell motility (Tano et al., 2010; Gutschner et al., 2013). Since the discovery of MALAT1, many other lncRNAs have been shown to play a direct role in nearly every major cancer type. Tumor suppressive lncRNA transcripts are commonly observed to be downregulated, for instance the lncRNA TARID, which promotes the expression of the tumor suppressor gene TCF21 through active promoter demethylation (Arab et al., 2014). Conversely oncogenic lncRNAs can be overexpressed, such as NEAT1, an architectural lncRNA associated with metastasis in many cancer types (Hirose et al., 2014; Blume et al., 2015; Yamazaki et al., 2018; Zhang et al., 2018). As such, lncRNAs may represent critical regulators of oncogenic-driver pathways that could serve as undiscovered clinical intervention points in LUAD.

While lncRNAs have been observed to be important in cancer biology, functional prediction of newly-discovered lncRNAs remains a major challenge. However, many lncRNAs expressed from pseudogene loci have been shown to regulate the specific genes with which they have sequence homology. Pseudogenes are DNA sequences that are defunct relatives of functional proteincoding genes (herein referred to as parent genes) and arise during either gene duplication events, or the reverse transcription of an mRNA transcript into a new genomic location. Through evolution these duplicated genes have acquired mutations such as premature stop codons and frameshifts, which results in the loss of protein coding ability, while still retaining a high degree of sequence homology with the original parent gene (Khachane and Harrison, 2009). Recently, pseudogene-derived lncRNAs have been shown to regulate their parent genes and this novel mechanism has been observed in many tumor types, including lung cancer (Sun et al., 2017; Hu et al., 2018). A prominent example is the tumor suppressor gene PTEN (chromosome 10), regulated both positively by PTENP1 (chromosome 9), a lncRNA transcribed from the sense strand of the pseudogene locus, and negatively by the lncRNA PTENP1-AS1, which is transcribed from the strand antisense to the parent gene (Johnsson et al., 2013).

Pseudogenes have been continually omitted from large RNAsequencing datasets due to the complexity of separating highly similar pseudogene sequences from parent genes. However, Milligan et al. (2016) recently generated an atlas of lncRNAs overlapping pseudogenes, which has provided a foundation for their analysis in RNA sequencing datasets. We hypothesize that the functions of pseudogene-derived lncRNAs are an underexplored mechanism of gene regulation that occurs more broadly than previously realized, and that these events contribute to the tumorigenesis of LUAD. We performed next generation RNAsequencing on microdissected LUAD tumors and matched nonmalignant tissue to identify deregulated lncRNAs expressed from pseudogene loci (herein referred to as 9-lncs). We then explored the relationships of these 9-lncs with their parent genes, and explored their significance in relation to patient clinical features in our discovery dataset as well as a validation dataset.

### RESULTS

### 9-lnc Expression Is Deregulated in Lung Adenocarcinoma

Pseudogenes vary widely in terms of length, gene fraction, and identity to parent genes, and can be expressed as lncRNAs that are sense, antisense, partial overlapping, or internal to the parent. In light of this variation, 9-lncs are observed to have vastly different regulatory effects on downstream target genes (**Figure 1A**). In our curation of 9-lncs in LUAD, we have included those that have exonic overlap with a pseudogene (partial or full length) and considered both sense and antisense transcripts (**Supplementary Table S1**). 9-lncs were analyzed in an in-house discovery (BCCA, n = 72) and external validation (TCGA, n = 108) cohort of LUAD and paired non-malignant lung tissues (**Table 1**). We identified aberrantly expressed 9-lncs that are significantly deregulated in both the discovery and validation datasets with the same direction of expression alteration (9-lncs upregulated or downregulated in tumors compared to matched non-malignant tissue).

We found 104 lncRNAs expressed from 102 pseudogene loci to be significantly deregulated in LUAD (**Supplementary Table S2**). To our surprise, we found that the majority of these deregulated 9-lncs were downregulated in tumors (**Figures 2A,B**). Most of these were unannotated lncRNAs, such as RP11-1007O24.3, which was downregulated in tumors, with only 24 of the total deregulated 9-lncs having been previously described in scientific literature annotated in PubMed, albeit none in the field of pseudogene-mediated deregulation (**Supplementary Table S3** and **Figure 2B**). Twenty of these 24 have been described in the context of cancer, with only four in lung cancer. This includes DGCR5, a lncRNA we found to be overexpressed in

(A) Summary of the regulatory mechanisms of pseudogene-derived lncRNAs (9-lncs) that retain sequence homology with the parent gene. Overall, lncRNAs have been shown to function through a variety of regulatory mechanisms, acting on the DNA, RNA, and protein levels. (B) Flow diagram description of the analysis pipeline applied for the identification of deregulated 9-lncs. Patient LUAD samples were collected and subjected to next generation sequencing to quantify RNA expression. Gene expression was then compared between tumors and matched non-malignant tissue to identify significantly deregulated transcripts. LncRNAs with exonic overlap to known pseudogenes were then identified and confirmed in a 2nd set of LUAD and matched non-malignant tissue. Deregulated 9-lncs were then assayed to determine associations with clinical features.



<sup>a</sup> Ever smokers is a term that includes both current and former smokers.

tumors. DGCR5 has been reported to promote LUAD progression by sequestering a variety of miRNAs involved in cell cycle regulation, although it has not been investigated with regard to its pseudogene-derived nature (Chen et al., 2017; Dong et al., 2018; Luo et al., 2018).

We were interested in examining the genetic events that could impact pseudogene loci, and thus affect 9-lnc expression. We mapped the chromosomal distribution of the deregulated 9-lncs, finding them to be distributed throughout the genome and detected on most chromosomes, except for chromosomes 4 and Y (**Figure 3**). The locations of each of the parent genes of deregulated 9-lncs are similarly distributed through the genome (**Supplementary Table S4**). We then determined the overlap of these genes with regions of recurrent chromosomal amplification and deletion as determined by The Cancer Genome Atlas (TCGA) for LUAD (Zack et al., 2013). While some 9-lncs overlap with regions of recurrent deletion, the majority do not, indicating that they may be regulated by mechanisms other than copy number alteration (**Figure 3**).

### Global Patterns of 9-lnc and Parental Gene Expression

As a first step to identify deregulated 9-lncs that may function through regulation of their respective parent genes, we explored whether 9-lncs with significantly deregulated expression were associated with altered parent gene transcript levels (Poliseno et al., 2010; Johnsson et al., 2013; Feng et al., 2017; Huang et al., 2018). We obtained parent gene information for the 95 deregulated 9-lncs and determined that they shared 104 parent genes. Some pseudogenes contained multiple lncRNAs, and some lncRNAs overlapped multiple pseudogenes, constituting a total number of 116 9-lnc-parent gene pairings. For each 9-lncparent pair we compared groups of tumors with high levels of 9-lnc expression to those with low levels of 9-lnc expression.

FIGURE 3 | Genome-wide distribution of deregulated pseudogene-derived lncRNAs in lung adenocarcinoma. Circular representation of the genomic distribution of the deregulated pseudogene-derived lncRNAs discovered in our study, as well as known regions of copy number alterations in lung adenocarcinoma as described by TCGA. The outer concentric circle represents the human karyotype from the genomic build hg19. The blue concentric circle contains known regions of copy number amplification (red boxes) and deletion (blue boxes) that have been previously published. The inner green circle represents the specific genomic location of our pseudogene-derived lncRNAs found to be either upregulated (green circles) or downregulated (purple circles) in lung adenocarcinoma. Finally, the inner connecting lines represent the interaction between the deregulated pseudogene-derived lncRNAs and the locations of their respective protein-coding parent genes. Chromosome 9 (magnified region) highlights that some of the downregulated pseudogene-derived lncRNAs overlap with genomic regions frequently deleted in lung adenocarcinoma.

We found that 33 9-lncs have a significant expression relationship with their parent gene in at least 1 dataset (**Supplementary Table S5**). This included 21 sense 9-lnc-parent-gene pairs, and 13 antisense 9-lnc-parent gene pairs. Having identified 9-lncs with expression associated with parent gene expression, we investigated whether parent genes had known oncogenic or tumor suppressive roles. We performed a literature search to determine if any had been previously described in the context



<sup>a</sup> denotes number of entries in PubMed.

of cancer. Interestingly, we found that 65 of these parent genes had been previously described in cancer, and of those, 33 had been described in lung cancer (**Table 2**). Of the 34 significantly differentially expressed 9-lnc-parent gene pairs, 25 parent genes were described in cancer. This includes lung cancer associated genes like CS, which affects tumor drug response, as well as RCN1, which is associated with poor prognosis and tumor progression in lung cancer (Chen et al., 2014, 2018).

As the vast majority of the deregulated 9-lncs that were correlated with their parent gene had positive associations, we were interested whether this was a global phenomenon or exclusive to deregulated genes. We performed a Spearman's correlation analysis on every 9-lnc-parent-gene pair with expression data in our dataset irrespective of deregulation status (n = 390 gene pairs). We plotted the distribution of Spearman's rho (ρ) values for the 9-lnc-parent-gene pairs and compared them to the rho values for 9-lncs paired to randomly selected genes. We found that the 9-lnc-parentgene pairs have significantly more positive relationships than the random gene pairs in both the BCCA (Mann Whitney U-test, p < 0.0001) and TCGA datasets (Mann Whitney U-test, p < 0.0001; **Figure 4A**). Studies have shown that lncRNAs transcribed from opposite strands can have different regulatory effects on target genes (Johnsson et al., 2013; Balbin et al., 2015). To determine if transcriptional orientation has an effect on 9-lnc-parent relationships we compared the Spearman's rho values of sense 9-lnc-parent pairs to antisense 9-lnc-parent pairs. In both datasets we observed that the sense 9-lncparent pairs to have significantly more positive relationships than the antisense 9-lnc-parent pairs (Mann Whitney U-test, TCGA set (p < 0.0001), and BCCA set (p < 0.0025) (**Figure 4B** and **Supplementary Table S6**). Strongly positively

correlated 9-lnc-parent pairs include TPT1-AS1/RCN1 and LINC00887/CS (**Figure 4C**).

#### 9-lncs and Their Parent Genes Are Associated With Patient Survival

If the aberrant expression of 9-lncs is biologically relevant, it follows that they may be relevant in tumor aggressiveness, stage, and patient survival. We performed a two-group analysis using a Mann Whitney U-test between Stage I tumors and Stage II-IV tumors, as the majority of our tumors fell into these categories (**Table 1**). Of the deregulated 9-lncs we found CTC-250I14.3 to be associated with Stage 1 disease and downregulated in both the BCCA and TCGA LUAD cohorts (**Supplementary Figure S1**). Our discovery datasets were limited in sample size for survival association analysis; therefore, we examined a third cohort of 719 LUAD from the KM Plotter database (Kaplan–Meier Plotter)<sup>1</sup> (**Table 1**). This cohort was limited to genes with probe coverage on microarray platforms. A total of 19 of the deregulated 9-lncs were represented on this platform, yet, the majority of these 9-lncs (16 of 19) were significantly associated with poor overall survival (log-rank p < 0.05, **Table 3**).

While we were not able to investigate the survival associations of all deregulated 9-lncs as many were not covered by the microarray platforms, the majority (72 out of 103) of their parent genes were represented. We discovered that 67 of these parent genes were associated with patient survival. Twenty-eight of these survival associated parent genes were also significantly associated with the expression of their paired deregulated 9-lnc. Furthermore, we found 11 pairs where both 9-lnc and parent gene are associated with patient survival. For example RP11-1007O24.3, a 9-lnc downregulated in tumors, is positively associated with expression of survival-associated parent gene ARIH1 in both the BCCA and TCGA datasets (**Table 3** and **Figure 5A**). Further examples also include 9-lnc-parent pairs such as ZBED3-AS1 and HMGB1, which are positively correlated at the expression level, and both significantly associated with survival (**Figure 5B**). We also observe 9-lncparent pairs that are associated with survival, but do not share an expression relationship such as LINC00667 and parent gene TACC3 (**Figure 5C**). Collectively, our discovery of the broad deregulation of 9-lncs, many of which are survival-associated and associated with parent gene expression, may indicate that 9-lncs impact LUAD biology through trans regulation of their cancer-associated parent genes.

#### DISCUSSION

Here we expand upon the work done by Milligan et al. (2016), completing the first large-scale analysis of lncRNA expression from pseudogene loci in LUAD and paired non-malignant lung

<sup>1</sup>http://kmplot.com/analysis/

TABLE 3 | Associations between 9-lncRNA, parent gene expression, and patient outcome.


<sup>a</sup> Denotes poor overall survival associated with high gene expression.

<sup>b</sup> Denotes poor overall survival associated with low gene expression.

<sup>c</sup> No information for the parent gene.

tissue. We discovered a broadly positive association between 9-lncs and parent-gene expression, suggestive of an alternative mechanism of cancer gene regulation. While there have been singular examples of deregulated pseudogene-derived lncRNAs in cancer, we show that this phenomenon is widespread in LUAD. In addition to being correlated with 9-lnc expression, we find that many of the parent genes of these deregulated lncRNAs are annotated cancer genes and are significantly associated with patient survival, highlighting how these previously unappreciated non-coding genes may affect LUAD biology.

While the identification of lncRNAs associated with cancer phenotypes is increasing, a great challenge in the field remains the accurate downstream prediction of lncRNA function. Unlike protein-coding genes or small non-coding RNAs, features like complex folding patterns and unknown binding motifs have contributed to the challenging functional characterization of lncRNAs. We utilized the sequence similarity found between lncRNAs expressed from pseudogene loci, and their parent genes to predict the function of this subset of lncRNAs in LUAD. We identified a set of 104 9-lncs deregulated in LUAD in two independent datasets. This greatly increases the number of deregulated lncRNAs known to be expressed from pseudogene loci in LUAD. Interestingly, the majority of these 9-lncs were under-expressed in tumors compared to non-malignant tissues, suggesting that they may have tumorsuppressive roles, and that their downregulation is advantageous to LUAD tumorigenesis.

Under-expression was not significantly associated with regions of recurrent copy number deletion in LUAD, although a subset of deregulated 9-lnc loci were localized to these regions (**Figure 3**). These observations suggest that they may be regulated by alternative molecular mechanisms, including broad chromosomal aberrations that affect whole chromosome arms, or epigenetic mechanisms. For example, endogenous retroviruses and repetitive elements often become aberrantly expressed in cancer due to deregulated methylation patterns (Kassiotis, 2014). We did not observe enrichment of 9-lncs or their parent genes on any chromosomes, despite the fact that pseudogenes are known to be overabundant on the human X chromosome (Drouin, 2006).

The direction of transcription often affects lncRNA function (Balbin et al., 2015). For example, the PTEN pseudogene (PTENP1) expressed in the sense direction can function as a decoy for inhibitory miRNAs that would otherwise cause translational inhibition of the PTEN parent mRNA. Conversely, when the antisense lncRNA is expressed from the PTENP1 locus, the transcript is able to localize to the PTEN parent locus and recruit chromatin-remodeling machinery, which leads to the silencing of PTEN transcription (Poliseno et al., 2010; Johnsson et al., 2013). Both mechanisms have been coopted by cancer cells for their respective tumor suppressive and oncogenic roles (Hu et al., 2018). We found sense 9-lnc-parent pairs, (which account for 208 out of 391 9-lncs examined) to be more positively correlated than antisense 9-lnc-parent pairs in both cohorts (**Figure 4B**). This may imply that 9-lncs are more likely to regulate their parent gene in a positive manner which may occur through mechanisms such as miRNA sponging or transcript stabilization when transcribed in the sense direction (Glenfield and McLysaght, 2018; Hu et al., 2018). The distribution of Spearman's ρ-values for antisense-parent gene pairs suggests a more even split between positive and negative regulation. A limitation of this study is that we cannot discount the possibility that sequencing reads for sense overlapping 9-lncs that have sequence homology with their parent gene are being mapped to the parent gene instead of the 9-lnc. This potential issue warrants further investigation considering both the large number of annotated pseudogenes in the genome (n = 13,000) and the possibility of false interpretation of sequencing data for both protein-coding and non-coding genes.<sup>2</sup> While these alignment errors could affect sense 9-lnc-parent gene pairs,

<sup>2</sup>https://www.genenames.org/cgi-bin/statistics

HMGB1, and (C) LINC00667 and parent gene TACC3. Expression associations (upper row) for each pseudogene-derived lncRNA and parent gene pair were found by stratifying samples into tertiles by high (red) and low (black) expression of the pseudogene-derived lncRNA, and plotting the expression of the parent-gene (RPKM) on the y-axis. Survival associations were found for pseudogene-derived lncRNAs (middle row) and their respective parent genes (bottom row). Samples were stratified into tertiles with high (red) and low (black) expression of the gene-of-interest, and the significance of the associations were assessed using the logRank method through GraphPad Prism 8 software on data obtained from kmPlot (n = 673). Survival information was not available for RP11-1007O24.3 and it was thus omitted from this analysis.

antisense 9-lnc-parent gene pairs are not subjected to this same technical artifact. Recently, RNA-sequencing analysis strategies have emerged that begin to address this issue, and long read RNAsequencing could be used to reduce errors in sequence alignment.

When looking at the parent genes of these deregulated 9-lncs we were interested to find that many had previously described roles in cancer (**Table 2**). This includes EGLN1, a well described cancer gene involved in regulation of tumor hypoxia, and CDC42, an oncogene involved in cell cycle control (Chan and Giaccia, 2010; Maldonado and Dharmawardhane, 2018). Many of these deregulated 9-lncs were also associated with clinical parameters such as patient survival and patient stage, in addition to the correlated expression between 9-lncs and their parent genes (**Table 3**). For example, ZBED3-AS1 and HMGB1 were positively correlated at the expression level, and low expression of both genes was associated with poor patient survival (**Figure 5B**). We also observed 9-lnc-parent pairs where both genes are associated with survival, but do not share an expression relationship. LINC00667 and parent gene TACC3, for example, are both survival-associated, but not correlated at the expression level. TACC3 is a component of the TACC3/ch-TOG/clathrin protein complex, and roles in complex assembly have been previously observed for lncRNAs (Munschauer et al., 2018; **Figure 5C**). Thus, it is possible that LINC00667 is involved in a form of regulation that would not affect transcript levels, including protein-complex assembly. While we were unable to assess survival associations for many of the deregulated 9-lncs, they may still impact patient survival

through regulation of their parent genes. We found 28 out of 34 expression-associated 9-lnc-parent gene pairs to have parent genes associated with patient survival. RP11-1007O24.3, for example, was positively correlated with survival-associated parent gene ARIH1 expression in both cohorts. ARIH1 has been previously described to be a mediator of DNA-damage response and mitophagy in cancer cells (**Figure 5A**; Perdomo et al., 1988; von Stechow et al., 2015). The potential regulatory impact of 9-lncs on their clinically relevant parent genes is considerable and may represent a novel avenue for targeted therapies. While this study focused on the broad effects of 9-lnc deregulation, future studies utilizing in vitro and in vivo experiments will be necessary to determine the specific mechanisms of parent gene regulation.

As 9-lncs may represent an unexplored area of cancerassociated parent gene regulation, their therapeutic relevance should be further explored. LncRNAs make ideal targets for therapies that target RNA products such as Antisense Oligonucleotide (ASO) therapies, since RNA is their final functional state, rather than the intermediate product for proteincoding genes (Crooke et al., 2018). In addition, ASOs are easier and less costly to develop than small molecule inhibitors, and are in development as aerosol sprays that may be ideal for lung cancer treatment (Kjems and Howard, 2012; Li and Chen, 2013). However, as ASOs target through complementary sequence pairing, they would have to be designed in such a way as to not interfere with the parent gene, especially in the case of 9-lncs expressed from the sense strand.

This strategy of identifying lncRNAs aberrantly expressed from pseudogene loci may be useful when applied to other cancer types. Indeed, we see that several of our deregulated 9-lncs have been described in other tumor types, such as TPT1- AS1 in cervical cancer, and HGC11 in both prostate cancer and hepatocellular carcinoma (Zhang et al., 2016; Xu et al., 2017; Jiang et al., 2018). Additionally, as lncRNA expression is highly tissue specific, the application of this approach to other cancer types may yield novel disease-specific 9-lnc-parent pairs, highlighting the clinical utility of examining these previously underappreciated transcripts. Overall, 9-lnc-cancer-parent-gene axes represent alternative mechanisms of cancer gene regulation, and their identification is a critical step toward the functional characterization of lncRNAs.

#### CONCLUSION

There is a growing need to functionally characterize lncRNAs. Pseudogene-derived lncRNAs have been shown to be involved in cancer and regulate the expression of their parent genes. We show here how pervasive this gene regulatory mechanism is in LUAD samples. We identify a large set of deregulated 9-lncs, with aberrant expression observed in RNA-sequencing data from two LUAD cohorts of paired tumor and non-malignant lung tissue samples. We show that these deregulated 9-lncs have clinical value and that the parent genes, many of which are correlated with 9-lnc expression, have been implicated in cancer phenotypes and are associated with clinical outcome Together, our results highlight the important roles of the non-coding transcriptome in cancer cellular biology.

#### MATERIALS AND METHODS

#### Next Generation Sequencing of Lung Adenocarcinoma Patient Samples Discovery Cohort

We performed Next Generation Illumina HiSeq RNA sequencing on a set of 36 micro-dissected LUAD tumors and matched adjacent non-malignant tissue (n = 72). Our British Columbia Cancer Agency cohort (BCCA) was composed of fresh-frozen LUAD tumors and matched non-malignant lung parenchymal tissue collected from 36 patients at the Vancouver General Hospital with approval from the University of British Columbia-BCCA Research Ethics Board. Consent obtained from the tissue donors of this study was both informed and written. To avoid field effects non-malignant samples were collected from areas > 2 cm away from the tumor. In order to reduce contaminating sequences derived from alternative cell types, tissue microdissection was guided by a pathologist. Samples used in this study contained > 80% tumor cell or > 80% nonmalignant cell content. Total RNA was extracted using Trizol reagent and standard procedures.

#### Processing of RNA-Sequencing Data

Total RNA was used for library construction at the Genome Sciences Center (GSC, Vancouver, BC, Canada). Briefly, samples were first analyzed using Agilent Bioanalyzer RNA nanochip, and samples that passed quality check were arrayed into a 96-well plate. PolyA+ RNA was purified using the 96-well MultiMACS mRNA isolation kit on the MultiMACS 96 separator (Miltenyi Biotec, Germany) from 2 µg total RNA with on-column DNase I-treatment as per the manufacturer's instructions. Doublestranded cDNA was synthesized from the purified polyA+-RNA using the Superscript Double-Stranded cDNA Synthesis kit (Life Technologies, United States) and random hexamer primers at a concentration of 5 µM. The paired-end sequencing library was prepared following the GSC paired-end library preparation protocol, which is strand specific. Sequencing was performed using the Illumina HiSeq 2000 platform. Raw sequencing reads were subject to a quality control process. Reads with a length < 50 nt (under two thirds of maximum read length of 75 nt) and quality level (Phred) < 20 were discarded. High quality reads (.fastq files) were aligned to the NCBI GRCh37 reference human genome build using the STAR aligner (v 2.4.1d) under default parameters (Dobin et al., 2013). Aligned reads (.bam files) were quantified using Ensembl Transcripts (Release 75) reference annotations (Flicek et al., 2014). Raw RNA sequencing reads from each patient (tumor and corresponding non-malignant tissue) were deposited at BioProject.<sup>3</sup> Quantification was performed using the Partek Flow platform as reads per kilobase per million (RPKM). RNA sequencing (.bam files) and clinical

<sup>3</sup>http://www.ncbi.nlm.nih.gov/bioproject/516232

data for a secondary set of LUAD tumors and matched nonmalignant tissue (n = 108) were downloaded from The Cancer Genome Atlas (TCGA) Data Portal for validation purposes<sup>4</sup> (**Table 1**). Expression profiles from TCGA were processed as described above.

#### Identification of lncRNAs Expressed From Pseudogene Loci and Corresponding Parent Genes 9-lnc Annotation

Milligan et al. (2016) recently published a global atlas of lncRNAs that have exonic overlap with positionally nonredundant (unique) pseudogenes from 3 major pseudogene databases. Using this resource we obtained a list of lncRNAs overlapping pseudogene loci (**Supplementary Table S1**) that we used as a foundation for our expression analysis. As the degree of sequence overlap required for a pseudogenederived lncRNA to regulate its parent gene is unknown, we did not restrict our analysis to full-length, expressed pseudogenes, and included lncRNAs with any exonic overlap, including sense and antisense transcripts in order to annotate the most comprehensive list of 9-lncs (lncRNAs overlapping pseudogene loci).

#### Parent Gene Annotation

The parent gene information was also extracted for all pseudogenes overlapping our list of 9-lncs that have parentgene annotations in the YaleHuman60 and Retroali5 databases (**Supplementary Table S7**). Manual literature search was performed for parent genes of deregulated 9-lncs that were not contained in these databases.

#### Statistical Analysis Identification of Significantly Deregulated 9-lncs in Paired LUAD and Non-Malignant Lung Tissue

Gene expression for protein-coding and non-coding genes was compared between tumors and non-malignant tissue and significantly deregulated genes were identified using a Wilcoxon signed-rank test (p < 0.05) and subjected to a Benjamini-Hochberg (BH) FDR correction. 9-lncs as identified previously were extracted, and those that were significantly deregulated between tumors and non-malignant tissue, in both our discovery (BCCA) and validation (TCGA) were selected for further analysis (**Figure 1B**).

#### 9-lncs and Parent Gene Expression

Tumors were sorted by 9-lnc expression for each 9-lncparent gene pair, and grouped into top and bottom 9-lnc expressing tertiles. Parent gene expression was then compared between the two groups using the Mann Whitney U-test (pvalue ≤ 0.05). We performed a global expression analysis to determine whether 9-lnc-parent pairs were more positive or negatively correlated than random chance. For all 9-lncs with expression data in both datasets, Spearman's correlation rho values were calculated for 9-lnc-parent gene pairs (n = 390) and compared to 9-lnc-random gene pairs. Random genes in our expression matrices were selected to pair with each 9-lnc. Each gene was assigned a number and pairs were chosen by using a random number generator.<sup>5</sup> Spearman's rho values were then plotted, smoothed, and compared (Mann Whitney U-test, p-value ≤ 0.05). Rho value distribution was also compared between sense lnc-parent gene pairs (n = 208) and antisense lnc-parent pairs (n = 182).

#### Literature Searches

To determine if each gene of interest (9-lnc or parent gene) had been previously described in the context of tumors we searched Pubmed using the terms "gene + cancer" or "gene + lung cancer."

#### Hierarchical Clustering and Data Visualization

Unsupervised hierarchical clustering was performed in order to visualize and examine the expression of the 9-lncs in individual samples (**Figure 2A**). Average Linkage was used as a cluster distance metric, while Pearson Correlation was used as a point distance metric. To visualize the expression patterns of the most highly expressed 9-lncs, those with an average expression value of ≥ 10 RPKM in either tumor or non-malignant samples were included in the analysis.

#### Distribution of Deregulated 9-lncs Across Genome

Locations of 9-lncs and parent genes were compared to identify their genomic position (**Supplementary Table S4**). Circular plot visualization was performed using the R-package Circlize (**Figure 3**; Gu et al., 2014). LUAD-specific regions of significant recurrent somatic copy number alterations had been previously identified by TCGA, and were used in this study to determine if the deregulated 9-lncs overlapped with frequently altered regions (Zack et al., 2013). All genomic coordinates correspond to the NCBI GRCh37 reference human genome build.

#### Clinical Features

#### Survival Analysis

A large public clinical database (Kaplan–Meier Plotter)<sup>6</sup> comprised of 719 LUAD samples was used to determine the association between both protein-coding and non-coding gene expression with patient outcomes. Similar to the BCCA and TCGA cohorts, the patient samples in this 3rd dataset were mostly comprised of Stage I and Stage II tumors (**Table 1**). Of the 104 deregulated 9-lncs, 19 were represented in this database, while 70% of parent genes (72 out of 103) were present. Default settings were used and a log-rank (Mantel–Cox) test was applied to compare survival between groups of tumors with high and low expression of each gene tested, where p < 0.05 was considered statistically significant. The optimal expression cut-off was selected for each gene.

<sup>4</sup>https://portal.gdc.cancer.gov/

<sup>5</sup>https://www.random.org/

<sup>6</sup>http://kmplot.com/analysis/

#### Association With Tumor Stage

fgene-10-00138 March 4, 2019 Time: 19:43 # 12

The majority of tumors for BCCA and TCGA fell into the categories of Stage I and Stage II (**Table 1**). We compared expression of deregulated 9-lncs between Stage I tumors and tumors classified as Stage II and above using a Mann Whitney U-test (p-value ≤ 0.05).

#### AUTHOR CONTRIBUTIONS

GS was responsible for the project design. GS, KE, AS, VM, BM, MP, EM, and WL contributed to data acquisition, data analysis, interpretation of results, and manuscript preparation. WL was the principal investigator of this project. All authors have read, edited, and approved the final manuscript.

#### FUNDING

This study was supported by grants from the Canadian Institutes for Health Research (CIHR FDN-143345) and Scholarships from CIHR. EM is a Vanier Canada Graduate Scholar, and is additionally supported by UBC Faculty of Medicine.

### REFERENCES


Crooke, S. T., Witztum, J. L., Bennett, C. F., and Baker, B. F. (2018). RNA-targeted therapeutics. Cell Metab. 27, 714–739. doi: 10.1016/j.cmet.2018.03.004


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00138/full#supplementary-material

FIGURE S1 | Expression of lncRNA CTC-250I14.3 is significantly associated with Stage I disease in both the BCCA and TCGA LUAD cohorts (Mann Whitney U-test).

TABLE S1 | Detailed information of all pseudogene-derived lncRNAs.

TABLE S2 | Statistical analysis of the deregulated lncRNAs in the BCCA and TCGA cohorts (BHC-corrected p-values).

TABLE S3 | Number of PUBMED entries for lncRNA cancer-association.

TABLE S4 | Genomic locations of significant lncRNAs and their respective parent genes.

TABLE S5 | Expression correlations between the lncRNAs and their resective parent genes in the BCCA and TCGA cohorts.

TABLE S6 | Spearman's correlation coefficients for global expression analysis of ψ-lnc-parent pairs.

TABLE S7 | Parent gene information for pseudogenes contained in Retrolali5 and Yale60 databases.



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

Copyright © 2019 Stewart, Enfield, Sage, Martinez, Minatel, Pewarchuk, Marshall and Lam. 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.

# LncRNAs GIHCG and SPINT1-AS1 Are Crucial Factors for Pan-Cancer Cells Sensitivity to Lapatinib

Zhen Xiang1†, Shuzheng Song1†, Zhenggang Zhu<sup>1</sup> , Wenhong Sun<sup>2</sup> \*, Jaron E. Gifts <sup>3</sup> , Sam Sun<sup>3</sup> , Qiushi Shauna Li <sup>3</sup> , Yingyan Yu<sup>1</sup> \* and Keqin Kathy Li <sup>3</sup> \*

*<sup>1</sup> Department of Surgery of Ruijin Hospital, and Shanghai Institute of Digestive Surgery, Shanghai Key Laboratory for Gastric Neoplasms, Shanghai Jiao Tong University School of Medicine, Shanghai, China, <sup>2</sup> Guangxi Key Laboratory of Processing for Non-ferrous Metal and Featured Materials, Research Center for Optoelectronic Materials and Devices, School of Physical Science Technology, Guangxi University, Nanning, China, <sup>3</sup> Department of Chemistry, Center for Diagnostics and Therapeutics, Georgia State University, Atlanta, GA, United States*

Edited by: *Yujing Li, Emory University, United States*

#### Reviewed by:

*Feng Wang, Emory University School of Medicine, United States Kesheng Wang, East Tennessee State University, United States Haidong Huang, Cleveland Clinic, United States*

#### \*Correspondence:

*Wenhong Sun 2018001@gxu.edu.cn Yingyan Yu yingyan3y@sjtu.edu.cn Keqin Kathy Li kli@epigentic.us*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

Received: *15 November 2018* Accepted: *16 January 2019* Published: *19 February 2019*

#### Citation:

*Xiang Z, Song S, Zhu Z, Sun W, Gifts JE, Sun S, Li QS, Yu Y and Li KK (2019) LncRNAs GIHCG and SPINT1-AS1 Are Crucial Factors for Pan-Cancer Cells Sensitivity to Lapatinib. Front. Genet. 10:25. doi: 10.3389/fgene.2019.00025* Lapatinib is a small molecule inhibitor of EGFR (HER1) and ERBB2 (HER2) receptors, which is used for treatment of advanced or metastatic breast cancer. To find the drug resistance mechanisms of treatment for EGFR/ERBB2 positive tumors, we analyzed the possible effects of lncRNAs. In this study, using CCLE (Cancer Cell Line Encyclopedia) database, we explored the relationship between the lncRNAs and Lapatinib sensitivity/resistance, and then validated those findings through *in vitro* experiments. We found that the expression of EGFR/ERBB2 and activation of ERBB pathway was significantly related to Lapatinib sensitivity. GO (Gene Oncology) analysis of top 10 pathways showed that the sensitivity of Lapatinib was positively correlated with cell keratin, epithelial differentiation, and cell-cell junction, while negatively correlated with signatures of extracellular matrix. Forty-four differentially expressed lncRNAs were found between the Lapatinib sensitive and resistant groups (fold-change > 1.5, *P* < 0.01). Gene set variation analysis (GSVA) was performed based on 44 lncRNAs and genes in the top 10 pathways. Five lncRNAs were identified as hub molecules. Co-expression network was constructed by more than five lncRNAs and 199 genes in the top 10 pathways, and three lncRNAs (GIHCG, SPINT1-AS1, and MAGI2-AS3) and 47 genes were identified as close-related molecules. The three lncRNAs in epithelium-derived cancers were differentially expressed between sensitive and resistant groups, but no significance was found in non-epithelium-derived cancer cells. Correlation analysis showed that SPINT1-AS1 (*R* = −0.715, *P* < 0.001) and GIHCG (*R* = 0.557, *P* = 0.013) were correlated with the IC50 of epithelium-derived cancer cells. In further experiments, GIHCG knockdown enhanced cancer cell susceptibility to Lapatinib, while high level of SPINT1-AS1 was a sensitive biomarker of NCI-N87 and MCF7 cancer cells to Lapatinib. In conclusions, lncRNAs GIHCG and SPINT1-AS1 were involved in regulating Lapatinib sensitivity. Up-regulation of GIHCG was a drug-resistant biomarker, while up-regulation of SPINT1-AS1 was a sensitive indicator.

Keywords: pan-cancer, computational analysis, LncRNAs, lapatinib, targeted therapy

### INTRODUCTION

Lapatinib is a small molecular drug that has been shown to be a dual tyrosine kinase inhibitor, which is involved in the EGFR/HER1 and ERBB2/HER2 pathways and suppresses the autophosphorylation of these receptors. Clinically, it has been used in combination therapy with capecitabine in patients with advanced or metastatic breast cancer that overexpressed ERBB2/HER2 in the cases of previous treatment with anthracyclines, taxanes, or trastuzumab (Herceptin) (Geyer et al., 2006). In addition, a satisfactory response rate has also been found with Lapatinib treatment for ERBB2-positive progressive gastric cancer (Cetin et al., 2014; Satoh et al., 2014). However, in patients with head and neck squamous cell carcinoma, Lapatinib combined with radiotherapy did not show therapeutic effects (Harrington et al., 2015). Similarly, in ERBB2/EGFR positive metastatic bladder cancer patients who underwent first-line chemotherapy didn't get benefit from Lapatinib maintenance treatment (Powles et al., 2017). Therefore, uncovering the drug-resistant mechanism of Lapatinib will help improve the therapeutic effects of Lapatinib targeted therapy and find new sensitive biomarkers.

Long non-coding RNAs (lncRNAs) are a large class of transcribed RNA molecules that are longer than 200 nucleotides but do not encode proteins. In addition to the regulation of diverse cellular processes, such as epigenetics, cell cycle, and cell differentiation, they have been found to play important roles in carcinogenesis, tumor development, and treatment resistance (Heery et al., 2017; Peng et al., 2017; Hahne and Valeri, 2018; Wang et al., 2018; Wu et al., 2018). For instance, Ma et al. found that lncRNAs CASC9 and EWAST1 were two crucial molecules associated to EGFR-TKIs resistant in non-small cell lung cancer (Ma et al., 2017).

The Cancer Cell Line Encyclopedia (CCLE) database (https:// portals.broadinstitute.org/ccle) is an open access resource with the most completely integrated datasets of cancer cells genomes and drug effectiveness. It includes the experimental datasets of drug treatment of 24 kinds of chemical compounds in almost 1,000 cancer cell lines of various human cancers (Barretina et al., 2012). Kim et al. used CCLE database in their recent publication. They found that high levels of FGFR and integrin β3 are resistant to crizotinib treatment, suggesting that FGFR, and integrin β3 could be predictive markers for Met-targeted therapy (Kim et al., 2015). To date, there is a limited number of studies (Jiang et al., 2014; Niknafs et al., 2016; Bester et al., 2018; Li D. et al., 2018; Sun et al., 2018) to explore lncRNAs by CCLE database. In this study, we analyzed the lncRNAs of whole-genome datasets of CCLE after treatment with Lapatinib on pan-cancer cell lines, and proposed crucial lncRNAs GIHCG and SPINT1-AS1 involved in regulating Lapatinib sensitivity.

### MATERIALS AND METHODS

#### Data Extraction From CCLE

There are 5,344 lncRNA probes and 49,331 non-lncRNA probes in the whole-genome gene expression profile chip used in CCLE (Barretina et al., 2012). There are 1,037 cell lines of various cancer types in the database. Among those, 504 cell lines had been treated with Lapatinib and got IC50 (half maximal inhibitory concentration) data and 501 cell lines were examined by microarrays. Since the study focused on solid tumors, we deleted cell lines of hematopoietic and lymphoid cell lines. Finally, 420 solid tumor cell lines were enrolled in the study (**Table 1**).

### Cancer Cell Lines and Cell Culture

Nineteen cancer cell lines were used for validating experiments in vitro. Four of those were gastric cancer cell lines (NCI-N87, SGC-7901, AGS, and MKN-45), three were melanoma cell lines (MuM-2C, MV3, and A-375), three were hepatocarcinoma cell lines (LM3, 97L, and Huh7), three were thyroid cancer cell lines (KHM-5M, CAL-62, and C643), two were breast cancer cell lines (MCF7 and SK-BR-3), two were pancreatic cancer cell lines (TCC-PAN2 and BxPC3), and two were colorectal cancer lines (DLD-1 and NCIH-747). Cell lines NCI-N87, MuM-2C, LM3, MV3, Huh7, SGC-7901, CAL-62, AGS, MCF7, C643, 97L, SK-BR-3, KHM-5M, A-375, TCC-PAN2, MKN-45, and BxPC3 were purchased from the Cell Bank of Type Culture Collection of Chinese Academy of Sciences (Shanghai, China). Cell lines DLD-1 and NCIH-747 were purchased from The Global Bioresource Center ATCC (Maryland, USA). The cell lines were cultured in RPMI-1640 supplemented with 10% fetal bovine serum in a humidified incubator at 37◦C with 95% air and 5% CO2.

### Transient Transfection of siRNAs

SPINT1-AS1 and GIHCG siRNAs were transfected into cancer cells by Lipofectamine 2000 (Invitrogen, Carlsbad, California,



TABLE 2 | Lapatinib IC50 of 420 cancer cell lines.

#### TABLE 2 | Continued


*(Continued)*

#### TABLE 2 | Continued



*(Continued)*

#### TABLE 2 | Continued



*(Continued)*

#### TABLE 2 | Continued




\**Extracted from CCLE database (https://portals.broadinstitute.org/ccle).*

*IC50 (*µ*M) is half maximal inhibitory concentration (IC50), which is defined as a drug concentration producing absolute 50% inhibition of growth in cell proliferation assay. By definition, this metric relies on the assumption, that at a high concentration of the drug, 100% effect is achieved as all cells die in a proliferation assay.*

USA) according to the manufacturer's instructions. The siRNA sequences are shown in Table S1.

#### RNA Extraction and Quantitative Real-Time PCR Analysis

Total RNA was isolated using the TRIzol solution (Invitrogen, California, USA). The cDNA was synthesized using Reverse Transcription kit (TOYOBO, Japan). Real-time PCR was performed in 10 µl reaction mixtures with the HT 7900 (Applied Biosystems, Foster City, USA) using SYBRTM Select Master Mix (Applied Biosystems, Foster City, USA). The sequences of primers were designed and synthesized by Sunny Biotech (Shanghai, China): The primer sequences are shown in Table S1.

#### Cell Viability Assay

Five thousand cells of different cancers were placed in each well of 96-well plates (100 µl/well). Different concentrations of Lapatinib (Selleck, Houston, USA) were incubated for 48 h. After adding 10 µl CCK-8 for 2 h, OD value was measured at 450 nm by spectrophotometry (BioTek, Vermont, USA).

#### Data Analysis

The "corrplot" and "pheatmap" package in R software were utilized for visualizing pearson correlation analysis and cluster analysis by "euclidean" method. The Benjamini and Hochberg method was used to calculate P. adjust value. By means of "clusterProfiler" package in R, GSEA (Gene Set Enrichment Analysis) and GO (Gene Ontology) analyses were carried out to explore involved gene clusters. GSEA is a computational method based on previous publication by Subramanian et al. (2005). GO analysis is a kind of gene enrichment analysis to classify gene set on three aspects: molecular function, cellular component and biological process (Ashburner et al., 2000). Differentially expressed lncRNAs and genes with difference larger than 1.5-fold were obtained by "limma" package, which is often used to explore differentially expressed genes between two phenotypes (Ritchie et al., 2015). The top 10 gene clusters of all cancer cell lines were scored using "GSVA" package (Gene Set Variation Analysis,) in R language, which utilizes non-parametric unsupervised method for evaluating gene set enrichment (GSE) in transcriptomic data (Hanzelmann et al., 2013). Cytoscape software was applied to establish co-expression network and determine hub lncRNAs. The inhibiting ratio and Lapatinib IC50 were calculated according to OD value by GraphPad Prism 6.0 (Inc., La Jolla, CA, USA). The relative RNA levels were calculated by 2−11CT (1CT = LncRNACT value − GAPDHCT value , 11CT= 1CT−1CTmin , 1CTmin: minimum 1CT of expression levels of lncRNA GIHCG or SPINT1-AS1 in cell line). Student's t-tests were performed by GraphPad Prism 6.0. P < 0.05 was considered statistically significant.

#### RESULTS

#### Lapatinib IC50 From Pan-Cancer Cell Lines Analysis

The CCLE data of Lapatinib IC50 of the selected 420 cell lines was shown in **Table 2**. The upper limit of IC50 was originally determined as 8µM for those cancer cell lines in the database. There were 302 cancer cell lines with IC50 higher than 8µM, which were insensitive to Lapatinib drug. There were 118 cancer cell lines with IC50 lower than 8µM, which were relatively sensitive to Lapatinib drug. Taking 8µM of IC50 as a threshold, we categorized 420 cancer cell lines into two groups, high\_IC50 (n = 302) and low\_IC50 (n = 118). Since EGFR and ERBB2 are the targets of the Lapatinib drug, the expression levels of EGFR, and ERBB2 in high\_IC50 and low\_IC50 groups were analyzed. The expression levels of EGFR and ERBB2 were significantly higher in low-IC50 group than in high\_IC50 (**Figure 1A**, P = 0.006 and P < 0.001, respectively). The distribution tendency of 22 types of solid cancer cell lines in high-IC50 (up to 8µM) and low\_IC50 (lower than 8µM) groups is presented in **Figure 1B**. GSEA analysis showed that ERBB pathway-related genes were enriched in low\_IC50 group (**Figure 1C**, ERBB signaling pathway NES = −1.81, P < 0.002, p. adjust = 0.064; regulation of ERBB signaling pathway NES = −1.69, P < 0.002, p. adjust = 0.064).

### Pathway Analysis Involved in Lapatinib Sensitivity

To illustrate the mechanism of Lapatinib resistance, we selected genes with fold-change >1.5 times to perform GO analysis (Table S2). In the top 10 involved pathways, Lapatinib sensitivity was positively associated with cell keratin, epithelial differentiation,

FIGURE 1 | The correlation of mRNA expression levels of EGFR and ERBB2 and Lapatinib IC50. (A) The bar charts of mRNA expression levels of EGFR (left) and ERBB2 (right) of cancer cell lines between the high\_IC50 and low\_IC50 groups of Lapatinib drug. The expression levels of EGFR and ERBB2 are significantly higher in the low\_IC50 group than that in the high\_IC50 group (*p* < 0.01). (B) The distribution tendency of 22 types of solid cancer cell lines in high-IC50 (up to 8µM) and low\_IC50 (lower than 8µM). The red lines represent mean value of Lapatinib IC50. (C) The enrichment analysis of ERBB signaling pathway reveals that ERBB signaling pathway is significantly enriched in Lapatinib low\_IC50 group. "Y" axis indicates the enrichment score (ES) value, and "X" axis indicates genes according to differential expression value between high\_IC50 and low\_IC50 groups. The blue and red dot curves represent ES value. The bottom barcodes represent the leading gene set that strongly contributed to ES value. The positive ES value represents positive correlation to Lapatinib IC50, and minus ES value represents negative correlation to Lapatinib IC50.

and cell-cell junction, while negatively related to signatures of extracellular matrix (**Figure 2**, P < 0.001, P. adjust < 0.001).

#### Analysis of LncRNAs Involved in Lapatinib Sensitivity

We further screened the differentially expressed lncRNAs, and 44 lncRNAs were identified between the high\_IC50 group and low\_IC50 group (**Figure 3A** and **Table 3**, fold-change >1.5, P < 0.01). Then, we selected genes in the top 10 pathways and 44 differential lncRNAs for the construction of the co-expression network. The enrichment scores of the top 10 pathway genes in every cancer cell lines were calculated and determined by GSVA analysis. Five lncRNAs were highlighted as the hub factors in the top 10 regulating pathways (**Figure 3B**). The association of the 5 lncRANs with 199 genes in the top 10 pathways was further analyzed, and a molecular network of coexpression was established, which included top 50 key molecules closely associated to Lapatinib sensitivity. Three crucial lncRNAs, GIHCG, SPINT1-AS1, and MAGI2-AS3, still remained in the co-expression network (**Figure 3C**).

#### Differential Expressing Analysis of Three LncRNAs Between Epithelial and Non-epithelial Cancer Groups

We divided the 420 cancer cell lines into epithelium derived group (n = 278) and non-epithelium derived group (n = 142; including nervous system, bone, cartilage, and pleura). The differential expression levels of the three lncRNAs between the

two groups are presented in **Figure 4A**. In the epitheliumderived group, the differential expression levels of the three lncRNAs between Lapatinib high\_IC50 and low\_IC50 groups were significantly different (**Figure 4B**, P < 0.05). In the nonepithelium groups, there was no significant difference of the three lncRNAs between Lapatinib high\_IC50 and low\_IC50 groups. Higher expressing level of SPINT1-AS1 was found in epitheliumderived cancer cells, and higher expressing levels of MAGI2-AS3 and GIHCG were observed in the non-epithelium group.

Differentially expressed genes (1.5-fold change) between the Lapatinib high\_IC50 and low\_IC50 groups in epithelial group (Table S3) were utilized to perform GO analysis. Enhanced signatures of cell keratin, epithelial differentiation, and cell-cell junction were observed in Lapatinib low\_IC50 group, and decreased signature of extracellular matrix were observed in Lapatinib low\_IC50 group (**Figure 5**, P < 0.001, P. adjust < 0.001).

#### Correlation of LncRNAs SPINT1-AS1, GIHCG, or MAGI2-AS3 and Lapatinib Sensitivity in Epithelial Group

Correlation analysis revealed that Lapatinib IC50 of the nonepithelial group was higher than that of the epithelial group (**Figure 6A**). Of the three critical lncRNAs, SPINT1-AS1, and GIHCG were the lncRNAs most correlated to Lapatinib sensitivity (**Figure 6B**). SPINT1-AS1 and GIHCG were selected as key factors of affecting Lapatinib sensitivity of epithelial cancers. The up-regulation of SPINT1-AS1 was found in low\_IC50 group and increased GIHCG was found in high\_IC50 group (**Figure 6C**).

#### Validating Study of GIHCG and SPINT1-AS1 on Regulating Lapatinib Sensitivity in vitro

In validating experiments, we examined expression levels of GIHCG and SPINT1-AS1 in seven types of cancer cell lines (thyroid cancer, pancreatic cancer, liver cancer, melanoma, gastric cancer, breast cancer, and colorectal cancer) and Lapatinib IC50 of the same cancer cell lines. Correlation analysis showed that higher expression levels of SPINT1-AS1 were significantly associated with lower Lapatinib IC50 (**Figure 7A**, R = −0.715, P < 0.001), while higher expression levels of GIHCG were significantly related to higher Lapatinib IC50 (**Figure 7A**, R = 0.557, P = 0.013).

The sensitive cancer cell lines of NCI-N87 (gastric cancer) and MCF7 (breast cancer), as well as the resistant cancer cell lines of NCIH-747(colon cancer) and BxPC3 (pancreatic cancer)




*log FC, log2 of fold-change. Positive value indicates increased expression in high\_IC50 group, and negative value indicates decreased expression in high\_IC50 group. NA, Not available.*

were selected for a subsequent validating study. After knockingdown expression levels of GIHCG and SPINT1-AS1 by small interfering RNAs, Lapatinib IC50, and inhibitory rate of cancer cells were detected. Among three small interference sequences of GIHCG and SPINT1-AS1 mRNAs, siRNA sequence 3 of GIHCG (Si3, **Figure 7B**), and siRNA sequence 1 of SPINT1- AS1 (Si1, **Figure 7C**) were identified as effective siRNAs for further experiments.

Knocking-down of GIHCG could significantly enhance the sensitivity to Lapatinib in MCF7 and BxPC3 cancer cell lines (**Figure 7D**), while down-regulation of SPINT1-AS1 could promote resistance to Lapatinib in NCI-N87 and MCF7 cancer cell lines (**Figure 7E**). To clarify whether there is a mutual regulatory relationship between GIHCG and SPINT1- AS1, we detected the expression level of SPINT1-AS1 after GIHCG knockdown and vice versa. As shown in **Figures 7F,G**,

suppression of GIHCG in Lapatinib resistant cancer cell lines NCIH-747 and BxPC3 could induce up-regulation of SPINT1- AS1 (P < 0.05), while knockdown of SPINT1-AS1 did not change the expression level of GIHCG (P > 0.05).

### DISCUSSION

LncRNA is an important regulatory molecule in drug resistance during chemotherapy or gene targeted therapy (Li et al., 2016; Dong et al., 2018; Wu et al., 2018; Zhou et al., 2018). In this study, we analyzed Lapatinib sensitivity to EGFR and ERBB2 targeted therapy pan-cancer cell line wide. We noticed that Lapatinib sensitivity was not only positively correlated to the activation of EGFR and ERBB2 signaling pathways, but also positively associated to cell keratin, epithelial differentiation, and cell-cell junction. The Lapatinib sensitivity of cancer cell lines was negatively associated to extracellular matrix signature. By screening differentially expressed lncRNAs and establishing coexpression network between Lapatinib high\_IC50 and low\_IC50 groups, three key lncRNAs, SPINT1-AS1, GIHCG, and MAGI2- AS3, were found. Of those, GIHCG and SPINT1-AS1 were only differentially expressed in epithelial derived cancers. SPINT1- AS1 was negatively related to Lapatinib IC50, whereas GIHCG was positively associated to Lapatinib IC50. By siRNAs treatment, downregulation of SPINTA-AS1 could promote Lapatinib resistance, while downregulation of GIHCG promoted Lapatinib sensitivity. The combination of bioinformatical approach and experimental study confirmed that lncRNAs were involved in regulating sensitivity to Lapatinib targeted therapy.

PI3K/Akt, Ras/Raf/MEK/ERK1/2, and PLCγ pathways are downstream pathways of EGFR and ERBB2 and play important roles in cell proliferation and survival of multiple cancers

(Roskoski, 2014). The expression levels of EGFR and ERBB2 are positively correlated to Lapatinib sensitivity (Rusnak et al., 2007; Xiang et al., 2018). Trastuzumab (Herceptin) is a molecular targeted drug of ERBB2-positive metastatic/advanced breast cancer and gastric cancer (Bang et al., 2010; Loibl and Gianni, 2017). Lapatinib is a small molecule chemical, which proved effective for ERBB2-positive advanced or metastatic breast cancer when combined with capecitabine after previous treatment with anthracyclines, paclitaxel, or trastuzumab (Geyer et al., 2006). In gastric cancer, treatment with Lapatinib plus capecitabine and oxaliplatin also revealed anti-cancer effects on HER2 amplified gastroesophageal adenocarcinoma, especially in Asian and younger patients (Hecht et al., 2016). LncRNAs emerged as one of the new resistance mechanisms to chemotherapy or

molecule targeted therapy. By bioinformatics analysis, Lapatinib sensitive cancer cells exhibited enrichment of genes related to cell keratin, epithelial differentiation, and cell-cell junction. The ERBB family plays an important role in regulating cell differentiation (Pellat et al., 2017). We noticed that Lapatinib sensitivity is positively correlated to ERBB pathway activation. It means that cancer cells sensitive to Lapatinib drug often showed enrichment of cell differentiation-related genes, while Lapatinib-resistant cancer cells are often accompanied by enrichment of extracellular matrix pathway (D'Amato et al., 2015; Khan et al., 2016; Lin et al., 2017; Watson et al., 2018). Furthermore, increases of extracellular matrix could further induce epithelial-mesenchymal transition of cancer cells (Tzanakakis et al., 2018).

in the gene clusters.

Although the role of lncRNAs in cancer progression and Lapatinib resistance have been reported in other studies (Russell et al., 2015; Li et al., 2016; Liang et al., 2018; Ma et al., 2018), this is the first study that proved that lncRNAs GIHCG and SPINT1-AS1 are involved in regulating therapeutic sensitivity to Lapatinib. Based on pan-cancer cell lines analysis, Lapatinib IC50 is significantly different between non-epithelial cancer cell lines, and epithelial cancer cell Lines. As the inhibitor of miR-200b/200a/429, LncRNA GIHCG was shown effectively promoting the progression of liver cancer through inducing methylation of miR-200b/200a/429 promoter (Sui et al., 2016). GIHCG is also involved in promoting cancer proliferation and migration in tongue and renal cancers (D'Aniello et al., 2018; Ma et al., 2018). However, there is no study \$om whether or not GIHCG could regulate Lapatinib drug sensitivity in cancers. LncRNA SPINT1-AS1 is a Kunitz type 1 antisense RNA1, belonging to serine peptidase inhibitor. An increased expression of SPINT1-AS1 has been observed in colorectal cancer (Li C. et al., 2018). It is also the first time that lncRNA SPINT1-AS1 has been found regulating Lapatinib drug sensitivity on multiple cancer cells. In validating experiments, the knockdown of SPINT1-AS1 did not result in the up-regulation of GIHCG. We speculated that GIHCG may regulate SPINT1-AS1 expression through regulating promoter methylation or by manner of competitive endogenous RNA (ceRNA) (Zhang G. et al., 2018; Zhang L. et al., 2018). However, the mutual regulatory mechanisms of lncRNA GIHCG and SPINT1-AS1 remain to be studied in the future.

### CONCLUSION

In conclusion, the current study proposed a group of lncRNAs related to Lapatinib sensitivity based on pan-cancer cell lines analysis. By subsequent experimental study, lncRNAs GIHCG and SPINT1-AS1 were firstly identified as crucial lncRNAs in regulating Lapatinib resistance or sensitivity in epithelium-derived cancer cell lines. SPINT1-AS1 is a Lapatinib sensitivity predictor, while GIHCG is a predictive molecule for Lapatinib resistance.

#### ETHICS STATEMENT

The protocols used in this study were approved by Rui Jin Hospital Ethics Review Boards. Written

informed consents were obtained from all human material donors in accordance with the Declaration of Helsinki. Animals were used according to the protocols approved by Rui Jin Hospital Animal Care and Use Committee.

## AUTHOR CONTRIBUTIONS

KL and YY conceived and designed the experiments. ZX, ShS, ZZ, JG, and QL performed the experiments. ZX, ZZ, SaS, WS, YY, and KL analyzed the data. ZX, ShS, ZZ, SaS, WS, YY, and KL

0.001.

contributed reagents, materials, and analysis tools. ZX, YY, and KL wrote the paper.

#### FUNDING

This project was supported by the National Natural Science Foundation (NSF 81270622 and 81772505), Bagui Talent Foundations (T3120097921, T3120099202, A3120099201, and C31200992001), Innovation Foundation for Key Laboratory of Processing for Non-ferrous Metal and Featured Materials (AE3390003605), National Key R&D Program of China (2017YFC0908300, 2016YFC1303200), China 973 Program

#### REFERENCES


(2013CB733700 and 2011CB510102), Shanghai Science and Technology Committee (18411953100), the Cross-Institute Research Fund of Shanghai Jiao Tong University (YG2017ZD01, YG2015MS62), Innovation Foundation of Translational Medicine of Shanghai Jiao Tong University School of Medicine (15ZH4001, TM201617, and TM 201702), and Technology Transfer Project of Science & Technology Dept. Shanghai Jiao Tong University School of Medicine.

#### ACKNOWLEDGMENTS

We acknowledge open database of CCLE.

or metastatic gastric, esophageal, or gastroesophageal adenocarcinoma: TRIO-013/LOGiC–a randomized phase III trial. J. Clin. Oncol. 34, 443–451. doi: 10.1200/JCO.2015.62.6598


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

Copyright © 2019 Xiang, Song, Zhu, Sun, Gifts, Sun, Li, Yu and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# Molecular Mechanisms in Clear Cell Renal Cell Carcinoma: Role of miRNAs and Hypermethylated miRNA Genes in Crucial Oncogenic Pathways and Processes

Eleonora A. Braga<sup>1</sup> \* † , Marina V. Fridman<sup>2</sup>† , Vitaly I. Loginov1,3, Alexey A. Dmitriev<sup>4</sup> and Sergey G. Morozov<sup>1</sup>

1 Institute of General Pathology and Pathophysiology, Moscow, Russia, <sup>2</sup> Vavilov Institute of General Genetics, Russian Academy of Sciences, Moscow, Russia, <sup>3</sup> Research Center of Medical Genetics, Moscow, Russia, <sup>4</sup> Engelhardt Institute of Molecular Biology, Russian Academy of Sciences, Moscow, Russia

Clear cell renal cell carcinoma (ccRCC) is the third most common urological cancer, and it has the highest mortality rate. The increasing drug resistance of metastatic ccRCC has resulted in the search for new biomarkers. Epigenetic regulatory mechanisms, such as genome-wide DNA methylation and inhibition of protein translation by interaction of microRNA (miRNA) with its target messenger RNA (mRNA), are deeply involved in the pathogenesis of human cancers, including ccRCC, and may be used in its diagnosis and prognosis. Here, we review oncogenic and oncosuppressive miRNAs, their putative target genes, and the crucial pathways they are involved in. The contradictory behavior of a number of miRNAs, such as suppressive and anti-metastatic miRNAs with oncogenic potential (for example, miR-99a, miR-106a, miR-125b, miR-144, miR-203, miR-378), is examined. miRNAs that contribute mostly to important pathways and processes in ccRCC, for instance, PI3K/AKT/mTOR, Wnt-β, histone modification, and chromatin remodeling, are discussed in detail. We also separately consider their participation in crucial oncogenic processes, such as hypoxia and angiogenesis, metastasis, and epithelial-mesenchymal transition (EMT). The review also considers the interactions of long non-coding RNAs (lncRNAs) and miRNAs of significance in ccRCC. Recent advances in the understanding of the role of hypermethylated miRNA genes in ccRCC and their usefulness as biomarkers are reviewed based on our own data and those available in the literature. Finally, new data and perspectives concerning the clinical applications of miRNAs in the diagnosis, prognosis, and treatment of ccRCC are discussed.

Keywords: clear cell renal cell carcinoma, microRNA, target genes, angiogenesis, epithelial-mesenchymal transition, metastasis, hypermethylated miRNA genes, long non-coding RNA

#### Edited by:

Ge Shan, University of Science and Technology of China, China

#### Reviewed by:

Alessio Paone, Sapienza University of Rome, Italy Graziella Curtale, The Scripps Research Institute, United States

> \*Correspondence: Eleonora A. Braga eleonora10\_45@mail.ru

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 15 November 2018 Accepted: 22 March 2019 Published: 24 April 2019

#### Citation:

Braga EA, Fridman MV, Loginov VI, Dmitriev AA and Morozov SG (2019) Molecular Mechanisms in Clear Cell Renal Cell Carcinoma: Role of miRNAs and Hypermethylated miRNA Genes in Crucial Oncogenic Pathways and Processes. Front. Genet. 10:320. doi: 10.3389/fgene.2019.00320

### INTRODUCTION

fgene-10-00320 April 17, 2019 Time: 16:21 # 2

Renal cell carcinoma (RCC) is diagnosed in approximately 90% of patients with kidney cancer. RCC has the highest mortality rate among urogenital cancers (Vasudev et al., 2012). Approximately 270,000 new cases of RCC and 116,000 RCC-related deaths are reported globally every year (Randall et al., 2014). RCC is a heterogeneous group of epithelial tumors, among which clear cell RCC (ccRCC) is the most common and accounts for 70–80% of the reported cases of RCC (Cairns, 2010). The severity of ccRCC is higher than that of papillary kidney cancer and chromophobe tumors (Cairns, 2010). Lack of effective diagnostics in the early stages of the disease, increasing mortality rate, and resistance to therapies in patients with metastatic ccRCC emphasize the need to discover new biomarkers that are applicable for the early diagnosis of ccRCC and detection of metastasis.

Epigenetic regulatory mechanisms involving DNA methylation at the genomic transcriptional level and the interaction of non-coding RNAs, in particular microRNAs (miRNAs), with target messenger RNA (mRNA) at the post-transcriptional level are important in the regulation of genes and proteins (Baylin and Jones, 2016). miRNAs are involved in the regulation of cell proliferation, differentiation, and stress responses, and in the regulation of other fundamental biological processes and signaling pathways (Cora et al., 2017). Alterations in expression and regulatory functions of miRNA can be one of the key factors of various pathogeneses. miRNAs are involved in the development of more than 300 diseases, including oncological diseases. The number of publications aimed at identifying target genes and signaling pathways that involve miRNAs has increased (Reddy, 2015; Dragomir et al., 2018; Gattolliat et al., 2018). miRNAs are important positive or negative regulators of all processes characteristic of the pathogenesis of tumors, which include control of the cell cycle, apoptosis, neo-angiogenesis, tissue invasion, and metastasis (Exposito-Villen et al., 2018; Goradel et al., 2018; Kashyap et al., 2018; Mens and Ghanbari, 2018).

miRNAs are classified as being oncogenic or oncosuppressive (tumor suppressive) based on their stimulating or inhibiting effects, respectively, on tumor development. The targets of oncogenic miRNAs usually include the mRNAs of the tumor suppressor genes. In contrast, the targets of miRNAs that act to suppress tumor development are oncogenes and genes involved in tumor progression. The regulation of tumor suppressor miRNAs involves methylation of the promoter regions of their genes. The methylation suppresses their expression and subsequently inhibits their suppressor function. These processes occur more often in genes encoding tumor suppressor miRNAs than in genes encoding tumor suppressor proteins (Kunej et al., 2011; Piletic and Kunej, 2016). Methylation of tumor suppressor miRNA genes and the interaction of miRNAs with target mRNAs have a systemic effect on key processes and signaling pathways involved in carcinogenesis (Lopez-Serra and Esteller, 2012; Loginov et al., 2015; Baylin and Jones, 2016). Studies of methylation and profiling of miRNA expression have driven the design of minimally invasive diagnostics, and recent advances in the field of cancer epigenomics have highlighted the profound possibilities of these approaches in clinical practice (Li et al., 2015; Xing and He, 2016; Morris and Latif, 2017).

Epigenetic mechanisms involved in ccRCC genesis have received less attention than other cancers, such as colorectal, lung, breast, and prostate carcinomas. Nevertheless, many genes susceptible to hypermethylation have been identified, including VHL, RASSF1A, CDH1, and APAF1, and reported as promising biomarkers of ccRCC (for example, Dmitriev et al., 2014; Braga et al., 2015; Xing and He, 2016; Morris and Latif, 2017). Data on the expression profiles of miRNAs, their target genes in ccRCC, and their potential use in the clinic are accumulating and have been covered in earlier reviews (for example, Li et al., 2015; He Y.H. et al., 2018). However, the contribution of methylation of miRNA genes to epigenetic regulatory mechanisms in ccRCC remains unclear, although interesting results have been reported.

This review examines the expression profiles, targets, and functions of oncogenic and tumor suppressor miRNAs, their role in molecular mechanisms and signaling pathways of ccRCC, as well as their clinical potential. In addition, based on data obtained in our studies and the published literature, we consider in detail the advances made in studies of hypermethylated miRNA genes in ccRCC and their usefulness as biomarkers.

Several long non-coding RNAs (lncRNAs) have been recently described in RCC, as has their involvement in regulatory interactions as competing endogenous RNA (ceRNA), whereby lncRNAs can act as miRNA sponges to modulate gene expression (Li et al., 2017; Chan and Tay, 2018). Moreover, integrative analyses have enabled construction of a ccRCC regulatory ceRNA network comprising 89 lncRNAs, 10 miRNAs, and 22 mRNAs (Yin et al., 2018). In general, however, very few publications have addressed lncRNAs in relation to ccRCC (60 papers on the topic were found indexed in PubMed as of January 2019). In this review, we also consider some data on lncRNAs and their interactions with miRNAs that are relevant to ccRCC.

#### ONCOGENIC AND SUPPRESSIVE MIRNAS IN MAIN SIGNALING PATHWAYS OF CCRCC

Expression microchips and quantitative PCR have been used to identify miRNAs with decreased expression levels in kidney tumors, presumably oncosuppressor miRNAs, and with increased expression levels in ccRCC, presumably oncogenic miRNAs (for example, Li et al., 2015). The expression of the miRNA genes is generally evaluated by determining the levels of mature miRNA, which is very stable and thus an effective biomarker (Andorfer et al., 2011).

Tumor suppressor miRNAs include miR-34a (Yamamura et al., 2012; Zhang C. et al., 2014; Toraih et al., 2017), miR-30c (Huang J. et al., 2013), miR-30d (Wu et al., 2013), miR-99a (Cui et al., 2012), miR-133a (Kawakami et al., 2012), miR-133b (Wu et al., 2014), miR-138 (Ding et al., 2018), miR-141 (Li W. et al., 2014), miR-143 (Yoshino et al., 2013), miR-182 (Kulkarni et al., 2018), miR-187 (Zhao et al., 2013), miR-199a (Qin et al., 2018), miR-200c (Ding et al., 2018), miR-205 (Xu et al., 2018), and others. These miRNAs target mRNAs of oncogenes

or genes encoding proteins mediating the progression of kidney tumors. For example, the targets of the crucially important miR-34a are c-MET (Toraih et al., 2017), c-MYC (Yamamura et al., 2012), and NOTCH1 (Zhang C. et al., 2014) oncogenes. All are involved in the proliferation and activation of the cell cycle. Many targets of suppressor miRNAs are associated with invasion and migration. For example, matrix metalloproteinase 9 (MMP-9) is a validated target of miR-133b (Wu et al., 2014). VIM, EZH2, ZEB2 (Ding et al., 2018), and HIF1A (Song et al., 2011) are regulated by miR-138.

Typical oncogenic miRNAs include miR-7 (He H. et al., 2018), miR-21 (Li X. et al., 2014; Chen J. et al., 2017; Cui et al., 2017), miR-155 (Ji et al., 2017), miR-590-5p (Xiao et al., 2013), and others. Their targets, on the contrary, are tumor suppressor genes, which include PDCD4 (Li X. et al., 2014), PTEN (Cui et al., 2017), TIMP3 (Chen J. et al., 2017), FOXO3A (Ji et al., 2017), PBRM1 (Xiao et al., 2013), non-coding MEG3 (He H. et al., 2018), and others.

Modern methodologies have provided a great deal of information concerning the miRNA expression profiles in ccRCC (He H. et al., 2015; Wotschofsky et al., 2016). Sixty-three differentially expressed miRNAs have been identified by analyzing the massive sequencing data published in The Cancer Genome Atlas (Liang et al., 2017).

To date, a substantial amount of data has been obtained on the role of miRNA in the regulation of target genes in ccRCC and its pathogenesis (Jung et al., 2009; Redova et al., 2011; Heinzelmann et al., 2014). A study of miRNA and mRNA gene networks constructed on the basis of their expression profiles in ccRCC reported a key role for miR-106a-5p. The loss of this miRNA led to the increased expression of 49 putative targets (Müller and Nowak, 2014). Other miRNAs implicated in this study were miR-135a-5p (32 targets), miR-206 (28 targets), miR-363-3p (22 targets), and miR-216b (13 targets) (Müller and Nowak, 2014). The targets included genes that affect apoptosis, metastasis, cellular mobility, and oncogenes (c-MET, VEGFA, NRP2, and FLT1). A similar study (Butz et al., 2015) pinpointed miR-124-3p, miR-30a-5p, and miR-200c-3p as the most significant miRNAs affecting protein expression in ccRCC; the expressions of these miRNAs were often reduced.

Notably, since the von Hippel Lindau (VHL) gene has an important role in both familial and sporadic kidney cancer, the genes associated with VHL-dependent regulation are important candidates in studying the spectrum of miRNAs that alter VHL expression (Neal et al., 2010). Loss of VHL function results in constitutive activation of the hypoxia-inducible factor (HIF) pathway, which leads to hypoxia and subsequent expression of angiogenic factors. Oncogenic miRNAs that enhance the development of hypoxia and angiogenesis and their targets have been identified. The next section of this review is devoted to a detailed look at the VHL/HIF pathways and the involved miRNAs.

The significance of the phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT)/mechanistic target of rapamycin (mTOR) pathway (Joosten et al., 2018) associated with the escape from apoptosis, growth, and proliferation of cells in kidney cancer must be noted. The Cancer Genome Atlas Research Network indicates frequent mutations in the genes of this pathway in ccRCC. The PI3K/AKT/mTOR pathway can also stimulate HIF through mTOR modulation (Cancer Genome Atlas Research, 2013).

miR-148a targets AKT2, and this interaction leads to the suppression of cell growth, colony formation, migration, and invasion and tumor growth in xenografts. The miR-148a level is decreased in RCC and is inversely correlated with tumor size, stage, and metastasis to lymph nodes (Cao et al., 2017). In cultured RCC cells, the inhibition of FLOT1 by miR-182-5p reduces the phosphorylation and activation of AKT2 and subsequent phosphorylation of FOXO3a (Xu et al., 2014). Oncosuppressor and transcription factor FOXO3A is translocated out of the nucleus upon phosphorylation by proteins such as Akt (protein kinase B). FOXO3a decreases proliferation and arrests cells in the G1 phase. The level of miR-182-5p is often reduced in RCC (Xu et al., 2014).

On the contrary, miR-122 mimic increases the level of phosphorylation of AKT2 and mTOR. In addition to its enhanced expression in RCC, miR-122 reduces the expression of the Sprouty RTK signaling antagonist 2 (SPRY2) gene. The product of this gene is the inhibitor of Ras/MAPK signaling pathway (Wang Z. et al., 2017). Thus, while miR-148a and miR-182-5p act as tumor suppressors (Fan Y. et al., 2016; Cao et al., 2017), miR-122 acts as an oncogene (Lian et al., 2013).

The expression of miR-99a was first shown to be downregulated in RCC tumor tissues and correlated with poor survival (Cui et al., 2012). The authors also reported that the use of its mimic suppressed cell growth, migration, invasion in vitro and in vivo, and induced cell arrest at the G1 phase (Cui et al., 2012). However, in contrast to these results, which indicated tumor suppression activity of miR-99a, a recent study (Oliveira et al., 2017) described the upregulation of miR-99a and downregulation of its target gene mTOR in most of the ccRCC samples examined. These findings indicated the oncogenic activity of miR-99a. Both studies implicated mTOR as the direct target of miR-99a in renal cancer cells. Further studies are needed to definitively determine the suppressive or oncogenic activity of miR-99a in ccRCC.

miR-144 also targets mTOR in RCC. The reduced expression of miR-144 has been inversely correlated with the stage and size of the tumor, while increased miR-144 expression can suppress cell growth and arrest cells in the G1 phase (Xiang et al., 2016). However, another study (Xiao et al., 2017) identified miR-144-3p as an oncomiRNA. The authors described that the overexpression of miR-144-3p promoted proliferation, migration, invasion, and chemoresistance in ccRCC cells. They also identified AT-rich interactive domain 1A (ARID1A) as a direct target gene of miR-144-3p in ccRCC. ARID1A is a transcription regulator that is a part of the SNF/SWI remodeling complex. The data to date highlight the lack of clarity concerning the role of miR-144 in ccRCC. Further in-depth studies are required.

miR-137 expression is reduced in RCC and miR-137 suppresses the activation of the PI3K/AKT signaling pathway. This leads to reduced proliferation, migration, and invasiveness of cells, enhanced apoptosis, and suppressed tumor growth in

xenografts (Zhang and Li, 2016). The lncRNA SNHG1 also negatively regulates this miRNA in RCC (Zhao S. et al., 2018).

The target of most oncogenic miRNAs in ccRCC is the tumor suppressor phosphate and tensin homolog (PTEN) gene, which is the negative regulator of the PI3K/AKT/mTOR pathway. In particular, PTEN directly interacts with miR-23b (Zaman et al., 2012), miR-193-3p (Liu L. et al., 2017), miR-21 (Cui et al., 2017), and miR-22 (Fan W. et al., 2016), although miR-22 acts as a tumor suppressor. Appropriate interactions that affect the expression of PTEN may be significant in the development of the disease. Aggressive ccRCC specimens are characterized by a reduced level of PTEN (Cancer Genome Atlas Research, 2013). There are some indications that miR-193a-3p and miR-224 can affect the PI3K/AKT pathway by targeting the ST3 beta-galactoside alpha-2,3-sialyltransferase 4 (ST3GalIV) gene in RCC (Pan et al., 2018a).

The influence of some miRNAs on the PI3K/AKT pathway is depicted in **Figure 1**.

It is believed that the effect of the reduced expression of miR-124-3p in ccRCC is due to an increase in the expression of its putative targets, CAV1 and FLOT1, which are cytoplasmic membrane proteins associated with caveolae-mediated endocytosis (Butz et al., 2015). CAV1 expression causes the AKT-dependent formation of lamellipodia, which increases cell migration and invasion. In addition, it activates the Ras-extracellular signal-regulated kinase (Ras-ERK) pathway and suppresses anoikis, a form of programmed cell death specific to non-malignant cells induced by the detachment of anchorage-dependent cells. FLOT1 is also associated with endocytosis and activation of epidermal growth factor receptor, and subsequent activation of MAPK (Butz et al., 2015).

Studies have also shown the importance of the Wnt/β-catenin signaling pathway in RCC and indicated that many WNT antagonist genes are inhibited by methylation of their promoters in ccRCC (Joosten et al., 2018). In ccRCC, miR-106b-5p can activate this pathway, having as its targets the following negative regulators of the Wnt/β-catenin pathway: LZTFL1, SFRP1, and DKK2 (Lu et al., 2017). Thus, being an oncogene, miR-106b-5p supports the aggressiveness and stemness phenotypes of ccRCC (Lu et al., 2017). miR-1260b also targets DKK2 and stimulates the

proliferation and invasiveness of RCC cells, and its expression is enhanced in RCC (Hirata et al., 2013).

miR-203a targets the glycogen synthase kinase-3β (GSK3β) gene (Hu et al., 2014), whose product is involved in β-catenin degradation (Joosten et al., 2018). Thus, miR-203a also acts as an oncogene in RCC, being a predictor of poor prognosis (Hu et al., 2014). IGF-II mRNA binding protein 1 (IGF2BP1), which participates in the Wnt/β-catenin pathway, is a target of miR-372 in the RCC cell lines, thus miR-372 acts as a tumor suppressor (Huang et al., 2015).

The roles of miRNAs in important processes in ccRCC, which include cell migration, invasion, and proliferation, were highlighted in a recent review (He Y.H. et al., 2018).

Notably, mutations in known oncogenes and tumor suppressors, such as RAS, TP53, RB, and PTEN, are not typical of RCC (Joosten et al., 2018). However, sporadic mutations occur with high frequency in the genes encoding proteins associated with histone modification and chromatin remodeling. These genes include SETD2 (3–12%), KDM5C (3–8%), KDM6A (1%), BAP1 (8–15%), and PBRM1 (21–41%) (Cancer Genome Atlas Research, 2013). miRNA-mediated repression also contributes to the altered expression of these genes in RCC. For example, miR-106a-5p, which targets SETD2, functions as an oncogene in this case. SETD2 is instrumental in suppressing and halting the cell cycle at the transition from G0 to G1, and also inhibits proliferation and stimulates apoptosis in RCC (Xiang et al., 2015). SETD2 increases the binding of H3K36me3 to the p53 promoter that enhances the expression of p53. In ccRCC, miR-590-5p acts as an oncogene and its PBRM1 target acts as a tumor suppressor (Xiao et al., 2013).

Immune checkpoint molecules play an important role in the control of carcinogenesis. In ccRCC, the B7 homolog 3 (B7-H3) immune checkpoint molecules are validated targets of miR-187. This miRNA exhibits the characteristics of a tumor suppressor, and its concentration is reduced in both tumor tissue and plasma that provides opportunities for use in diagnosis of patients with ccRCC. A low level of miR-187 is associated with advanced stages of the disease and a worse 5-year survival rate. In vitro, it reduces cell proliferation and mobility and suppresses tumor growth (Zhao et al., 2013). Non-classical human leukocyte antigen G (HLA-G) inhibits the cytotoxic activity of T-lymphocytes and natural killer cells. HLA-G is often expressed in RCC and is a target of miR-148a and miR-133a, and overexpression of these miRNAs suppresses its effect (Jasinski-Bergner et al., 2015).

It is interesting to note the dual functions of some miRNAs in ccRCC. For instance, one of the targets of miR-106a-5p is the P21 (RAC1) Activated Kinase 5 (PAK5) gene. PAK5 is a serine/threonine kinase (activated by RAC) that regulates cytoskeleton dynamics, cell proliferation, and survival. miR-106a-5p is reportedly a tumor suppressor that reduces metastasis in xenografts and reduces migration and invasiveness of RCC cell lines (Pan et al., 2017). These data contradict the studies that reported oncogenic effects of miR-106a-5p, such as its ability to target SETD2 and indirectly downregulate TP53 (Xiang et al., 2015) and VHL protein levels (Oliveira et al., 2017). The contradictory nature of the influence of miR-106a-5p on ccRCC is summarized in **Figure 2**.

Thus, some miRNAs (for example, miR-106a and miR-144) can show both tumor suppressor and oncogenic properties depending on the context, i.e., their effect is apparently dependent on the target gene and pathways involved. However, some contradictory data, for instance, for miR-99a (Cui et al., 2012; Oliveira et al., 2017), may be the result of inconvenient methodology or inadequate sampling of examined tumors.

The next few sections of this review will be devoted to discussing the regulation of angiogenesis and epithelial-mesenchymal transition (EMT), given the great importance of these processes in RCC.

### MOLECULAR MECHANISMS IN HYPOXIA AND ANGIOGENESIS IN CCRCC: THE ROLE OF MIRNA

In most clear cell carcinomas of the kidney, the expression of the VHL gene is dysregulated in a similar manner (98% of cases feature gene deletion, with hypermethylation and point mutations in the remaining 2%). VHL binds HIF1A and degrades it only in the presence of oxygen. In the absence of oxygen, HIF1A dimerizes with HIF1B and leads to the activation of transcription of more than 40 genes, including the vascular endothelial growth factor (VEGF) family and erythropoietin (EPO) genes, genes encoding growth factors (e.g., EGF, PDGF) and glycolytic enzymes, and genes involved in glucose metabolism (e.g., GLUT1/4) (Roskoski, 2017). In the presence of VHL mutations downregulating its expression, the aforementioned regulatory influences of VHL on HIF1 are disrupted. Inactivation by oxygen is also a characteristic of HIF2A (Serocki et al., 2018). The VEGF-A protein of the VEGF family stimulates the dimerization and activation of the VEGF receptor 2 (VEGFR2). This interaction regulates, in particular, the angiogenesis process, ensuring the supply of oxygen and nutrients to the tumor and encouraging metastasis. The activation of VEGFR2, in turn, leads to the activations of the ERK1/2, AKT, Src, and p38

FIGURE 2 | Multiple features of the effects of miR-106a-5p in ccRCC signaling pathways. The overwhelming effect of miR-106a-5p on tumor suppressors TP53 and VHL and oncogene PAK5 is presented. ↑ indicates activity stimulation and > indicates activity suppression according to any of the mechanisms.

MAPK pathways, which are important for tumor development and metastasis. These activations occur through a cascade of phosphorylation events involving the tyrosine moiety of various proteins, which further recruit phosphotyrosine-binding proteins. These pathways are associated with key angiogenesis events that include cell proliferation, migration, survival, and vascular permeability (Roskoski, 2017). In patients with ccRCC being treated with VEGF and mTOR blockers, angiogenesis can be stimulated by angiopoietin 2, c-MET, or interleukin (Malouf et al., 2016).

While HIF1 responds to acute hypoxia, the adaptation to chronic hypoxia is mediated by the expression of HIF2 and HIF3 in endothelial tissues, while the expression of HIF1 is downregulated. miRNAs play a significant role in this adaptation. While there is a significant overlap of HIF1 and HIF2 targets, the role of HIF3 is much less clear. It has at least six isoforms. One suppresses the activity of HIF1 and HIF2, while others may activate the transcription of a number of genes (Serocki et al., 2018).

In response to prolonged hypoxia, 786-O, a VHL-defective cell line of ccRCC, displays upregulation of the expression of L1CAM and FBN1 genes. In addition to the downregulated expression of miR-100 and miR-378, 786-O hypoxic cells have downregulated expression levels of AUTS2, MAPT, AGT, and USH1C genes. Expression of miR-100 and miR-378 was also found to be significantly reduced in bone metastases of ccRCC (Chen S.C. et al., 2017).

In ccRCC, the response to hypoxia and metastasis is also modulated by the androgen receptor (AR). AR increases the expression of miR-185-5p, which targets VEGF-C (a gene encoding lymphangiogenetic factor). At the same time, binding of miR-185-5p to the promoter region of HIF2A mRNA increases HIF2A expression and the subsequent expression of VEGFA (an angiogenetic factor). This explains why AR-positive ccRCC metastasizes to the lung rather more often than to lymph nodes (Huang et al., 2017).

In RCC patients, especially those diagnosed with ccRCC and VHL disorders caused by familial mutations, disruption of signaling pathways inhibited by tyrosine kinase inhibitors and other related inhibitory drugs is most effective (Mickley et al., 2015). The possibility of using drugs based on miRNA mimics and anti-miR is also being considered for these purposes (Schanza et al., 2017; Serocki et al., 2018).

As mentioned before, miRNAs contribute significantly to regulation of the main hypoxia response pathways. At least 40 miRNAs affect the expression of HIF alone in various tissues, including miR-687 in embryonic renal tissue and miR-429 and miR-19a in endothelial cells (Serocki et al., 2018). Upregulation of miR-106a and miR-106b and downregulation of their target gene VHL has been described for ccRCC (Oliveira et al., 2017).

VEGFR2 is the putative target of miR-221 in metastatic RCC (Khella H.W.Z. et al., 2015). Accordingly, poor survival outcome in patients treated with the receptor tyrosine kinase inhibitor sunitinib correlates with a high level of miR-221 accompanied by low levels of VEGFR2. The effect of miR-221 is different for endothelial and tumor cells; it lowers proliferation and angiogenesis in human umbilical vein endothelial cells

(HUVECs) but increases the proliferation of ACHN kidney cancer cells (Khella H.W.Z. et al., 2015).

Responses to hypoxia are also influenced by lncRNA. Antisense transcripts of the 3<sup>0</sup> and 5<sup>0</sup> non-coding regions of HIF1A are 3<sup>0</sup> aHIF-1α and 5<sup>0</sup> aHIF-1α, respectively. In human lung epithelial cells, the expression of these lncRNAs is induced by hypoxia, and they can suppress the expression of HIF1A (Uchida et al., 2004). In addition, increased expression of the HOX transcript antisense intergenic RNA (HOTAIR) lncRNA is observed in RCC, and it correlates with the tumor progression. It has a direct inhibitory effect on miR-217, the target of which is HIF1A. Suppression of miR-217 reduces apoptosis, increases proliferation and migration of tumor cells, and stimulates EMT in RCC (Hong et al., 2017).

In RCC and ovarian cancer, miR-192 has anti-angiogenic properties due to the effect on its EGR1 and HOXB9 targets (Wu et al., 2016). In lung cancer, EGR1 binds to the VEGFA proximal promoter, activating its expression (Shimoyamada et al., 2010). The effect of HOXB9 on vascularization in breast cancer has been documented (Seki et al., 2012). Moreover, low levels of miR-192 in various malignant tumors are a predictor of poor prognosis (Wu et al., 2016).

The influence of some miRNAs on the VHL/HIF pathway is summarized in **Figure 3**.

In turn, the expression of certain miRNAs is affected by the proteins of the VHL-HIF-VEGF pathway (for example, Schanza et al., 2017). Thus, knockdown or genetic elimination of the VHL gene in RCC cells leads to an increase in miR-28-3p expression. The target of this miRNA is the gene for mitotic spindle checkpoint protein (Mad2), which blocks cells with unformed spindles from entering mitosis. Dysregulation of Mad2 expression leads to chromosomal instability (Hell et al., 2014). The expression of miR-204 in ccRCC cells is, on the contrary, induced by VHL. The target of miR-204 is MAP1LC3B (LC3B), which is associated with macro-autophagy (Mikhaylova et al., 2012). LC3B-mediated macro-autophagy is necessary for nutrient supply to the tumor during the period of metabolic

stress and is important for cell survival, tumor growth, and increased aggressiveness (Sato et al., 2007). At the same time, hypoxia in RCC suppresses miR-30c under the influence of HIF (Huang J. et al., 2013).

As a result of the interaction of different proteins of the VHL/HIF/VEGF pathway, along with the participation of miRNAs, positive and negative feedback loops may appear that prolongs the initial effect of hypoxia.

In mice, renal ischemia/reperfusion procedure was reported to reduce the concentration of 40 miRNAs by almost half, while the concentration of 36 other miRNAs more than doubled (Liu et al., 2012). Quantitative RT-PCR performed 4–12 h after the procedure confirmed an increase in miR-210 concentration. In HUVEC-12 cells, angiogenesis is stimulated by increases in the expression of VEGF and VEGFR2, which are triggered by an increased miR-210 concentration (Liu et al., 2012). Since the target of miR-210 in chondrocytes and hepatocellular carcinoma cells is HIF3A, it is highly likely that this miRNA may regulate the switch between acute hypoxia and chronic hypoxia responses in these cells (Serocki et al., 2018). HIF1A may regulate miR-210 and has been shown to enhance its expression in RCC (Nakada et al., 2011). HIF1A-mediated enhancement of miR-210 expression was also shown in breast cancer, while HIF2A showed no regulatory effect (Camps et al., 2008). However, in 786-O RCC cells, HIF2A also influenced miR-210 expression (Zhang Z. et al., 2009). Increased expression of miR-210 is most typical for ccRCC (Valera et al., 2011).

A feedback loop has been constructed for the cardiomyocytes. HIF1 increases transcription of the MIR-21 gene, and miR-21, in turn, induces the expression of HIF1A via the regulation of PTEN/AKT signaling pathway (Liu et al., 2014). The expression of miR-21 is also increased in oral squamous cell carcinoma (OSCC). It is highly concentrated in the exosomes of OSCC cells and enhances the migratory and invasive properties of tumor cells in a mechanism that is dependent on HIF1A and HIF2A (Li et al., 2016).

During hypoxia, exosomes containing miR-135b are formed in multiple myeloma cells. Suppression of the HIF1AN by miR-135b in HUVEC cells leads to increased HIF1 activity and angiogenesis (Umezu et al., 2014). This mechanism remains turned on even after the restoration of the normal supply of oxygen, which prolongs the antihypoxic response.

Enhanced miR-18a expression after 24 h of hypoxia was observed in human choroidal endothelial cells (Han et al., 2015). Since HIF1A mRNA is the direct target of this miRNA, this interaction may act as the switch between the responses to acute hypoxia and chronic hypoxia (Han et al., 2015).

Hypoxia suppresses the expression of DICER, which is associated with miRNA processing. However, it is not understood how this does not lead to the total disruption of miRNA activity (Serocki et al., 2018). Some miRNAs, such as the tumor suppressor miR-182-5p, are more sensitive to hypoxic suppression of DICER expression than others. In addition to the aforementioned effect on the AKT pathway, miR-182-5p acts as an inhibitor of its target HIF2A, even though it is primarily manifest in VHL-deficient ccRCC cells (Fan Y. et al., 2016). The miR-200 family is another example of miRNAs that are sensitive

to disruption of DICER expression. DICER is the direct target of miR-122, a miRNA that is highly expressed in ccRCC, while the loss of DICER decreases the expression of the miR-200 family, leading to EMT (Fan Y. et al., 2018).

Most tumor cells respond to hypoxia by shifting their energy metabolism to glycolysis (the Warburg Effect). Many miRNAs participate in this shift, which is a characteristic of many solid tumors (Morais et al., 2017). In particular, in RCC, miR-1291 directly inhibits the expression of the GLUT1 glucose transporter gene, and its levels in tumor tissue are reduced as compared to the levels in normal tissue (Yamasaki et al., 2013). In ccRCC, hypoxia-induced miR-101 expression can be a mechanism for activating glycolysis, as miR-101 inhibits the TIGAR (TP53-Induced Glycolysis and Apoptosis Regulator) gene, which is a member of the p53 pathway that is involved in mediating the exchange to the pentose phosphate pathway (Xu et al., 2017). However, TIGAR-mediated regulation of glycolysis has an anti-apoptotic effect, protecting against DNA damage.

lncRNAs also affect the transition to glycolysis (Serocki et al., 2018). In cell cultures obtained from solid tumors, HIF1A-inducible long intergenic non-coding (linc) RNA-p21 was shown to bind to HIF1A, disrupting its interaction with VHL and attenuating the ubiquitination of HIF1A. As a result, both HIF1A and lincRNA-p21 accumulate that causes a transition to glycolysis (Yang et al., 2014).

### METASTASIS AND EMT IN KIDNEY CANCER: THE ROLE OF MIRNA

The incidence of metastasis is high in ccRCC (25–30%) and increases to over 50% after surgery (Rydzanicz et al., 2013). Metastatic ccRCC is extremely resistant to therapy, with responses to chemo-, radio-, and immunotherapy observed in no more than 10% of patients, with 5-year survival rates as low as 9% in the presence of distant metastases (Randall et al., 2014).

Metastasis of primary tumors occurs in several stages – from changes in biochemistry, morphology, and migration abilities of tumor cells to the appearance of surface receptors that provide directed migration to target organs, followed by the formation of specific microenvironment in the target organ, into which metastatic cells can enter and survive.

Epithelial-mesenchymal transition is an important stage in metastasis, during which changes occur in the properties of cancer cells. The changes contribute to metastasis. The process of EMT is characterized by a loss of cell polarity and intercellular bonds, as well as an increase in the migration activity of cells and their invasive properties. A critical event in EMT is the loss of E-cadherin and an increase in the activity of its transcriptional repressors, which include ZEB1, ZEB2, TWIST, SNAIL, and SLUG (Piva et al., 2016). EMT causes the development of sarcomatoid ccRCC (the most unfavorable form of cancer) due to the switched expression of E-cadherin to N-cadherin, dissociation of β-catenin from the membrane, and increased expression of SNAIL and SPARC (Conant et al., 2011).

An evaluation of the expression of 11 EMT markers in nephrectomized samples of RCC revealed that E-cadherin, clusterin, TWIST, and vimentin levels were significant predictors of recurrence (Harada et al., 2012).

Wilms' tumor, WT1, causes mesenchymal-to-epithelial transition (MET) in kidneys during their development. During development of other organs, such as the heart, this gene causes EMT. WT1 is absent in normal renal tissue but is expressed in ccRCC due to the decreased expression of VHL or expression of mutated VHL. WT1 expression in ccRCC in the presence of SNAIL and E-cadherin starts a "hybrid" process in which EMT and MET traits are simultaneously expressed (Sampson et al., 2014). The loss of VHL activity also stimulates EMT, thereby increasing the expression of HIF1A, which subsequently activates nuclear factor-kappa B (NF-κB). The cytoplasmic expression of NF-kB has been correlated with the invasiveness of ccRCC (Kankaya et al., 2015).

Chronic oxidative stress-stimulated malignant transformation of kidney cells is accompanied by their acquisition of stem characteristics and EMT (Mahalingaiah et al., 2015). This leads to a significant increase in the expression of genes such as BCL2, CCND1 (CyclinD1), BIRC5 (Survivin), and PCNA, as well as VIM (Vimentin), ACTA2 (α-Smooth muscle actin, α-SMA) and SNAIL1, as well as a significant decrease in the expression of CDH1 (E-cadherin), cytokeratin, and CTNNB1 (β-catenin). The suppression of OCT4 and SNAIL with small interfering RNA (siRNA) can partially reverse these changes (Mahalingaiah et al., 2015). The effect of hypoxia on EMT can also be partially achieved by reducing the expression of miR-30c and increasing the expression of its target, SLUG, which also leads to a decrease in E-cadherin expression and stimulation of cell migration in RCC (Huang J. et al., 2013).

Downregulation of FOXO3a in ccRCC also leads to an increase in expression of SNAIL that stimulates EMT (Ni et al., 2014). In ccRCC, FOXO3a is a target of oncogenic miR-155, which increases proliferation, colony formation, migration, and invasion (Ji et al., 2017). SNAIL1 in ccRCC is the target of miR-30e-3p (minor variant), which inhibits invasion and migration of tumor cells (Wang D. et al., 2017).

The level of miR-720 is elevated in RCC, while its downregulation suppresses tumor growth (Bhat et al., 2017). miR-720 in RCC targets CTNNA1 (αE-catenin) and CDH1 (E-cadherin), thereby stimulating cell migration and invasiveness.

ZEB2 suppresses the expression of E-cadherin. In RCC, ZEB2 enhances the migratory and invasive ability of cells and the expression of ZEB2 correlates with more advanced forms of the disease and worse survival rates (see the review by Piva et al., 2016). One of the targets of miR-141 (a member of the miR-200 family) in RCC is ZEB2 and thus it can suppress EMT (Li W. et al., 2014). Reduced expression of miR-141 has been correlated with a poor response to sunitinib (Berkers et al., 2013). In addition, in RCC, miR-141 binds to the lncRNA HOTAIR causing its cleavage by the Argonaute 2 (Ago2) complex (Chiyomaru et al., 2014). As already mentioned, HOTAIR is also capable of stimulating EMT (Hong et al., 2017). The high expression of miR-122 in ccRCC stimulates EMT as a result of the targeting of DICER

and OCLN (occludin), which plays an important role in tight junctions, which are characteristic of renal tissue. The loss of tight junctions leads to an increase in cell mobility (Jingushi et al., 2017). Downregulated expression of DICER results in the suppression of the maturation of the miR-200 family.

In RCC, three miRNAs – miR-192 (which has anti-angiogenic effects), miR-194, and miR-215 – target ZEB2, MDM2, and TYMS oncogenes (Khella H.W. et al., 2013). miR-30a is specifically downregulated in metastatic tumors compared to its expression in primary tumors (Butz et al., 2015). Other studies on ccRCC have implicated the ZEB2 and glucose-regulated protein 78 (GRP78) genes as targets of miR-30a (Chen Z. et al., 2017; Wang C. et al., 2017).

Expression of the transcription factor SOX4, which is normally involved in embryogenesis, is limited to a small number of cell types in which the mature cells retain stem cell characteristics. Due to the influence of SOX4 on the Wnt/β-catenin, Notch1, and p53 pathways, as well as the components of the miRNA processing machinery (DICER, Argonaute 1, RNA helicase A), upregulation of the expression of SOX4 can cause the development of different types of cancer and stimulate EMT. In contrast, the high expression of SOX4 may be associated with a more favorable disease course in other types of cancer. In different cancer types, many miRNAs target SOX4. They usually act in an anti-oncogenic manner and are characterized by reduced expression in ccRCC (Geethadevi et al., 2018). In ccRCC, miR-138 has a tumor suppressor role via the targeting of SOX4. The effects include the inhibited proliferation, migration, and invasion of tumor cells, increased expression of E-cadherin, and decreased expression of vimentin (Liu F. et al., 2017).

In ccRCC, miR-210 targets TWIST1. When the expression of this miRNA is turned off, the cancer cells begin to display the EMT morphology and increased tumor growth is observed in xenografts. High TWIST1 expression along with low expression of miR-210 is associated with a low survival rate in ccRCC. The aforementioned association of miR-210 with a response to hypoxia and its angiogenic behavior is important. It cannot be ruled out that miR-210 may show varied roles at different stages in the development of ccRCC (Yoshino et al., 2017).

The reduced expression of miR-203 in ccRCC primary tumors correlates with poor prognosis and metastasis, while increased expression of miR-203 suppresses RCC cell lines growth and ccRCC metastasis (Xu M. et al., 2015). The authors attributed this to the action of miR-203 on fibroblast growth factor 2 (FGF2) as a target. The FGF2 protein is able to interact with the dimer of growth factor receptor-bound protein 2 (GRB2) involved in the PI3K/AKT and Ras/MAPK signaling pathways. Moreover, FGF2 is involved in the fine regulation of the Ras/MAPK pathway via negative feedback. FGF2 is overexpressed in many cancers (Ornitz and Itoh, 2015). The interaction between lncRNA HOTAIR and miR-203 leads to the suppression of miR-203 and stimulates EMT; when this interaction is eliminated, the expression of E-cadherin, claudin, PTEN, p21, and p27 increase, while vimentin expression is reduced (Hong et al., 2017). Methylation of MIR-203 is a specific characteristic of metastatic forms of ccRCC (see below and Pereyaslova et al., 2016). The influence of some miRNAs and lncRNA HOTAIR on EMT-related pathways is summarized in **Figure 4**.

The role of the cell environment, in particular the intercellular matrix, is important in metastasis. CD44 is one of the important targets of miR-34a in the RCC lines. The CD44 protein is a cell surface glycoprotein that is involved in the intercellular interaction, cell adhesion, and migration. CD44 is a receptor for hyaluronic acid, osteopontin, collagen, and it can interact with matrix metalloproteinases. Ectopic expression of miR-34a stops cellular growth, migration, and invasion, and also significantly inhibits the growth of carcinoma in xenografts and metastasis in nude mice (Yu et al., 2014).

ADAM metallopeptidase domain 17 (ADAM17, also known as disintegrin and A metalloproteinase 17) is a target of miR-145 in RCC. The level of ADAM17 is significantly increased in RCC, whereas the level of miR-145 is reduced. ADAM17 stimulates cell proliferation and migration, and also negatively regulates miR-145 through tumor necrosis factor-alpha (TNFα). The result is the formation of a positive feedback loop. In healthy kidneys, this metalloproteinase is found in various types of cells capable of giving rise to RCC (Doberstein et al., 2013).

The expression of miR-186 is greatly reduced in RCC. This miRNA suppresses cell growth, colony formation, invasion, and arrests the cell cycle at the G0/G1 stage. It is believed that the effects of miR-186 are due to the action on SENP1 (sentrin/SUMO-specific protease 1) as a target (Jiao et al., 2018). The SENP1 protein is the regulator of the SUMO pathway, which triggers the detachment of SUMO from homeodomain interacting protein kinase 2 (HIPK2, a transcriptional regulator), histone deacetylase 1 (HDAC1), and metastasis associated 1 (MTA1). The influence of the mimic miR-186 on SENP1 can suppress the pathway of NF-kB, which reduces the expression of p-IkBa and p-p65, as well as the underlying cyclin D1 and MMP9, and increases the expression of p21 and BCL-2 (Jiao et al., 2018).

Reduced expressions of miR-149-5p and miR-149-3p are correlated in ccRCC with increased tumor aggressiveness and

metastases (Okato et al., 2017). The authors attributed this to the fact that the target of both miRNAs is forkhead box protein M1 (FOXM1), a transcriptional activator that regulates the expression of cyclins, such as cyclin B1 and cyclin D1.

### ABERRANTLY METHYLATED MIRNA GENES IN CCRCC

One of the critical means of regulating the expression of miRNA genes is the methylation of the CpG island that is adjacent to or overlaps with the miRNA gene (Han et al., 2007; Lopez-Serra and Esteller, 2012). It is assumed that the percentage of miRNA genes that are aberrantly methylated in tumors is several times higher than the genes encoding proteins, which increases their prospects as biomarkers (Kunej et al., 2011; Baylin and Jones, 2016; Piletic and Kunej, 2016). Methylation profiles of miRNA genes have been constructed for epithelial tumors that have different localizations, including the colon, lung, breast, prostate, and ovaries. These are being considered as new potential markers and marker systems for the diagnosis and prognosis of these malignant cancers (Pronina et al., 2017b; Torres-Ferreira et al., 2017; Heller et al., 2018; Loginov et al., 2018a,b; Shi et al., 2018; Zare et al., 2018).

In contrast to the most common and widely studied types of cancer, such as colon, lung, breast, and prostate cancer, less is known about the methylation of the miRNA genes in ccRCC. What is known mainly concerns genes of the miR-9 and miR-34 families (Hildebrandt et al., 2010; Vogt et al., 2011; Schiffgen et al., 2013). These studies have shown that the promoter of the gene coding for miR-34a is significantly methylated in RCC cell lines and that the expression of this miRNA is downregulated. Treatment with 5-aza-2<sup>0</sup> -deoxycytidine increases the expression of miR-34a and reduces the expression of its target (Yu et al., 2014).

The last 5 years have seen a significant increase in the understanding of the role of hypermethylation of miRNA genes in the pathogenesis of ccRCC. In ccRCC, miR-10b is characterized by a highly methylated promoter and significantly reduced expression (He C. et al., 2015). Transfection by lentivirus carrying this miRNA or treatment with demethylating agents inhibits cell proliferation, migration, and invasiveness in cell culture. However, the target genes that are significant in RCC are unknown (He C. et al., 2015). In stomach cancer, miR-10b acts as a tumor suppressor and targets microtubule-associated protein RP/EB family member 1 (MAPRE1) (Kim et al., 2011). In non-small-cell lung cancer and bladder cancer, it acts as an oncogene by enhancing cellular migration and invasion. The putative targets associated with this effect are KLF4 and HOXD10 (Xiao et al., 2014).

A high miR-21/miR-10b ratio in metastatic ccRCC correlates with a poor prognosis (Fritz et al., 2014). Interestingly, an increased level of MIR-21 expression that correlates with a poor prognosis is associated with hypomethylation of its promoter (Cancer Genome Atlas Research, 2013).

The aforementioned miR-182-5p tumor suppressor miRNA can also be regulated by methylation. An upstream CpG island (8–10 kb) contains a putative transcription start site. Its methylation level is significantly increased in RCC cell cultures and is slightly elevated in ccRCC samples compared to the normal tissue, which may explain the decreased expression of this miRNA. One of the targets of miR-182-5p is MALAT-1. This gene produces a precursor transcript from which a ncRNA is derived by RNase P cleavage of a tRNA-like small ncRNA (known as mascRNA) from its 3<sup>0</sup> end. Its ribonucleoprotein complexes may act as a transcriptional regulator for numerous genes, including some genes involved in cancer metastasis and cell migration. Downregulation of MALAT-1 leads to upregulation of p53 and downregulation of CDC20 and AURKA, which are the drivers of the cell cycle mitotic phase (Kulkarni et al., 2018).

The gene encoding miR-766-3p is highly methylated in RCC tissues compared to the normal tissues. This miRNA behaves like a suppressor. Its direct target is SF2, whose repression also reduces the expression of AKT and ERK (Chen C. et al., 2017). The promoter of miR-145, which targets the metallopeptidase gene ADAM17, is also strongly methylated in RCC. Treatment with 5-aza-2<sup>0</sup> -deoxycytidine increases the expression of miR-145 and reduces the expression of ADAM17 (Doberstein et al., 2013). The promoter region of the MIR-200c gene is hypermethylated in RCC cell lines, which correlates with decreased expression, and it is not methylated in the normal renal cell line (Gao et al., 2014). The MIR-492 gene has a strongly methylated promoter in ccRCC and reduced expression. The use of 5-aza-2<sup>0</sup> -deoxycytidine or the histone deacetylase inhibitor 4-phenylbutyric acid increases the expression of miR-492 in ccRCC cell cultures, inhibits cell proliferation, and stimulates apoptosis and adhesion (Wu et al., 2015). The upstream promoter of MIR-106a gene is hypermethylated in RCC, which might be responsible for its downregulation (Pan et al., 2017).

Over the past 5 years, we have systematically analyzed the methylation of up to 20 miRNA genes in various cancers, including lung, breast, ovarian, and kidney cancer. These analyses have identified new markers and marker systems that are useful for the diagnosis and prognosis of these malignant diseases (Beresneva et al., 2013; Rykov et al., 2013; Pronina et al., 2017b; Braga et al., 2018; Loginov et al., 2018a,b; Varachev et al., 2018). Moreover, we have shown a correlation between the levels of expression and methylation, which has confirmed the functional role of methylation of a group of miRNA genes in the pathogenesis of breast, ovarian, and kidney cancer (Pronina et al., 2017b; Loginov et al., 2018a,b; Varachev et al., 2018).

We first found that the frequency of methylation of six miRNA genes (MIR-124-2, MIR-124-3, MIR-9-1, MIR-9-3, MIR-34b/c, and MIR-129-2) was significantly higher in malignant tumors of patients with ccRCC than in normal kidney tissues (Beresneva et al., 2013). Subsequent studies by our group established that 16 miRNA genes (MIR-124-1/-2/-3, -125b-1, MIR-129-2, MIR-132, MIR-137, MIR-193a, MIR-34b/c, MIR-375, MIR-203, MIR-9-1, MIR-9-3, MIR-107, MIR-130b, and MIR-1258) were hypermethylated and two miRNA genes (MIR-191, MIR-212) were hypomethylated in ccRCC (Pereyaslova et al., 2016; Loginov et al., 2017; Varachev et al., 2018). Moreover, seven miRNA genes (MIR-124-3, MIR-125b-1, MIR-129-2,

MIR-137, MIR-34b/c, MIR-375, MIR-9-3) were downregulated and correlated with altered methylation. The findings provided evidence of the functional significance of methylation in the deregulation of miRNA genes and in the pathogenesis of ccRCC (Varachev et al., 2018).

Besides hypermethylation, some miRNA genes studied (MIR-124-2/-3, -34b/c, MIR-129-2, MIR-107, MIR-148a, MIR-203) have been associated with aspects of ccRCC progression that include advanced pathologic stage, tumor size, and differentiation grade. In particular, hypermethylation of six miRNA genes (MIR-125b-1, MIR-129-2, MIR-203, MIR-375, MIR-107, and MIR-1258) significantly correlated with the presence of metastasis (Pereyaslova et al., 2016; Varachev et al., 2018), and most significantly for MIR-375 and MIR-1258 (p-value ∼ 10−7–10−<sup>8</sup> ).

On the basis of hypermethylated miRNA genes, novel marker systems have been suggested for ccRCC diagnosis (MIR-125b-1, MIR-375, MIR-137, MIR-193a) and prediction of metastasis (MIR-125b-1, MIR-375, MIR-107, MIR-1258, MIR-203) (Loginov et al., 2017; Varachev et al., 2018).

A Russian team of researchers demonstrated the correlation of the methylation of the MIR-129-2 gene with increased expression of genes, such as RARB(2), RHOA, NKIRAS1, and CHL1 (Pronina et al., 2017a). Preliminary data from another study indicated a negative correlation between the expression levels of miR-375 and the pro-apoptotic gene APAF1 (Varachev et al., 2018). The targets of most hypermethylated miRNAs, identified by us recently remain unclear and further research is needed.

However, the roles of some of the novel hypermethylated miRNAs in the pathogenesis of kidney cancer have been studied. For instance, the reduced expression of miR-124 is considered a highly significant factor in ccRCC genesis and may be an essential predictor of survival (Butz et al., 2015). Moreover, the miRNA and gene expression data from 458 ccRCC and 254 normal kidney specimens were used to construct the integrated miRNA-target interaction network, which revealed miR-124 as a key miRNA contributing to the aggressive behavior of ccRCC (Butz et al., 2015).

**Table 1** presents examples of miRNAs encoded by genes that are hypermethylated and downregulated in ccRCC. Data on target genes and functions in RCC are also provided. The information presented pertains mainly to those genes whose role of hypermethylation was first explored by our group and data from other authors concerning the functional roles, target genes, and in some cases the signaling pathways involved.

For five miRNAs we identified (miR-124-3p, miR-129-3p, miR-137, miR-203, and miR-375), functional studies by other authors confirmed suppressor properties and/or association with metastasis. The tumor suppressor and anti-metastatic functions of the hypermethylated MIR-129-2 gene we discovered (Beresneva et al., 2013), are consistent with the data that miR-129-3p downregulates multiple metastasis-related genes in RCC cells, including SOX4, and also decreased phosphorylation of focal adhesion kinase and expression of MMP-2/9 (Chen et al., 2014). In addition, miR-129-3p is involved in miR-129-3p/TRPM7/AKT/FOXO1 signaling pathway (Zhao Z. et al., 2018). The collective data confirm the ability of miR-129-3p to restrain metastasis in ccRCC.

In ccRCC, the hypermethylated MIR-137 gene (Varachev et al., 2018) abrogates the tumor-promoting effect of overexpressed ncRNA Small Nucleolar RNA Host Gene 1 (SNHG1), which is associated with metastasis and poor prognosis of ccRCC patients (Zhao S. et al., 2018).

Data on the nature of the role of miR-203 in RCC, either as a tumor suppressor or an oncogene, is inconsistent (Hu et al., 2014). However, in other cancer types (hepatocellular carcinoma, multiple myeloma, lung cancer, esophageal cancer, bladder cancer), miR-203 acts as a suppressor (Hu et al., 2014). It is possible that this miRNA has different roles at different stages of the development of ccRCC and/or in its different variants. Overall, we observed significantly more frequent hypermethylation of the MIR-203 gene in tumors of patients with ccRCC with metastases. Moreover, our data on the anti-metastatic activity of miR-203 are consistent with data from three other papers (see **Table 1**), including the results on the role of miRNA-203 in inhibition of EMT and metastatic genes (Dasgupta et al., 2018).

Our data concerning the correlation of miR-375 hypermethylation with metastasis strengthens the evidence that the suppression of the tumor aggressive phenotypes of ccRCC mediated by this miRNA regulates the oncogene YWHAZ, which is its direct target (Zhang X. et al., 2018). As discussed earlier, aberrantly methylated miR-148a was suggested to function as a tumor suppressor in RCC by targeting AKT2 (Cao et al., 2017). A recent paper (Reustle et al., 2018) implicated the role of aberrantly methylated miR-212-3p and miR-132-3p in ccRCC in the post-transcriptional regulation of BCRP/ABCG2 (breast cancer resistance protein), which contributes to the multi-drug resistance seen in cancer. Hypermethylated miR-193a-3p can affect the PI3K/AKT pathway in RCC (Pan et al., 2018a).

However, our data on the frequent hypermethylation of the MIR-125b-1 gene and the hypermethylation linkages with progression and metastasis (Varachev et al., 2018) do not agree with the increased level of this miRNA and its role in promoting cell (Jin et al., 2017). The properties of this miRNA are also ambiguous in other types of cancer, such as breast cancer (Loginov et al., 2018a). Additional studies are required to determine the role and mechanism of miR-125b in RCC.

Tumor suppressor or/and anti-metastatic properties were found for some miRNAs, encoded by 16 aberrantly methylated genes that discovered by us, including miR-124, miR-129, miR-132, miR-137, miR-148a, miR-203, and miR-375, as well as for miR-9 and miR-34 genes. However, further analyses should explore the expression and functional properties of some other miRNAs, which genes were also found as hypermethylated (miR-107, miR-125b, miIR-1258, miR-130b, and miR-193a) or showed hypomethylation (miR-191, miR-212) in cancers of the kidney. Overall, our group has provided novel data on the contribution of methylation to the regulation of more than 10 miRNA genes in ccRCC (Beresneva et al., 2013; Loginov et al., 2017; Varachev et al., 2018). Further research and validation of target genes for these aberrantly methylated miRNAs are required. Moreover, hypermethylated


TABLE 1 | Examples of the most studied miRNAs encoded by hypermethylated genes in ccRCC, focusing on the role of aberrant methylation in miRNA deregulation, target genes, and functions.

miRNAs could someday be used as convenient markers for ccRCC in the clinic, as has been suggested for non-invasive diagnosis and prognosis of prostate cancer using miR-193b, miR-129-2, and miR-34b/c methylation in tissue and urine samples (Torres-Ferreira et al., 2017).

In **Table 1**, it is interesting to note that among the seven most studied hypermethylated miRNAs, four have lncRNAs as one or more targets. For example, miR-124-3p interacts with the lncRNA HOTAIR, miR-137 with lncRNA SNHG1, miR-182-5p with lncRNA MALAT1, and miR-203 with the HOTAIR and SNHG14 lncRNAs. These data reinforce the role of lncRNAs in the regulatory functions of miRNA, apparently by acting primarily as ceRNA to reduce the content of specific miRNAs depending on other factors.

### CLINICAL APPLICATION OF MIRNAS AS MARKERS FOR DIAGNOSIS AND PROGNOSIS

In this review, we included only the most convincing results obtained from studies featuring large sample sizes and detailed

statistical evaluations. Initially, we examined the features of miRNAs that are relevant concerning their diagnostic potential. These features are summarized in **Table 2**.

The diagnostic value of elevated miR-210 levels is well established. This may be due to chronic hypoxia observed in ccRCC (Zhang Z. et al., 2009). However, no correlation of miR-210 expression with tumor stage has been reported (Iwamoto et al., 2014). The expression of miR-1233 in RCC cell lines is also stimulated by hypoxia (Dias et al., 2017).

The target of miR-451 is an mRNA encoding the PSMB8 protein, which has a pro-inflammatory function, and which is presumably significant in RCC (Zhu et al., 2016). In RCC, miR-193a-3p affects the PI3K/AKT pathway (Liu L. et al., 2017; Pan et al., 2018a). miR-28-5p is associated with chromosomal instability (Hell et al., 2014). In addition, its target is Ras-related small GTP-binding oncoprotein RAP1B (Wang et al., 2016). miR-141 is involved in the suppression of EMT (Li W. et al., 2014).

Targets of miR-144 include mTOR (Xiang et al., 2016) and MAP3K8 (Liu F. et al., 2016). Thus, miR-144 can suppress cellular proliferation, EMT, and metastasis. However, another study (Xiao et al., 2017) implicated miR-144 as having an oncogenic role, with involvements in the stimulation of proliferation, migration, invasion, and resistance to sunitinib. Furthermore, miR-144 acts as an upregulated marker in the plasma of ccRCC patients, particularly those with advanced pT stage, which is a characteristic of oncogenes (Lou et al., 2017). The inconsistent functional properties of miR-144 in ccRCC revealed to date highlight the need for further studies before its clinical application can be realized.

miR-378 stands apart from other miRNAs. In a study that used samples from patients at all stages of ccRCC (Redova et al., 2012), the diagnostic value of the increased expression of miR-378 was found. However, when samples were used only from patients at the I-II stage, the decreased expression of miR-378 had diagnostic value (Wang et al., 2015). This dichotomy of miRNA – with an oncogenic role in some types of cancer or at certain stages of the same disease, while being a tumor suppressor in other cases – is quite common. In ccRCC, miR-378 could be a potential marker of the disease stage (Li H.C. et al., 2014; Fedorko et al., 2015). However, Hauser et al. (2012) rejected the correlation between its expression in serum and pathologic stage. The authors also did not find evidence of differential expression of miR-378 levels in diseased and healthy individuals. The expression of miR-378 is likely associated with angiogenesis (Li H.C. et al., 2014). Indeed, in the ccRCC cell line, the expression of miR-378 declines with prolonged hypoxia (Chen S.C. et al., 2017). However, the mechanisms of action of miR-378 in ccRCC have not yet been adequately studied.

The genes discussed above, including some with inconsistent features (miR-144 and miR-378), have been suggested for the diagnosis of ccRCC (**Table 2**). Thus, the methods of ccRCC diagnosis need further verification.

The level of some miRNAs can be valuable as prognostic markers of survival in ccRCC (**Table 3**). Reviews describing miRNA markers used to predict survival and metastasis in ccRCC, have been published (Fedorko et al., 2016; Ran et al., 2017; He Y.H. et al., 2018). In the present review, we present only the most recent studies that featured large sample sizes (**Table 3**). Unfortunately, the data obtained so far has not always been consistent. For example, an increase of miR-210, which serves as a diagnostic criterion for ccRCC, was associated with prolonged overall survival in one study (McCormick et al., 2013) and


TABLE 2 | Non-invasive markers suggested for use in ccRCC diagnosis: miRNA level and diagnostic value.

Note: ↓ – decreased level of miRNA; ↑ – increased level of miRNA; AUC – area under the receiver operating characteristic curve, Sn – sensitivity, Sp – specificity. Only the most convincing results were included, with sample sizes ≥ 30 and detailed statistical evaluations: AUC, sensitivity, and specificity.

TABLE 3 | Prognostic markers for poor survival and metastasis in ccRCC: miRNA level and predictive value.


Note: ↓ – decreased level of miRNA; ↑ – increased level of miRNA; OS – overall survival; DFS – disease-free survival; CSS – cancer-specific survival; PFS – progression-free survival, <sup>∗</sup> – ratio. Only the most recent (since 2013) works in which sets of >35 ccRCC samples were included.

decreased overall survival in another study (Samaan et al., 2015). For most miRNAs, the nature of interactions in ccRCC is not well understood.

Despite the relative simplicity of miRNA gene methylation studies, the approach has barely been used instead of the expression levels of miRNAs, to evaluate the prognosis of survival or predict metastasis in ccRCC. The only exception is a study (Gebauer et al., 2013) that described that the methylation of the gene encoding miR-124-3 was associated with a more advanced RCC stage, metastasis, and an increased risk of relapse in 111 RCC samples (80 ccRCC, with 77 paired with histologically normal tissue). Also, a recent review did not mention studies involving miRNA gene methylation in RCC (Ran et al., 2017). Studies we undertook identified novel biomarker systems by using hypermethylated miRNA genes as an indicator in ccRCC, both for the diagnosis (MIR-125b-1, MIR-375, MIR-137, and MIR-193a) and for the prediction of metastasis (MIR-125b-1, MIR-375, MIR-107, MIR-1258, and MIR-203) (Loginov et al., 2017; Varachev et al., 2018). The use of both biomarker systems in 70 patients was characterized by a high discrimination value

[area under the curve (AUC) 0.93], sensitivity (86%), and specificity (95%).

Few studies have explored the use of the miRNA level as an indicator to predict the response to ccRCC treatment. Most of the studies have focused on predicting the success of the response to tyrosine kinase inhibitors that are the first-line treatment for metastatic ccRCC. However, approximately 20% of the patients who are treated rapidly develop resistance to these drugs. For example, for 74 patients with metastatic ccRCC who were treated with tyrosine kinase inhibitors, early progression of the disease was observed in 16 (Garcia-Donas et al., 2016). Selection of increased miR-1307-3p and miR-425-5p levels as the biomarker predicted the risk of poor outcome to treatment with an AUC value of 0.75, which is better than the risks predicted using parameters based on clinicopathological factors. This result was revalidated in a group of 64 patients (Garcia-Donas et al., 2016). The predictors of the outcome of sunitinib treatment were studied in 123 patients with ccRCC (the dataset included metastatic and non-metastatic ccRCC patients). Of the 123 patients, 97 were characterized by prolonged (>22 months) progression-free survival (PFS) and the remaining 26 showed rapid progression of the disease in the first 3 months (Puente et al., 2017). It must be noted that, along with the other characteristic clinical indices, the former group was distinguished by the presence of high levels of the transcription factors HEYL, HEY, and HES, accompanied by elevated levels of miR-27b, miR-23b, and miR-628-5p, although miR-23b is considered as a typical oncogene (Zaman et al., 2012). In another study, 56 sunitinib-treated patients with metastatic ccRCC (24 responsive patients with PFS > 18 months and 32 non-responsive patients with PFS < 6 months) (Kovacova et al., 2018). A good response to the drug was positively associated with miR-942 and miR-133 levels, and the model based on using their expression levels as indicators gave predictions with an AUC value of 0.81.

Another aspect of the use of miRNA as biomarkers is the predictive capabilities of miRNA polymorphisms (variations or polymorphisms in sequences of miRNA, target mRNA, miRNA genes, and target miRNA genes and their pathways). The polymorphism in the MIR-34b/c promoter region has been amply studied. This polymorphism reduces the expression of this miRNA (allele C for rs4938723) in the homozygotic condition, increases the risk of RCC in the Chinese population, especially in elderly, men, smokers, and alcohol abusers (Zhang S. et al., 2014). Polymorphisms in the miRNA-binding sites of 102 genes belonging to the VHL-HIF1α pathway and their effects on RCC risk have been analyzed (Wei et al., 2014). The most significant effect of two polymorphisms in the MAPK1 gene, and the presence of four of the five adverse alleles in the MAPK1, CDCP1, TFRC, and DEC1 gene set, was a more than twofold increase in the risk of RCC.

The genotype CC, a variant of the 3<sup>0</sup> untranslated region (UTR) of the SET domain containing lysine methyltransferase 8 (SET8) gene, significantly reduced the risk of ccRCC (the odds ratio = 0.318). This single nucleotide polymorphism, also designated rs16917496, is located in the miR-502 binding site within the 30UTR of the SET8 gene and is associated with its downregulation in ccRCC. Knockdown of SET8 inhibited proliferation, migration, and invasiveness of ccRCC cells in culture (Zhang et al., 2017). Many ccRCC-influencing genes are associated with remodeling, and SET8 may be one of them (Joosten et al., 2018).

### CONCLUSION

In recent years, the role of miRNA in the pathogenesis of ccRCC has been amply studied. In ccRCC, miRNAs involved in processes that include the response to hypoxia, EMT, and chromatin remodeling play significant roles. Nevertheless, the picture is far from complete. The search for new diagnostic and prognostic markers is still largely empirical and the associated pathways and processes are not clearly elucidated. For example, in a study involving prediction and correlation analyses, miRNA–mRNA pairs displayed a significant inverse relationship. miR-30b was one of these miRNAs, and its expression was negatively correlated with the highest number of genes in ccRCC (Liu et al., 2018). Among the latter, there are, for example, genes that are significant for tumor invasion, such as Integrin Subunit Alpha 5 (ITGA5). Nevertheless, almost nothing is known concerning the influence of miR-30b on ccRCC. The many potential biomarkers include some differentially expressed miRNAs or differentially methylated sites. However, only a few have been experimentally verified (Wang Z. et al., 2018).

For the diagnosis of ccRCC, the use of miRNAs like miR-144 has been proposed since their targets may be relevant for this cancer. However, this use is hampered by inconsistent findings, with both tumor suppressive and oncogenic activity reported in separate studies. Results of various studies for other miRNAs, including miR-106a, miR-125b, miR-203, and miR-378, have been contradictory. These discrepancies may be due to the varying roles of these miRNAs at different stages of cancer or in different forms of the disease or to their ability to interact with various targets depending on the cellular context.

Our recent contributions have revealed the potential for the use of the methylation profiles of miRNA genes in the search for reliable diagnostic and prognostic marker systems (i.e., high AUC). The results have suggested the need to expand the search for the selection of new miRNA genes regulated by methylation. Moreover, in-depth studies of the identified hypermethylated miRNA genes associated with metastasis with respect to their target genes and functions are necessary.

The recently discovered lncRNAs represent a novel class of potential prognostic biomarkers in ccRCC. Data on the function and clinical applications of lncRNA in ccRCC is accumulating, with 60 articles identified in PubMed as of January 2019. However, only a few of these studies (10%) explored the interactions of lncRNAs with miRNAs. Nevertheless, a detailed analysis of the molecular mechanisms of ccRCC in original articles allowed us to identify at least seven miRNAs (miR-122, miR-124-3p, miR-137, miR-141, miR-182-5p, miR-203, and miR-217) that interact with the lncRNAs (e.g., HOTAIR, MALAT-1, SNHG1, and SNHG14). Further research is needed to elucidate these interactions in ccRCC and to assess their clinical potential.

#### AUTHOR CONTRIBUTIONS

fgene-10-00320 April 17, 2019 Time: 16:21 # 15

EB, MF, VL, AD, and SM wrote the manuscript. All the authors revised the work critically for important intellectual content, approved the version to be published, and agreed to be

#### REFERENCES


accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

### FUNDING

The work was carried out within the framework of fundamental research for state academies for 2013-2020 (No. 0520-2019-0030).



prognostic outcomes in patients undergoing radical nephrectomy. BJU Int. 110(11 Pt C), E1131–E1137. doi: 10.1111/j.1464-410X.2012.11297.x



through diminished MALAT-1 expression. Mol. Cancer Res. 16, 1750–1760. doi: 10.1158/1541-7786.MCR-17-0762




of renal cell carcinoma. Oxid. Med. Cell. Longev. 2017:3269379. doi: 10.1155/ 2017/3269379


at the time of nephrectomy. Clin. Chim. Acta 416, 5–10. doi: 10.1016/j.cca.2012. 11.010


c-Myc transcriptional complexes in human renal cell carcinoma. Carcinogenesis 33, 294–300. doi: 10.1093/carcin/bgr286



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

Copyright © 2019 Braga, Fridman, Loginov, Dmitriev and Morozov. 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.

# Transcriptomics Analysis of Circular RNAs Differentially Expressed in Apoptotic HeLa Cells

#### Bilge Yaylak, Ipek Erdogan and Bunyamin Akgul\*

Non-Coding RNA Laboratory, Department of Molecular Biology and Genetics, Izmir Institute of Technology, Izmir, Turkey

Apoptosis is a form of regulated cell death that plays a critical role in survival and developmental homeostasis. There are numerous reports on regulation of apoptosis by protein-coding genes as well as small non-coding RNAs, such as microRNAs. However, there is no comprehensive investigation of circular RNAs (circRNA) that are differentially expressed under apoptotic conditions. We have performed a transcriptomics study in which we first triggered apoptosis in HeLa cells through treatment with four different agents, namely cisplatin, doxorubicin, TNF-α and anti-Fas mAb. Total RNAs isolated from control as well as treated cells were treated with RNAse R to eliminate the linear RNAs. The remaining RNAs were then subjected to deep-sequencing to identify differentially expressed circRNAs. Interestingly, some of the dys-regulated circRNAs were found to originate from protein-coding genes well-documented to regulate apoptosis. A number of candidate circRNAs were validated with qPCR with or without RNAse R treatment as well. We then took advantage of bioinformatics tools to investigate the coding potential of differentially expressed RNAs. Additionally, we examined the candidate circRNAs for the putative miRNA-binding sites and their putative target mRNAs. Our analyses point to a potential for circRNA-mediated sponging of miRNAs known to regulate apoptosis. In conclusion, this is the first transcriptomics study that provides a complete circRNA profile of apoptotic cells that might shed light onto the potential role of circRNAs in apoptosis.

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Yulin Jin, Emory University, United States Argyris Papantonis, Universität zu Köln, Germany

> \*Correspondence: Bunyamin Akgul bunyaminakgul@iyte.edu.tr

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 14 November 2018 Accepted: 18 February 2019 Published: 13 March 2019

#### Citation:

Yaylak B, Erdogan I and Akgul B (2019) Transcriptomics Analysis of Circular RNAs Differentially Expressed in Apoptotic HeLa Cells. Front. Genet. 10:176. doi: 10.3389/fgene.2019.00176 Keywords: apoptosis, circular RNA, RNA-seq, transcriptomics, HeLa

### INTRODUCTION

Apoptosis is a tightly regulated mechanism of type 1 programmed cell death that mediates balance between survival and cell death. Apoptosis mainly proceeds with the extrinsic death receptor pathway or the intrinsic mitochondrial dysfunction pathway (Vince and Silke, 2009; Fuchs and Steller, 2011). Functional apoptosis pathway is vital for tissue homeostasis and organ development. Accordingly, dysregulation of apoptosis is associated with a wide range of diseases such as cancer, autoimmune diseases, neurodegenerative diseases, and acute pathologies (Favaloro et al., 2012). Therefore, much effort has been made to unravel molecular pathways that trigger and regulate apoptosis.

During past decades, a plethora of studies revealed that considerable amount of DNA called "dark matter of genome is transcribed but not translated" (Bertone et al., 2004). MicroRNAs (miRNAs) and long non-coding RNAs (lncRNAs) are well-known examples of non-coding RNAs

Yaylak et al. Circular RNA Profiles in Apoptosis

(ncRNAs) (Lewis et al., 2005; Lander, 2011). Recent studies revealed yet another type of ncRNAs, circRNAs, as a novel type of endogenous ncRNAs that has the potential to regulate not only the protein-coding genes but also ncRNAs as well (Salzman et al., 2013; Jeck et al., 2013; Memczak et al., 2013; Huang S. et al., 2017).

Circular RNAs are typically generated by back-splicing of exons from pre-mRNAs with a 3'-to-5' phosphodiester bond at the junction site (Chen, 2016). Since they do not possess a 5' cap or a poly(A) tail, they are frequently missed in poly(A)<sup>+</sup> RNA profiling studies. Also, the unusual back-splicing generates a read that has been historically filtered out during conventional RNA-seq studies. The recent development of non-polyadenylated sequencing technology followed by exonuclease-mediated degradation of linear RNAs by RNAse R resulted in the discovery of thousands of circular RNAs widely expressed in human (Suzuki et al., 2006). The majority of circular RNAs are conserved across species, and generally reveal tissue/developmental-stagespecific expression (Qu et al., 2015). Searching for out-of-order arrangement of exons known as a "backsplice" in combination with computational pipelines facilitated the genomewide identification of circRNAs (Jeck and Sharpless, 2014). Nonetheless, identification of backsplicing reads can be challenging because of high false positive rates of computational tools. These apparent backsplicing sequences can originate from different mechanisms such as tandem duplication, RNA-trans splicing and reverse transcriptase template switching rather than exonic circular RNAs (Cocquet et al., 2006; McManus et al., 2010; Jeck and Sharpless, 2014).

Circular RNAs can regulate gene expression through various mechanisms. Perhaps, one of the most prominent functions of circRNAs is to modulate endogenous miRNA activities by acting as miRNA sponges (Hansen et al., 2013; Memczak et al., 2013). CircRNAs might also regulate transcription by interacting with RNA polymerase II and U1 snRNP or various RNA binding proteins such as Mbl (Ashwal-Fluss et al., 2014). Moreover, the canonical pre-mRNA splicing and circularization may compete with each other, resulting in a decrease in the amount of the linear mRNA (Ashwal-Fluss et al., 2014). Besides, if the translation start site of an mRNA is included in the circularized exon, the remaining portion of the mRNA will miss a proper translation initiation site, negatively affecting its translation (Huang G. et al., 2017). Although circular RNAs were initially categorized as non-coding RNA, it was obvious that a portion of circRNAs held the potential for translation, especially those that retain an intact open reading frame. Spectacularly, cap-independent translation of a subset of circular RNAs has been reported recently, paving the way for truncated translation products from circRNAs (Pamudurti et al., 2017).

There are only a few studies that report the involvement of circRNAs in apoptosis. Circ\_016423 has been reported to have a high possibility of regulating apoptosis in ischemic prenumbral cortex, potentially through harboring binding sites for various miRNAs (Mehta et al., 2017). Hsa\_circ\_0010729 was reported to regulate proliferation and migration as well as apoptosis in hypoxia-induced human umbilical vein endothelial cells (HUVECs) by targeting miR-186-HIF-alpha axis (Dang et al., 2017). MFACR possesses a binding site for miR-652-3p, which in turn regulates the MTP18 mRNA, which codes for a nuclear-encoded mitochondrial membrane protein associated with apoptosis in cardiomyocytes (Wang et al., 2017).

Holdt et al. (2016) reported that overexpression of circANRIL, the circular form of lncRNA ANRIL, induces apoptosis in HEK-293 cells. Further, circular Rar1, circRar1, was shown to regulate apoptosis in N2a mouse neuroblastoma cells (Nan et al., 2016). Pro-apoptotic lncRpa and apoptosis-related circRar1 were shown to be regulated by miR-671. However, a complete genomewide profile of dysregulated circular RNAs under apoptotic conditions remains unknown. Here we report the first genome-wide profile of differentially expressed circular RNAs in apoptotic HeLa cells and provide bioinformatics data that suggest the potential translatability of certain candidate circRNAs.

## MATERIALS AND METHODS

### Cell Culture, Drug Treatment and Measurement of Apoptosis

HeLa cells were obtained from DKFZ GmbH (Germany). Cells were cultured in RPMI 1640 (with L-Glutamine, Gibco, United States) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco, United States) and 1% penicillinstreptomycin (Gibco) in an atmosphere of 5% CO<sup>2</sup> at 37◦C. Based on the initial dose and time kinetics, the subsequent treatments were carried out in the following conditions (1) cisplatin (Santa Cruz, United States) at a concentration of 80 µM for 16 h; (2) doxorubicin (Cell Signaling, United States) at a concentration of 0.5 µM for 4 h; (3) anti-Fas mAb (Millipore, United States) at a concentration of 0.5 µg/ml for 16 h; and (4) TNF-α (Millipore, United States) at a final concentration of 125 ng/ml TNF-α with 10 µg/ml cycloheximide (Applichem, Germany) for 8 h. Untreated and dimethylsulfoxide (DMSO)-treated (0.1%) cells were used as negative control.

Following three biological replicates of drug treatments, 0.5 × 10<sup>6</sup> cells were trypsinized by 1× Trypsin-EDTA (Gibco, United States) and washed in 1X cold PBS (Gibco, United States), followed by resuspension in 1× Annexin binding buffer (Becton Dickinson, United States). The resuspended cells were stained with Annexin V-PE (Becton Dickinson, United States) and 7AAD (Becton Dickinson, United States) for 15 min in dark at room temperature and analyzed by FACS Canto (Becton Dickinson, United States).

Control and apoptotic cells were also stained with NucRedTM Dead 647 ReadyProbesTM Reagent (Invitrogen) to visualize apoptotic cells under fluorescence microscopy. HeLa cells (0.4 × 10<sup>6</sup> cells/well) were seeded in a six-well plate and treated with 4 drugs as explained above. DMSO (5%) (v/v) was used as positive control. At the end of treatment period, 2 drops/ml of medium was applied on the cells and cells were monitored by a fluorescent microscope (Zeiss Observer Z1) after 30 min of incubation in dark.

### CircRNA Deep-Sequencing and Bioinformatics Analyses

fgene-10-00176 March 12, 2019 Time: 16:38 # 3

Control and treated cells were sent to NOVAGENE (Hong Kong) in RNAlater for RNA isolation and deep sequencing using a previously published method optimized for circular RNA identification (Szabo and Salzman, 2016). Three biological replicates of total RNAs from control and treated cells were then subjected to poly(A) tail elimination, rRNA depletion and RNase-R treatment for circRNA enrichment prior to deep sequencing (Szabo and Salzman, 2016). The original raw data obtained from high throughput sequencing (Illumina HiSeqTM2500) were first transformed to Sequenced Reads containing read sequence and corresponding base quality (in FASTQ format) through Base Calling. Novel circRNAs were identified and annotated using CIRCexplorer (Zhang et al., 2014), which remapped the unmapped reads to the genome by tophat's fusion-search to find back-spliced reads. The normalized expression of circRNAs in each sample was calculated based on the transcript per million (TPM) method (Zhou et al., 2010) using DESeq2 version 1.6.3 (Love et al., 2014) where the threshold was set as padj <0.05.

Log2 ratios were clustered by K-means and self-organizing maps (SOM) algorithms as well as TPM. CircRNA and miRNA pairing were annotated by miRSystem (Lu et al., 2012) for further analysis. All differentially expressed circular RNAs, origin mRNAs and potential miRNA binding sites were classified based on their fold change, pathway-drug specificity and source genes. Translation potential of circRNAs was examined by CircInteractome<sup>1</sup> , Ensembl<sup>2</sup> , NCBI ORF Finder<sup>3</sup> , IRESitefootnoteiresite.org/, and SRAMP: prediction of mammalian N6-methyladenosine (m6A) sites<sup>4</sup> .

Initially, we focused on cisplatin-treated samples due to the widespread use of cisplatin in cancer treatment. Thus, 109 differentially expressed circular RNAs in cisplatin-treated samples were subjected to further bioinformatics analyses. StarBase<sup>5</sup> was used to investigate potential miRNA-circRNA interactions. miRNAs with potential binding sites were then fed into miRSystem<sup>6</sup> to identify putative target mRNAs as well as pathways that might be affected by the potential miRNA-circRNA interactions. To examine the translation potential of circRNA candidates, exon sources of circular candidates were downloaded from Ensembl (see text footnote 2). Those that originate from the first exons were searched for the presence of a potential internal ribosome entry site using IRESite (see text footnote 4). SRAMP (see text footnote 5) was used to analyze the presence of m6A sites on hsa\_circ\_0029693.

#### Validation of circRNAs by Cloning

Total RNA was isolated from cisplatin (80 µM) and DMSO (0.1%) (Applichem, Germany) treated cells using TRIzol (Thermo Fisher Scientific, United States) according to the


manufacturer's instructions. For all candidates, divergent primers were designed by using the CircInteractome web tool (see text footnote 1). These primers were specific to amplify backsplice junctions, excluding the potential for the amplification of linear mRNAs (**Table 1**). Total RNAs were then treated with RNAse R (Epicenter) according to manufacturer's instructions to eliminate the linear RNAs and to enrich the circRNAs. Briefly, 2 µg of total RNAs, 1 µl buffer and 5 units of RNase R were mixed in a final volume of 10 µl and incubated for 30 min at 37◦C. Subsequently, RNA clean-up was performed with Nucleospin RNA kit (Macherey Nagel, Germany) according to the manufacturer's instructions.

cDNAs were prepared from RNAse R<sup>+</sup> and RNAseR<sup>−</sup> RNA samples with ProtoScript <sup>R</sup> first strand cDNA synthesis kit (NEB, United States) with random primers in a total volume of 50 ul according to the manufacturer's instructions. The cDNA reactions contained 200 ng of total RNAs or equivalent of RNAsetreated samples. The candidate circRNAs were amplified by PCR using the following conditions: Initial denaturation at 95◦C for 30 s, 35 cycles of denaturation at 95◦C for 30 s, annealing at 60◦C for 1 min, extension at 68◦C for 1 min, 1 cycle of final extension at 68◦C for 5 min. PCR products were run on 1% agarose gel to examine the size and purity of the amplified products. Bands of the correct size were extracted from the gel using the gel extraction kit (Macherey Nagel, Germany). The purified fragments were cloned into the pCR <sup>R</sup> II TA vector using the TA cloning kit (Thermo Fisher Scientific, United States) and sequenced at the BIOMER center (IZTECH, Turkey) to validate the backsplice junction of target circRNAs.

#### qPCR Analyses

cDNAs prepared from RNAse R<sup>+</sup> and RNAseR<sup>−</sup> RNA samples were used as templates for qPCR with GoTaq q-PCR Master Mix (Promega, United States). Two step PCR amplification conditions were as follows: initial denaturation at 95 ◦C for 2 min, (40 cycles) denaturation at 95◦C for 15 s and annealing/extension at 60◦C for 1 min.

#### TABLE 1 | List of divergent primers.


<sup>a</sup>Convergent primer, <sup>b</sup> for qPCR amplification, <sup>c</sup> for cloning.

<sup>1</sup> circinteractome.nia.nih.gov

<sup>2</sup> ensembl.org/index.html

<sup>3</sup>ncbi.nlm.nih.gov/orffinder/

group for 16 h (A), 80 µM cisplatin for 16 h (B), 0.5 µM doxorubicin for 4 h (C), 0.5 µg/ml anti-Fas mAb for 16 h (D) and 125 ng/ml TNF-α for 8 h (E). Cells were then stained with Annexin V-PE (Becton Dickinson, United States) and 7AAD (Becton Dickinson, United States) and analyzed by FACS Canto (Becton Dickinson, United States). Annexin V-positive cells were regarded as early apoptotic cells (F). (G) The overall TPM cluster analysis. The normalized expression of circRNAs in each sample is calculated based on the TPM method (Zhou et al., 2010). The normalized circRNA expression levels are clustered based on log10(TPM+1) value where the red color represents circRNAs with high expression level while the blue color represents circRNAs with low expression level. The color from red to blue represents the log10(TPM+1) value from large to small. 1a, 2a, and 3a refer to each of the three biological replicates (C is control DMSO). (H) Distribution of differentially expressed circRNAs in apoptotic HeLa cells. CP, cisplatin; FAS, anti-Fas mAb; TNF, TNF-α, and DOX, doxorubicin. All experiments were performed in triplicates. p < 0.005.

#### Statistical Analyses

All experiments were carried out at least in three biological replicates. Student's t test was used to analyze the data. A P-value of 0.05 was considered statistically significant unless indicated otherwise.

### RESULTS AND DISCUSSION

#### Induction of Apoptosis and Deep Sequencing Analyses

Cisplatin and doxorubicin treatment were performed to induce apoptosis by intrinsic pathway while anti-Fas mAb and TNF-α treatments were carried out to trigger the extrinsic pathway (Boatright and Salvesen, 2003; Morgan et al., 2013; Huang et al., 2018). In order to attain an apoptosis rate of approximately 50%, drug /ligand concentrations and incubation periods were optimized through dose and time kinetics (data not shown). **Figures 1A–E** shows the dot blot analyses of cell populations after treatment at optimum doses and times. All drug and ligand treatments caused cells to shift to the Annexin V+/7AAD– region (**Figures 1B,D,E**) except doxorubicin (**Figure 1C**), which caused cells to be included in Annexin V+/7AAD+ population. Apoptosis rates ranged between 27 and 58% (**Figure 1F**), depending on the treatment agent. Cisplatin and TNF-α caused the highest apoptotic rate (58%) while anti-Fas-treated population displayed a lower rate of apoptosis (27%) due to the desensitization of HeLa cells to FasR stimulation (Holmström et al., 1999). Control and apoptotic cells were also stained with NucRedTM Dead 647 as a qualitative visualization of cell death (**Supplementary Figure S1**). The Annexin V+/7AAD– state of the cells were interpreted as cells being caught at the early apoptotic phase and thus they were deemed suitable for downstream steps.

To identify differentially expressed circRNAs, total RNAs were subjected to circRNA enrichment as explained in "Materials and Methods." The number of raw reads ranged from 81,360,972 to 106,246,444 whereas clean reads ranged from 79,147,032 to 103,277,920 with an error rate of 0.02 to 0.03%. Q30 score was between 91 and 94.43%, indicating fairly a low error rate. The lowest percentage of total mapped reads was 68.77% whereas the highest percentage of total mapped reads was 94.3%. For all samples, the primary source of circRNAs was introns while getting reads from exons and intergenic sequences as well. TPM cluster analysis was conducted to obtain normalized circRNA expression data. Read count value from expression level analysis of circRNA was used as input data for K-means and SOM clustering (**Supplementary Figure S2**). This clustering revealed that circRNAs within the same cluster exhibited the same changing trend of expression levels under

different conditions. Based on the TPM cluster analysis, it is apparent that circRNA expression differs among control, drugand ligand-treated groups (**Figure 1G**). TNF-α and cisplatin treatment groups exhibited the highest number of dys-regulated circRNAs. However, the number of circRNAs differentially expressed in doxorubicin and anti-Fas mAb-treated HeLa cells were relatively low (**Figure 1H**). Venn diagram and size graph show that 109 circular RNA were differentially expressed in cisplatin-treated HeLa cells. Likewise a total of 236 circular RNAs showed differential expression in the TNF-α treated HeLa cells. Five and four circular RNA were differentially expressed in doxorubicin and anti-Fas mAb-treated HeLa cells, respectively. Consistent raw read numbers in three biological replicates of doxorubicin and anti-Fas mAb-treated samples suggest a drugspecific response rather than a technical problem, which begs for further investigation.

#### Candidate Selection

Because the drugs and ligands used in our studies are potent inducers of not only apoptosis but also other cellular phenotypes such as stress, proliferation or cell cycle, we applied a number of filtering criteria to select candidate circRNAs for further analyses. Basically we selected the dys-regulated circRNAs that (1) had the highest log change difference between control and drug treated cells, (2) had a statistical significance of differential expression, and (3) originate from mRNAs that were reported or have the potential to regulate apoptosis. Based on these criteria, we selected 10 candidates that met our criteria the most. Among these candidates, hsa\_circ\_0099768, hsa\_circ\_0083543, hsa\_circ\_0043795, hsa\_circ\_0051666, and hsa\_circ\_0014824 were upregulated under apoptotic conditions whereas the rest of them were downregulated (**Table 2**). All candidates exhibited differential expression under cisplatin treatment whereas hsa\_circ\_0099768 was differentially expressed under all treatment conditions. hsa\_circ\_0083543 was the other candidate differentially expressed in cisplatin-, doxorubicin-, and anti-Fas-treated HeLa cells.

### RNAse R Treatment and qPCR Validation

Although the bioinformatics algorithms eliminate artifacts from the circRNAs, additional experimental analyses are required to call the RNA-seq reads as circRNAs. Typically, RNAse R treatment is used to eliminate linear RNAs to prevent the potential amplification of tandem duplications (Panda et al., 2017). Additionally, the amplified products are sequenced to conclusively state the presence of a circRNA. To this extent, we used the total RNAs from control and cisplatin-treated cells (**Figure 2A**) and chose circ-HIPK3 as a positive control to test circRNA validation. Li et al. (2017) reported that circ-HIPK3 is indeed expressed in HeLa cells. We designed divergent primers using the Circinteractome tool (Dudekula et al., 2016) to amplify the positive control circ-HIPK3. qPCR was then performed with cDNAs that were reverse transcribed from both RNAse R<sup>+</sup> and RNAse R<sup>−</sup> samples.



Circular RNAs that were significantly down- and up-regulated upon treatment with all agents is listed below with their parental coding mRNAs. The majority of top 5 upregulated and downregulated circular RNAs originated from protein coding genes well-documented to regulate apoptosis. P-value, log change and apoptotic regulator function of the origin linear mRNA were considered as criteria in candidate selection. <sup>∗</sup> denotes the candidate common in all drug/ligand groups, ∗∗ denotes the candidate common in cisplatin, doxorubicin and anti-Fas group.

First, we examined the efficiency of RNAse R treatment to eliminate linear RNAs using DMSO-treated control RNAs. As shown in **Figure 2B**, RNAse R treatment resulted in the loss of the linear beta actin mRNA (Lane 3 vs. 4). However, the positive control circ-HIPK3 was resistant to the treatment with RNAse R (**Figure 2B**, Lane 2). We also validated the resistance of our candidates, hsa\_circ\_0012992, hsa\_circ\_0014824, and hsa\_circ\_0029693 to RNAse R (**Figure 2B**, Lanes 1 and **2C**). All these analyses showed that we were able to eliminate the linear RNAs and enrich the circular forms under our experimental conditions.

We then performed qPCR analyses with total RNAs to quantitatively measure the circRNA transcripts in cisplatintreated cells to validate circRNA-seq data. To this extent, we subjected three candidates, hsa\_circ\_0012992, hsa\_circ\_0099768, and hsa\_circ\_0029693 to qPCR analyses, which were shown to exist in the circular form (**Figure 3**). As shown in **Figure 3**, hsa\_circ\_0012992 and hsa\_circ\_HIPK3 were downregulated 3.3- and 2-fold, respectively, while hsa\_circ\_0099768 and hsa\_circ\_0029693 were upregulated 6.7- and 5.6-fold, respectively. We then carried out cisplatin treatments in two other cell lines, Jurkat and MCF-7, to examine whether this expression is cell-specific (**Figure 3**). Hsa\_circ\_HIPK3 expression was 2-fold downregulated in HeLa cells whereas its expression was upregulated in MCF7 and Jurkat cells 1.2 and 2-fold respectively. Expression of hsa\_circ\_0099768 was upregulated in HeLa and Jurkat cells. HeLa cells possessed the highest differential expression with a 6.7-fold change. Hsa\_circ\_0029693

was upregulated 5.6- and 2.1-fold in HeLa and MCF7 cells, respectively, while its expression was downregulated by 2.7 fold in Jurkat cells. All these results show that the induction of apoptosis results in differential expression of a number of circRNAs and their expression appears to be cell-specific at least for the candidate that we have analyzed in this study. Although our data provide an indirect relationship between the differential expression of certain circRNAs and apoptosis, further experiments are required to establish a direct link between an individual circRNA and apoptosis. For example, these candidates should be silenced or over-expressed to demonstrate the direct genotype-phenotype relationship.

Molecular details of circRNA biogenesis are still in progress. It needs to be resolved whether canonical splicing precedes back-splicing or vice versa (exon skipping vs. direct backsplicing) (Chen and Yang, 2015). Recent evidence points to the use of constitutive exons in circRNA biogenesis at the expense of their linear counterpart (Aufiero et al., 2018). Thus, we subjected linear counterparts of the candidate circRNAs to qPCR analyses to examine whether circRNA expression is correlated with the template mRNA expression (**Figure 4**). LATS2, USP33, and FGF14 were the linear counterparts of hsa\_circ\_0029693, hsa\_circ\_0012992, and hsa\_circ\_0099768, respectively. In cisplatin-treated HeLa cells, USP33, the gene encoding a ubiquitin specific peptidase, was downregulated 2-fold in parallel with its circular counterpart hsa\_circ\_0012992, while hsa\_circ\_0099768 was upregulated 6.7-fold in apoptotic samples in contrary to its linear counterpart FGF14, the gene encoding fibroblast growth factor (**Figure 4**). The other candidate, has\_circ\_0029693 was also subjected to qPCR analysis with LATS2 in four drug/ligand-treated apoptotic samples. Both circular and linear counterparts were upregulated in TNF-alphainduced apoptotic samples in 3.5- and 5.9-fold, respectively (**Figure 4**). These results suggest that has\_circ\_0099768 biogenesis is probably independent from its template mRNA biogenesis while the biogenesis of the other two candidates is correlated with the expression level of their template mRNAs. However, further analyses are required to examine the correlation between alternative splicing and circRNA biogenesis.

## CircRNA-miRNA Interactions and Coding Potential

One potential function of differentially expressed circRNAs is to serve as miRNA sponges to regulate the functions of miRNAs and their target mRNAs (Chen, 2016). To examine whether our circRNAs might regulate apoptosis through miRNA-circRNA interactions, we checked the potential miRNA binding sites on the candidate circRNAs. Origin mRNAs of candidate circular RNAs are obtained from CircInteractome tool by entering circular RNA ID. Potential miRNA binding sites were obtained by using starBase v3.0 which is an open-source platform for studying the miRNA-circRNA interactions from CLIP-seq (Li et al., 2014). To qualify as a miRNA sponge, a circRNA is expected to harbor multiple binding sites for the same miRNA as illustrated in the case of cdr1as-miR-7 interaction (Li et al., 2014; Chen, 2016).

We prioritized the cisplatin-treated samples due to the cancer therapeutic potential of cisplatin. Interestingly, 71 out of 109 circRNAs possessed miRNA binding sites. Those 71 circRNAs were then classified based on the number of binding sites. miRNAs that were likely to bind to the same circRNA at least five times were analyzed by miRSystem. The target mRNAs of potentially sponged miRNAs were subjected to KEGG functional annotation analysis and the resulting top 10 pathways were presented in **Figure 5**. This analysis showed that the potential miRNA-circRNA interaction under our experimental setting is likely to modulate cellular processes such as pathways in cancer, cell adherence function and MAPK signaling pathway (**Figure 5**).

Perhaps, one of the most interesting functions of circRNAs is their translatability (Pamudurti et al., 2017). Potential translation of circRNAs could generate a truncated protein that could have a diverse array of functions in the cell. Thus, we also examined the translation potential of our candidates. Hsa\_circ\_0029693 originates from the 2nd exon that includes half of the 5'UTR and the first translated exon of the linear source gene LATS2. The resulting circRNA is 444 nt that harbors a 147-aa open reading frame and an in-frame stop codon (**Figure 6A**). The 3D structure of putative truncated protein product of has\_circ\_0029693 was created by SWISS-MODEL<sup>7</sup> . It might be a homodimer protein that contains caspase-3 and caspase-7 cleavage site motif but loses its protein kinase domain **(Figure 6A**). We used SRAMP (see text footnote 5) to check for potential m6A sites as these sites are potential indicator of translatability (Zhou et al., 2016). When we subjected has\_circ\_0029693 to such an analysis, a very high confident m6A site motif was detected in a site close proximity to translation start codon. Additionally, an IRES site was detected nearby the start codon (**Figure 6B**). All these analyses suggest that some of our candidates hold the potential for translation that could generate truncated products. Further experimental evidence would be required to conclusively demonstrate the translatability of circRNAs. For example, the polysome association of circRNAs would be a good support for their translatability. Perhaps, the most direct evident would be to detect these products (either the

<sup>7</sup> swissmodel.expasy.org

endogenous product or the over-expressed product) through mass spectroscopy. It will be quite interesting to investigate the potential role of these truncated products as they might potentially compete with the full- length proteins or have an entirely different function.

#### CONCLUSION

Apoptosis is a highly regulated cellular process that is important for cell survival and cell death and it is involved in various physiological as well as pathological conditions. In this study, we provide a comprehensive profile of circRNAs in HeLa cells to another (**Figure 1G**). At least in HeLa cells, TNF-α and cisplatin are more potent in circRNA induction. However, doxorubicin and anti-Fas mAb treatment caused a moderate level of circRNA differential expression. It is difficult to account for this difference in the number of differentially expressed circRNAs. The consistent read number in all three biological replicates potentially point to a drug-specific response rather than a technical problem but this point requires further investigation. We then validated the existence of at least three circRNA candidates by PCR-amplification (**Figure 2**) and sequencing of the amplified fragments. qPCR analyses showed cell specific expression of circ-HIPK3 and hsa\_circ\_0029693 in HeLa, MCF-7 and Jurkat cells (**Figure 3**).

Interestingly, circRNAs, differentially expressed under apoptotic conditions, house binding sites for miRNAs reported to regulate apoptosis, suggesting a circRNA-miRNA-mRNA regulatory loop (**Figure 5**). Although we do not provide any experimental evidence at this point, our bioinformatics analyses provide an indirect link for a potential interaction between candidate circRNAs and miRNAs known to regulate apoptosis. However, more direct evidence would involve coimmunoprecipitation of circRNAs and miRNAs followed by PCR-amplification of the putative miRNAs. Also, silencing and/or over-expression would be helpful in demonstrating such direct interactions and their effect on the apoptotic phenotype. Our bioinformatics analyses also yielded interesting data at least with respect to the translatability of one candidate circRNA, has\_circ\_0029693. However, more solid evidence is required to claim the translation of a protein product from this circRNA.

In conclusion, although more experimental evidence is needed to demonstrate a direct link between differentially expressed candidate circRNAs and apoptosis, our data provide the first transcriptomics profile of circRNAs in HeLa cells. These data can be used in future studies to establish a direct link between candidates and apoptosis and also to study the effects of miRNA-circRNA interactions and putative truncated proteins on apoptosis.

#### DATA AVAILABILITY

fgene-10-00176 March 12, 2019 Time: 16:38 # 9

Raw sequence reads have been deposited in the GEO database (GSE125249).

### AUTHOR CONTRIBUTIONS

BA obtained the funding and designed the experiments. BY and IE performed the experiments. BY performed bioinformatics

#### REFERENCES


analyses. BA, BY, and IE analyzed the data and wrote and approved the manuscript.

#### FUNDING

This study was funded by TUBITAK (Project No: 215Z081 to BA).

#### ACKNOWLEDGMENTS

The authors would like to thank BIOMER (Iztech, Turkey) and the specialists Ozgur Okur and Dane Ruscuklu for flow cytometry analyses and sequencing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00176/full#supplementary-material

HeLa cells to fas receptor-mediated apoptosis. Mol. Cell. Biol. 19, 5991–6002. doi: 10.1128/MCB.19.9.5991


focal ischemia. Stroke 48, 2451–2548. doi: 10.1161/STROKEAHA.117. 017469


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

Copyright © 2019 Yaylak, Erdogan and Akgul. 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.

# Analyses of a Panel of Transcripts Identified From a Small Sample Size and Construction of RNA Networks in Hepatocellular Carcinoma

Zhiyong Sheng1,2, Xiaolin Wang<sup>2</sup> , Geliang Xu<sup>1</sup> \*, Ge Shan2,3 \* and Liang Chen<sup>2</sup> \*

<sup>1</sup> Anhui Province Key Laboratory of Hepatopancreatobiliary Surgery, First Affiliated Hospital of University of Science and Technology of China, Hefei, China, <sup>2</sup> Hefei National Laboratory for Physical Sciences at Microscale, The Chinese Academy of Sciences (CAS) Key Laboratory of Innate Immunity and Chronic Disease, School of Life Sciences, University of Science and Technology of China, Hefei, China, <sup>3</sup> Chinese Academy of Sciences (CAS) Centre for Excellence in Molecular Cell Science, Shanghai Institutes for Biological Sciences, CAS, Shanghai, China

#### Edited by:

Subbaya Subramanian, University of Minnesota Twin Cities, United States

#### Reviewed by:

Daisuke Kaida, University of Toyama, Toyama, Japan Dominique Belin, Université de Genève, Switzerland

#### \*Correspondence:

Geliang Xu xugeliang2007@163.com Ge Shan shange@ustc.edu.cn Liang Chen anqingcl@ustc.edu.cn

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 14 November 2018 Accepted: 23 April 2019 Published: 08 May 2019

#### Citation:

Sheng Z, Wang X, Xu G, Shan G and Chen L (2019) Analyses of a Panel of Transcripts Identified From a Small Sample Size and Construction of RNA Networks in Hepatocellular Carcinoma. Front. Genet. 10:431. doi: 10.3389/fgene.2019.00431 Hepatocellular carcinoma (HCC) is one of the most common cancers in the world. Dysregulation of mRNAs and non-coding RNAs (ncRNAs) plays critical roles in the progression of HCC. Here, we investigated HCC samples by RNA-seq and identified a series of dysregulated RNAs in HCC. Various bioinformatics analyses established long non-coding RNA (lncRNA)-mRNA co-expression and competing endogenous RNA (ceRNA) networks in circRNA-miRNA-mRNA axis, indicating the potential cis and/or trans regulatory roles of lncRNAs and circRNAs. Moreover, GO pathway analysis showed that these identified RNAs were associated with many biological processes that were related to tumorigenesis and tumor progression. In conclusion, we systematically established functional networks of lncRNA-mRNA, circRNA-miRNA-mRNA to further unveil the potential interactions and biological processes in HCC. These results provide further insights into gene expression network of HCC and may assist future diagnosis of HCC.

Keywords: hepatocellular carcinoma, lncRNA, miRNA, circRNA, network

## INTRODUCTION

Hepatocellular carcinoma (HCC) is one of the most frequent malignancies worldwide, and the highest incidence rates in the world are reported in Asia and Africa, with China alone accounting for half of the cases in Asia and Africa (McGlynn and London, 2011). Clinical investigation shows that HCC is the sixth most common cancer and third main cause of cancer mortality in the world (Forner et al., 2018). The overall 5 years survival rate is less than 20% due to lack of early and timely detection and treatment (Lee et al., 2012). So far, the available treatment approaches for HCC mainly include resection, liver transplantation, image-guided tumor ablation, and systemic therapy. The 5 years survival rates of 60–70% can be achieved in early-stage HCC patients (Llovet and Bruix, 2003; Llovet et al., 2003). However, systematic therapies cannot improve the survival rates of the patients with advanced stage of HCC due to limited specific and effective biomarkers or targets for clinical treatments (Llovet and Bruix, 2003; Llovet et al., 2003; Marquardt et al., 2012).

Dysregulated gene expression is a common theme underneath human disease (Adams and Cory, 2007; Aguirre-Ghiso, 2007; Amaravadi and Thompson, 2007; Azam et al., 2010; Hanahan and Weinberg, 2011). The resulted abnormal mRNAs and non-coding RNAs (ncRNAs) play roles in the occurrence and progress of diseases such as cancers (Prensner et al., 2011; Kung et al., 2013; Iyer et al., 2015; Rusk, 2015; Klingenberg et al., 2017). Many mRNAs are encoded by tumor genes or tumor suppressor genes, and ncRNAs such as microRNAs (miRNAs), long ncRNAs (lncRNAs), and circular RNAs (circRNAs) are also known to be associated with tumorigenesis and tumor progression (Panzitt et al., 2007; Wang et al., 2010; Wei et al., 2012; Zhang et al., 2012; Li G. et al., 2015; Palazzo and Lee, 2015; Zhou et al., 2016; Cui et al., 2017; Shen et al., 2018; Zhang and Wu, 2018).

In HCC, tumor genes such as procollagen-lysine, 2-oxoglutarate 5-dioxygenase 3 (PLOD3), and splicing factor 3b subunit 4 (SF3B4) are overexpressed in HCC tumor tissues compared with adjacent tissues, and knocking down their expression levels can effective suppress tumor invasion (Shen et al., 2018). Conversely, tumor suppressor genes such as Alcohol dehydrogenase 4 (ADH4) is significantly downregulated in tumor tissues, and patients with lower ADH4 expression levels have worse prognosis and lower overall survival rate (Wei et al., 2012).

In HCC, many miRNAs are found to play oncogenic or tumor suppressive roles (Zhang et al., 2012; Li G. et al., 2015; Zhou et al., 2016). For example, miR-421 can promote proliferation and migration in HCC cell by downregulating farnesoid X receptor (FXR), which is usually highly expressed in normal liver (Zhang et al., 2012). On the other hand, miR-708 is markedly downregulated in HCC tumor tissues compared with adjacent tissues. Low miR-708 level is related to tumor node metastasis (TNM) in advanced stage patients, and over expression of miR-708 can suppress the invasion and migration of in HCC cell lines in vitro (Li G. et al., 2015).

LncRNAs are also correlated to HCC (Chen et al., 2018). The lncRNA HULC is upregulated in HCC with high specificity (Panzitt et al., 2007; Wang et al., 2010). It can be detected both in tumor tissues and blood of HCC patients. Aberrantly elevated HULC promotes HCC invasion and progression by increasing expression of the oncogene HMGA2. The lncRNA PVT1 enhances the tumorigenesis and metastasis of HCC by acting as a competing endogenous RNA (ceRNA) for miR-186-5p, and the knockdown of PVT1 can effectively impede HCC invasion (Karreth et al., 2015; Cui et al., 2017).

CircRNAs are a large class of transcripts in multicellular organisms with emerging importance (Chen and Shan, 2015; Chen et al., 2015; Haque and Harries, 2017). One of functional mechanisms of circRNAs is acting as ceRNA (Hansen et al., 2013; Memczak et al., 2013). For example, the circRNA CDR1as functions as sponge of miR-7 in neuronal tissues (Memczak et al., 2013; Piwecka et al., 2017). Alterations in circRNA expression are found to be related to aberrant physiology and human diseases (Hansen et al., 2013; Memczak et al., 2013; Chen and Shan, 2015; Chen et al., 2015; Holdt et al., 2016; Piwecka et al., 2017).

In this study, we have identified differentially expressed mRNAs, miRNAs, lncRNAs, and circRNAs in fresh HCC tissues through high-throughput RNA sequencing. Functional networks of lncRNA-mRNA and circRNA-miRNA-mRNA have been established to provide new insights for biomarkers and treatments in HCC.

## MATERIALS AND METHODS

## Clinical Samples

All fresh HCC patient tumor samples and adjacent tissues were collected from The First Affiliated Hospital of University of Science and Technology of China, which was approved by the Human Research Ethics Committee of University of Science and Technology of China (USTCEC201700007). Written informed consent was obtained from each patient for this study. All samples were rinsed with DEPC water and then kept in RNAhold (TransGen) within 30 min after removing from the operation. HCC patient tumor sample and adjacent tissue pairs were collected from 21 patients (12 males and 9 females with advanced stage HCC, all of them were HBsAg positive, and did not have anti-tumor therapy before surgery).

### Total RNA Extraction

The clinical samples were cut into small pieces and homogenized in TRizol reagent (Life Technologies) by homogenizer. Total RNA was extracted by using TRizol reagent according to the manufacturer's instructions.

### Transcriptome Data Analysis

Total RNAs from four pairs of HCC patient tumor tissues were extracted for high-throughput sequencing. Whole transcriptome libraries were constructed by the TruSeq Ribo Profile Library Prep Kit (Illumina, United States), according to the manufacturer's instructions. In brief, 10 µg total RNA was depleted rRNA with an Illumina Ribo-Zero Gold kit and purified for end repair and 5<sup>0</sup> -adaptor ligation. Then, reverse transcription was performed with random primers containing 3 0 adaptor sequences and randomized hexamers. Finally, the cDNAs were purified and amplified with thermo cycler. The PCR products of 300–500 bp were purified, quantified and stored at -80◦C before sequencing. The libraries were subjected to 151 nt paired-end sequencing with an Illumina Nextseq 500 system (Novogene, China). Each library was generated a depth of 50–100 million read pairs and then adapters were removed with cutadapt to obtain clean reads. For mRNA and lncRNA analyses, the expression levels were calculated using TopHat2 and Cufflinks followed by the annotation references of Refseq and Ensemble transcript databases with the genome release Homo sapiens, hg19. The differentially expressed mRNA and lncRNA were determined by DEseq2 with the corresponding cutoff (P < 0.001, RPKM ≥ 10, | log2(fold change)| ≥ 1 for mRNA and P < 0.05, RPKM ≥ 1, | log2(fold change)| ≥ 1 for lncRNA). For circular RNA (circRNA) prediction, we identified the candidates with find\_circ (Memczak et al., 2013) and the junction reads were calculated as Transcripts Per Kilobase Million (TPM). A criterion of P ≤ 0.05, TPM ≥ 0.1 and | log2(fold change)| ≥ 1 among four pairs of samples was used to identify differentially expressed

#### TABLE 1 | Primers for RT-qPCR validation.

fgene-10-00431 May 7, 2019 Time: 16:50 # 3


Sheng et al. Transcripts and RNA Networks in Hepatocellular Carcinoma

#### TABLE 1 | Continued


SL, Stem loop RT primer.

circRNAs. RNA-seq data were deposited in NCBI with the GEO accession code GSE128274.

#### Small RNA Data Analysis

For small RNA (sRNA) sequencing, eight sRNA libraries were generated with TruSeq small RNA (Illumina, United States) according to the manufacturer's instructions. Then the prepared libraries were sequenced with an Illumina Nextseq 500 system (Novogene, China). After filtering out the reads shorter than 15 nt, the remaining reads were mapped to the human genome (hg19) and the miRNA database in miRBase with bowtie (-v 1). The differentially expressed miRNAs were determined by DEseq2 with the cutoff of P < 0.05, TPM ≥ 1, | log2(fold change)| ≥ 1.

### Construction of Co-expression and CeRNA Network

For the co-expression network of significantly dysregulated lncRNAs and mRNAs, Pearson's correlations were calculated the co-expression analysis according to the expression levels in eight samples. A criterion of the coefficient parameter R-squared more than 0.99 was used for the remaining RNAs to further construct the network. For the competing endogenous RNAs (ceRNA) network of significantly dysregulated circRNAs and mRNAs, the miRNA/mRNA and miRNA/circRNA interaction were predicted with TargetScanHuman7.2. The above networks were both performed with Cytoscape<sup>1</sup> .

#### GO Analysis

The significantly dysregulated mRNAs in co-expression and ceRNA network were both analyzed using GOrilla web-server with default parameters (Eden et al., 2009).

(Continued)

<sup>1</sup>www.cytoscape.org

#### Statistical Analysis

In all experiments, Student's t-tests were used to calculate P-values, as indicated in the figure legends. The values reported in the graphs represent averages of actual number of independent experiments, with error bars showing SD. After analysis of variance with F-tests, the statistical significance and P-values were evaluated with Student's t-tests.

### Reverse Transcription and Real-Time Quantitative PCR

cDNA was prepared using GoScript Reverse Transcription System (Promega) according to the manufacturer's protocol. Notably, for the first strand cDNA synthesis of miRNA, stem-loop method was used (Kramer, 2011). Quantitative real-time PCR was performed with GoTaq SYBR Green qPCR Master Mix (Promega) on a PikoReal 96 real-time PCR system followed by 40 amplification cycles (Thermo Fisher Scientific) according to standard procedures. Actually, all amplification curves already reached stationary stage before 35 amplification cycles, and the readings of Ct value were obtained at the exponential stage. Relative RNA expression was normalized to ACTB expression level. All primers are shown in **Table 1**.

## RESULTS

### Identification of Differentially Expressed RNAs in HCC Samples

To identify HCC-related RNAs, we used four HCC patients' fresh tumor tissues and paired adjacent non-tumor tissues for RNA sequencing. Differential expression of mRNAs, lncRNAs,

significant difference.

circRNAs, and miRNAs were then analyzed (**Figure 1A**). Pearson's correlation coefficient analysis showed that tumor tissues were positively correlated with each other, and control tissues also showed strong positive correlation (**Figure 1B**). It seems evident that the 4 tumor samples differ more from normal samples than among themselves, maybe due to the tumor heterogeneity. We noticed that previously reported transcripts with aberrant levels in HCC such as mRNAs (PLOD3, SF3B4, ADH4, and COLEC10), lncRNAs (HULC and SNHG7), and miRNAs (miR-421 and miR-761) were also identified in our RNA-seq (**Figure 1C**; Panzitt et al., 2007; Wang et al., 2010; Wei et al., 2012; Zhang et al., 2012; Li G. et al., 2015; Zhou et al., 2016; Shen et al., 2018; Zhang and Wu, 2018).

Next, we performed volcano plots for mRNAs, lncRNAs, miRNAs and circRNAs in HCC paired tissues (fold change ≥ 2.0, P < 0.05) (**Figures 2A–D**). 919 differentially expressed mRNAs, 207 lncRNAs, 216 miRNAs, and 152 circRNAs were identified (**Figures 2A–D**). Among them, 452 mRNAs (49.18%), 116 lncRNAs (50.04%), 138 miRNAs (63.89%), and 50 circRNAs (32.89%) were upregulated, while 467 mRNAs (50.82%), 91 lncRNAs (43.96%), 78 miRNAs (36.11%), and 102 circRNAs (67.11%) were downregulated (**Figures 2A–D**). We also conducted a hierarchical cluster analysis to display the differential expression of four types of RNAs across eight samples (**Figures 2A–D**). Tumor and adjacent non-tumor samples respectively were classified into different branch (**Figures 2A–D**).

### RT-qPCR Verification of the Dysregulated RNAs in Clinical Samples

To validate the results of the RNA-seq, we selected 23 dysregulated RNA candidates including mRNAs, lncRNAs, miRNAs and circRNAs for RT-qPCR validation with 21 HCC patient sample pairs (**Figures 3A–D**). Housekeeping gene ACTB was used as the endogenous control. In the mRNA group, ZIC5, C12orf75, C1QL1, TMEM74, and GNAZ were significantly upregulated in tumor tissues compared to the adjacent control; PZP and FAM65C were significantly downregulated compared to the adjacent control (**Figure 3A**). In the lncRNA group, lncRNA-TOB2P1, LOC100499489, lnc-NMRAL2p, and lnc-NBPF22P were markedly upregulated in tumor tissues, while LINC01093, LOC100130899, LOC200772, and lnc-FENDRR were significantly downregulated in tumor tissues, compared to the adjacent controls (**Figure 3B**). In the miRNA group, miR-10-3p was significantly upregulated in tumor tissues, while miR-200a-3p, miR-200b-3p, and miR-139-5p were significantly downregulated in tumor tissues (**Figure 3C**). In the circRNA group, circAKR1B10 and circAKR1C3 were significantly upregulated in tumor tissues, while circHMGCS1 and circC3P1

miRNAs. Boxes represent dysregulated mRNA, triangles represent dysregulated miRNA, circles represent dysregulated circRNAs. Red, upregulated; blue, down-regulated. (B) GO analyses of affected pathways in three main categories: cellular components, biological processes and molecular functions.

were markedly downregulated, compared to the adjacent controls (**Figure 3D**). For these 23 RNAs examined with 21 pairs of patient samples, the RT-qPCR verification was in accordance with RNA-seq results with high confidence (**Figure 3E**).

We also validated those transcripts with aberrant levels in HCC such as mRNAs previously identified by other studies (**Figure 1C**), with the same 21 HCC patient sample pairs with RT-qPCR (**Figure 4**). Among the eight transcripts (PLOD3, SF3B4, ADH4, COLEC10, HULC, SNHG7, miR-421, and miR-761), only two (ADH4 and COLEC10) transcripts were in accordance with RNA-seq results and previous reports (**Figure 4**; Wei et al., 2012; Zhang and Wu, 2018).

### DISCUSSION

#### Co-expression Network of lncRNAs/mRNAs and GO Analysis

We then set out to investigate the identified lncRNAs associated with HCC. LncRNA functions can be predicted based on the functions of their co-expressed protein-coding genes, and alterations in the associations between these genes in clinical samples can be used to identify key lncRNAs in HCC (Cabili et al., 2011; Liao et al., 2011; Guo et al., 2013). We constructed a co-expression network of lncRNAs and co-expressed mRNAs based on the RNA-seq data to investigate their interactions (**Figure 5A**). Our analysis demonstrated that 98 lncRNAs interacted with 175 mRNAs (**Figure 5A**). The results of GO pathway analyses of the differentially RNAs showed that most co-expressed lncRNAs were closely related to several important pathways, including biological processes such as gene silencing, chromatin silencing, and response to stress, cellular components such as extracellular region, and molecular functions such as protein dimerization activity (**Figure 5B**).

#### Construction of CeRNA Network

One of molecular functions of circRNAs is ceRNA (Hansen et al., 2013; Memczak et al., 2013; Piwecka et al., 2017). In order to investigate the potential circRNAs acting as ceRNAs in HCC through regulating miRNAs and consequently modulating mRNAs, we constructed a ceRNA network among differentially expressed circRNAs, miRNAs, and mRNAs in HCC (**Figure 6A**). 15 circRNAs (**Table 2**), 17 miRNAs, and 89 mRNAs were found to be correlated in this ceRNA network. The results of GO pathway analyses showed that mRNAs in this network were correlated to regulations of biological processes such as protein phosphorylation, signal transduction, and cell proliferation, molecular functions such as transcription regulator activity and double-stranded DNA binding, and cellular components such as Rb-E2F complex (**Figure 6B**).

#### Survival Curves of Identified Genes

To explore the relationship between our identified targets and clinical observations, we then examined with survival curves in online database, UALCAN analysis<sup>2</sup> . Total of 9 mRNAs, 7 of them TABLE 2 | Information of circRNAs in Figure 5A.


identified in our RNA-seq and qRT-PCR verification and 2 of them previously identified and further verified by this study, were analyzed with survival curves. 6 (ZIC5, C12orf75, PZP, FAM65C, ADH4, and COLEC110) out of the 9 mRNAs were correlated to survival curves with significance in HCC (**Figure 7**). Those genes we found upregulated in HCC (ZIC5, C12orf75) were positively correlated with survival curves (patients with higher expression levels in HCC survive shorter). Genes downregulated in HCC (PZP, FAM65C, ADH4, and COLEC110) were negatively correlated with survival curves (patients with higher expression levels in HCC survive longer). For other identified genes including significantly upregulated/downregulated mRNAs, lncRNAs, miRNAs, and circular RNAs, the UALCAN database does not have HCC-relevant information about them.

The incidence of HCC has been increasing, and the consequent mortality is also rising for the past decades (Forner et al., 2018). For the early-stage patients, it is amenable to potentially curative treatments such as resection, liver transplantation, image-guided tumor ablation and systemic therapy, which can increase the 5 years survival rates to 60–70% (Llovet and Bruix, 2008; Yu, 2016). However, patients who are diagnosed at the advanced stage of HCC are treated poorly due to lack of effective treatment options for potential liver disease. Early diagnosis and effective surveillance are required for the treatment of HCC patients to reduce the disease-related mortality. Future diagnosis and treatments call for novel HCC biomarkers and potential targets.

RNA biomarkers, as measurable clinical indicators, can be used to predict and detect some diseases state and symptoms outside the body of patient with unique advances. To provide effective treatment for HCC patients and insights for future diagnosis, several potential RNA biomarkers for HCC have also been investigated (Klingenberg et al., 2017). Accumulating evidences have shown that lncRNA and miRNAs are suitable potential markers for HCC (Xie et al., 2016; Birgani et al., 2018). It has been reported that more than half of the miRNAs genes are located in cancer-associated genomic regions or in fragile sites

<sup>2</sup>http://ualcan.path.uab.edu/analysis.html

(Greene et al., 2017). Aberrant expression of lncRNAs, miRNAs, and circRNAs along with mRNAs may directly or indirectly lead to the progression of cancers due to their massive involvement (Ryan et al., 2010; Cheetham et al., 2013; Guarnerio et al., 2016).

RNA-seq is a powerful tool to study and detect the global transcriptome in tissues and cells (Sharma et al., 2010; Adey et al., 2013). In this study, we have identified 919 differentially expressed mRNA, 207 lncRNAs, 216 miRNAs, and 152 circRNAs in HCC through RNA-seq, these dysregulated RNAs, especially those validated with 21 patient HCC samples can be highlighted as potential biomarkers or therapeutic targets for HCC (**Figure 3**).

Although increasing pieces of evidence have demonstrated the role of aberrant expression of mRNA, miRNA, lncRNA, and circRNA in HCC, not many studies have systematically investigated the crosstalk among transcripts in this context. The co-expression network between lncRNA and mRNA (**Figure 5A**) and the ceRNA network of differential expression circRNA-miRNA-mRNA (**Figure 6A**) in our study provide insights for further investigation. Of course, both networks have their own limitation. For example, circRNAs can function not only as ceRNAs but also as transcriptional regulators (Li Z. et al., 2015; Hu and Zhou, 2018). Another interesting point is that circRNAs seem to be more often downregulated in tumor tissues as shown in this study as well as several other studies (**Figure 2D**; Zheng et al., 2016; Greene et al., 2017).

We examined the 8 RNAs reported in other studies, however only two were found consistent with our RT-qPCR results from 21 patients (**Figure 4**). The inconsistence may be due to the fact that all patients in this study were HBsAg positive with advanced stage HCC, and our patient cohort may be distinct from previous studies. We did not have an opportunity to investigate the potential exposure of the main HCC carcinogen aflatoxin of these patients, which may be a weakness of this study. Another limitation of the present study is that, just like most Chinese HCC patients, all patients in this study are already in the advanced stage upon their first diagnosis, due to limited coverage of preclinic screening. We also explored the correlation of the RNAs identified in our study with patient survival curves. 6 out of the 9 mRNAs were correlated to survival curves of HCC, indicating multiple transcripts identified in this study may play critical roles in the tumorigenesis and advance of HCC (**Figure 7**).

#### CONCLUSION

In conclusion, we have provided a comprehensive identification and analyses of the differentially expressed mRNAs,

#### REFERENCES


miRNAs, lncRNAs, and circRNAs using RNA-seq, and some of these transcripts have been verified with clinic HCC samples. Functional network of lncRNA-mRNA and circRNA-miRNA-mRNA ceRNA network have been systematically established to further indicate potential interactions in HCC. GO pathway analyses also facilitate future studies on the specific mechanisms of HCC. We expect this work will serve as a valuable resource in future clinical diagnosis and therapy of HCC.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the Human Research Ethics Committee of University of Science and Technology of China (USTCEC201700007) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Human Research Ethics Committee of University of Science and Technology of China (USTCEC201700007).

### AUTHOR CONTRIBUTIONS

GX, GS, and LC designed and initiated this project. GS provided the major funding. GS and LC supervised the experiments. ZS and XW performed all the experiments. ZS, XW, LC, and GS analyzed the data and wrote the manuscript. All authors have discussed the results and made comments on the experiment.

### FUNDING

This work was supported by the National Basic Research Program of China (2015CB943000), the National Key R&D Program of China (2018YFC1004500), the National Natural Science Foundation of China (31725016 and 31600657), and the Strategic Priority Research Program (Pilot study) "Biological basis of aging and therapeutic strategies" of the Chinese Academy of Sciences (XDPB10).

### ACKNOWLEDGMENTS

We thank the Bioinformatics Center of the USTC, School of Life Sciences, for providing supercomputing resources.

aneuploid HeLa cancer cell line. Nature 500, 207–211. doi: 10.1038/nature 12064



hepatocellular carcinoma: a single-center experience. Korean J. Hepatol. 18, 48–55. doi: 10.3350/kjhep.2012.18.1.48


following liver transplantation accounting for within-patient heterogeneity. BMC Med. Genomics 9:18. doi: 10.1186/s12920-016-0179-4


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

Copyright © 2019 Sheng, Wang, Xu, Shan and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(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.

# Long Non-coding RNA in Neuronal Development and Neurological Disorders

Ling Li1,2, Yingliang Zhuang1,2, Xingsen Zhao1,2 and Xuekun Li1,2 \*

<sup>1</sup> The Children's Hospital, School of Medicine, Zhejiang University, Hangzhou, China, <sup>2</sup> Institute of Translational Medicine, School of Medicine, Zhejiang University, Hangzhou, China

Long non-coding RNAs (lncRNAs) are transcripts which are usually more than 200 nt in length, and which do not have the protein-coding capacity. LncRNAs can be categorized based on their generation from distinct DNA elements, or derived from specific RNA processing pathways. During the past several decades, dramatic progress has been made in understanding the regulatory functions of lncRNAs in diverse biological processes, including RNA processing and editing, cell fate determination, dosage compensation, genomic imprinting and development etc. Dysregulation of lncRNAs is involved in multiple human diseases, especially neurological disorders. In this review, we summarize the recent progress made with regards to the function of lncRNAs and associated molecular mechanisms, focusing on neuronal development and neurological disorders.

#### Edited by:

Zhao-Qian Teng, Institute of Zoology (CAS), China

#### Reviewed by:

Jing He, Guangzhou Women and Children Medical Center, China Chang-Mei Liu, Chinese Academy of Sciences, China

> \*Correspondence: Xuekun Li xuekun\_li@zju.edu.cn

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 14 November 2018 Accepted: 27 December 2018 Published: 23 January 2019

#### Citation:

Li L, Zhuang Y, Zhao X and Li X (2019) Long Non-coding RNA in Neuronal Development and Neurological Disorders. Front. Genet. 9:744. doi: 10.3389/fgene.2018.00744 Keywords: long non-coding RNA, gene expression, neuronal development, neurological disorders, mechanism

### INTRODUCTION

Over the past several decades, advances in genomic sequencing technology and findings from largescale consortia have facilitated our understanding of the complexity and flexibility of mammalian genomes. The majority of mammalian genomes are transcribed, whereas only a few transcripts encode proteins, the majority of transcripts are non-coding RNAs (ncRNAs) (Roberts et al., 2014). Based on the length of transcripts, ncRNAs are usually classified into two categories: small noncoding RNAs and long non-coding RNAs (lncRNAs). Small ncRNAs are usually <200 nucleotides, including microRNAs, Piwi-interacting RNAs and small nuclear RNAs (snoRNAs). lncRNAs are >200 nucleotides and frequently transcribed by polymerase II, and share some features, e.g., 5<sup>0</sup> capping, 3<sup>0</sup> -polyadenylation, alternative splicing and sequence conservation with mRNA (Ponting et al., 2009; Nagano and Fraser, 2011).

Although lncRNAs generally lack protein coding capacity, spatiotemporal-specific expression patterns have highlighted the diverse functions and complicated mechanisms of lncRNAs (Cao et al., 2018). Currently, it is widely accepted that lncRNAs play an important function in a variety of biological processes, including regulating gene expression, both at the transcriptional and the post-transcriptional level, shaping the chromatin conformation and imprinting the genomic loci (Lee and Bartolomei, 2013; Chen, 2016; Cao et al., 2018), and multiple diseases such as neurological disorders, cancer, and immunological diseases (Bian and Sun, 2011; Huarte, 2015; Wan et al., 2017).

In this review, we summarize the recent progress made regarding the functions of lncRNAs, especially the functions and associated mechanisms related to neurological disorders.

### CHARACTERIZATION OF LncRNA

LncRNAs are generally transcribed from various genomic contexts and tend to have fewer exons than protein-coding transcripts (Iyer et al., 2015). Although there are still many challenges in annotation and interpretation of lncRNAs, because of the lack of an unambiguous classification framework, the existing lncRNAs can be subdivided into several categories based on their positional relation to protein coding genes, DNA elements or diverse mechanisms of processing (St Laurent et al., 2015; Kopp and Mendell, 2018) (**Figure 1**).

Sense lncRNAs are transcribed from the sense DNA strand, and have overlapping regions with protein-coding genes, including un-spliced sense partially intronic RNAs (PINs) and spliced transcripts resembling mRNAs (St Laurent et al., 2015). Further, natural antisense transcripts (NATs) of protein-coding genes have also been identified and many NATs share some opposite strand DNA sequences with the sense transcripts (Katayama et al., 2005). Some studies also indicate that NATs have either positive or negative effects on the corresponding sense transcripts or nearby protein-coding transcripts (Faghihi et al., 2010; Modarresi et al., 2012). For example, human brain-derived neurotrophic factor antisense RNA (BDNF-AS) was originally identified as natural antisense transcripts of neuronal transcriptional factor BDNF, shares 225 complementary nucleotides with BDNF mRNA and regulates the expression of BDNF both in vivo and in vitro (Modarresi et al., 2012;Fatemi et al., 2015).

Other studies indicate that intronic regions of coding genes produce a lot of lncRNAs. These intronic lncRNAs form the largest class of lncRNAs and are expressed independently from the pre-mRNA of protein coding genes. Many intronic lncRNAs fail to be debranched after splicing and form a covalent circle without 3<sup>0</sup> linear appendages, these circular intronic ncRNAs (ciRNAs) were found to play a regulatory role on their host genes (Zhang et al., 2013). In addition, circRNAs derived from the internal exons of pre-mRNAs through backsplicing, have also been found in various cell lines and tissues (Wu H. et al., 2017). These circular ncRNAs usually present tissue- and developmental stage-specific expression, such as the intensively studied cerebellar degeneration-related protein 1(CDR1as) (Memczak et al., 2013).

A relatively well-characterized subclass of lncRNAs is large/long intergenic or intervening non-coding RNAs (lincRNAs), and transcribed from the intergenic regions. LincRNAs have no overlapping sequences with transcripts of either protein-coding genes or other types of genes (Clark and Blackshaw, 2014). At the molecular level, most annotated lincRNAs have mRNA-like features including 5<sup>0</sup> -cap structures, 3 0 -poly(A) tails, exon–exon splice junctions and association with ribosomes (Cabili et al., 2011). Compared with mRNA counterparts, lincRNAs exhibit a more tissue-specific expression, a greater nuclear localization and less evolutionary conservation (Djebali et al., 2012).

Promoter upstream transcripts (PROMPTs) localize in a fairly narrow region between ∼500 and ∼2500 nucleotides upstream of transcription start sites of nearby active proteincoding genes (Preker et al., 2011; Lloret-Llinares et al., 2016). It was reported that the expression levels of certain PROMPTs are altered in stress conditions, such as DNA damage responses and osmotic responses (Lloret-Llinares et al., 2016; Song et al., 2018). Enhancer-related lncRNAs (eRNAs) are bidirectional transcripts of enhancers and have enhancer-like functions. Increased binding of DNA hydroxylase Tet1 and histone methyltransferases Mll3/Mll4 and DNA hypomethylation and H3K27ac modifications at enhancers, may activate eRNAs transcription. Both PROMPTs and eRNAs are targets of the RNA exosome and display similarities during processing (Andersson et al., 2014; Wu H. et al., 2017).

Emerging evidence indicates that telomeric repeat-containing RNA (TERRA) is a heterogeneous lncRNA consisting of a combination of subtelomeric and telomeric sequences. These sequences are mostly transcribed from intrachromosomal telomeric repeats by pol II and polyadenylated at 3<sup>0</sup> region (Luke and Lingner, 2009). The length and expression level of human TERRA is influenced by the telomere length. The vast majority of mouse TERRA-binding sites were found in distal intergenic and intronic regions, where TERRA may regulate expression of target genes (Chu et al., 2017; Diman and Decottignies, 2018).

SnoRNA-ended lncRNAs (sno-lncRNAs) are transcripts of one intron flanked by two snoRNA genes that can be further processed to form snoRNA. sno-lncRNAs can be stabilized by snoRNPs formed by snoRNAs and specific protein components. SLERT is a representative Box H/ACA snoRNA-ended lncRNA and has been reported to be translocated to the nucleus by snoRNAs to function in pre-rRNA biogenesis (Wu H. et al., 2017).

### PHYSIOLOGICAL FUNCTIONS OF LncRNA

Loss- and gain-of-function studies revealed that many lncRNAs are involved in various biological processes during development. Many lncRNAs have been found to regulate transcription via chromatin modulation, by working as molecular scaffolds for protein–protein interactions or interacting with chromatin modifying complexes and recruiting chromatin modifying complexes to specific loci, to activate or repress target gene expression. Some lncRNAs could affect transcription by modulating the binding of the general transcription machinery and regulatory factors (Wang and Chang, 2011; Fang and Fullwood, 2016; Wan et al., 2017; Lekka and Hall, 2018). Aside from modulating chromatin states, nuclear lncRNAs are involved in the RNA processing (Tripathi et al., 2010), turnover, silencing, translation and decay of mRNAs (Gong and Maquat, 2011; Carrieri et al., 2012; Geisler and Coller, 2013), or act as

miRNA decoys to neutralize miRNA-mediated mRNAs silencing and interact with signaling molecules, to modulate signaling pathways (Faghihi et al., 2010; Liu et al., 2015). In addition, some lncRNAs are determined to be precursors of certain miRNAs at particular stages of development (Dykes and Emanueli, 2017) (**Table 1**).

### LncRNAs and Stem Cell Pluripotency and Differentiation

Accumulating evidence suggests that lncRNAs exert critical functions in pluripotency maintenance, reprogramming and lineage differentiation of stem cells (Wang and Chang, 2011; Ghosal et al., 2013). The long intergenic non-protein coding RNA regulator of reprogramming (lincRNA-ROR), increases the reprogramming efficiency of human induced pluripotent stem cells (iPSCs) and promotes the maintenance of embryonic stem cells (ESCs) pluripotency (Loewer et al., 2010). Similar to a miRNA sponge, lincRNA-ROR forms a regulatory feedback loop with miR-145 and OCT4, SOX2, and NANOG, and regulates ESC pluripotency (Wang et al., 2013). MIAT (myocardial infarction associated transcript) is a co-activator of Oct4 and participates in OCT4 and NANOG regulatory networks in mouse ESCs. Loss of MIAT reduces the expression of Oct4, Sox2, and Klf4, and inhibits ESCs proliferation (Sheik Mohamed et al., 2010).

#### LncRNAs and Development

Genomic imprinting is an important epigenetic mechanism and is crucial for normal development in mammals. It restricts gene expression on one of the two parental chromosomes in diploid cells and affects both male and female descendants (Barlow and Bartolomei, 2014). H19, a maternally expressed 2.3 kb lncRNA, is generated from the highly conserved and imprinted vertebrate gene cluster insulin-like growth factor 2 (Igf2)/H19. H19 transcripts are the precursors of miR-675-3p and miR-675-5p (Cai and Cullen, 2007). Before parturition, H19 slows the growth of the placenta down partially, by down-regulating the RNA binding protein HuR. The decreased HuR cannot block the processing of miR-675, which further decreases the growth regulator Igf1r with Igf2 as its main ligand (Keniry et al., 2012). Another two well-characterized lncRNAs that have been found to regulate genomic imprinting are Kcnq1ot1 (KCNQ1 opposite strand transcript 1) and Airn (antisense of IGF2R non-protein coding RNA), both are paternally expressed and regulate transcriptional silencing through a multilayered silencing pathway (Perry and Ulitsky, 2016).

Besides genomic imprinting, dosage compensation plays a vital role in equalizing the dosage of X-linked genes between males and females in heterogametic species. Xist is a 17 ∼ 20 kb lncRNA and transcribed from the X inactivation center. During female development, Xist initiates X-chromosome inactivation (XCI), by progressively coating the future inactive X chromosome (Xi) and then utilizing its conserved A-repeat domain to bind PRC2, to form a transcriptionally silent nuclear compartment. The compartment is enriched by H3K27me3 and responsible for the chromosome-wide gene repression in the Xi (Chery and Larschan, 2014). TSIX, transcribed from the active chromosome (Xa), represses Xist at the early steps of X inactivation (Gendrel and Heard, 2014). Another lncRNA JPX/ENOX, which is transcribed from Jpx/Enox gene, that resides 10 kb upstream of

#### TABLE 1 | Diverse mechanisms of lncRNAs playing function.

fgene-09-00744 January 21, 2019 Time: 17:54 # 4

Frontiers in Genetics | www.frontiersin.org


**121**

TABLE 1 | Continued


Xist, also involves in XCI through repressing the TSIX expression from the Xi and evicting nuclear protein CCCTC-binding factor (CTCF) away from promoter of XIST to activate the XIST expression from Xi (Sun et al., 2013).

#### LncRNAs in Neurodevelopment

A great number of lncRNAs are expressed during neural development and in the brain. Using high-throughput technologies, in situ hybridization, microarray analysis and RNA sequencing (RNA-seq), researchers have found that most of the lncRNAs examined (849 out of 1328) are expressed in specific cell types, subcellular compartments and different regions of the brain (Ng et al., 2013b; Shi et al., 2017). LncRNAs display differential expressions across the cortical layers, and region-specific expressions in the subventricular zone, dentate gyrus and olfactory bulb of mice (Belgard et al., 2011; Ramos et al., 2013). Based on a unique custom microarray platform, 8 lncRNAs were identified to be expressed in an age-dependent manner, from 36 surgically resected human neocortical samples, ranging from infancy to adulthood (Lipovich et al., 2014). However, the function of those lncRNAs need to be validated, through loss-of-function assays, RNA-protein association assays or assessments of RNA-chromatin association.

Some lncRNAs are also found to participate in neural cell fate determination, neuronal-glia fate switching and oligodendrocyte elaboration. An antisense transcript of the distal-less homeobox 1 (Dlx1), Dlx1AS, was discovered to be up-regulated during GABAergic differentiation and downregulated during oligodendrocyte differentiation (Mercer et al., 2010). A subsequent study found that Dlx1AS participated in neurogenesis, implying its function in neuronal differentiation by regulating expression of its homeobox gene neighbors (Ramos et al., 2013). Evf2, a cloud-forming Dlx5/6 ultra-conserved eRNA, influences the formation of GABAergic interneurons in both a mouse and human forebrain. In the developing ventral forebrain, Evf2 regulates Dlx5, Dlx6 and glutamate decarboxylase 1 (Gad1) expression by recruiting DLX1 and/or DLX2 and methyl CpG binding protein 2 (MeCP2) to specific DNA regulatory elements. GAD1 is an enzyme responsible for catalyzing glutamate to form GABA. Evf2 mouse mutants reduced GABAergic interneurons in the early postnatal dentate gyrus and hippocampus (Bond et al., 2009; Berghoff et al., 2013; Cajigas et al., 2018).

The vertebrate retina is comprised of three well-organized cell type-specific neuron layers, interconnected by synapses (Ng et al., 2013b). Six3OS is the long non-coding opposite strand transcript (lncOST) of the homeodomain factor Six3. During mammalian eye development, Six3 regulates both early eye formation and postnatal retinal cell specification. Knockdown of Six3OS leads to a decrease of bipolar cells and an increase of Müller glia, similar to the results in the knockdown of Six3. In contrast, overexpression of Six3OS decreased syntaxin positive cells. Gene perturbation studies revealed that Six3OS participates in retinal cell specification as a molecular scaffold, to regulate Six3 activity rather than expression (Rapicavoli et al., 2011). TUG1 (taurine upregulated gene 1), a spliced and polyadenylated lncRNA, is highly conserved in humans and mice. TUG1 may participate in rod-photoreceptor genesis and inhibits cone-photoreceptor gene expression globally, by altering the chromatin configurations of photoreceptor-specific transcription factors Crx and Nrl (Young et al., 2005).

### LncRNAs IN NEUROLOGIC DISORDERS

Emerging evidence has shown that the dysregulation of lncRNAs is related to multiple neurological disorders, such as schizophrenia (Scholz et al., 2010), autism spectrum disorder (ASD) (Wang et al., 2015c), Parkinson's (Ni et al., 2017), Huntington's (Sunwoo et al., 2017) and Alzheimer's diseases (Faghihi et al., 2008).

Schizophrenia (SCZ) is a debilitating mental disorder with a broad spectrum of neurocognitive impairments. Abundant data suggests that both genetic and environmental factors contribute to the pathophysiology of SCZ (Seidman and Mirsky, 2017). Several lncRNAs have been used as biomarkers and therapeutic targets for SCZ (Chen et al., 2016). lncRNA MIAT, also known as Gomafu or RNCR2, is down-regulated in SCZ upon neuronal activation (Sun et al., 2018). Previous studies found MIAT either acts as a competitive endogenous RNA (ceRNA) for miR-150-5p, miR-24, miR-22-3p or miR-150, to influence cell proliferation, apoptosis and migration, or participates in various signaling pathways by enhancing Nrf2 (nuclear factor erythroid 2-related factor 2) and Oct4 expression. Subsequent studies revealed that MIAT can directly bind to the splicing regulator quaking homolog (QKI) and splicing factor 1 (SF1), to modulate several gene expressions in the neuron. In SCZ patient brains, DISC1 (disrupted in schizophrenia 1), ERBB4 (v-erb-a erythroblastic leukemia viral oncogene homolog 4) and their alternatively

spliced variants were all down-regulated due to MIAT upregulation. MIAT could act as a scaffold to affect alternative splicing of those SCZ-associated genes (Roberts et al., 2014; Liu et al., 2018; Sun et al., 2018).

Autism spectrum disorder (ASD) is a heterogeneous group of neurodevelopmental disorders characterized by impaired reciprocal social interactions, communication, and repetitive stereotyped behaviors (Tang et al., 2017). 222 differentially expressed lncRNAs have been identified from autistic brain tissues. 90% of these lncRNAs are oriented in or around known genes related to neurodevelopmental and psychiatric diseases, such as UBE3A (ubiquitin protein ligase E3A), which is associated with Angelman syndrome, that shares common features with ASD. At the same time, it has been found that the number of lncRNAs differentially expressed within a control sample, was much greater than that within an autistic sample (1375 lncRNAs vs. 236 lncRNAs, respectively) (Ziats and Rennert, 2013). A genome-wide association study (GWAS) of ASD identified a 3.9 kb lncRNA designated MSNP1AS, which is encoded by the opposite strand of the moesin pseudogene 1 (MSNP1). The sense transcript MSN encodes the moesin protein that regulates neuronal architecture and immune responses. MSNP1AS was found to be significantly upregulated in a postmortem ASD temporal cortex, and overexpression of MSNP1AS led to significant decreases in MSN, moesin, neurite number and length in cultured neurons. Thus, MSNP1AS contributes to ASD risk, by possibly influencing the sense transcript MSN expression negatively (Wilkinson and Campbell, 2013).

BACE1-AS is a conserved non-coding antisense transcript of β-secretase 1 (BACE1) and has been shown to be closely associated with Alzheimer's disease (AD). BACE1 is responsible for the generation of β-amyloid and the amyloid plagues in the brain, which are the primary pathophysiology of AD. BACE1- AS is markedly up-regulated in AD brains and promotes the stability of BACE1 through stabilizing BACE1 mRNA, thereby increasing the BACE1 protein and Aβ1–42 levels (Faghihi et al., 2008). Knock down of BACE1-AS in vivo resulted in the downregulation of both BACE1 and BACE1-AS, along with reduced β-amyloid in the brain. In addition, the brain cytoplasmic RNA BC200 (BCYRN1), GDNF gene antisense transcript (GDNF-AS) and Sox2 overlapping transcript (Sox2OT), all participate in progress and development in AD brains (Wan et al., 2017).

Huntington disease (HD) is a hereditary neurodegenerative disease with symptoms including dementia, chorea, and psychiatric disturbances. HD is caused by a CAG trinucleotide abnormal expansion in the first exon of the huntingtin gene and its probability of occurrence is 1/10000. Microarray data found that the expression of four lncRNAs significantly changed in HD brains: NEAT1 (nuclear paraspeckle assembly transcript 1) and TUG1 are upregulated, and DGCR5 (DiGeorge syndrome critical region gene 5) and MEG3 (maternally expressed 3) are downregulated. The up-regulation of NEAT1 in HD might contribute to the pathogenic alteration of the transcriptional status, by sequestrating various paraspeckle proteins (Sunwoo et al., 2017), whereas TUG1 is possibly activated by p53 and then interacts with PRC2, to affect downstream HD-associated genes. DGCR5 and MEG3, are both direct targets of REST. Their down-regulation may result from the aberrant accumulation of REST in the nuclei of striatal neurons in HD (Johnson, 2012; Hwang and Zukin, 2018). Using the whole genome chromatin immunoprecipitation sequencing (ChIP-Seq) method, HAR1, a deeply conserved genomic region that is directly bound by REST was confirmed. This region encodes a pair of structured lncRNAs as well, HAR1F and HAR1R. Both HAR1F and HAR1R were downregulated in the striatum of HD patients (Johnson et al., 2010).

Parkinson's disease (PD) is one of most prevalent neurodegenerative disorders, characterized by progressive impairments of motor abilities caused by the loss of dopamine-producing cells in the brain. Antisense ubiquitin carboxy-terminal hydrolase L1 (AS-Uchl1) was discovered to upregulate the translation of UchL1 protein at a post-transcriptional level depending on a 5<sup>0</sup> overlapping sequence and an embedded inverted SINEB2 sequence (Carrieri et al., 2012). AS-Uch1 is strongly down-regulated in neurochemical models of PD as a component of the Nurr1-dependent gene network and the subsequent reduced translation of UCHL1 protein, lead to the perturbation of the ubiquitin-proteasome system (Carrieri et al., 2015). H19 upstream conserved 1 and 2 (Huc1 and Huc2), lincRNA-p21, MALAT1, SNHG1, and TncRNA were differentially expressed in PD patients (Kraus et al., 2017). As these lncRNAs are associated with synaptogenesis, proliferation and apoptosis, the expression of these lncRNAs precede the course of PD, suggesting they may be biomarkers of PD (Kraus et al., 2017).

### MECHANISMS OF LncRNAs IN BIOLOGICAL PROCESSES

lncRNAs could provide functions through differential mechanisms, including serving as molecular scaffolds, molecular signals, guiding chromatin modifiers, and miRNA sponges, etc. (**Figure 2**).

### Molecular Scaffolds

Xist, a 17 ∼ 20 kb lncRNA transcribed from the X inactivation center, initiates X-chromosome inactivation (XCI) by progressively coating the future inactive X chromosome (Xi). Xist can bind to chromatin-modifying complexes PRC2 through the conserved A-repeat domain and form a transcriptionally silent nuclear compartment. This compartment is responsible for the chromosome-wide gene repression in the Xi (Chery and Larschan, 2014). Another example of the lncRNA acting as a molecular scaffold is the HOTAIR, which are transcripts of the antisense strand of HOXC gene cluster, which can modulate nearby gene expression by interacting with PRC2 and lysine specific demethylase 1 (LSD1) (Tsai et al., 2010).

lncRNA tsRMST, an isoform of RMST (rhabdomyosarcoma 2 associated transcript) was highly expressed in human iPSCs and ESCs. Further studies revealed that tsRMST down-regulation leads to NANOG and the PRC2 complex component SUZ12 fail to bind to the promoters of several inactive genes. These

genes are thereby activated and promote ectoderm and endoderm differentiation (Ng et al., 2012; Yu and Kuo, 2016).

### Molecular Signals

Genomic imprinting restricts gene expression on one of the two parental chromosomes and the parental-specific gene expression in diploid cells, and affects both male and female descendants (Barlow and Bartolomei, 2014). H19 transcripts are the precursors of miR-675-3p and miR-675-5p (Cai and Cullen, 2007). Before parturition, H19 slows the growth of the placenta down partially, through down-regulating the RNA binding protein HuR. The decreased HuR fails to block the processing of miR-675, which further decreases the growth regulator Igf1r (Keniry et al., 2012).

MALAT1 (metastasis-associated lung adenocarcinoma transcript (1) is initially found as an abundant lncRNA in nuclear speckles to regulate processes of mRNA alternative splicing by modulating the levels of serine/arginine splicing factors (Tripathi et al., 2010). Recent studies confirmed that MALAT1 is a sensitive prognostic marker for lung cancer metastasis and linked to several other human cancers (Gutschner et al., 2013).

## Guiding Chromatin Modifiers

Genome regulation via DNA methylation and post-translational histone modifications by the activity of chromatin modifiers, is a well-documented function of lncRNA in eukaryotes (Bohmdorfer and Wierzbicki, 2015; Nanda et al., 2016). Altered DNA methylation patterns at CpG islands and mutations in chromatin modifiers, may result in oncogenesis (Haladyna et al., 2015; Nanda et al., 2016).

The first lncRNA identified to interact with both maintenance and de novo methylases, Dum, is tightly associated with myogenesis and transcriptionally induced by MyoD upon myoblast differentiation. Dum can recruit DNA methyltransferase1/3a/3b complex to the Dppa2 promoter, through intra-chromosomal looping, mediated by RAD21 and NIPBL, resulting in two CpG loci hypermethylation and Dppa2 silencing (Wang et al., 2015a). Recent studies of HOTAIR suggest that under heypoxia, HOTAIR expression is up-regulated in several cancer cells induced by the hypoxia-inducible factors (HIFs), recruiting hypoxia-response elements (HRE) to bind on the HOTAIR promoter. Along with HIFs, histone acetyltransferase CREB-binding protein (CBP/p300) and histone H3K4 specific methyltransferases, mixed lineage leukemia (MLL) family, are enriched in the HRE region of the HOTAIR promoter (Bhan et al., 2017).

Additionally, N-Myc can directly bind to the JMJD1A promoter to upregulate JMJD1A expression in neuroblastoma cells. The upregulated JMJD1A then directly binds to the MALAT1 promoter to demethylate H3K9, to activate MALAT1 expression (Tee et al., 2014; Peng et al., 2018). Furthermore, H19 and mir-675 were found to participate in the adipogenesis through mir-675, targeting the histone deacetylase (HDAC) 4–6 3<sup>0</sup> untranslated regions and inducing HDACs 4–6 down-regulation. The reduced HDAC 4–6 then reduced H19 expression, possibly by reducing the levels of CTCF occupancy in the H19 imprinting control region. H19 inhibition then facilitates the bone marrow mesenchymal stem cells differentiating into adipocytes (Huang et al., 2016).

#### miRNA Sponges

fgene-09-00744 January 21, 2019 Time: 17:54 # 8

A handful of microRNAs have been reported to influence the mRNA stability of protein-coding genes on post-transcriptional level. Recent studies on lncRNAs discovered several miRNAlncRNA interactions based on in silico and experimental analyses (Paraskevopoulou and Hatzigeorgiou, 2016). One of the wellstudied lncRNAs which act as miRNA sponges, is lincRNA-ROR, which decoys miR-145 in self-renewing human ESCs. A regulatory feedback loop formed by lincRNA-RoR, miR-145 and the core transcription factors OCT4, SOX2, and NANOG is closely related to the ESCs pluripotency (Wang et al., 2013). MALAT1 not only interacts with several splicing factors, but also binds to miRNAs including miR-101, miR-9, miR-125b, and miR217 to regulate the interactions between miRNA and mRNAs (Paraskevopoulou and Hatzigeorgiou, 2016).

### Other Mechanisms

Some lncRNAs exert their function by maintaining DNA looping between enhancer and promoter regions, or by recruiting chromatin regulatory proteins to establish high affinity interactions between different regions of the DNA, resulting in closely positioned promoters and enhancers (Yang et al., 2013). For example, linRNA-p21 regulates the expression of its neighboring gene p21, by cis-regulatory enhancer-like DNA elements, which embed within the p21 gene body (Dimitrova et al., 2014). lncRNAs can also interact with DNA directly through nucleic-acid hybridization, and regulate nearby gene expression, such as Airn (antisense of IGF2R non-protein coding RNA), which produces transcription interference on paternal allele, by spanning the Igf2 gene promoter (Latos et al., 2012). Additionally, lncRNAs may influence the threedimensional organization of the mammalian nucleus. FIRRE (firre intergenic repeating RNA element) is transcribed from the X chromosome, and interacts with hnRNPU in the process of nuclear organization (Yang et al., 2015). Trans-acting lncRNAs also modulate the protein activity and RNA stability by directly binding in the nucleus or cytoplasm. lncRNA NORAD functions as a negative regulator of the RNA binding protein PUMILIO1

### REFERENCES


and 2 (PUM1 and 2) in the cytoplasm. The knockdown of NORAD leads to the degradation of PUM1/2 targeted mRNAs (Lee et al., 2016).

### PERSPECTIVES

Previous studies have revealed that lncRNAs play an important role in neuronal development and function through differential mechanisms. The dysregulation of lncRNAs could result in neurological diseases. Advances in sequencing technologies and their applications will contribute substantially to uncovering and investigating novel lncRNAs and their functions.

One challenge in the lncRNA field is whether lncRNAs can be used as diagnostic biomarkers or therapeutic targets for diseases. Considering that the down-stream targets of lncRNAs could be broad, it is hard to use it as a specific "key" to one "lock." Although the validation of their functions could be performed in vitro and in vivo, it is difficult to claim a specific target. Another challenge is to better understand the mechanisms of lncRNAs functions. It is highly necessary to develop proper genetic tools and to establish animal models to dissect the regulatory networks of lncRNAs, and their interaction with other epigenetic modifications.

### AUTHOR CONTRIBUTIONS

LL, YZ, XZ, and XL wrote the manuscript. All authors commented on the manuscript.

### FUNDING

LL was supported by the Natural Science Foundation of Zhejiang Province (Grant No. LQ18C060001). XL was supported in part by the National Key Research and Development Program of China (Grant No. 2016YFC0900400) and the National Natural Science Foundation of China (Grant Nos. 31771395 and 31571518).


Cai, B., Song, X. Q., Cai, J. P., and Zhang, S. (2014). HOTAIR: a cancer-related long non-coding RNA. Neoplasma 61, 379–391. doi: 10.4149/neo\_2014\_075


fgene-09-00744 January 21, 2019 Time: 17:54 # 9




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

Copyright © 2019 Li, Zhuang, Zhao and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# MicroRNAs and Androgen Receptor: Emerging Players in Breast Cancer

#### Erika Bandini and Francesca Fanini\*

Biosciences Laboratory, Department of Clinical and Experimental Oncology and Hematology, Istituto di Ricovero e Cura a Carattere Scientifico, Istituto Scientifico Romagnolo per lo Studio e la Cura dei Tumori (I.R.S.T.) S.r.l. IRCCS, Meldola, Italy

Breast cancer (BC) is the most common cause of cancer among women, with a high incidence rate occurrence every year worldwide despite advances in its management. BC is characterized by a spectrum of subtypes which respond differently to treatments due to their biological features, representing the main issue in the control of this type of malignancy. Androgen receptor (AR) is emerging as a target to investigate among hormone receptors, since it seems to play a role at various stages of development of specific BC subsets. For this reason, in recent years AR has become very important in the clinical practice, although its role remains controversial. A number of studies have proposed a correlation between microRNAs (miRNAs), a class of gene expression modulators, and AR in prostate cancer (PC), but there are still few evidences about the relationship between miRNAs and AR in BC. The purpose of this review is to present a state of the art scenario with consideration to the most recent discoveries about miRNAs involved in the AR associated pathogenesis of BC, in order to provide new insights into the role of miRNAs as key drivers in the modulation of AR, and possible actors in the development and progression of BC. Moreover, we consider findings about involvement of AR signaling in all stages of BC, highlighting its association with different subsets of breast carcinomas and with pre- and postmenopausal state of patients.

Keywords: microRNAs, androgens, receptor, breast, cancer

### INTRODUCTION

Francesca Fanini francesca.fanini@irst.emr.it

Edited by: Yujing Li,

Reviewed by: Peter Igaz,

Roberto Gherzi,

United States \*Correspondence:

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Emory University, United States

Semmelweis University, Hungary

University of California, San Diego,

Received: 18 October 2018 Accepted: 26 February 2019 Published: 19 March 2019

#### Citation:

Bandini E and Fanini F (2019) MicroRNAs and Androgen Receptor: Emerging Players in Breast Cancer. Front. Genet. 10:203. doi: 10.3389/fgene.2019.00203

Breast cancer (BC) is the most common cause of cancer among women and in 2018 in the United States 266,120 new cases and 40,920 deaths have been estimated (Siegel et al., 2018). Despite advances in the management of the disease, a high incidence rate occurs every year worldwide as a consequence of several factors such as socioeconomic differences in the population (Lukong et al., 2017; Dean et al., 2018), ethnicity (DeSantis et al., 2017; Foy et al., 2018), dietary habits (Tabung et al., 2016; Nattenmüller et al., 2018), and disparities in screening programs (Miglioretti et al., 2015). The study of BC has highlighted a substantial tissue heterogeneity, showing several molecular profiles each with distinct clinical and biological features (Perou et al., 2000) which make this tumor differently responsive to treatments, and adverse in its management. In the last years, molecular profiling by gene expression and transcriptional studies has provided an important tool to classify BCs into four well-established subtypes: Luminal A, Luminal B, Basal-like, and

**Abbreviations:** AR, androgen receptor; BC, breast cancer; DFS, disease-free survival; DHT, dihydrotestosterone; E2, β-estradiol; ER, estrogen receptor; miRNAs, microRNAs; MA, molecular apocrine; OS, overall survival; PC, prostate cancer; PGR, progesterone receptor; RFS, relapse-free survival; TNBC, triple negative breast cancer.

human epidermal growth factor receptor 2 (Her2)-enriched (Parker et al., 2009). Among Basal like group, TNBC represents a heterogeneous category of cancer whose immunohistochemical classification lacks of ER, PGR, and HER2 protein expression. Up to day, several studies have been conducted in order to better identify molecular-based therapies. Integrated molecular analyses have significantly enhanced the knowledge about genomic drivers of the most common BC subtypes, giving prominence to the discovery of novel subtype-specific targets that can be exploited in the future especially for the treatment of TNBCs (Koboldt et al., 2012; Burstein et al., 2015). Lehmann et al. analyzed 587 TNBC cases and identified 6 TNBC subtypes displaying unique gene expression profile: two basal-like (BL1 and BL2), an immunomodulatory (IM), a mesenchymal (M), a mesenchymal stem-like (MSL) and a Luminal Androgen Receptor subtype (LAR), composed of AR driven tumors, and that has been suggested to be correlated to the previously identified by Farmer et al. "Molecular Apocrine" subtype (MA) (Farmer et al., 2005; Lehmann et al., 2011; Lehmann-Che et al., 2013). Among these subtypes, LAR type was found to be associated with older patient age, apocrine histologic features, low density of stromal tumorinfiltrating lymphocytes (TIL), and low Ki-67 labeling index (Kim et al., 2018). Sex steroid hormone receptors, ER and PGR, have always played a leading role in the development and progression of BC; however, in the last years, AR has emerged as a prominent player to focus attention on. In terms of therapeutic options, AR may provide a further strategy to counteract breast malignancy, especially in patients with ER negative (ER−) tumors that do not benefit from endocrine or Her2 targeted treatments. In the last decades several small molecules have been identified as critical regulators of transcription and translation of proteins involved in tumorigenesis, and among them miRNAs are the most studied. They belong to a broad family of small noncoding RNAs, and they deserve great attention since they can modify the expression of tumor suppressor genes or oncogenes, affecting signaling pathways of cancer cells. The function of miRNAs in BC has been deeply explored, with the first miRNA signature reported by Iorio et al., and followed by a plethora of studies that have determined a functional role of miRNAs in the disease (Iorio et al., 2005). Anyway, little is known about the emerging dysregulation mechanisms of miRNAs in the context of hormonal signaling, especially for androgens. In this review, we present a state of the art scenario about the role of AR in BC, highlighting the main issues about this "player" that is very debating especially for what concerns its function in different subsets of breast carcinomas. Moreover, we focus on the most recent discoveries about miRNAs involved in the AR associated pathogenesis of BC, since so far this topic has been considered almost exclusively in PC.

#### THE ROLE OF ANDROGEN RECEPTOR IN BREAST CANCER

Androgen receptor is a member of the family of steroid nuclear receptors which mediates the biological effects of androgens. It is well established that AR is considered an oncogenic driver at all stages of PC, but its role in BC remains controversial. In fact, expression of AR splice variants (ARVs) has been elucidated in PC, but only recently the presence of multiple known and novel ARV transcripts has been demonstrated in a panel of BC cell lines and human tissues: AR-V1, -V3, -V4, -V7, and -V9 (Hu et al., 2014). In particular, AR-V7 was observed to be constitutively active and involved in androgen deprivation resistance in more than 50% of BC cases (Hickey et al., 2015). Whereas AR and ER-α have a quite similar structure and are co-expressed by many BCs, the role of AR may be different depending on the levels of both hormone receptors in the tumor environment. This becomes important in the evaluation of clinical practice, as in pre-clinical models of BC AR was able either to stimulate or inhibit cell proliferation (Macedo et al., 2006; Lin et al., 2009).

In the last years several studies have focused on the role of AR in ER-α positive (ER+) BCs, since there would seem to be a correlation between its expression grade and some clinical advantages. In fact high levels of AR are associated to reduced lymph node involvement, better DFS, RFS, and OS, response to endocrine therapies and chemotherapy, lower tumor grade, Ki67 expression, smaller tumor size and less necrosis, suggesting for AR a possible role as a tumor-suppressor in malignant breast epithelial cells (Peters et al., 2009; Castellano et al., 2010; Vera-Badillo et al., 2014). Rangel et al. investigated the prognostic impact of AR/ER ratio in 402 ER+ BC patients, showing its inverse relation with aggressiveness of biological features and worse prognosis (Rangel et al., 2018b). Similarly, Basile et al. reported that a high AR/ER ratio seems to be detrimental in BC treated with endocrine therapy (Basile et al., 2017), while in 2 validated BC cohorts, ER+ patients with AR positivity ≥78% had the best survival, and among them those with a ratio of AR: ERα >0.87 exhibited the best outcomes (Ricciardelli et al., 2018). In a study involving 479 BC women, it has been evidenced that in ER+ patients the expression of forkhead box A1 (FOXA1), a pioneer factor which helps the recruitment of ER and AR to their response elements on the genome, was directly correlated to the presence of AR and to better outcome, providing additional knowledge about recurrence (Rangel et al., 2018a). Also Park et al. confirmed that ER+ patients with low expressed AR and FOXA1 tumors were significantly correlated to worse RFS (Park et al., 2017). Moreover, recent data showed that over 90% of metastasis from luminal tumors preserved FOXA1 expression (Ross-Innes et al., 2012), and the concomitant expression of AR and FOXA1 in metastatic lesions may promote the luminal to MA transition. In TNBCs, AR is expressed in 10–43% of cases but its prognostic value remains still unclear. Actually, larger cohort numbers should be needed to determine a role for AR in this peculiar subtype. In some studies involving TNBC cases, the presence of AR appeared correlated to an increase in overall mortality, lymph node metastasis and higher tumor stage (Hu et al., 2011; McGhan et al., 2014). Conversely, another group demonstrated that androgen pathways are associated with reduced aggression TNBC, and that AR loss may have a role in the progression of the tumor (McNamara et al., 2014). A meta-analysis involving 13 studies with 2826 TNBC cases, suggested a potential role of AR in a lower risk of recurrence highlighting that AR positive women showed prolonged DFS (Wang et al., 2016). After analysis of 135

invasive TNBC cases, AR and epidermal growth factor receptor (EGFR) expression was evaluated in order to stratify TNBCs into three risk groups: low risk (AR+ EGFR−) characterized by better prognosis and beneficial from anti-androgen therapies; high risk (AR− EGFR+) with worst prognosis, but better responsiveness to chemotherapy; and intermediate-risk (AR+ EGFR+, AR−, EGFR−) (Astvatsaturyan et al., 2018).

Molecular apocrine subtype has been studied in vitro using BC cell lines whose growth was promoted by AR expression. Robinson et al. demonstrated that in the absence of ER-α more than a half of AR binding events showed an analogous pattern to that of ER-α in ER+ cells, promoting the expression of ER target genes, and suggesting a role of AR as a ER-α mimic (Robinson et al., 2011). Anyway, the biological interaction between ER-α and AR still needs to be clarified. Curiously, in a transcriptomic study involving male BC, chromatin binding landscape of ER in relation to steroid hormone receptors including AR, was determined. Results showed that AR pathway was the only hormonal signaling more associated with the ER-α binding genes, confirming that genomic functions of ER-α and AR in BC are largely overlapping (Severson et al., 2018). For what concerns HER2-enriched BC subtype, it has been found strongly related to MA and studies have suggested a strong evidence of the proliferative role of AR (Ni et al., 2011; Chia et al., 2015). Lehmann-Che et al. tried to characterize MA tumors and found that they were all defined ER−, AR+, FOXA1+, with an overexpression of HER2 or prolactin induced protein (GCDFP15), useful for discriminating MA from basal-like (BL) in the context of ER− tumors. This distinction can be useful to include MA patients in specific "AR pathway" trials, being this subtype rather aggressive (Lehmann-Che et al., 2013). There are evidences that AR can promote ERK activation up-regulating HER2 gene transcription, therefore contributing to the growth of Her2+ BC (Naderi and Hughes-Davies, 2008; Chia et al., 2011). More recently, the functional role of AR was investigated by silencing assays and a reduction in the growth of Her2+ BC cells HCC1954 and SKBr3 was observed, also after treatment with the androgen antagonist Enzalutamide, highlighting a function of AR in promoting the growth of Her2+ BC cells (He et al., 2017). Daemen and Manning explored HER2 amplification in 3155 breast tumors and found that the HER2–enriched (HER2E) subtype had a distinct transcriptional landscape independent of HER2-amplificated (HER2A) that reflected and confirmed how AR signaling can replace ER-driven tumorigenesis (Daemen and Manning, 2018). In a study involving 1297 primary tumors and 336 paired axillary lymph node metastases, Kraby et al. found a highest proportion of AR positivity in the Luminal B subtype while the lowest was observed in the basal phenotype. Interestingly, in 60/72 cases a changeover from AR− primary tumor to AR+ lymph node metastasis occured. Moreover, in primary tumors AR expression was an independent and favorable prognostic marker, particularly in the Luminal A subtype, and in grade 3 tumors (Kraby et al., 2018). All these observations underline the need for a more detailed classification of tumor samples aimed at a more targeted and personalized treatment of patients. The role of AR in BC subtypes is resumed in **Table 1**.

AR is expressed in all stages of BC (in situ, primary and metastatic). In fact, it is estimated that up to 90% of primary BC and up to 75% of metastatic lesions expressed AR (Hickey et al., 2012), as well as in the 50–80% of invasive BCs and in the 85% of ductal carcinoma in situ (DCIS) (Lim et al., 2014), although among the BC subtypes the frequency appears variable. Nevertheless, its role in breast carcinogenesis remains a debated topic as its contribution to the different tumor stages development and progression still needs to be clarified. Feng et al. reported the involvement of DHT in the initiation of epithelial-to-mesenchymal transition (EMT) of BC cells in an AR-dependent but ER-independent manner, indicating the role of androgens in cancer invasion and metastasis (Feng et al., 2017), Schrijver et al. investigated receptor conversion in 91 effusion metastasis, pleural and peritoneal, of 69 patients by immunohistochemistry and in situ hybridization. AR receptor status changed from positive in the primary tumor to negative in the effusion metastases or vice versa in 46–51% of cases, and


this was more often associated in patients previously treated with ET (Schrijver et al., 2017). This new finding could be relevant for investigating AR-targeted therapies in ER− and endocrine resistant BC. RNA sequencing was performed to investigate CTCs isolated from blood samples of patients with metastatic ER+ BC, and a comparison between cases with progression in bone vs. visceral organs was made. Results showed that the most activated pathway in CTCs from bone was that of AR, especially involving splice variant AR-V7. Curiously, AR expression within CTCs was associated with the duration of treatment with aromatase inhibitors (AIs), proposing a possible mechanism in the contribution of acquired resistance to ET, and underlying the role of AR in BC bone metastasis together with the therapeutic option of its targeting in patients with metastatic setting (Aceto et al., 2018). Usually, the maintenance of the balance between DHT, the most potent endogenous AR ligand derived from testosterone (Labrie et al., 2003; Gao et al., 2005), and E2 ensures the physiological response of the breast tissue, including BC tissue, depending on the hormonal needs and the menopausal status. In fact, the circulating androgens concentration varies in woman in relation to pre and postmenopause state (Giovannelli et al., 2018). Whereas after menopause circulating level of E2 decrease dramatically up to 10-fold, androgens begin to acquire an important function (Rothman et al., 2011). Several studies have tried to analyze the correlation between circulating androgens and BC growth since this relationship remains unclear, although up to now a high serum testosterone level has been associated with an increased risk in postmenopausal women. It follows that an additional complication in understanding the role of AR is to be attributed to the menopausal state of patients, which seems to be a more significant variable than age. It would be important to distinguish between the intratumoral estrogen or androgen production, and to take into consideration the balance between these different sex hormones. The most of breast tumors are estrogen-dependent and are characterized by a high expression of ER that could interfere with the activity of AR and vice versa. Premenopausal patients BC tissues are characterized by higher production of estrogen, and in these individuals ovary is the main source of E2. Otherwise, in postmenopausal state estrogens derived from circulating adrenal androgens, such as androstenedione, and in these patients BC tissues presents lower levels of E2 and higher androgen levels (Takagi et al., 2018). How hormonal changes influence cancer development is still a discussed issue. Curiously, data showed that in recent decades incidence rates of advanced BC have increased for premenopausal women (Fahlén et al., 2018).

### THE INTERACTION BETWEEN miRNAs AND ANDROGEN RECEPTOR IN BREAST CANCER

MicroRNAs are the most explored non-coding RNAs, and give rise to a large family of short (19–24 nucleotides) single-strand RNAs which take part in a variety of biological processes, such as cell proliferation, death, differentiation, and stress response (Bartel, 2004; Kozomara and Griffiths-Jones, 2014). They operate recognizing a 2–7 nucleotides "seed-region" in the target mRNA, which can be localized in the 3<sup>0</sup> -UTR (Lewis et al., 2005), in the 5<sup>0</sup> -UTR (Lytle et al., 2007), or in the coding region (Forman et al., 2008). Their regulatory function on gene expression is performed through the control of translation of the mRNA target, which can result in downregulation but also in upregulation of the encoded protein (Ambros, 2004; Vasudevan et al., 2007). A decisive turning point was given by Fabbri et al., who highlighted for the first time the ability of miRNAs secreted by tumor-derived exosomes (TEX) to act as paracrine agonists of a specific receptor family suggesting an involvement in the tumor microenvironment interaction and a new possible target for cancer treatment (Fabbri et al., 2012). On this trail, other groups started to analyze the implication of miRs in tumor communication, growth and spread, and recently it has been demonstrated how breast-cancer TEX are able to carry precursor miRNAs (pre-miRNAs) complexed with Dicer, TRBP and AGO2 proteins displaying a cell-independent capacity to process premiRNAs into mature form, contributing to the comprehension of a cell-autonomous process occurring in exosomes when secreted into the extracellular space (Melo et al., 2014).

A number of studies have proposed a correlation between miRNAs and AR in PC (Shi et al., 2007; Epis et al., 2009; Ribas et al., 2009; Cao et al., 2010; Nadiminty et al., 2012), but there are still few evidences about the possible role of miRNAs in regulating AR expression in BC. For the first time, Nakano et al., through miRNAs Polymerase Chain reaction (PCR) Arrays, identified miR-363 as an androgen-inducible miRNA. In MCF-7 BC cells they highlighted a possible androgens-related feedback loop involving the gene IQWD1 (IQ motif and WD repeats-1) and miR-363: under low androgens levels IQWD1 was downregulated by miR-363, but this negative modulation did not occurred after DHT administration (**Figure 1A**). Interestingly, IQWD1 has a role in protecting AR proteins from degradation via proteasome (Nakano et al., 2013). In AR+/ER− models, androgens seemed to mediate a negative correlation between miR-let-7a and the expression of its target oncogenes CMYC and KRAS. In particular, in the MA MDA-MB 453 and in the TNBC MDA-MB 231 cell lines treated with DHT a significant increase in let-7a expression was observed together with a decrease of CMYC and KRAS (**Figure 1B**). Similarly, in BC tissues the negative correlation was confirmed by IHC, highlighting a new androgen-induced AR activating signal pathway that directly upregulates let-7a and negatively regulates CMYC and KRAS, inhibiting proliferation of AR+/ER− cells (Lyu et al., 2014). Results about the tumor suppressive role of let-7a were confirmed also in AR+/ER+ BC cells, where DHT stimulation led to an AR translocation to the nucleus with transcriptional upregulation of let-7a, decreased cell proliferation, self-renewal capacities, invasion and migration (Zhang et al., 2018). Moreover, in order to deepen the effects of let-7a/AR pathway on breast tumor-initiating cells, Zhang et al. examined the expression of AR, let-7a and CD44+/CD24−/low in invasive BC tissues. AR was significantly correlated to let-7a and CD44+/CD24−/low, highlighting that patients expressing AR and let-7a could have a better outcome, unlike those with a CD44+/CD24−/low phenotype which showed a worse prognosis. These findings put

by miR-363. (B) In MA and TNBC cells, the DHT administration results in an androgen-induced AR activating signal pathway which upregulates let-7 expression and negatively regulates CMYC and KRAS that are targets of let-7. (C) In TNBC cells, the lncRNA ARNILA is negatively regulated by AR after DHT treatment, causing a decreased adsorption of miR-204 which in turn inhibits Sox4 expression, a gene known to promote EMT. (D) In TNBC cells, DHT induces upregulation of miR-328-3p with concomitantly decrease of its target CD44, diminishing EMT, migration and adhesion.

in evidence that DHT-induced AR activation plays a critical role in BC, and that AR/let-7a signaling could be exploited as a new optional therapeutic target (Zhang et al., 2018). Another study evidenced the interaction between AR and miRNAs in controlling BC cells behavior. Three BC cell lines (Luminals and MA subtypes) were screened for 84 miRNAs showing each of

them a distinct basal miRNAs expression profile. High level of let-7a and -7b found in MA-MDA-453 appeared to be distinctive for MA subtype, whereas miR-205 seemed to represent a marker in the luminal T47D and MCF-7 cells. Furthermore, treatment with the AR agonist CI-4AS-1 led to alterations in the expression profile of other micro-RNAs, such as miR-100 and miR-125 which were found significantly downregulated simultaneously with the increase and extracellular release of metalloprotease-13 (MMP13). Interestingly, the transfection of miR-100 and -125b abrogated the induction of MMP13, suggesting a correlation between these micro-RNAs and AR in the control of BC growth (Ahram et al., 2017). In TNBC, the gene SRY-box 4 (Sox4) is known to promote EMT, thereby progression, invasion and metastasis, and is found abnormally overexpressed. Yang et al. identified an AR negatively induced long non-coding RNA (lncRNA) ARNILA that correlated to poor PFS, tightly connected to AR and able to sequester miR-204, in turn facilitating the expression of its target Sox4. Particularly in AR+ carcinomas, ARNILA is suppressed by the action of DHT and AR, resulting in the decreased adsorption of miR-204 thus favoring Sox4 expression inhibition. On the other hand, in AR− tumors, the action of ARNILA leads to Sox4 expression by increasing the sequestration of miR-204, leading to the induction of EMT and metastatic propagation (**Figure 1C**). All together these findings spread light to the discovery of new lncRNA/miRNA/AR mechanisms correlated with poor clinical outcome by regulating EMT, migration, and invasion in TNBC (Yang et al., 2018). Again in TNBC, investigating the modulation of miRNAs after treatment with DHT a group reported miR-328-3p as the most upregulated. Concomitantly, CD44 target of miR-328- 3p, decreased, diminishing cell adhesion, migration and EMT, and this result was confirmed also after miR-328-3p mimic transfection (**Figure 1D**) (Al-Othman et al., 2018). In another study, a total of 153 miRNAs were found to be differentially expressed in AR+ vs. AR− cell lines. The authors identified

miR-143, -145, -31, -30c, -30b-3p, 199a, and -181 as significantly downregulated in AR+ cells, while miR-933 and -5793 appeared as the most upregulated, suggesting a role for these miRNAs in the regulation of AR in BC (Shi et al., 2017). Again, the interaction between miR-30a and AR was explored, and miR-30a role was investigated in ER−, PR−, AR+, MDA-MB-453 BC cells. After DHT treatment, which activates Androgeninduced AR signal, a miRNAs profile was identified by miRNAs array, showing a downregulated expression of miR-30a, b and c (among which the most downregulated was miR-30a), and an upregulated expression of AR. Interestingly, in the AR mRNA 3 0 -untranslated region resides a bioinformatic putative miR-30a, b and c binding site confirming AR as a direct target of miR-30a. Nevertheless, AR does not bind miR-30a promoter region which could be downregulated through other AR-induced cell signaling pathways. This study identified a positive feedback mechanism of regulation which could be explained by two effects. First, the activation of AR expression and AR-induced signal downregulates mir-30a expression that in turn promotes AR availability. Second, the downregulation of mir-30a expression has a negative effect on the inhibition of cell growth induced by itself, being miR-30a a cancer suppressor gene (Lyu et al., 2017). Interestingly, Casaburi et al. reported that androgens can reduce BC cells proliferation by negative modulation of the onco-miR-21. By treatment with synthetic androgen miboleron (Mib) and Chromatin immune-precipitation (ChIP) analysis, they provided evidence that activated AR works as a transcriptional inhibitor of miR-21 expression. In particular, AR was able to bind the proximal promoter of miR-21 in a specific ARE sequence, involving the recruitment of HDAC3 as cofactor in the AR-mediated transcriptional repression (**Figure 2**). This hypothesis was also supported by a significant reduction of PolII binding in Mib treated extracts, providing further evidence about the protective role of androgens in BC cells (Casaburi et al., 2016).

#### DISCUSSION AND FUTURE HORIZONS

Many studies support the idea that BC is a heterogeneous pathology and this consideration is mainly motivated by the existence of different subtypes classified on the basis of hormone receptor expression. About 70–80% of BC express considerable level of ER and are estrogen-dependent (Takagi et al., 2018), but, as well, approximately 70–90% of them express AR, and close to 75% are considered to express both AR and ER (Vera-Badillo et al., 2014) suggesting a role of androgen hormones in the pathogenesis of BC. Although up today the association between levels of circulating androgens and BC risk is still under discussion, several studies identified AR as a marker of favorable prognosis and have demonstrated the anti-proliferative effects of androgens in BC cells (Andò et al., 2002; Lanzino et al., 2010; Takagi et al., 2010). Definitely, in this moment the scientific community can only state that the AR positivity makes more intricate the BC molecular outlook. It has been suggested that circulating androgens may have a role both as independent molecules and as substrate for estrogen synthesis, but limited to AR+/ER+ BC since in AR+/ER− BC they may act in a more homogenous way (Giovannelli et al., 2018). On the basis of different androgens effects on BC, several approaches having AR as a target have been evaluated, including AR agonists and antagonists.

Since the findings that miRNAs play a role in carcinogenesis and are deregulated in several types of tumor, they have obtained a lot of interest about their potential use as therapeutic agents in chemotherapy. Furthermore, the supposing of the existence of

#### REFERENCES


a specific uptake can help the purpose to customize miRNAs as therapeutics alone or, more likely, in combination with today's anti-cancer therapies. Surely more stress should be placed on understanding the balance between AR and ER in relation to the different subtypes, which gives rise to main questions regarding a different response to endocrine therapies in BC. For instance, the interesting positive feedback mechanism identified by Lyu et al. between AR and miR-30a could be a starting point for further studies about the role of miRNAs as a therapeutic predictive markers, besides the identification of other miRs that are able to target AR, or molecules involved in the AR pathway, can certainly help to find more answers about this interaction (Lyu et al., 2017). Also, not to be underestimated is the recent intriguing branch of miRceptor which fits very well in the context of BC as a hormone-dependent tumor, and which is linked to broader themes such as the study of the tumor microenvironment. Consequently, it is reasonable to foresee how the interaction between miRNAs and AR in BC can become in the future an extensively investigated field, in order to increase the treatment chances and try to get much closer to BC personalized therapies.

#### AUTHOR CONTRIBUTIONS

EB wrote the first draft of the manuscript. FF contributed to the writing of the manuscript. FF and EB drew figures and tables. Both authors critically reviewed the manuscript and approved the final version of the manuscript.

subtypes and targets of triple-negative breast cancer. Clin. Cancer Res. 21, 1688–1698. doi: 10.1158/1078-0432.CCR-14-0432



subtypes. Breast Cancer Res. Treat. 172, 283–296. doi: 10.1007/s10549-018- 4904-x



common in malignant pleural and peritoneal effusions of breast cancer patients treated with endocrine therapy. Oncotarget 8, 55550–55561. doi: 10.18632/ oncotarget.15548


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

Copyright © 2019 Bandini and Fanini. 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.

# Non-coding RNA in Fragile X Syndrome and Converging Mechanisms Shared by Related Disorders

#### Yafang Zhou1,2, Yacen Hu1,2, Qiying Sun1,2 and Nina Xie1,2 \*

<sup>1</sup> Department of Geriatric Neurology, Xiangya Hospital, Central South University, Changsha, China, <sup>2</sup> National Clinical Research Center for Geriatric Disorders, Central South University, Changsha, China

Fragile X syndrome (FXS) is one of the most common forms of hereditary intellectual disability. It is also a well-known monogenic cause of autism spectrum disorders (ASD). Repetitive trinucleotide expansion of CGG repeats in the 5<sup>0</sup> -UTR of FMR1 is the pathological mutation. Full mutation CGG repeats epigenetically silence FMR1 and thus lead to the absence of its product, fragile mental retardation protein (FMRP), which is an indispensable translational regulator at synapsis. Loss of FMRP causes abnormal neural morphology, dysregulated protein translation, and distorted synaptic plasticity, giving rise to FXS phenotypes. Non-coding RNAs, including siRNA, miRNA, and lncRNA, are transcribed from DNA but not meant for protein translation. They are not junk sequence but play indispensable roles in diverse cellular processes. FXS is the first neurological disorder being linked to miRNA pathway dysfunction. Since then, insightful knowledge has been gained in this field. In this review, we mainly focus on how non-coding RNAs, especially the siRNAs, miRNAs, and lncRNAs, are involved in FXS pathogenesis. We would also like to discuss several potential mechanisms mediated by non-coding RNAs that may be shared by FXS and other related disorders.

Keywords: non-coding RNA, Fragile X syndrome, RNAi mediated epigenetic silencing, microRNA mediated translational regulation, potential converging mechanisms

#### INTRODUCTION

Non-coding RNA is a kind of transcript from DNA but not meant for protein translation. Accumulating evidence has shown that non-coding RNA molecules play indispensable roles in diverse cellular processes. Understanding mechanisms mediated by these non-coding RNAs are of great importance for understanding the pathogenic process of relevant diseases. Functional molecules classified into this category have included micro-RNA (miRNA), long non-coding RNA (lncRNA), small-interfering RNA (siRNA), transfer RNA (tRNA), ribonucleoprotein RNA (rRNA), and piwi- interacting RNA (piwi-RNA), etc. (Cao et al., 2006; Beermann et al., 2016; Treiber et al., 2018).

Fragile X syndrome (FXS) is one of the most common forms of hereditary intellectual disability. It is also a well-known monogenic cause of autism spectrum disorders (ASD) (Wang et al., 2012). FMR1 is the responsible gene. The repetitive trinucleotide expansion of CGG repeats in the 5<sup>0</sup> -untranslated region (5<sup>0</sup> -UTR) of FMR1 is the pathological mutation. Normal individuals

Edited by:

Thomas S. Wingo, Emory University, United States

#### Reviewed by:

Stefania Filosa, Institute of Bioscience and Bioresources, National Research Council, Italy Woan-Yuh Tarn, Academia Sinica, Taiwan

> \*Correspondence: Nina Xie xienina@csu.edu.cn

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 14 October 2018 Accepted: 11 February 2019 Published: 01 March 2019

#### Citation:

Zhou Y, Hu Y, Sun Q and Xie N (2019) Non-coding RNA in Fragile X Syndrome and Converging Mechanisms Shared by Related Disorders. Front. Genet. 10:139. doi: 10.3389/fgene.2019.00139

usually bear CGG repeat expansion ranging from 6 to 55, while in FXS patients, this expansion often reaches beyond 200, known as the full mutation (Santoro et al., 2012). The FMR1 gene was first cloned in 1991, rendering FXS the first discovered disease caused by trinucleotide expansion mutation (Verkerk et al., 1991). In addition to this mutation, conventional mutations including gross deletions, small indels, and missense or nonsense mutations have also been reported (Luo et al., 2014, 2015; Myrick et al., 2014). Over the last decade, FXS caused by the full mutation CGG repeat expansion receives the most intense attention. The full mutation CGG repeats lead to epigenetic silencing of FMR1 and absence of its product, fragile mental retardation protein (FMRP) (Xie et al., 2016). FMRP is a complex RNA binding protein which plays indispensable roles in synaptic plasticity. It has four important RNA binding motifs including one arginine-glycine-glycine (RGG box) and three K homology domains (KH0, KH1, and KH2), recognizing special RNA secondary structures, such as the kissing complex and G quadruplex (Darnell et al., 2001, 2005; Myrick et al., 2015). By binding with target mRNA, FMRP mainly functions as a translation repressor at synapsis, regulating local translation spatially and temporarily to shape synaptic structure and plasticity (Nakamoto et al., 2007; Bassell and Warren, 2008; Huang et al., 2014). Most of FXS patients' phenotypes could be attributed to the loss of FMRP. Clinically, the most common symptom is intellectual retardation. Other neurological symptoms include ASD, attention deficit hyperactivity disorder (ADHD), and epilepsy. Non-neurological symptoms include macroorchidism, distinct facial features (elongated faces, protruded ears, and big forehead), and connective tissue abnormalities (mitral valve prolapse, flat fee, joint hyperextensibility, and high arched palate) (Jacquemont et al., 2007).

FXS is the first neurological disorder found to be linked to the miRNA pathway (Jin et al., 2004a). In this review, we mainly focus on how non-coding RNAs, especially the siRNAs, miRNAs, and lncRNAs, are involved in FXS pathogenesis. We would also like to discuss several potential mechanisms mediated by non-coding RNAs that may be shared by FXS and other related disorders.

### NON-CODING RNA MEDIATED MECHANISMS IN FXS PATHOPHYSIOLOGY

### How RNAi Is Involved in the Epigenetic Silencing of FMR1?

RNA interference (RNAi) refers to the process of mRNA degradation or translation inhibition mediated by small RNAs. It is a mechanism regulating gene expression at the post-transcriptional level in eukaryotic cells. siRNA and miRNA are the two most important types of small RNAs involved. Although their origins are different, they share similar downstream machinery when encountered with Dicer, an RNase III-like enzyme initiating RNAi. Precursors of siRNA or miRNA are cleaved by Dicer to be short double-stranded RNAs (dsRNAs). In these duplexes, only one functional strand is kept, while the other one is degraded. The functional strand is the mature siRNA or miRNA that finally assemblies with Argonaute proteins (AGO) to form the RNA-induced silencing complex (RISC). A major difference lies in that siRNA requires perfect base-pairing with the target sequence to guide AGO to the targeted locus while miRNA could tolerate several mismatches. AGO associated with siRNA usually induces mRNA degradation, in contrast, AGO loaded with miRNA tends to cause translation inhibition (Ha and Kim, 2014; Holoch and Moazed, 2015; Hu et al., 2017).

Full mutation CGG expansion triggers extensive DNA methylation, repressive histone modification, and chromatin condensation in the 5<sup>0</sup> -UTR of FMR1, transcriptionally silencing the gene and leading to loss of FMRP (Coffee et al., 1999; Biacsi et al., 2008; Alisch et al., 2013). How are the abnormal epigenetic markers triggered and maintained? Although this epigenetic silencing process has been the focus of study over the last two decades, detailed mechanisms are still mysterious. The consensus is that full mutation CGG repeat is the prerequisite trigger to initiate and maintain the repressive epigenetic changes of FMR1. Several models have been indicated to date, albeit it is still difficult to form an integrated one. The first model is DNA based. Secondary structures formed by the CGG repeats serve as substrates for DNA methyltransferase to initiate de novo DNA methylation, or as targets bound by repeat binding proteins to recruit repressor complexes (Smith et al., 1994; Bulut-Karslioglu et al., 2012). The second model is RNA based, where hairpin structures in mRNA formed by CGG repeats exceeding a certain threshold trigger the RNAi pathway to deposit repressive epigenetic markers (Kim et al., 2006; Usdin et al., 2014). The third model is a blended one, where the DNA:RNA hybrid is at play. During transcription, hybridization of the nascent RNA to its unzipped DNA template forms a special R-loop, which may act as a structural block or nucleosome analogy to induce epigenetic silencing (Colak et al., 2014; Groh et al., 2014). Our discussion below is focused on the RNA based model. What may be the role of RNAi in the epigenetic silencing of FMR1?

RNAi has been suggested as a conservative mechanism participating in the formation of heterochromatin in fission yeast. In this scenario, the siRNA serves as a localizer for the RISC complex to achieve site-specific epigenetic modulation. RISC is associated with Ago1 (the yeast argonaute homolog), Chp1 (a protein required for methylation of H3K9), and Tas3 (a protein required for localization of the chromatin). This RISC-like heterochromatin-targeting complex is termed RITS (Reinhart and Bartel, 2002; Volpe et al., 2002; Verdel et al., 2004).

In the context of FXS, it was observed that stable hairpin-structured RNAs containing pre-mutation CGG repeats could be processed by Dicer to generate small RNAs. Based on this observation and the role siRNA has in heterochromatin formation. A model for the RNAi mediated methylation of full mutation CGG repeats was proposed in 2004. In early embryo development, FMR1 gene containing the full mutation CGG repeats is being transcribed actively. Bi-directional transcription of the DNA template generates dsRNAs bearing full mutation CGG repeats. Dicer further cleaves these dsRNAs to produce small RNAs. After being incorporated into the RITS complex,

similar to siRNAs, these small RNAs guide RITS to the CGG expansion region, recruiting epi-effectors, such as histone methyltransferase and DNA methyltransferase, to induce FMR1 silencing epigenetically (Jin et al., 2004a; **Figure 1**). Thus, RNAi may play a critical role in the epigenetic silencing process of FMR1.

### How miRNA Is Involved in FMRP Mediated Translational Regulation?

miRNA is a short non-coding single-stranded RNA molecule consisting of 20–25 nucleotides. Its biogenesis experiences a series of events. The Primary microRNA (pri-miRNA) was first converted to the precursor microRNA (pre-miRNA) by a microprocessor comprising of DGCR8 and DROSHA proteins. Next, Dicer cleaves the pre-miRNA to generate short dsRNAs, the functional strand of which further assemblies with AGO proteins (also known as PIWI/PAZ-domain proteins) to form the RISC complex. This functional strand is the mature miRNA. It functions as a localizer as well as a translation repressor by binding to the 3<sup>0</sup> -untranslated region (30 -UTR) of the mRNA targets (Carthew and Sontheimer, 2009; Holoch and Moazed, 2015).

FXS is the first neurological disease being linked to miRNA pathway dysfunction (Jin et al., 2004a). FMRP is a RNA binding protein which mainly functions as a translational repressor. Studies from several groups suggested that FMRP is associated with critical components of the miRNA pathway. Loss of FMRP may cause aberrant miRNA profiling (Caudy et al., 2002; Ishizuka et al., 2002). The hippocampus of post-natal day 7 FMR1 knockout mice showed significantly different miRNA expression profiles compared to the WT (Liu et al., 2015). In drosophila, dfmr1 is associated with Ago1. Absence or partial loss of Ago1 impairs FMRP-mediated regulation of neural development and synaptogenesis (Jin et al., 2004b). Recently, in a FXS mice model, FMRP was shown to participate in pri-miRNA processing by upregulating DROSHA expression at the translational level. Loss of FMRP was associated with accumulation of specific pri-miRNAs and reduction of pre-miRNAs (Wan et al., 2017).

Efforts have been made to clarify detailed mechanisms of how FMRP regulates protein translation via the miRNA pathway. For example, FMRP could stabilize the binding of miRNA to its target mRNA via the KH domain (Plante et al., 2006). Besides, partly dependent on miR-125b, FMRP negatively regulates the expression of NR2A, a NMDA receptor subunit affecting synaptic plasticity (Edbauer et al., 2010). Moreover, FMRP regulates axon guidance gene via the miRNA pathway by repressing the RE-1 silencing transcription factor (REST) (Halevy et al., 2015). FMRP and miR-181d cooperatively regulate the axon elongation process by repressing translation of Map1 (a microtubule-associated protein) and Calm1 (a calcium signaling regulator). Upon nerve growth factor stimulation, Map1 and Calm1 mRNAs are released from granules suppressed by FMRP and miR-181d to translate actively for axon elongation (Wang et al., 2015; **Figure 2A**).

Later studies in succession provided further insight into how FMRP represses target mRNA translation via RNAi. A possible mechanism is FMRP interacts with MOV10, a RNA helicase implicated in miRNA pathway. MOV10 has a dual function in translational control. It assists the miRNA-mediated translational repression for specific RNAs but inhibits others. In the former case, MOV10 binds to the 3<sup>0</sup> -UTR of target mRNA, unwinds specific RNA secondary structures, and exposes the miRNA recognition elements (MRE) to facilitate the interaction between MRE and RISC. In the latter scenario, both FMRP

histone methyltransferase and DNA methyltransferase, to induce epigenetic silencing of FMR1 (the CGG expansion region is light blue colored).

and MOV10 bind to the target mRNA in proximity. The association between MRE and RISC is disrupted (Kenny et al., 2014; **Figure 2B**). Another mechanism is that FMRP utilizes miRNA to control translation at synapses temporally and spatially. For example, phosphorylation of FMRP promotes assembly between miR-125a-AGO2 complex and PSD-95 mRNA. Upon mGluR stimulation, dephosphorylation of FMRP facilitates disassembly between miR-125a-AGO2 complex and PSD-95 mRNA (Muddashetty et al., 2011; **Figure 2C**).

Endogenous FMRP expression level is also regulated by the miRNA pathway. For example, the expression level of miR-130b was negatively correlated with the FMRP level in mice embryonic neural precursor cells (NPC) (Gong et al., 2013). In humans, miR-130b also seems to be a negative regulator of FMRP. Utilizing a patient-derived cell line, researchers identified a RNA binding protein HuR which may function as a RISC antagonist, as its binding site overlaps with that of miR-130b. This patient has no full mutation, however, the c.<sup>∗</sup> 746T > C variant in 3<sup>0</sup> -UTR renders him with significantly decreased FMRP level and FXS phenotypes. A likely mechanism is that the c.∗ 746T > C variant abolishes the binding between HuR and FMR1 mRNA. Instead, miR-130b binds to the 3<sup>0</sup> -UTR of FMR1 mRNA, resulting in decreased FMRP expression (Suhl et al., 2015; **Figure 2D**). Additionally, in a zebrafish FXS model, researchers transgenically overexpressed rCGG (CGG trinucleotide repeats) motif via high titer retroviral delivery. As the expression level of rCGG increased, the level of FMR1-miRNA was increased, oppositely, the level of FMR1 transcription was decreased, indicating that rCGG derived-miRNA participated in FMR1 transcription suppression (Lin, 2015).

In sum, miRNA facilitates the FMRP mediated translational regulation via diverse mechanisms to shape synaptic plasticity and morphology, to achieve translational control temporally and spatially, and to regulate endogenous FMRP level (**Figure 2**).

#### lncRNA Participates in FXS Pathophysiology

Long non-coding RNA (lncRNA) is a genome transcript longer than 200nt. As a major category of non-coding RNA, lncRNA participates in formation of RNA-protein complexes, gene regulation, modulation of protein localization, and X

chromosome inactivation (Wu R. et al., 2016). In FXS, lncRNA may affect the cell proliferation process and may serve as novel clinical biomarkers. Similar to miRNA and siRNA, its role in FXS pathogenesis should not be ignored.

FMR4 is a 2.4 kb lncRNA, which is primate-specific and plays an anti-apoptotic role in human cells. It is transcribed from FMR4, which resides upstream FMR1. FMR4 and FMR1 a bi-directional promoter. In FXS, both of them are silenced due to the CGG repeat expansion, which may contribute to FXS pathogenesis (Khalil et al., 2008). FMR4 is also a trans-acting element which regulates hundreds of genes involved in neural development. By altering chromatin state epigenetically, FMR4 selectively modulates the cellular proliferation of human NPC (Peschansky et al., 2016). Later on, combining the technology of rapid amplification of cDNA ends (RACE) and next generation sequencing, researchers investigated the FMR1 gene locus systemically and found two novel lncRNAs, FMR5, and FMR6. FMR5 is transcribed in the sense direction, the transcription start site (TSS) of which resides upstream the FMR1 TSS. FMR6 is transcribed in the antisense direction, beginning from the 3<sup>0</sup> -UTR of FMR1. FMR5 and FMR6 may be novel clinical biomarkers due to their distinct expression patterns between full mutation and premutation patients (Pastori et al., 2014). A recent study provides new clues in how lncRNA is involved in FXS pathophysiology. TUG1 is a lncRNA which prevents axonal growth by negatively regulating the SnoN-Ccd1 pathway. In WT mice, FMRP directly binds with TUG1 to decrease its stability. However, in FXS mice, absence of FMRP leads to TUG1 overexpression. TUG1 interacts with SnoN, a crucial transcriptional regulator which controls axonal growth via regulating the actin-binding protein Ccd1. By affecting the transcriptional activity of SnoN, overexpression of TUG1 decreases the level of Ccd1 mRNA and protein, leading to abnormal axonal growth (Guo Y. et al., 2018). Albeit reference regarding this section is scarce, the importance of lncRNA is non-negligible, as evidenced by the research findings described above.

#### miRNA Links FMRP to Glia Function

Glutamate is a major excitatory neurotransmitter in brains. Dysregulation of glutamate in neurons participates in various neuropsychiatric disorders by affecting synaptic plasticity. Interestingly, abnormal regulation of glutamate in cells surrounding neurons may also be an important pathogenic process. For the encircled neurons may become more exciting. For example, in the astrocytes isolated from the cortex of FMR1-deficient mice, the GLT1 expression, a major glutamate transporter of glia, is significantly reduced compared to the WT, so is the uptake of glutamate. Additionally, treating the cortical slices with GLT1 inhibitors results in significantly enhanced neuronal excitability in FMR1-deficient mice but not in controls. Moreover, based on an astroglia-specific conditional FMRP knockout and restoration mice model, researchers found that selective re-expression of FMRP in astrocytes of FMR1-deficient mice rescues phenotypes of the decreased GLT1 expression in cortical astrocytes and the abnormal spine morphology in cortical pyramidal neurons. Upregulation of GLT1 expression alleviates enhanced neuronal excitability and corrects spine abnormality in FMR1-deficient mice. These results together indicate that absence of FMRP may impair the ability of astrocytes, via dysregulated GLT1, to remove excessive glutamate from neurons and thus lead to neuronal hyperexcitability in FXS (Higashimori et al., 2013, 2016).

Notably, the regulation of GLT1 may be mediated by exosomes. Exosomes are membrane vesicles secreted by cells. They contain various signaling molecules including miRNAs. After being secreted into extracellular space, exosomes could be fused into surrounding cells to assist intercellular communication (Xiao et al., 2017). miR-124a has a positive effect on the expression level of GLT1 both in vivo and in vitro. Transfer of miR-124a from neurons to astrocytes via exosomes selectively increases the astroglial GLT1 protein level without affecting the mRNA level, suggesting the exosome-derived miR-124a is a regulator of the GLT1 expression (Morel et al., 2013).

It is likely that absence of FMRP may cause aberrant miRNA function in the neuron/astroglia interaction and thus affect GLT1 expression and glutamate uptake ability of astrocytes. More studies are needed to shed light on this topic.

#### NON-CODING RNA MEDIATED MECHANISMS SHARED BY FXS AND RELATED DISORDERS

#### miRNA of FXTAS May Also Be Pivotal to FXS

Fragile X tremor ataxia syndrome (FXTAS) is an aging-related neurodegeneration disorder, featured by progressive tremor, ataxia, and cognition decline. In FXTAS, FMR1 containing the premutation CGG repeat expansion (60–200 repeats) is not silenced. Instead, it is transcribed actively, resulting in significantly higher mRNA level than normal but less FMRP protein. It is thought that the gain-of-function RNA toxicity is to blame, of which the miRNA pathway is at play. For instance, in neuronal cells and brain tissues from FXTAS patients, RNA aggregates containing pre-mutation CGG repeats decrease mature miRNA levels by sequestering RNA-binding proteins which are crucial to miRNA biogenesis, such as DGCR8 and DROSHA (Sellier et al., 2013). In olfactory neurons of C. elegans, expression of the expanded CGG repeats weakens the olfactory plasticity formation by interacting with the C. elegans specific argonaute ALG-2 (Juang et al., 2014). Additionally, miRNA-277, 424, 101, 129-5p, and 221 have also been indicated in the pathogenesis of FXTAS (Tan et al., 2012; Alvarez-Mora et al., 2013; Zongaro et al., 2013).

FXTAS is the neurodegenerative disorder most closely related to FXS. FXTAS and FXS may have overlapping mechanisms in the earlier neurodevelopmental stage or the later neurodegenerative stage, as either the premutation or the full mutation FMR1 is transcribed actively at the very beginning, and both of them could lead to cognition impairment in late-life. miRNAs that are critical to FMRP function in

FXTAS may also be pivotal to FXS. Discovery of such miRNAs is necessary.

### miRNA Regulating mTOR Activity Might Play a Role in FXS

Synaptic plasticity refers to the adaptability of synaptic transmission in an activity-dependent manner. Long-term potentiation (LTP) and long-term depression (LTD) are the two most common forms of synaptic plasticity. When the transmission is being strengthened for more than 1 h, it is termed LTP. When the transmission is being weakened for more than 1 h, it is termed LTD. LTP and LTD is the basis for learning and memory, the mGluR-LTD of which plays a central role in FXS pathogenesis (Heller et al., 2014).

Normally, the mGluR-LTD consists of two phases. In the early phase, when mGluR is stimulated, PP2A is instantly activated to dephosphorylate FMRP, derepressing mRNAs targeted by FMRP to cause a rapid synaptic translation burst. In the late phase, the mammalian target of rapamycin (mTOR) is activated. mTOR inhibits PP2A but activates S6K1, which phosphorylates FMRP again to counterbalance the previous translation burst. The overall effect of mGluR stimulation on the surface AMPA receptor is internalization (Nakamoto et al., 2007; Narayanan et al., 2007, 2008; Bassell and Warren, 2008).

In FXS, on the one hand, the signaling cascade from mGluR stimulation to mTOR activation comprises of homer, PIKE, PI3K, PDK1/2, AND Akt, etc. (Santoro et al., 2012). Since FMRP represses the activity of PIKE and PI3K, loss of FMRP results in increased mTOR signaling activity (Sharma et al., 2010). On the other hand, the translation repression at basal state mediated by phosphorylated FMRP is abolished due to FMRP deficiency. As a result, no matter whether there is a stimulus or not, mRNAs targeted by FMRP are always being transcribed actively. The temporarily released translation burst described above is lost, resulting in persistent and increased AMPA receptor internalization. Collectively, the mGluR/mTOR/LTD signaling is exaggerated in FXS (Bassell and Warren, 2008; Santoro et al., 2012).

Theoretically, any miRNA regulating components along the mGluR/mTOR/LTD pathway might play a role in FXS pathogenesis. As described below, accumulating evidence has suggested a bunch of miRNAs could either increase or decrease the activity of mTOR signaling. Although most of them are not directly linked to the pathophysiology of FXS, it is still worth noting that some of them might help unveil novel clues for studying overlapping mechanisms between FXS and other disorders.

Firstly, some miRNAs increase mTOR activity. For example, MiR-125b is an oncogenic miRNA inhibiting the tumor suppressor p53. An important downstream target of p53 is PIK3CA, which encodes the p110α subunit of PI3K. As the negative regulation of PIK3CA via p53 is weakened by miR-125b, the mTOR activity is increased (Astanehe et al., 2008; Zeng et al., 2012). miR-451 is also an oncogenic miRNA, which has been implicated in glioma. It may activate the mTOR pathway by inhibiting AMPK signaling in glioma and colorectal cancer cells (Godlewski et al., 2010; Chen et al., 2014). Additionally, both miR-21 and miR-93 are oncogenic and manifest significantly differential expression profiling in ASD. They may increase the mTOR activity by inhibiting the putative target PTEN which is a negative regulator of PI3K (Sarachana et al., 2010; Kawano et al., 2015; Wu Y.E. et al., 2016; Guo X. et al., 2018).

On the contrary, miR-7 and miR-155 are miRNAs downregulating mTOR activity. miR-7 is a tumor suppressor targeting PIK3CD, Akt, and mTOR. PIK3CD is an integral catalytic subunit of PI3K. By inhibiting PI3K, Akt, and mTOR itself, the overall impact of miR-7 on mTOR pathway is suppression (Kefas et al., 2008; Fang et al., 2012). miR-155 is a negative regulator of mTOR via targeting multiple signals including RHEB. It has been implicated both in cancer and autism (Wan et al., 2014; Wu Y.E. et al., 2016).

Moreover, miR-199 seems to play a dual role in mTOR signaling. In cancer cells, it usually acts as a tumor suppressor by downregulating the mTOR pathway activity. In Rett syndrome, miR-199 is a positive regulator of the mTOR pathway activity. However, the MeCP2 deficiency hinders generation of the precursor-miR-199a. As a result, the positive regulation of mTOR signaling via miR-199 is weakened by MeCP2 deficiency, leading to decreased mTOR pathway activity and Rett syndrome phenotypes (Tsujimura et al., 2015).

Last but not least, in turn, the mTOR signaling pathway also regulates the biogenesis and activity of miRNAs. For instance, miR-21 is positively regulated by mTOR, while miR-125b is negatively regulated by mTOR (Ge et al., 2011; Bornachea et al., 2012; Ye et al., 2015).

miRNAs described here are mainly linked to cancer, ASD, and Rett syndrome. Although there is no evidence supporting a direct relationship between them and FXS pathophysiology, it is still possible that several of them may have aberrant functions in FXS. By increasing or decreasing mTOR activity, they might participate in the regulation of mGluR/mTOR/LTD signaling, and further contribute to the shaping of synaptic plasticity (**Figure 3**). More research on this topic is needed to clarify the linkage between miRNAs and mTOR pathway activity in FXS pathogenesis.

### miRNA May Function as a Broad-Spectrum Therapeutic Agent

The increased mTOR pathway activity is a central pathogenic mechanism shared by FXS, aging, and several other neurological disorders such as autism and tuberous sclerosis, evidenced by reproducible results from yeast to mammalian animal models (Fabrizio et al., 2001; Jia et al., 2004; Johnson et al., 2013; Ebrahimi-Fakhari and Sahin, 2015; Kilincaslan et al., 2017). It is hopeful that the progress of mTOR activity regulation in one disorder may shed light on others.

The mTOR inhibitor has the potential to be a broad-spectrum therapeutic agent. For example, in FMR1 KO mice, pharmacological inhibition of mTOR pathway rescues multiple FXS phenotypes, including excessive synaptic protein synthesis, persistent AMPA receptor internalization, and increased spine density (Gross et al., 2010). Optimistically, mTOR inhibitors have

spawned several clinical trials for FXS (Luo et al., 2016). In aging mice, rapamycin, a classic mTOR inhibitor, was able to slow the aging process in multiple organs (Wilkinson et al., 2012). In tuberous sclerosis patients, everolimus was capable of controlling refractory epilepsy and alleviating autistic symptoms by inhibiting the mTOR pathway activity (Kilincaslan et al., 2017).

Theoretically, miRNAs that inhibit mTOR activity may have therapeutic effects similar to the chemicals described above. In this sense, understanding miRNA mediated mTOR regulation in FXS may help develop new therapeutic agents not only for FXS but also for other disorders where the mTOR dysregulation is an essential pathogenic mechanism.

#### miRNA Links FMRP to Alzheimer's Disease

Alzheimer's disease (AD) is pathologically featured by the Aβ plaques and neurofibrillary tangles containing hyperphosphorylated tau protein (Zhang et al., 2016). miR-132 and miR-125b are well established FMRP-associated miRNAs regulating synaptic structures and functions (Edbauer et al., 2010). Intriguingly, they are also involved in AD pathogenesis. miR-132 has protective effects on neurons both in vivo and in vitro, the mechanisms of which include reducing Aβ production and glutamate toxicity, targeting tau modifiers to decrease hyperphosphorylated tau proteins, inhibiting cell apoptosis, and strengthening hippocampal LTP (Smith et al., 2015; Hernandez-Rapp et al., 2016; Salta et al., 2016; Wang et al., 2017; El Fatimy et al., 2018). The expression level of miR-132 in cholinergic nucleus basalis Meynert is stable at early stages but decreases significantly as the disease course develops, suggesting miR-132 may be a facilitator of neurodegeneration (Zhu et al., 2016). In contrast, miR-125b was shown to be a deleterious factor in AD pathogenesis. miRNA profiling of AD brains showed that miR-125b was significantly increased in the hippocampus (Lukiw, 2007). In vitro overexpression of miR-125b was shown to induce cell apoptosis, enhance oxidative stress, and promote tau hyperphosphorylation in the cell and animal models of AD (Banzhaf-Strathmann et al., 2014; Jin et al., 2018). As aforementioned, miRNAs assist FMRP mediated translational repression in FXS. Suboptimal ratio and assembly between FMRP, miRNA, and mRNA results in

TABLE 1 | Non-coding RNA involved in FXS pathogenesis.


translation dysregulation. Meanwhile, mRNAs of Aβ precursors and Aβ receptors, including NMDARs, mGluR5, and PSD-95, are classic targets negatively regulated by FMRP. Therefore, establishing AD animal models lacking or overexpressing FMRP may help elucidate how FMRP interplays with miR-132 or miR-125b to regulate Aβ related proteins at different disease stages in specific brain regions.

FMRP is also implicated in AD pathogenesis by interacting with lncRNA. It has been controversial whether the BC1 RNA, a lncRNA, functions as an adapter molecule for FMRP in FXS. Some researchers support the view that BC1 directly binds with FMRP to increase its affinity to target mRNAs (Zalfa et al., 2003, 2005; Napoli et al., 2008). Others argue that FMRP directly binds to the target mRNA. The interaction between FMRP and BC1 is non-specific. These two molecules act independently (Iacoangeli et al., 2008; Zhong et al., 2010). Notably, in the context of AD, this controversial issue seems to be reconciled to some extent. In the Tg2576 AD mice model, when FMRP binds to APP mRNA, APP translation is suppressed, subsequently producing less Aβ protein. However, when BC1 binds to FMRP, the interaction between FMRP and APP mRNA is disrupted, resulting in active APP translation and increased Aβ production, which further causes the spatial learning and memory deficits (Zhang et al., 2018). This study is a good example where FMRP indeed binds directly with either BC1 RNA or target mRNA. However, this binding does not necessarily include BC1 RNA and target mRNA simultaneously. Therefore, investigating how BC1 and other lncRNAs, APP, and FMRP interact with each other at synapsis may help understand the pathogenesis of both AD and FXS.

#### FMRP Regulates Stem Cell Fate via miRNA

The first evidence indicating FMRP may regulate germline stem cell fate via the miRNA pathway came in 2007. In drosophila, The interaction between FMRP and Ago1 may facilitate maintenance but suppress differentiation of the germline stem cells (Yang et al., 2007). FMRP may also function via the bantam miRNA to regulate the germline stem cell fate (Yang et al., 2009). In testicle cells, FMRP was found to be associated with a bunch of miRNAs including miR-383, which is mainly expressed in spermatogonia. miR-383 is negatively regulated by FMRP during spermatogenesis, failure of which may contribute to male infertility (Tian et al., 2013). In the brain, miR-510, located near the fragile site on X chromosome, was associated with CGG expansion into full mutation in the neurons derived from mesenchymal stem cells. Bioinformatics analysis indicated that enhanced miR-510 expression might facilitate CGG expansion via regulating target genes such as VHL and PPP2R5E (Fazeli

#### REFERENCES


et al., 2018). In mice embryonic NPC, the upregulation of miR-130b associates with decreased FMRP level and increased NPC proliferation tendency (Gong et al., 2013). The expression level of miR-181a, which translationally represses the expression of GluA2, a subunit of AMPA receptors, was increased in human NPCs derived from induced pluripotent stem cells. Since cells lacking the GluA2 subunit were more prone to differentiation, miR-181a may affect cell fate by interacting with the AMPA signaling pathway (Achuta et al., 2018). According to these data, it is fair to say that the FXS stem cell model provides an excellent monogenic platform for studying how miRNAs are involved in stem cell fate regulation, especially in testis and brain.

### CLOSING REMARKS AND OUTLOOK

In summary, Non-coding RNA participates in FXS pathophysiology via multiple aspects. The RNAi mechanism provides a RNA-based model for explaining the epigenetic silencing of FMR1. The miRNA mainly facilitates FMRP's role as a translational repressor, controlling axon growth, shaping synaptic plasticity, and regulating endogenous FMRP level. The lncRNA also plays non-negligible roles (**Table 1**). Besides, understanding how non-coding RNA functions in FXS may help uncover converging mechanisms shared by FXS and related disorders. This list may well include FXTAS, ASD, Rett syndrome, AD, tuberous sclerosis, infertility, cancer, and aging, etc. In the future, studies dissecting the non-coding RNA profiling temporally and spatially are needed to identify the non-coding RNA regulation pattern in specific neurodevelopmental or neurodegenerative stages. The non-coding RNA that may function as a nexus between FXS and related disorders should be given more attention. As such research findings will help unveil converging mechanisms shared by different diseases and further contribute to the development of a broad-spectrum non-coding RNA therapeutic agent.

#### AUTHOR CONTRIBUTIONS

All authors contributed to the manuscript writing, revision, and figure design.

### ACKNOWLEDGMENTS

We thank all the kind support from Department of Geriatric Neurology, Xiangya Hospital, Central South University.

syndrome is specific to the FMR1 locus. BMC Med. Genet. 14:18. doi: 10.1186/ 1471-2350-14-18


transcription in ovarian surface epithelium and in ovarian cancer. J. Cell Sci. 121(Pt 5), 664–674. doi: 10.1242/jcs.013029




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

Copyright © 2019 Zhou, Hu, Sun and Xie. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# MiR-378a-5p Regulates Proliferation and Migration in Vascular Smooth Muscle Cell by Targeting CDK1

Shaoyan Liu<sup>1</sup>† , Yanyan Yang<sup>2</sup>† , Shaoyan Jiang<sup>3</sup> , Hong Xu<sup>4</sup> , Ningning Tang<sup>2</sup> , Amara Lobo<sup>1</sup> , Rui Zhang<sup>1</sup> , Song Liu<sup>1</sup> , Tao Yu<sup>2</sup> \* and Hui Xin<sup>1</sup> \*

<sup>1</sup> Department of Cardiology, The Affiliated Hospital of Qingdao University, Qingdao, China, <sup>2</sup> Institute for Translational Medicine, Qingdao University, Qingdao, China, <sup>3</sup> Department of Cardiology, The Affiliated Cardiovascular Hospital of Qingdao University, Qingdao, China, <sup>4</sup> Department of Orthodontic, The Affiliated Hospital of Qingdao University, Qingdao, China

Objective: Abnormal proliferation or migration of vascular smooth muscle cells (VSMCs) can lead to vessel lesions, resulting in atherosclerosis and in stent-restenosis (IRS). The purpose of our study was to establish the role of miR-378a-5p and its targets in regulating VSMCs function and IRS.

#### Edited by:

Zhao-Qian Teng, Chinese Academy of Sciences, China

#### Reviewed by:

Chi-Ming Wong, The University of Hong Kong, Hong Kong Alessio Paone, Sapienza University of Rome, Italy

#### \*Correspondence:

Tao Yu yutao0112@qdu.edu.cn; dachao1201@hotmail.com Hui Xin xinhuiqy@163.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 10 October 2018 Accepted: 15 January 2019 Published: 19 February 2019

#### Citation:

Liu S, Yang Y, Jiang S, Xu H, Tang N, Lobo A, Zhang R, Liu S, Yu T and Xin H (2019) MiR-378a-5p Regulates Proliferation and Migration in Vascular Smooth Muscle Cell by Targeting CDK1. Front. Genet. 10:22. doi: 10.3389/fgene.2019.00022 Methods: EdU assays and Cell Counting Kit-8 (CCK-8) assays were applied to evaluate VSMCs proliferation, wound healing assays and transwell assays were applied to assess cells migration. Furthermore, quantitative reverse transcription–polymerase chain reaction (qRT-PCR) was performed to investigate the expression level of miR-378a-5p IRS patients and healthy individuals. Target genes were predicted using Target Scan and miRanda software, and biological functions of candidate genes were explored through bioinformatics analysis. Moreover, RNA-binding protein immunoprecipitation (RIP) was carried out to analyze the miRNAs interactions with proteins. We also used Immunofluorescence (IF) and fluorescence microscopy to determine the binding properties, localization and expression of miR-378a-5p with downstream target CDK1.

Results: The expression of miR-378a-5p was increased in the group with stent restenosis compared with healthy people, as well as in the group which VSMCs stimulated by platelet-derived growth factor-BB (PDGF-BB) compared with NCs. MiR-378a-5p over-expression had significantly promoted proliferative and migratory effects, while miR-378a-5p inhibitor suppressed VSMC proliferation and migration. CDK1 was proved to be the functional target of miR-378a-5p in VSMCs. Encouragingly, the expression of miR-378a-5p was increased in patients with stent restenosis compared with healthy people, as well as in PDGF-BB-stimulated VSMCs compared with control cells. Furthermore, co-transfection experiments demonstrated that miR-378a-5p overexpression promoted proliferation and migration of VSMCs specifically by reducing CDK1 gene expression levels.

Conclusion: In this investigatory, we concluded that miR-378a-5p is a critical mediator in regulating VSMC proliferation and migration by targeting CDK1/p21 signaling pathway. Thereby, interventions aimed at miR-378a-5p may be of therapeutic application in the prevention and treatment of stent restenosis.

Keywords: miR-378a-5p, vascular smooth muscle cell, stent-restenosis, proliferation, migration, atherosclerosis, CDK1

## INTRODUCTION

fgene-10-00022 March 13, 2019 Time: 17:9 # 2

Coronary artery disease (CAD) is a serious disease threatening human health with its high mortality rate. Percutaneous Coronary Intervention (PCI) and stent implantation are commonly applied in the treatment of these obstructive diseases (Feinberg, 2014; Buccheri et al., 2016). However, there will be new atherosclerosis around stent implantation, and this will further increase the rate of restenosis after stent implantation (Wang H. et al., 2015; Liu et al., 2018). As a foreign substance, a scaffold can cause damage to the tunica intima, followed by inflammation and platelet aggregation in the damaged areas; resulting in the formation of plaque and thrombosis in the long term (Liu et al., 2018; Tang et al., 2018). And after stent implantation, endothelialization occurs gradually which cannot be removed from the blood vessels again; so, if in-stent restenosis occurs, stent implantation must be repeated (Alfonso et al., 2006; Finn et al., 2007). Aberrant proliferation and migration of Vascular smooth muscle cells (VSMCs) were identified as the main causes of these adverse events (Krist et al., 2015; Afzal et al., 2016; Huang et al., 2017). When VSMCs are stimulated, they promote the transfer of VSMCs from tunica media to tunica intima, from the contractile to the secretory, while stimulating free VSMCs and fibroblasts to secrete a large amount of extracellular matrix; the extracellular matrix is continuously deposited in the blood vessels, causing the intima to gradually thicken, resulting in stenosis (Braun-Dullaeus et al., 1998). Consequently, investigating key regulators and understanding the molecular mechanisms of VSMC biology has become a major method of treating atherosclerosis and stent restenosis.

MicroRNA (miRNA) is a type of small non-coding RNA that negatively modulates gene expression through mRNA translation repression or the induction of target mRNA instability (Gareri et al., 2016). Mounting shreds of evidence suggested that several miRNAs play critical roles in regulating VSMC proliferation and migration, such as miR-133, miR-221, miR-222, miR-663, miR-143, and miR-145 (Liu et al., 2009; Xin et al., 2009; Li et al., 2013; Mcdonald et al., 2015). Altering the expression of miRNA may have therapeutic potential in the prevention and treatment of stent restenosis.

It has been reported that the expression levels of miR-378a-5p in cardiac myocytes increased under some external stimuli (Wang et al., 2017; Rui et al., 2018). MiR-378a-5p is involved in the biological functions of some tumor cells (Luo et al., 2012), promoting proliferation in some tumor cells, and inhibiting proliferation in some others (Wang Z. et al., 2015). Considering that there may be tissue specificity associated with miR-378a-5p. However, the effect of miR-378a-5p in the regulation of VSMC biology remains unknown. The objective of the study is to investigate the potential roles of miR-378a-5p, as well as the molecular mechanisms of VSMCs proliferation and migration. Firstly, we screened and identified differential expression of miR-378 in patients with stent restenosis, then we studied the effect of up-regulation and down-regulation miR-378a-5p on the biological function of VSMCs. We also found that CDK1 was a potential gene target for the miR-378a-5p. Meanwhile, p21 may be the downstream target of CDK1. Consequently, miR-378a-5p is a key modulator to regulate proliferation and migration of VSMC partly by modulating the level of CDK1 gene expression. In this way, we have enough reason to believe that miR-378a-5p could be used as a diagnostic marker for early diagnosis, monitoring, and treatment of molecular targets for stent restenosis.

### MATERIALS AND METHODS

### Blood Samples Acquisition and Baseline Clinical Characteristics Collection

Thirty-two persons were collected at the affiliated hospital of Qingdao University in Qingdao, China from June 2017 through February 2018. According to whether ISR was detected, they were classified into two groups: (1) The ISR group (n = 14): ISR is defined as a diameter stenosis greater than 50% in coronary angiography that occurs within the stent or 5 mm at the proximal or distal end of the stent; (2) The normal group (n = 18): 18 healthy persons without coronary heart disease as the control group. Basic information of all individuals collected, including age, gender, history of diabetes, drinking, hypertension, and smoking was noted. The research was supported by the Institutional Review Boards of Qingdao University Health Science Center. Paper version of informed consent was acquired from all subjects and the regional ethics committee in Qingdao, China approved the study protocol. The information of all clinical people is displayed in **Supplementary Table 3**.

#### Test Animals

All experimental laboratory animals were approved by the Animal Care and Use Committee. C57BL/6 and ApoE−/− mice were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. There were 3 mice in each group. The control group was given a normal diet, the experimental group was given a western diet (conventional mouse feed+0.15% cholesterol+21% fat) for 12 weeks, then cardiac blood was collected from mice weighing 25–30 g for further experiments.

#### Cell Culture

The VSMC was purchased from the Chinese Type Culture Collection (Chinese Academy of Sciences, Shanghai, China) and cultured in Dulbecco's modified Eagle's medium (GIBCO, Grand Island, NY, United States) containing 10% fetal bovine serum (ExCell Bio.) in a 5% CO2 humidified incubator at 37◦C. MiR-378a-5p mimics, miR-378a-5p inhibitor and negative control oligonucleotide (NC) (GenePharma, Shanghai, China) were transfected into the VSMCs using LipofectamineTM 2000 (Invitrogen, Grand Island, NY, United States).

#### Western Blot Analysis

Cells lysates were prepared in buffer mixture containing 1 ml RIPA (Solarbio, Beijing, China), 0.1 mM PMSF reagent and a protease inhibitor cocktail (Roche, Basel, Switzerland) for 10 min on ice, after which protein samples were separated by 10% SDS–PAGE, then transferred into 0.45 µm polyvinylidene

difluoride (PVDF) membrane, membranes were blocked with 5% not-fat milk in Tris-buffered saline-Tween 20 (TBS-T) for 1 h. And incubated with a rabbit anti-CDK1 monoclonal antibody (1:10000 dilution; Abcam, MA, United States) or anti-β-actin (1:2500 dilution; Cell Signaling Technology, United States). Then being washed three times with TBS-Tween 20, the secondary antibodies were added. Finally, the signals were visualized with Supersensitive ECL Chemiluminescent Kit, according to the directions of the manufacturer. The quantification of the protein bands was performed using ImageJ 1.8.0.

### RNA Extraction and qRT-PCR

Total RNA was extracted from the collected blood samples using TRIzol (Invitrogen, Grand Island, NY, United States), then treatment with DNase I (Takara, Otsu, Japan), then reverse RNA with reverse transcriptase kit (Takara) and mature miRNA levels were assessed using SYBR Green Real-time PCR Master Mix (Takara) according to the manufacturer's guidance. The following primers which used in the experiment showed in **Supplementary Tables 1, 2**. U6 and GAPDH are based on different detection genes as reference genes, respectively. Analysis of qRT-PCR results using the 2 −11Ct method.

### RNA Binding Protein Immunoprecipitation (RIP)

RNA-binding protein immunoprecipitation assays are performed to identify regions of the genome with RNA-binding proteins. In RIP assays, VSMCs were lysed in RIPA buffer containing 0.1 mM PMSF and 1% protease inhibitor cocktail on ice. After 10 min, the collecting cells were centrifuged at 12000 rpm for 20 min, the next step is to take 500 µg cell lysates incubated with the CDK1 antibody at 4◦C overnight. Then add protein A/G-agarose beads and incubate for 4 h at 4◦C with shaking. Immunoprecipitation separates RNA-binding proteins and their bound RNA. Furthermore, the combination of RIP

and quantitative qRT-PCR can present experimental results more intuitively.

## Cell Proliferation

Cell Counting Kit-8 (CCK-8) assay was performed to assess VSMC proliferation. Cells were incubated in DMEM at a density of 5 × 10<sup>3</sup> cells per well in 96-well plates for 24 h after transfection. And then the cells were maintained in 10 µl /well CCK-8 solution (7Sea-Cell Counting Kit, Shanghai, China) for an additional 1 h. Finally, the value was measured at 450 nm absorbance.

Another way to test cell proliferation is to use the EdU assay, VSMCs were cultured with EdU solution (50 nmol/L) (RiboBio, Guangzhou, China) for 2 h, then VSMCs were stained according to the product instructions. And finally, the pictures were obtained by a fluorescence microscope (Zeiss, LSM510, META). Image J 1.8.0 was used for analysis of the data.

### Cell Migration

Transwell assay and wound healing assay were performed to assess the migratory ability of VSMC.

Vascular smooth muscle cells were maintained in 6-well plates 12 h before transfection. After transfected, cells were cultured for 24 h in normal medium and then an additional 24 h in serum-free DMEM. After resuspending VSMCs in serum-free DMEM (5 × 10<sup>5</sup> cells/ml), a mixture of 200 µl was added to the upper chamber of a transwell insert (Corning, Tewksbury, NY, United States) in a 24-well plate. Meanwhile, the lower chamber added 500 µl DMEM supplemented with 10% FBS. After 24 h incubation, PBS-rinsed cotton was used to wipe off the cells remaining on the upper side of the membrane. Then the cells were fixed with 4% paraformaldehyde for 1 h and dyed with 0.1% crystal violet for 30 min. After three washes, migrated VSMCs were recorded with a Zeiss LSM510 META microscope (20 × magnification), the 5 randomly fields were selected to count the cells.

#### Wound Healing Assay

Vascular smooth muscle cells were grown up to 60–70% in six-well plates. Then, cells were transfected with miR-378a-5p mimics, miR-378a-5p inhibitor, and NC. After 24 h, the wounds were made by a 1000-µl disposable pipette tip, which had reached almost 100% confluence. Distance on both sides of the scratch was visualized and photographed immediately and at different time points after wounding using a Zeiss LSM510 META microscope.

### Statistical Analysis

All data presented in this paper were the mean ± SD of at least three independent experiments, and an experiment performed with three samples for in vitro experiments. Data analyses were carried out using the GraphPad Prism 5 software. The quantitative data were presented as means ± SEM. Statistical analysis of the two groups by t-test, the different P-values indicate the different statistical significance: <sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001.

## RESULTS

fgene-10-00022 March 13, 2019 Time: 17:9 # 4

### MiR-378a-5p Expression in Stent-Restenosis Patients and ApoE −/− Mice

MiR-378a-5p expression were detected between the patients with stent-restenosis and control group, respectively, by qRT-PCR, in which we found that miR-378a-5p expression levels were upregulated in stent-restenosis patients compared with control group (**Figure 1A**); MiR-378a-5p expression levels were higher in atherosclerotic plaques of ApoE knockout (ApoE -/-) mice than in control subjects as measured by qRT-PCR (**Figure 1B**).

### MiR-378a-5p Promoted VSMCs Proliferation and Migration

MiR-378a-5p regulates biological function of VSMC. VSMCs were transfected with miR-378a-5p mimics (25 nM) and miR-378a-5p inhibitor (100 nM), both of them had visible transfection efficiency (**Figure 2A**). MiR-378a-5p mimics-transfected cells have stronger proliferative capacity compared with NC (25 nmol/L) (**Figure 2B**). Meanwhile, VSMC proliferation was assessed using the EdU assay; representative staining of the nucleus of proliferating VSMCs was shown in **Figure 2C**. EdU incorporation measured by confocal laser microscopy. In addition, we observed that miR-378a-5p significantly promoted cell migration in miR-378a-5p mimics-transfected cells compared with NC, the miR-378a-5p inhibitor has the opposite effect (**Figures 2D,E**). Sequence of RNAs used in this study showed in **Supplementary Table 4**.

### MiR-378a-5p Promoted Proliferation and Migration in PDGF-BB-Stimulated VSMCs

In order to detect the effects of miR-378a-5p in PDGF-BBstimulated VSMC, we transfect miR-378a-5p mimics, miR-378a-5p inhibitor and NC into VSMCs. MiR-378a-5p was upregulated after PDGF-BB treatment in a time-dependent manner (**Figure 3A**). MiR-378a-5p mimic-induce miR-378a-5p up-regulation significantly increased VSMC migration compared with NC, the miR-378a-5p inhibitors have the opposite effect, as demonstrated by wound closure assay (**Figures 3B,C**) and transwell assay (**Figures 3D,E**). To detect the effect of PDGF-BB on VSMC phenotype switching genes, VSMCs were treated with PDGF-BB (50 ng/ml) for 12 h, the expression levels of sm-MHC2 and α-SMA protein decreased after being transfected with miR-378a-5p mimics (**Figures 3F–I**). These results demonstrate that miR-378a-5p promoted proliferation and migration in PDGF-BB-stimulated VSMCs.

### Identification of CDK1 as a Direct Target of MiR-378a-5p in VSMCs

Target Scan algorithms found that CDK1 was a potential miR-378a-5p target. We found that the putative seed sequences

by qRT-PCR after transfection of miR-378a-5p mimics, miR-378a-5p inhibitor and negative control oligonucleotide (NC), ∗∗∗p < 0.001. (B) The CCK-8 assay was performed to investigate the proliferation of VSMCs with miR-378a-5p mimics, miR-378a-5p inhibitor transfection, at 0, 12, 24, and 36 h, respectively. (C) Representative micrographs of EdU staining of VSMCs, with control or miR-378a-5p transfection, scale bar = 20 µm. (D,E) miR-378a-5p effects on VSMC migration ability as determined via classic scratch assay and its quantification analysis, all images were taken under the same magnification, data are presented as mean ± SEM, <sup>∗</sup>p < 0.05 and ∗∗p < 0.01.

presented as mean ± SEM. <sup>∗</sup>p < 0.05 and ∗∗p < 0.01. (B,C) miR-378a-5p abrogated PDGF-BB-mediated effects on VSMC migration ability as determined via classic scratch assay and its quantification analysis, all images were taken under the same magnification. (D,E) The miR-378a-5p mimics significantly increased PDGF-BB-induced (50 ng/ml) VSMC migration, as determined by transwell assay (original magnification: × 200) and its quantification analysis, data are presented as mean ± SEM. <sup>∗</sup>p < 0.05. (F) Representative western blots of VSMC phenotype marker genes transfected with miR-378a-5p inhibitor compared with in NC under the condition with PDGF-BB (50 ng/ml) stimulated for 12 h. (G) Quantitative analysis of differentiation marker gene expression in the two groups, data are presented as mean ± SEM. <sup>∗</sup>p < 0.05 and ∗∗p < 0.01. (H) Representative western blots of VSMC phenotype marker genes transfected with miR-378a-5p mimics compared with NC under the condition with PDGF-BB (50 ng/ml) stimulated for 12 h. (I) Quantitative analysis of differentiation marker gene expression in the two groups, data are presented as mean ± SEM. <sup>∗</sup>p < 0.05.

for miR-378a-5p within the 3<sup>0</sup> -UTR of CDK1 were highly conserved and there has a potential seed sequence of miR-378a-5p in the 30UTR of CDK1 (**Figure 4A**). To illustrate the relationship between CDK1 and miR-378a-5p, we transfected VSMCs with miR-378a-5p mimics, inhibitor and NC, and investigated CDK1 expression using western blot analysis and qRT-PCR, overexpression of miR-378a-5p suppressed the protein expression of CDK1, as showed in **Figures 4B–D**, referring to the literature, p21 is one of the downstream targets of CDK1 (Kreis et al., 2016). In our experiment, we found that miR-378a-5p works by reducing CDK1 and then regulating p21. Then we performed immunofluorescence which showed that CDK1 protein expression was decreased with transfected miR-378a-5p mimics compared with NC, as showed in **Figure 4E**. The conclusion of these findings is: miR-378a-5p upregulation could inhibit CDK1 expression at the post-transcriptional level. RIP was used to analyze the protein interactions with CDK1 mRNA. The % input detected for CDK1 immunoprecipitation is above that detected for IgG immunoprecipitation, which means CDK1 antibody could pull down more miR-378a-5p than the nonspecific IgG antibody (**Figure 4F**). These results demonstrated that miR-378a-5p directly binds to the 3<sup>0</sup> -UTR of CDK1.

### CDK1 Is Involved in VSMC Proliferation and Migration

To determine the effect of CDK1 on the VSMC, expression of CDK1 under PDGF-BB stimulation gradient by western blot (**Figures 5A,B**). Two SiRNAs were designed to knock down CDK1, the sequences were shown in **Supplementary Table 2**. The transfection efficiency and expression efficiency of siCDK1 was detected by qRT-PCR. The result showed that CDK1 was significantly down-regulated when transfected with siCDK1(#1) and siCDK1(#2), respectively (**Figure 5C**). The influence of siCDK1 on the proliferation of cells was assessed by EdU assay (**Figure 5D**). Transwell assay was applied to investigate the migration ability of cells, showed in **Figures 5E,F**. The wound closure assay was performed to detect the migration ability of cells with or without PDGF-BB-induced (50 ng/ml) (**Figures 5G,H**). α-SMA and sm-MHC2 protein level were downregulated after transfected with siCDK1 (**Figures 5I,J**). The findings showed that the proliferation and migration activity of VSMCs transfected with siCDK1 was higher than that of cells transfected with an empty plasmid with or without PDGF-BB stimulation.

### MiR-378a-5p Targeted CDK1 Expression and Enhanced Migration of VSMCs

For further confirm whether CDK1 is a functional target gene of miR-378a-5p, wound closure (**Figures 6A,B**) and transwell assays (**Figures 6C,D**) were used to measure the migratory ability of VSMCs, compared with the NC, the migration rate of the miR-378a-5p mimic group was remarkably increased, while that of the si-CDK1 group has no significant increase, the group with miR-378a-5p mimic

+ siCDK1 restored the migration ability, the number of cells migrated significantly increased. From these results, we determined that CDK1 is a functional downstream target of miR-378a-5p.

#### CONCLUSION

In our study, we identified that miR-378a-5p is an important modulator in the PDGF-BB stimulated proliferation and

migration of VSMC by targeting, at least partly CDK1 pathway. Also, miR-378a-5p acts on CDK1 and then partly affects p21 to play a role in cell function. In addition, miR-378a-5p negatively regulates the expression of CDK1 after PDGF-BB serve as a stimulant to promote VSMC proliferation. SiCDK1 can partially recover the proliferation of VSMC by PDGF-BB. Furthermore, miR-378a-5p expression levels were upregulated in both human atherosclerotic vascular tissues and proliferation VSMC. The conclusion of these results is: miR-378a-5p/CDK1/p21 is a considerable pathway that can be used as a new therapeutic target in the prevention of atherosclerosis and stent restenosis.

### DISCUSSION

According to the National Health and Family Planning Commission, in the year 2015 more than 500,000 patients with coronary heart disease in mainland China need PCI. PCI has become the main revascularization strategy for unstable coronary artery disease. Despite this, PCI itself still has a problem that has not been overcome, ISR. DES placement does reduce the incidence of ISR, but its incidence is still as high as 10% (Pleva et al., 2018). ISR greatly limits the benefits of PCI. The prevention of ISR is still an important concern. A study found that the ISR of DES is mainly the result of the proliferation of VSMCs, and the high-pressure effect of post-stent expansion accelerates the proliferation of VSMCs. Proliferation, migration, and formation of the extracellular matrix of VSMCS in the middle of the blood vessels lead to intimal regeneration and stenosis of the lumen (Ko et al., 2012). Stent implantation must be performed again for patients with restenosis (Miziastec et al., 2009), so the targeted regulation of VSMC is of great significance for the treatment and prevention of post-stent restenosis.

Accumulating reports have been made to understand the effect of miRNAs in VSMCs biology (Choe et al., 2013), but the specific molecular mechanism is still unknown. In our study, we demonstrated that targeting of the miR-378a-5p/CDK1/p21 pathway may be a potential therapeutic method for stent restenosis.

MiR-378a is a small non-coding RNA molecule which has two mature chains: (1) miR-378a-3p, (2) miR-378a-5p (Krist et al., 2015). The early study of the miR-378a-5p is mainly based on its relationship with the occurrence and development of tumors. However, the role of miR-378a-5p in the regulation of VSMCs

requires deeper research. To explore the effects of miR-378a-5p in VSMCs in restenosis, we performed CCK-8 and EdU assays to detect VSMC proliferation, wound healing and transwell assays to evaluate VSMC migration. In the end, we come to this conclusion that miR-378a-5p plays a role in regulating the proliferation, migration and phenotypic transformation of VSMCs, that is miR-378a-5p participates in the abnormal VSMC biology functions that contribute to stent restenosis development.

Cyclin-dependent kinase 1 (CDK1) is a protein that regulates the cell cycle, which belongs to a serine/threonine kinase family (Malumbres and Barbacid, 2009); previous studies have proved that CDK1 acted as a key regulator for cell cycle (Yang et al., 2016), and its expression increases in several cancer growths, such as colon carcinoma (Meyer et al., 2009), non-small cell lung cancer tumor (Kim et al., 2008; Zhang et al., 2011); There are also researches that identify that inhibition of CDK1 can suppress the proliferation and migration of some tumor cells (Zhang et al., 2015). Interestingly, in oral squamous cell carcinoma (OSCC), the expression of CDK1 increases with the progression of tumor stage, but the expression of CDK1 is reduced at the stage IV and late stage of the tumor (Chen et al., 2015). There is also a report that suggests overexpression of CDK1 inhibits cell proliferation. CDK1 expression increased or decreased at different time intervals (Rui et al., 2011), this phenomenon is hard to explain from a biological view. So maybe miRNA plays different roles in regulating CDK1 at different time points; the effect of CDK1 on cell proliferation cycle needs to be further studied.

Although our research has demonstrated that miR-378a- 5p can target CDK1 to regulate proliferation and migration of VSMCs, but (1) there have been studies that show one miRNA could regulate many target genes, meanwhile, one gene could be modulated by different miRNA. Therefore, targeting miR-378a-5p to treat atherosclerosis and stent restenosis may also affect other genes related to the proliferation of VSMCs, establish an interactive network of non-coding RNAs related to restenosis for early prediction and prognostic evaluation purposes. (2) Non-coding RNA also has problems in application technology and security. (3) Non-coding RNAs also present application technique and safety issues, such as how to coat miR-378a-5p

#### REFERENCES


onto a scaffold, and to understand the effect of the release on human body. (4) Chemically synthesized miRNA inhibitor and mimics are used as scaffold coatings, considering factors such as in vivo concentration, half-life, dose, and sample specificity. (5) Another important consideration is whether external miRNAs will affect normal genes. In this research, there is still a lack of experiments on the downstream target p21 of CDK1, which needs further verification. Moreover, the selection of samples has some limitations, the number of samples is small, so the clinical sample size should be increased. The specific mechanism of miR-378a-5p for target regulation of atherosclerosis and stent restenosis remains to be further studied in the follow-up work.

### AUTHOR CONTRIBUTIONS

SL and NT carried out the cell and protein analysis. YY carried out the molecular experiments. SJ, HX, RZ, and SL performed the clinical analysis. TY and SL participated in the data analysis, performed the statistical analysis, and drafted the manuscript. TY and HX conceived and designed the study, participated in the data analysis and coordination, and helped to draft the manuscript. All authors read and approved the final manuscript.

### FUNDING

This work was supported by National Natural Science Foundation of China (31701208 and 81870331 to TY), China Postdoctoral Science Foundation (2017M612189), Natural Science Foundation of Shandong Province (ZR2017MC067), and The People's Livelihood Science and Technology Project of Qingdao (17-1-1-40-jch and 18-2-2-65-jch).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00022/full#supplementary-material


neointimal hyperplasia by targeting HMGB1 in arteriosclerosis obliterans. Cell Physiol. Biochem. 42, 2492–2506. doi: 10.1159/000480212


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

Copyright © 2019 Liu, Yang, Jiang, Xu, Tang, Lobo, Zhang, Liu, Yu and Xin. 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.

, Amara Lobo<sup>1</sup>

,

# Corrigendum: MiR-378a-5p Regulates Proliferation and Migration in Vascular Smooth Muscle Cell by Targeting CDK1

#### Approved by:

*Frontiers in Genetics Editorial Office, Frontiers Media SA, Switzerland*

#### \*Correspondence:

*Tao Yu yutao0112@qdu.edu.cn; dachao1201@hotmail.com Hui Xin xinhuiqy@163.com*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

Received: *21 February 2019* Accepted: *22 February 2019* Published: *15 March 2019*

#### Citation:

*Liu S, Yang Y, Jiang S, Xu H, Tang N, Lobo A, Zhang R, Liu S, Yu T and Xin H (2019) Corrigendum: MiR-378a-5p Regulates Proliferation and Migration in Vascular Smooth Muscle Cell by Targeting CDK1. Front. Genet. 10:193. doi: 10.3389/fgene.2019.00193* *<sup>1</sup> Department of Cardiology, The Affiliated Hospital of Qingdao University, Qingdao, China, <sup>2</sup> Institute for Translational Medicine, Qingdao University, Qingdao, China, <sup>3</sup> Department of Cardiology, The Affiliated Cardiovascular Hospital of Qingdao University, Qingdao, China, <sup>4</sup> Department of Orthodontic, The Affiliated Hospital of Qingdao University, Qingdao, China*

\*

\* and Hui Xin<sup>1</sup>

, Hong Xu<sup>4</sup>

, Ningning Tang<sup>2</sup>

Keywords: miR-378a-5p, vascular smooth muscle cell, stent-restenosis, proliferation, migration, atherosclerosis, CDK1

#### **A Corrigendum on**

Rui Zhang<sup>1</sup>

Shaoyan Liu1†, Yanyan Yang2†, Shaoyan Jiang<sup>3</sup>

, Tao Yu<sup>2</sup>

, Song Liu<sup>1</sup>

#### **MiR-378a-5p Regulates Proliferation and Migration in Vascular Smooth Muscle Cell by Targeting CDK1**

by Liu, S., Yang, Y., Jiang, S., Xu, H., Tang, N., Lobo, A., et al. (2019). Front. Genet. 10:22. doi: 10.3389/fgene.2019.00022

In the original article, we neglected to include the funder "National Natural Science Foundation of China," "81870331" to TY.

Additionally, there was an error in affiliation 3. Instead of "Department of Cardiology, The Affiliated Hospital of Qingdao University, Qingdao, China," it should be "Department of Cardiology, The Affiliated Cardiovascular Hospital of Qingdao University, Qingdao, China."

The authors apologize for these errors and state that this does not change the scientific conclusions of the article in any way. The original article has been updated.

Copyright © 2019 Liu, Yang, Jiang, Xu, Tang, Lobo, Zhang, Liu, Yu and Xin. 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.

# CircRNAs Are Here to Stay: A Perspective on the MLL Recombinome

Anna Dal Molin<sup>1</sup> , Silvia Bresolin<sup>2</sup> , Enrico Gaffo<sup>2</sup> , Caterina Tretti<sup>2</sup> , Elena Boldrin<sup>3</sup> , Lueder H. Meyer<sup>3</sup> , Paola Guglielmelli<sup>4</sup> , Alessandro M. Vannucchi<sup>4</sup> , Geertruij te Kronnie<sup>2</sup> \* and Stefania Bortoluzzi<sup>1</sup>

<sup>1</sup> Department of Molecular Medicine, University of Padua, Padua, Italy, <sup>2</sup> Department of Women's and Children's Health, University of Padua, Padua, Italy, <sup>3</sup> Department of Pediatrics and Adolescent Medicine, Ulm University Medical Center, Ulm, Germany, <sup>4</sup> CRIMM, Center for Research and Innovation of Myeloproliferative Neoplasms, AOU Careggi, Department of Experimental and Clinical Medicine, University of Florence, Florence, Italy

Chromosomal translocations harbored by cancer genomes are important oncogenic drivers. In MLL rearranged acute leukemia (MLLre) MLL/KMT2A fuses with over 90 partner genes. Mechanistic studies provided clues of MLL fusion protein leukemogenic potential, but models failed to fully recapitulate the disease. Recently, expression of oncogenic fusion circular RNAs (f-circ) by MLL-AF9 fusion was proven. This discovery, together with emerging data on the importance and diversity of circRNAs formed the incentive to study the circRNAs of the MLL recombinome. Through interactions with other RNAs, such as microRNAs, and with proteins, circRNAs regulate cellular processes also related to cancer development. CircRNAs can translate into functional peptides too. MLL and most of the 90 MLL translocation partners do express circRNAs and exploration of our RNA-seq dataset of sorted blood cell populations provided new data on alternative circular isoform generation and expression variability of circRNAs of the MLL recombinome. Further, we provided evidence that rearrangements of MLL and three of the main translocation partner genes can impact circRNA expression, supported also by preliminary observations in leukemic cells. The emerging picture underpins the view that circRNAs are worthwhile to be considered when studying MLLre leukemias and provides a new perspective on the impact of chromosomal translocations in cancer cells at large.

#### Edited by:

Zhao-Qian Teng, Institute of Zoology (CAS), China

#### Reviewed by:

Argyris Papantonis, Universität zu Köln, Germany Steven G. Gray, St. James's Hospital, Ireland

> \*Correspondence: Geertruij te Kronnie truustekronnie@unipd.it

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 09 October 2018 Accepted: 29 January 2019 Published: 13 February 2019

#### Citation:

Dal Molin A, Bresolin S, Gaffo E, Tretti C, Boldrin E, Meyer LH, Guglielmelli P, Vannucchi AM, te Kronnie G and Bortoluzzi S (2019) CircRNAs Are Here to Stay: A Perspective on the MLL Recombinome. Front. Genet. 10:88. doi: 10.3389/fgene.2019.00088 Keywords: circRNA, leukemia, MLL rearrangements, fusion-circRNA, translocation breakpoint region, blood cells

### THE ACUTE LEUKEMIA MLL RECOMBINOME

Cancer genomes often harbor chromosomal translocations and specific rearrangements are recurrent in specific cancers, most often involving two specific genes or one promiscuous gene translocated with an array of different genes. In acute leukemia, the MLL gene (official gene symbol KMT2A) on chromosome 11q23 breaks and fuses with more than 90 translocation partner genes (TPGs) (Ottersbach et al., 2018), with only a few of them recurrently found in MLL rearrangements (MLLre).

Genomic studies revealed that MLL fusions are clonal and are considered early initiating leukemogenic events (Yip and So, 2013; Sanjuan-Pla et al., 2015), and very few additional mutations are needed to generate infant MLLre (Dobbins et al., 2013). Mechanistic studies

of particular MLLre showed that fusion proteins are crucial for transforming potential (Reimer et al., 2017). In addition, recent data indicated that expression of the wild type KMT2A is dispensable for MLLre leukemic cells, whereas deletion of KMT2D alone or in combination with loss of KMT2A reduces proliferation and induces apoptosis of MLL-AF9 transformed cells (Chen et al., 2017).

The 2017 survey of the MLL recombinome consortium (Meyer et al., 2018), based on DNA sequence analysis of 2,345 MLLre acute leukemia cases and literature scrutiny, identified 135 different MLLre involving 94 TPGs, with only a few genes accounting for most cases. Association of specific partner genes and/or the breakpoint position with age classes and/or with leukemia subtypes has been described (Meyer et al., 2018). AF4 is almost exclusively (99%) found in BCP-ALL, whereas AF9 is slightly more frequent (60%) in myeloid leukemias. Specific MLL TPGs are detected exclusively in leukemias of the lymphoid (e.g., LAF4/AFF3) or of the myeloid lineages (e.g., SEPT6).

MLL breakpoints occur preferentially in three clusters (exin 9, ex-in 10, ex-in 11/12) (van Dongen et al., 1999; de Jonge et al., 2005). MLL encodes a Lysine Methyltransferase involved in tissue-specific epigenetic activation of developmental genes. Most TPGs encode proteins of complexes that affect transcriptional elongation. Even if the functional effect has not been investigated for all MLL fusions, MLLre leukemias display a deeply deregulated epigenetic and transcriptional state, and the contribution of MLL fusions to leukemia initiation and evolution, therapy resistance and relapse is still under active investigation. An array of cellular and animal models generated to study the leukemogenic mechanisms of MLL fusions furthered our understanding of MLLre, but failed to fully recapitulate the human disease features (Ottersbach et al., 2018). The current state-of-the-art is controversial regarding the cell-of-origin, the timing and level of MLL fusion protein expression. Even less is known of the transcripts expressed by MLL fusions. Recent reports of oncogenic fusion circular RNAs (f-circ) and data emerging on circular RNAs (circRNAs, transcripts in which the splice donor site is covalently bound to an upstream acceptor site by backsplicing) in general made the drawing more complex, but also very intriguing, and opened a series of new perspectives.

Rearranged cancer genomes (promyelocytic leukemia with PML-RARα, acute myeloid leukemia with KMT2A-MLLT3 and a model of the NPM1-ALK fusion) (Guarnerio et al., 2016; Babin et al., 2018) express f-circ. F-circ include two sequences not present in the normal genome: the fusion junction, in which two genomic regions far apart in normal genomes are juxtaposed, and the backsplice junction, connecting in reverse order two sequences of the fusion gene. A few studies showed that f-circ can be oncogenic but only started to disclose involved mechanisms. Concurrent expression of f-circM9 and MLL-AF9 protein contributed to leukemia progression in in vivo and ex vivo models, and f-circM9 expression increased drug resistance of leukemic cells (Guarnerio et al., 2016). F-circ generated from an EML4-ALK fusion promoted non-small cell lung cancer development (Tan et al., 2018). The oncogenic potential of f-circ was linked to mechanisms involving the fusion protein by the observation that proliferation and TKI resistance of BCR-ABL1 leukemic cells are enhanced by a f-circ that increases the fusion protein level. Nevertheless, none of these studies investigated the impact of rearrangements on circRNAs expressed by MLL and TPGs during normal hematopoiesis. In leukemic cells carrying translocations beyond generation of f-circ, also ablation or deregulation of circRNAs will occur with potential contribution to the disease.

Circularized transcripts of specific genes were reported since the 80's, but circRNAs were re-discovered (Salzman et al., 2012; Memczak et al., 2013, 2015; Bonizzato et al., 2016) when projects coupling RNA-seq with bioinformatics methods for reads mapping appropriate to detect backsplicing showed that thousands of circRNAs are expressed by many genes with cell type-specific expression regulation (Maass et al., 2017), and differentiation stage-specificity (Rybak-Wolf et al., 2015). As reviewed recently (Memczak et al., 2013, 2015; Bonizzato et al., 2016) circRNAs are abundantly expressed in the hematopoietic compartment. CircRNAs can play important and diverse functions, including some typical of non-coding RNAs. The first function assigned to circRNAs was sponging miRNAs (Salzman et al., 2012; Memczak et al., 2013, 2015; Bonizzato et al., 2016) and indirectly regulating miRNA-target expression. In this way, some circRNAs control key miRNA-involving axes in normal developmental processes and oncogenesis (Hansen et al., 2013; Li F. et al., 2015). Other circRNAs regulate cellular processes by interacting with RNA-binding proteins (Schneider et al., 2016), scaffolding molecular complexes, as shown for circFOXO3 that controls cell cycle progression by binding p21 and CDK1 (Du et al., 2016). As long as non-coding RNAs, also circRNAs have some coding potential, and can be translated into functional peptides according to recent reports (Legnini et al., 2017; Pamudurti et al., 2017; Yang et al., 2017).

Taken together these discoveries of circRNA pervasiveness and functions prompted us to study the circRNAs expressed by genes of the MLL recombinome (MLL-rec) in normal blood cells. In this perspective we examined how rearrangements can result in alteration of sequences and expression level of circRNAs normally generated by MLL and TPGs, and provided data of MLLre leukemia in support.

#### CircRNAs EXPRESSED IN NORMAL HEMATOPOIESIS BY GENES OF THE MLL RECOMBINOME

The MLL-rec analyzed here included MLL plus 94 TPGs (**Supplementary Table 1**): 75 genes of rearrangements disclosed by Meyer et al. (2018) and 19 genes previously reported in the literature collected in the same study. Almost 93% of 2,345 leukemia cases reported in Meyer et al. (2018) resulted from MLL fusions with one of 12 highly recurrent TPGs (AF4/AFF1, AF9/MLLT3, ENL/MLLT1, AF10/MLLT10, BCS1L/PTD, ELL, AFDN/AF6/MLLT4, EPS15, AF1Q/MLLT11, SEPT6, AF17/MLLT6, and SEPT9), that will be cited from now on using the official gene name.

To obtain a data-driven picture of the circular transcriptome of MLL-rec genes in the hematopoietic compartment, we

analyzed our RNA-seq dataset regarding normal hematopoiesis (available at GEO series ID: GSE110159 and upon request): 15 samples from healthy donors, including 3 of CD34+ cells purified from cord blood, and 12 of B-, T-cells and monocytes FACS sorted from PBMCs (4 different donors per cell type). RNA-seq data were obtained from ribo-depleted RNA with Illumina <sup>R</sup> HiSeq2000 (average depth of 145 M paired end 100 nt reads per sample, 66% of reads passed quality control). CircRNAs were detected and quantified using CirComPara v0.3 (Gaffo et al., 2017), implementing 6 circRNA detection methods (CIRI, Findcirc, CIRCexplorer2+Star, CIRCexplorer2+Segemehl, CIRCexplorer2+BWA, CIRCexplorer2+TopHat). CirComPara identified 41,515 circRNAs expressed from 8,138 individual genes. We focused on the subset of 16,606 circRNAs with high expression in at least one cell type (maximum of the averages per cell type in the top 40% highest values), which derived from 5,170 genes.

For 25 of the 95 genes in the MLL-rec, no circRNAs were detected in our dataset and 16 had only circRNAs expressed at low level. Interestingly, 54 genes of the recombinome expressed at least one abundant circRNA, for a total of 327 circRNAs (**Supplementary Table 2**), which were further investigated.

The expression level of these circRNAs was not different from the 16,005 circRNAs expressed by the 5,116 non-MLLrec genes of the human genome (**Supplementary Figure 1**). MLL-rec genes expressed several circRNAs at very high levels: the 33 most expressed circRNAs (**Supplementary Figure 2**) derived from 21 genes, including AFF1 and EPS15, which presented circRNAs overexpressed in stem cells; and SEPT6 and SEPT9, whose circRNAs were detected only in mature cells, with marked overexpression of circSEPT9 17:77402059–77402703:+ in monocytes. CircAFF3 2:100006632–100008932:- and circAFF4 5:132892164–132893118:-, both from MLL TPGs only found in leukemias of the lymphoid lineage, were upregulated in B-cells. Four genes contributed several circRNAs to the group of the most highly expressed: 10 circPICALM were mostly upregulated in monocytes, albeit circPICALM 11:86007542– 86026367:- and 11:86022367–86031611:- were more abundant in stem cells; 2 circME2 were upregulated in stem cells; 2 circARHGAP26 were upregulated in monocytes; 2 circPDS5A were upregulated in B-cells.

MLL-rec genes compared to non-MLL-rec genes had a significantly larger number of circular isoforms per gene (6.06 vs. 3.13 circRNAs per gene in average; t-test p-value 2.317 E−04), with a distribution of the number of circRNAs per gene significantly shifted toward higher values (Chi-squared test pvalue 1.791 E−06; **Supplementary Figures 3A,B**). Indeed, 9 of the recombinome genes expressed only one circRNA each, and 45 (83%) had multiple circular isoforms in blood cells (**Figure 1A**). The genes with the highest numbers of isoforms were PICALM (31 circRNAs), AKAP13 (19), EPS15 (16), PDS5A (16), ARHGAP26 (15), ITPR2 (15), EP300 (14), ME2 (14), and MYO1F (10). Moreover, MLLT10, SEPT6 and AFF1 had 6 circRNAs each, and MLL3 expressed 4 circRNAs.

Similar numbers of circRNAs were expressed in the different mature populations by MLL-rec (205, 244, and 210 in B-, T-cells and in monocytes, respectively), and only 94 circRNAs in the stem population (**Supplementary Figure 4A**). Apart from 44 circRNAs expressed in all cell types, 16 from 10 genes (CENPK, LPP, ITPR2, TCF12, CLTC, FNBP1, FRYL, SEPT11, SEPT6, and MLLT10) were detected only in stem cells (circCENPK 5:65528452–65529145:- was the most abundant), and 233 had expression restricted to one or more mature populations, being mostly detected in all the mature populations (80) or in lymphocytes (40).

Most genes displayed cell type-specific alternative circularization patterns (**Supplementary Figure 4B**). Of 16 stem cell-specific circRNAs, circCLTC 17:59668941–59669222:+ was expressed by a gene with circRNAs only in stem cells, and 15 derived from genes with other circular isoforms detected in mature cell types. The three most abundant of the 16 different circEPS15 had different expression profiles (**Supplementary Figure 4C**): circEPS15 1:51402435–51408332:- was expressed in all cell types with a slight upregulation in stem cells, whereas circEPS15 1:51394381–51448135:- was upregulated only in monocytes, and circEPS15 1:51405905–51408332:- was more abundant in B- and T-cells than in monocytes, and not detected in stem cells.

A scrutiny of the two main circRNA-indexing databases, CircBase and circRNADB, showed that 2,433 circRNAs were previously reported for 85 of 95 genes in the MLL-rec, including all the 54 with highly expressed circRNAs, and all but ACER1 and PFDN4 of the 70 genes with circRNAs detected in our data. Moreover, for 17 (out of 25) genes in the MLL-rec for which we did not observe appreciable circRNA expression in blood cells, the DBs reported circRNAs mostly in other tissues, such as brain (e.g., ARHGEF17) or muscle (e.g., FLNC). Even if a direct comparison of our data with data reported in the DBs is not feasible, findings concordantly indicate that circRNAs are expressed from the overwhelming majority of genes in the MLL-rec, with complex patterns of alternative circularization in many cases.

Of note, 63 new circRNAs were detected at high expression in blood populations investigated in this study, which are currently not annotated in the DBs (**Supplementary Table 2**). Newly detected circRNAs derived from 26 genes of the MLL-rec, including MLLT10, MLLT3, EPS15 and SEPT6, and comprised 15 circPICALM and circRNAs with expression restricted to stem cells or to a specific mature population.

### TRANSLOCATIONS CAN IMPACT CircRNA EXPRESSION FROM MLL/KMT2A AND TPGs

As described above, KMT2A and most of TPGs observed in acute leukemias do express circRNAs in normal blood cells (**Figure 1A**). In MLLre leukemias the rearrangement of KMT2A with a TPG leads to the formation of a "fusion gene." When KMT2A is expressed, also the downstream partner gene may be expressed resulting in a transcript that potentially is translated in a fusion protein (Ayton and Cleary, 2001; So et al., 2003). Alongside also f-circ may be produced from chimeric fusion genes (Guarnerio et al., 2016), as reviewed in Bonizzato et al.

(2016). In addition, the breaking and fusion of KMT2A and TPGs may alter the formation of circRNAs from exons proximal to the breakpoint of the translocated genes. Information on the KMT2A and TPGs specific exons that undergo circularization in normal hematopoiesis (**Figure 1B**) is critical to understand how a specific rearrangement can impact circRNA expression (**Figure 1C**). The breakpoint position defines which genomic regions are retained in the fusion gene, determining not only if and how f-circ are generated. The breakpoint position can also affect the circRNAs "normally" generated by the partner genes influencing their expression level or possibly activating cryptic backsplicing sites.

For KMT2A and of three of the main TPGs we further examined in detail circRNAs expression in normal blood cells in relation to potential breakpoint regions. In addition, we provided preliminary data of MLLre leukemia using RNAseq data of the THP1 cell line with KMT2A-MLLT3 (MLL-AF9) fusion (Guarnerio et al., 2016) and unpublished data of two specimens of infant ALL with KMT2A-MLLT1 (MLL-ENL) fusions (**Supplementary Table 3**).

In normal hematopoiesis, KMT2A produced 5 different circRNAs (**Figure 2A**). The most abundant circKMT2A (11:118481715–118482495:+; exons 7–8) was expressed in all mature populations, with upregulation in both B- and T-cells, while it was absent in the stem population, in which two other circKMT2A isoforms were detected. The formation of two wild type circKMT2A isoforms (exons 5–8 and 7–8) are potentially perturbed in MLLre leukemias, considering that the recurrent breakpoint region includes exon 8. The other three circKMT2A expressed in normal hematopoiesis (exons 12–16, 17–23, 21–23) will probably not be formed from the fusion gene, since they originate from exons not retained in most rearrangements. Thus, all circRNAs highly expressed in normal cells by KMT2A may be downregulated or even absent in the leukemic cells. In line, none of the circRNAs expressed in normal hematopoiesis by KMT2A was detected, neither in the THP1 cell line (in which two fusion circRNAs were previously detected by RT-PCR, and albeit supported by few reads, from RNA-seq data), nor in any of the two samples with KMT2A-MLLT1 fusion. Of note, KMT2A exons 5 or 7 are the starting ends backspliced both in normal hematopoiesis (joined with exon 8 in two of the five observed circKMT2A isoforms) and in THP1 cells (joined with MLLT3 exon 6 in the two f-circM9 isoforms) (Guarnerio et al., 2016). If and how the reported oncogenic potential of f-circM9 (KMT2A exon 7 – MLLT3 exon 6) is related to the function of the most expressed circKMT2A (KMT2A exons 7–8), that is fully contained in the f-circM9, remains to be determined.

The three TPGs most recurrent in acute leukemias (AFF1, MLLT3, MLLT1) expressed one to 5 circRNAs each in normal hematopoiesis (**Figures 2B–D**). AFF1 presented 5 circRNAs, with circAFF1 4:87046166–87047594:+ (exons 2–3) contributing alone to 90% of circular expression of the gene. The same circAFF1 was one of the most expressed in the sample set, it presented a remarkable upregulation in stem cells compared to mature cells according to our data, and was found upregulated in common lymphoid precursors and in monocytes (Nicolet et al., 2018). The AFF1 recurrent breakpoint region position in MLLre leukemias reveals that this circAFF1 will likely not be generated from the chimeric gene. The expression of the other 4 circAFF1 will depend on the exact position of the breakpoint (**Figure 2B**).

MLLT3 had 4 circular isoforms, almost all specific of T-cells. The second most expressed circMLLT3 contains part of intron 4, as observed for other circRNAs (Li Z. et al., 2015). These circRNAs will not be generated from the fusion gene due to the position of the breakpoints, either falling within the sequence undergoing circularization or eliminating all circularized exons from the fusion transcripts (**Figure 2C**). None of the T-specific circMLLT3 isoforms was detected in KMT2A-MLLT3 THP1 cells.

Also circMLLT1 9:6230570–6274759:- (exons 2–4), expressed in mature populations, will not be generated from the KMT2A-MLLT1 derivative, since the breakpoints of t(11,19) leukemias either abolish the MLLT1 exons undergoing circularization or fall in the intron flanking the backspliced exon 2. In line with our assumption, this circRNA was not detected in leukemic cells (**Figure 2D**) with two different KMT2A-MLLT1 fusions (KMT2A exon 9 – MLLT1 exon 6, and KMT2A exon 8 – MLLT1 exon 4 fusions, **Supplementary Table 3**).

Here, we analyzed one of the main KMT2A-TPG fusions, and the translocations indeed impacted circRNAs expressed from KMT2A and its TPG. Highly informative data on leukemias with KMT2A-MLLT1 fusions were fully in agreement with observations (based on a lower depth) in THP1 cells. Little is known about the other TPG-KMT2A derivatives.

To generalize, based on circRNA and breakpoint respective position, four scenarios can occur: (1) The exons undergoing circularization are retained in the KMT2A-TPG derivative (e.g., exons in the 5<sup>0</sup> region of KMT2A or in the 3<sup>0</sup> region of the TPG retained in the fusion gene), circRNAs can be still generated; (2) The exons undergoing circularization are not retained, the circRNAs can be abolished; (3) The breakpoint is located between the backspliced ends, the absence in the fusion gene of one or more exons normally undergoing circularization will prevent the formation of circRNAs, possibly favoring the generation of f-circ; (4) The breakpoint is located in an intron flanking one of the backsplice ends abolishing specific in cis sequence elements (Alu and inverted repeats) that favor the backsplicing (Jeck et al., 2013; Bonizzato et al., 2016), and/or sequences recognized by in trans regulatory factors, such as QKI and ADAR1, which promotes (Conn et al., 2015) and suppresses (Rybak-Wolf et al., 2015) circularization, respectively. We are well aware that our model of how translocations can impact expression of circRNAs from fused genes is hampered by simplicity. Position-effects, as well as deep genome-wide epigenetic and transcriptional deregulation in leukemic cells can affect circRNAs expressed from the fused genes as well as circRNAs from other loci.

### CONCLUSION AND FUTURE DIRECTIONS

Data emerging from recent literature and from the present study collectively show that KMT2A and TPGs express many circRNAs,

possibly playing important functions and being perturbed in cells bearing rearrangements. According to our observations, depending on the position of the breakpoint respective to the backspliced sequences, translocations can impact expression of circRNAs in addition to generating f-circ. Our data on circRNAs expressed from genes of the MLL recombinome are instrumental to analyze alterations of circRNAs from rearranged genes of MLLre leukemias and f-circ generation. It is known that f-circ can be oncogenic, but the involved mechanisms remain essentially unknown. F-circ and the other circRNAs expressed from KMT2A and TPGs might share part of their sequences and play similar functions (e.g., through common protein or nucleic acids interactors, or due to overlapping coding potential). F-circ were previously shown to reinforce the oncogenic potential of fusion proteins, perhaps cooperating to the same mechanisms (Babin et al., 2018). A parallel line of evidence showed that loss of both KMT2A and KMT2D impair survival pathways in leukemic cells (Chen et al., 2017), implicating them in oncogenesis. In this view, beyond f-circ, also circRNAs generated from the non-translocated alleles are critical candidates to be investigated for participation to disease mechanisms. Further, f-circ and circRNAs in general share exons with linear transcripts, and circularization is associated with exon skipping (Kelly et al., 2015). The equilibrium perturbed by translocations can involve a series of transcripts including "normal" as well as aberrant circular and linear transcripts, all linked in an intricate network of similar sequences and functions and interdependent biogenesis.

In conclusion, next to the studies of fusion proteins as oncogenic drivers of leukemias with MLL rearrangements, we stress the need of a molecular characterization of circRNAs expressed by fusion genes, KMT2A itself and its TPGs. A first direction could be to clarify the participation of circRNAs to the molecular complexes involving KMT2A, and to define circRNA molecular interactions with regulators, such as microRNAs, of KMT2A or of its interactors involved in leukemogenesis. The perspective emerging from this pilot study on MLLre acute leukemia is presumably valid for most of the driver chromosomal fusions occurring in cancer cells.

#### ETHICS STATEMENT

fgene-10-00088 February 13, 2019 Time: 20:15 # 7

All experiments involving human material followed the principles outlined in the Helsinki Declaration. The study has been approved by the ethics committees of Padova University Hospital, Ulm University Medical Center and of AOU Careggi (Florence) and written informed consent was obtained from all subjects.

#### AUTHOR CONTRIBUTIONS

GtK and StB conceived the study. ADM, EG, and StB contributed bioinformatics methods and performed data analysis. SiB, CT, and StB performed comparative public data analysis. ADM, StB,

#### REFERENCES


and GtK wrote the manuscript. StB, ADM, and SiB made the figures. EB, LM, PG, and AV provided data and revised the manuscript. All authors approved the final manuscript.

#### ACKNOWLEDGMENTS

We acknowledge for financial support: AIRC (IG #20052 to StB and GtK, MFAG #15674 to SiB, AIRC 5 × 1000 MYNERVA project, #21267 to AV and PG), Department of Molecular Medicine, University of Padua (PRID 2017 to StB), Cariparo (Pediatric research projects to GtK and StB), Ministero della Salute GR-2011-02352109 to PG.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00088/full#supplementary-material


in vivo. Haematologica 102, 1558–1566. doi: 10.3324/haematol.2017.16 4046


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

Copyright © 2019 Dal Molin, Bresolin, Gaffo, Tretti, Boldrin, Meyer, Guglielmelli, Vannucchi, te Kronnie and Bortoluzzi. 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.

# miR-499-5p Attenuates Mitochondrial Fission and Cell Apoptosis via p21 in Doxorubicin Cardiotoxicity

Qinggong Wan1,2† , Tao Xu<sup>1</sup>† , Wei Ding<sup>3</sup> , Xuejuan Zhang<sup>3</sup> , Xiaoyu Ji1,2, Tao Yu<sup>1</sup> , Wanpeng Yu<sup>1</sup> , Zhijuan Lin<sup>1</sup> and Jianxun Wang1,2 \*

<sup>1</sup> Center for Regenerative Medicine, Institute for Translational Medicine, College of Medicine, Qingdao University, Qingdao, China, <sup>2</sup> School of Basic Medical Sciences, Qingdao University, Qingdao, China, <sup>3</sup> Department of Comprehensive Internal Medicine, Affiliated Hospital, Qingdao University, Qingdao, China

#### Edited by:

Ge Shan, University of Science and Technology of China, China

#### Reviewed by:

Udayan Bhattacharya, Technion – Israel Institute of Technology, Israel Nithyananda Thorenoor, Pennsylvania State University, United States

> \*Correspondence: Jianxun Wang wangjx@qdu.edu.cn

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 06 October 2018 Accepted: 22 December 2018 Published: 21 January 2019

#### Citation:

Wan Q, Xu T, Ding W, Zhang X, Ji X, Yu T, Yu W, Lin Z and Wang J (2019) miR-499-5p Attenuates Mitochondrial Fission and Cell Apoptosis via p21 in Doxorubicin Cardiotoxicity. Front. Genet. 9:734. doi: 10.3389/fgene.2018.00734 Doxorubicin (DOX) is a broad-spectrum anti-tumor drug, but its cardiotoxicity limits its clinical application. A better understanding of the molecular mechanisms underlying DOX cardiotoxicity will benefit clinical practice and remedy heart failure. Our present study observed that DOX caused cardiomyocyte (H9c2) apoptosis via the induction of abnormal mitochondrial fission. Notably, the expression levels of p21 increased in DOX-treated cardiomyocytes, and the silencing of p21 using siRNA greatly attenuated mitochondrial fission and apoptosis in cardiomyocytes. We also found that miR-499- 5p could directly target p21 and attenuated DOX-induced mitochondrial fission and apoptosis. The role of the miR-499-5p-p21 axis in the prevention of DOX cardiotoxicity was also validated in the mice model. DOX treatment induced an upregulation of p21, which induced subsequent abnormal mitochondrial fission and myocardial apoptosis in mouse heart. Adenovirus-harboring miR-499-5p-overexpressing mice exhibited significantly reduced p21 expression, mitochondrial fission and myocardial apoptosis in hearts following DOX administration. The miR-499-5p-overexpressing mice also exhibited improved cardiomyocyte hypertrophy and cardiac function after DOX treatment. However, miR-499-5p was not involved in the DOX-induced apoptosis of cancer cells. Taken together, these findings reveal an emerging role of p21 in the regulation of mitochondrial fission program. miR-499-5p attenuated mitochondrial fission and DOX cardiotoxicity via the targeting of p21. These results provide new evidence for the miR-499-5p-p21 axis in the attenuation of DOX cardiotoxicity. The development of new therapeutic strategies based on the miR-499-5p-p21 axis is a promising path to overcome DOX cardiotoxicity as a chemotherapy for cancer treatment.

Keywords: doxorubicin, miR-499-5p, p21, cardiotoxicity, mitochondrial, apoptosis

## INTRODUCTION

Doxorubicin (DOX), also known as adriamycin (ADR), exerts a killing effect on a variety of tumors, and it is a widely used anti-tumor drug (Levis et al., 2017). However, a previous study found that DOX produced heart damage in many patients that eventually led to heart failure (Gorini et al., 2018). The development of treatment strategies that avoid DOX cardiotoxicity without

affecting its anti-tumor effects is urgently needed. A previous study found that DOX-induced cardiotoxicity primarily involved the production of reactive oxygen species (ROS) (Ichikawa et al., 2014), lipid peroxidation (Yarana et al., 2018), DNA damage (Li et al., 2009), mitochondrial dysfunction (Dhingra et al., 2014), apoptosis (Kumar et al., 2001), and autophagy dysregulation (Bartlett et al., 2017). These toxicities ultimately caused cell death of cardiomyocyte. Therefore, a better understanding of the molecular mechanisms underlying DOX cardiotoxicity will improve the clinical application of DOX during cancer therapy (Kluza et al., 2004).

MicroRNAs (miRNAs) are endogenous non-coding RNAs that are highly conserved in different species (Lim et al., 2003). miRNAs functionally participate in the developmental, physiological and pathological process via negative regulation of target gene (Bartel, 2009; Chistiakov et al., 2016). Increasing evidence suggests that miRNAs are actively involved in the regulation of cardiac functions, such as electrical signal conductance, heart muscle contraction, heart growth and morphogenesis (Kozomara and Griffiths-Jones, 2011). What's more, miRNAs are also powerful regulators of cardiovascular diseases, and manipulation of miRNAs is a promising strategy for the development of novel therapeutic agents (Mette et al., 2002; Liu S. et al., 2018; Tang et al., 2018).

miR-499-5p is a recently discovered member of myosinencoded miRNAs (Olivieri et al., 2013). Several studies reported that miR-499-5p was differentially regulated and functioned in the heart development process (Sluijter et al., 2010; Wilson et al., 2010; Fu et al., 2011; Wander et al., 2016). miR-499-5p is expressed at a high level in the heart under physiological conditions, and it attenuates the expression of the β-myosin heavy chain, which results in enhanced myocardial oxygen metabolism and tolerance. miR-499-5p is downregulated in human heart diseases and experimental models of heart failure, and it is involved in the transcriptional and posttranslational regulation of pathological hypertrophy (Chistiakov et al., 2016). Several lines of evidence suggest that miR-499-5p exerts cardioprotective effects via the protection of cardiomyocytes from stress-induced apoptosis. miR-499-5p inhibits calcineurin-mediated dephosphorylation of dynaminrelated protein-1 (Drp1) via the targeting of CnA α and CnA β, which reduces Drp-1 mitochondrial aggregation and attenuates Drp-1-mediated mitochondrial fission (Wang et al., 2011). miR-499-5p also targets several regulatory factors that inhibit mitochondrial cell apoptosis and increase cell survival, such as bispecific tyrosine phosphorylation regulatory kinase 2 (Dyrk2), programmed cell death protein 4 (Pdcd4) (Li Y. et al., 2016) and phosphoric acid Forint acid cluster sorting protein 2 (Pasc2) (Wang J. et al., 2014). However, the potential effects and underlying mechanisms of miR-499-5p in the protection against DOX-induced heart failure have not been revealed.

Cyclin-dependent kinase inhibitor 1a (CDKN1a), also known as p21, is a negative regulator that halts cell cycle progression at the G1/S and G2/M transition points via inhibition of CDK4,6/cyclin-D and CDK2/cyclin-E, respectively (Sharpless and Sherr, 2015). Mammalian p21 is expressed at very low levels in embryonic and neonatal hearts (Brooks et al., 1998). Numerous studies demonstrated that p21 was involved in the pathological process of myocardial injury (Wang et al., 2015a). Inhibition of p21 prevented endothelial cell apoptosis in arseniteinduced endothelial dysfunction-related vascular diseases (Nuntharatanapong et al., 2005). Increased p21 levels in cardiac fibroblasts surrounding an infarction area is a biomarker of hyperoxia perception (Sen et al., 2006). Hyperoxia causes cardiac remodeling via the induction of p21-dependent differentiation of cardiac fibroblasts (Roy et al., 2003a,b). Abnormal p21 expression is closely related to myocardial damage and cardiac hypertrophy (Wang R. et al., 2014). Transverse aortic constriction (TAC) surgery and DOX administration concomitantly increased p21 levels in rat and mouse cardiomyocytes (Terrand et al., 2011; Koga et al., 2013). Increased p21 expression in tissues mediates myocardial fibrosis and remodeling (Kuhn et al., 2007; Megyesi et al., 2015). However, the mechanism of p21 regulation of myocardial damage and the regulation of the p21 expression in cardiomyocytes under stress have not been elucidated.

Our present study investigated the molecular mechanisms underlying DOX cardiotoxicity. We found an emerging role of p21 in promoting mitochondrial fission and cardiomyocyte apoptosis via regulation of mitochondrial fission programming induced by DOX treatment. miR-499-5p inhibited DOXinduced mitochondrial fission and apoptosis in cardiomyocytes via the targeting of p21. DOX cardiotoxicity was prevented in adenovirus-harboring miR-499-5p-overexpressing mice, and miR-499-5p overexpression improved cardiac function. Our research demonstrated that the miR-499-5p-p21 axis constitutes a new antiapoptotic pathway to attenuate DOX-induced cardiotoxicity and may provide valuable insights to prevent DOX cardiotoxicity during cancer chemotherapy.

### MATERIALS AND METHODS

### Cell Culture and Treatment

H9c2 cells were purchased from the Shanghai Institutes for Biological Sciences (Shanghai, China). The human lung cancer cell line A-549, the human gastric cancer cell line SGC-7901, the human hepatocellular carcinoma cell line HepG-2 and the human colorectal cancer cell line SW-480 were purchased from the Chinese Academy of Sciences Cell Bank. Cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and antibiotics (100 IU/mL penicillin and 100 mg/mL streptomycin) in a humidified atmosphere of 5% CO2 at 37◦C. Cells were treated with 2 or 0.2 µM DOX, unless indicated otherwise.

#### Animal Experiments

Mice (C57BL/6, male, 7 weeks old) were injected with DOX or saline. Briefly, mice were treated twice per week with 10 mg/kg DOX or control saline solution for 1 week (20 mg/kg cumulative dose of DOX). Heart tissues were analyzed 1 week after the last treatment. DOX doses were based on previous reports (Zhu et al., 2009). Cardiac function and ventricular remodeling were investigated 1 week after the last treatment. All procedures involving animals were reviewed and approved by the Institutional Animal Care and Use Committee of Qingdao University Medical College.

### Quantitative Real-Time PCR (qRT-PCR)

Stem-loop qRT-PCR was performed in an Applied Biosystems ABI Prism 7000 sequence detection system. Total RNA was extracted using TRIzol reagent. DNase I (Takara, Otsu, Japan) was applied, and RNA was reverse-transcribed using a reverse transcriptase kit (Takara). Mature miR-499-5p levels were measured using SYBR Green Real-time PCR Master Mix (Takara) according to the manufacturer's instructions. The same reverse primer with the sequence 5<sup>0</sup> -GTGCAGGGTCCGAGGT-3<sup>0</sup> was used for all miRNAs. The primer used is as described in **Table 1**.

### Mitochondrial Staining

fgene-09-00734 January 17, 2019 Time: 18:39 # 3

We performed mitochondrial staining as described previously (Boman et al., 2000; Wang et al., 2015b). Briefly, we plated cells onto coverslips. Cells were stained after treatment for 20 min with 0.02 µM MitoTracker Red CMXRos (40743ES50; Yeasern, Shanghai, China). We imaged mitochondria using a laser scanning confocal microscope (Zeiss LSM510 META, Jena, Germany). We randomly measured at least 150 cells from each experiment to determine the percentage of cells undergoing mitochondrial fission.

### Immunoblotting

Immunoblotting was performed to determine the expression levels of p21 and actin. Briefly, cells were lysed for 1 h at 4◦C in lysis buffer (20 mM Tris pH 7.5, 2 mM EDTA, 3 mM EGTA, 2 mM DTT, 250 mM sucrose, 0.1 mM phenylmethylsulfonyl fluoride, and 1% Triton X-100) containing a protease inhibitor


TABLE 2 | The information of the antibodies used in his work.


cocktail. Protein samples were subjected to 12% SDS-PAGE and transferred to nitrocellulose membranes. Blots were probed using corresponding primary antibodies. Horseradish peroxidaseconjugated secondary antibodies were used. Antigen-antibody complexes were visualized using enhanced chemiluminescence. Equal protein loading was controlled using Ponceau Red staining of membranes. Antigen-antibody complexes were visualized using enhanced chemiluminescence. Bands were quantitated using Image J and α-actin was used as the loading control. Fold change was normalized to the indicated control. Information on the antibodies used is described in **Table 2**.

### Cell Death Assay

Cell death was determined using a trypan blue exclusion assay. Trypan blue-positive and trypan blue-negative cells were counted using a hemocytometer (Liu et al., 2009). We randomly measured 150 cells from each experiment to calculate the cell death rate.

### Apoptosis Assays and Histology

Apoptosis was determined using terminal deoxyribonucleotidyl transferase-mediated TdT-mediated dUTP nick-end labeling using a kit from Yeasern (Alexa Fluor 488). Harvested hearts were fixed in 4% paraformaldehyde, embedded in paraffin and sectioned at a 6-µm thickness. Detection procedures were performed in accordance with the kit instructions (He et al., 2017). We randomly measured 150 cells from each experiment to calculate the apoptotic rate.

### Echocardiographic Assessment

Echocardiography was performed as described previously. Generally, mice were mildly anesthetized, and transthoracic echocardiography was performed using a Vevo 2100 highresolution system (VisualSonics, Toronto, ON, Canada). Two-dimensional guided M-mode tracings were recorded in parasternal long and short axis views at the level of the papillary muscles. Systolic left ventricular internal diameter (LVIDs) and diastolic left ventricular internal diameter (LVIDd) were measured. We calculated the fractional shortening (FS) of the left ventricular diameter as (LVIDd – LVIDs)/LVIDd] × 100. All measurements were obtained for greater than three beats and averaged. Mice were euthanized after in vivo evaluations of cardiac function, and hearts were harvested and weighted prior to histological examination (Coppola et al., 2016).

### Electron Microscopy

Heart ultrastructural analysis was performed to quantify mitochondrial fission. Sample preparations and conventional electron microscopy were performed as described (Cadete et al., 2016). Samples were examined at a magnification of 15,000 using a JEOL JEM-1230 transmission electron microscope. Electron microscopy micrographs of thin sections were evaluated for comparisons of mitochondrial fission. The sizes of individual mitochondria were measured using Image-Pro Plus software. We defined mitochondria smaller than 0.6 µm<sup>2</sup> as fission mitochondria (Wang et al., 2015d).

FIGURE 1 | miR-499-5p attenuates mitochondrial fission and apoptosis in cardiomyocytes treated with DOX. (A) Measurement of miR-499-5p mRNA levels in H9c2 cells treated with 2 µM DOX at the indicated times. (B) Transfection with a miR-499-5p mimic forced the expression of miR-499-5p in cardiomyocytes. Cells were transfected with miR-499-5p mimic (miR-mimic) or negative control (NC) for 24 h. The expression levels of miR-499-5p were detected using real-time PCR. (C) miR-499-5p antagomiR knocked down endogenous miR-499-5p. Cells were transfected with a miR-499-5p antagomiR (Anti-miR-499) or antagomiR control for 24 h. The expression levels of miR-499-5p were detected using real-time PCR. (D) Overexpression of miR-499-5p inhibited 2 µM DOX-induced apoptosis in cardiomyocytes. Cardiomyocytes were transfected with miR-499-5p mimic or negative control (NC) for 24 h and treated with 2 µM DOX for 24 h. Cell death was detected using the Tunel assay. Representative images show Tunel staining results (blue, DAPI; green, Tunel). Scale bar: 100 µm. (E) Statistical analysis of Tunel-positive cells in each group. Data are expressed as the means ± SD, n = 3 experiment <sup>∗</sup>P < 0.05. (F) Knockdown of miR-499-5p sensitized

#### FIGURE 1 | Continued

fgene-09-00734 January 17, 2019 Time: 18:39 # 5

cardiomyocytes to undergo apoptosis induced by DOX (0.2 µM). Cell death was detected using the Tunel assay. Data are expressed as the means ± SD; n = 3 experiment. <sup>∗</sup>P < 0.05. (G) Overexpression of miR-499-5p inhibited DOX-induced mitochondrial fission in cardiomyocytes. Cells were transfected with a miR-499-5p mimic or negative control for 24 h. Representative images were captured using confocal microcopy. Scale bar: 2 µM. (H) Knockdown of miR-499-5p sensitized cardiomyocytes to undergo 0.2 µM DOX-induced cell death. Cell death was detected using the trypan blue assay. Data are expressed as the means ± SD, n = 3 experiment. <sup>∗</sup>P < 0.05. (I) Overexpression of miR-499-5p inhibited 2 µM DOX-induced cell death in cardiomyocytes. Cell death was measured using trypan blue assays. Data are expressed as the means ± SD, n = 3 experiment. <sup>∗</sup>P < 0.05. (J) Statistical analysis of the percentage of cells undergoing mitochondrial fission. Data are expressed as the means ± SD, n = 3 experiment. <sup>∗</sup>P < 0.05. Cardiomyocytes were transfected with a miR-499-5p mimic or negative control for 24 h and treated with 2 µM DOX for 24 h. (K) Statistical analysis of the percentage of cells undergoing mitochondrial fission. Data are expressed as the means ± SD, n = 3 experiment. <sup>∗</sup>P < 0.05. H9c2 cells were transfected with a miR-499-5p antagomiR or antagomiR control for 24 h and treated with 0.2 µM DOX for 24 h. The percentage of cells undergoing mitochondrial fission was determined.

#### Reporter Construction and Luciferase Assay

The p21 30UTR was amplified from mouse genomic DNA using PCR. The primers were as described in **Table 1**. PCR products were gel-purified and ligated into a pGL3 reporter vector (Promega) immediately downstream of the stop codon of the luciferase gene. Mutations of the p21 30UTR construct were introduced using a QuikChange II XL site-directed mutagenesis kit (Stratagene). The p21 30UTR-Mut (the wild-type p21 3 <sup>0</sup>UTR site: AGUCUUAA, p21 30UTR-Mut: AGACGGAA) was produced using a QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA, United States). A luciferase activity assay was performed as described previously (Wang et al., 2015c). Briefly, cells were cultured in 24-well plates, infected with miR-499-5p mimic or negative control and transfected with the plasmid construct pGL3-p21-30UTR or pGL3-p21-30UTR-Mut at a concentration of 200 ng/well using Lipofectamine 3000 (Invitrogen). The Renilla luciferase plasmid was cotransfected at 2.5 ng/well and served as the internal control. Cells were lysed 48 h after transfections, and luciferase activity was detected using a Dual Luciferase Reporter Assay kit (Promega). All experiments were performed in triplicate.

### Construction of Adenovirus and Overexpression Vector

miR-499-5p-overexpressing adenovirus and adenovirus β-galactosidase (β-gal) were prepared as described previously (Wang et al., 2011). All adenoviruses were amplified in HEK-293 cells. Adenoviral infection of cells was performed as described previously (Wang et al., 2009). The open reading frame (ORF) of the p21 gene was generated using RT-PCR, and p21 siRNA was purchased from Genepharma (Shanghai, China). P21 was cloned into the pcDNA3.1 expression vector according to the manufacturer's guidelines (Invitrogen). The constructed sequence was further confirmed using sequencing.

#### Data and Statistical Analysis

All values are expressed as the means ± standard error. n = 3. Statistical significance was defined as p < 0.05. One- or two-way analysis of variance (ANOVA) was used to test each variable for differences between treatment groups. If ANOVA demonstrated a significant effect, then pairwise post hoc comparisons were performed using Fisher's least significant difference test.

#### RESULTS

#### miR-499-5p Attenuates Mitochondrial Fission and Apoptosis in Cardiomyocytes Treated With DOX

miR-499-5p exerts a protective role in the pathogenesis of heart diseases and miR-499-5p mRNA levels are downregulated in cardiomyocytes during apoptotic stress and in the heart under pathological conditions (Matkovich et al., 2012). We detected miR-499-5p expression levels in cardiomyocytes exposed to DOX to investigate the role of miR-499-5p in DOX-induced cardiotoxicity. miR-499-5p expression was significantly downregulated after DOX (2 µM) treatment (**Figure 1A**). Cardiomyocytes were transfected with a miR-499-5p mimic. Real-time PCR demonstrated that miR-499-5p levels increased 4-fold compared to the negative control (**Figure 1B**). The miR-499-5p mimic efficiently inhibited mitochondrial fission (**Figures 1G,J**) and cell apoptosis (**Figures 1D,E,I**) in cardiomyocytes exposed to DOX (2 µM). We knocked down endogenous miR-499-5p using a miR-499-5p antagomiR to mimic the DOX-induced downregulation of miR-499-5p. Real-time PCR demonstrated that the miR-499-5p antagomiR efficiently knocked down the endogenous miR-499-5p (**Figure 1C**). Knockdown of miR-499-5p sensitized cardiomyocytes to DOX, which induced mitochondrial fission and apoptosis in cardiomyocytes at a lower concentration (0.2 µM) (**Figures 1F,H,K**). These results suggest the involvement of miR-499-5p in DOX cardiotoxicity and an attenuation of DOX-induced apoptosis in cardiomyocytes via inhibition of mitochondrial fission by miR-499-5p.

#### miR-499-5p Attenuates Mitochondrial Fission and Apoptosis in DOX Cardiotoxicity in vivo

We investigated the role of miR-499-5p in mice model to further validate its function in DOX-induced cardiotoxicity. miR-499-5p levels were significantly downregulated in DOX-treated mice heart (**Figure 2A**) while the serum miR-499-5p expression was significantly increased (**Figure 2B**). DOX administration induced abnormal mitochondrial fission and cell apoptosis in the mouse heart (**Figures 2C–F**). The adenovirus harboring miR-499-5p efficiently forced the expression of miR-499-5p in mouse heart (**Figure 2A**) and significantly attenuated cell mitochondrial fission and cell apoptosis in the heart following

DOX administration (**Figures 2C–F**). Notably, we observed a significant increase in the protein levels of p21 in DOX-treated mice, which was inhibited in the miR-499-5p-overexpressing mice (**Figure 2G**). In conclusion, miR-499-5p attenuated DOXinduced mitochondrial fission and cell apoptosis in the mouse heart in vivo.

### miR-499-5p Attenuates DOX Cardiotoxicity in Mice

fgene-09-00734 January 17, 2019 Time: 18:39 # 7

Doxorubicin treatment caused a high degree of cardiomyocytes apoptosis. Because cardiomyocytes exhibit a limited ability to regenerate, cardiomyocytes loss during treatment cannot be replenished. Cardiomyocytes undergo compensatory hypertrophy, and the heart develops dilated cardiac hypertrophy with a decrease in heart function under the treatment of DOX. ANP and β-MHC are two biomarkers for the cardiac hypertrophy (Wang et al., 2015b). We measured the expression levels of ANP and β-MHC to examine the role of miR-499-5p in DOX-induced cardiac hypertrophy. ANP and β-MHC expression levels were significantly upregulated in mice administered DOX, but these increases were significantly inhibited in the miR-499-5p-overexpressing mice (**Figures 3A,B**). These results indicate that miR-499-5p inhibits the cardiac hypertrophy induced by DOX cardiotoxicity. We also examined the effect of miR-499-5p on cardiac remodeling in mice. Our results demonstrated that the cardiac remodeling was ameliorated in miR-499-5p-overexpressing mice as assessed using echocardiography (**Figures 3C–E**). In conclusion, miR-499- 5p efficiently attenuated DOX cardiotoxicity in our mouse model.

#### p21 Is a Target of miR-499-5p

miRNAs primarily target the 30UTR region of target genes and negatively regulate target gene expression via the inhibition of translation or promotion of mRNA degradation (Bartel, 2009). We analyzed the potential targets of miR-499-5p using the bioinformatic program TargetScan to elucidate the molecular mechanisms by which it regulates cell apoptosis. miR-499-5p was predicted to bind to the 30UTR region of p21, which is conserved in rats, mice and humans (**Figure 4A**). We also observed that p21 protein levels increased in cardiomyocytes treated with DOX in a time-dependent manner while miR-499-5p expression decreased, which suggests that p21 may be a target of miR-499-5p (**Figures 1A**, **4C**). We forced the expression of miR-499-5p in cardiomyocytes to further analyze the regulatory role of miR-499-5p on p21. Overexpression of miR-499-5p significantly attenuated p21 expression in cardiomyocytes (**Figure 4B**). We used a luciferase assay system to examine whether miR-499-5p influenced p21 expression via a direct targeting of p21 30UTR. We cloned p21 30UTR containing the miR-499-5p binding site downstream of the luciferase reporter gene, p21 30UTR-Wt, to examine luciferase activity driven by the 30UTR of p21. We also generated a mutated luciferase construct, p21 30UTR-Mut, with mutations introduced into the miR-499-5p-binding site of p21 30UTR (**Figure 4D**). Both the wild-type 30UTR of p21 (p21 30UTR-Wt) and the mutated 30UTR of the p21 (p21 30UTR-Mut) showed a similar luciferase activity with the luciferase reporter gene (pGL3) (**Figure 4E**). However, the wild-type 30UTR of p21 exhibited low luciferase activity in the presence of miR-499-5p while the mutated 30UTR did not produce a significant response to miR-499-5p (**Figure 4E**). A luciferase activity assay in cardiomyocytes also demonstrated that DOX (2 µM) treatment increased the luciferase activity regulated by wildtype p21 30UTR. The mutation of miR-499-5p-binding site slowed the increase in these luciferase activity (**Figure 4F**). Taken together, these results suggest that p21 is a specific target of miR-499-5p.

### p21 Attenuates Mitochondrial Fission and Apoptosis in Cardiomyocytes Treated With DOX

Next, we investigated the potential role of p21 in DOXinduced cardiotoxicity. We knocked down endogenous p21 using siRNA (**Figure 5A**). Knockdown of endogenous p21 significantly decreased mitochondrial fission and cell death in cardiomyocytes exposed to DOX (2 µM) (**Figures 5C,D**). We also overexpressed p21 using the pcDNA3.1 eukaryotic expression vector with a CMV promoter (**Figure 5B**). p21 overexpression induced massive mitochondrial fission and sensitized cardiomyocytes to undergo apoptosis under a lower concentration of DOX (0.2 µM) (**Figures 5E,F**). Taken together, these results suggest that p21 promotes mitochondrial fission program and cell apoptosis in cardiomyocytes exposed to DOX.

### miR-499-5p Attenuates Mitochondrial Fission and Apoptosis via Targeting p21

We examined whether miR-499-5p attenuated mitochondrial fission and cell apoptosis via the targeting of p21. Overexpression of miR-499-5p inhibited DOX-induced mitochondrial fission and cell death while forced expression of p21 attenuated this inhibitory effect on mitochondrial fission (**Figure 6A**) and cell death (**Figure 6B**). Knockdown of the endogenous miR-499- 5p sensitized cardiomyocytes to undergo mitochondrial fission and cell death at a lower dose of DOX (0.2 µM) while simultaneously knockdown of p21 inhibited these mitochondrial fission (**Figure 6C**) and cell death (**Figure 6D**). These data suggest that miR-499-5p attenuates DOX-induced mitochondrial fission and apoptosis in cardiomyocytes via the targeting of p21.

#### miR-499-5p Is Not Involved in DOX-Induced Apoptosis in Cancer Cells

Our results showed that miR-499-5p is a promising factor for preventing DOX cardiotoxicity in cancer therapy, but it is important to exclude its role in the apoptosis of tumor cells. We compared differences in miR-499-5p expression levels between cardiomyocytes and tumor cells. The results revealed that miR-499-5p was expressed at very low levels in tumor cells, including SGC-7901, A-549, SW-480 and HepG-2s, compared to cardiomyocytes (**Figure 7A**). We examined miR-499-5p expression levels during cell apoptosis in tumor cells treated with DOX. The results demonstrated that miR-499-5p expression was insensitive to DOX treatment in tumor cells (**Figures 7B,C**). Notably, miR-499-5p overexpression did not affect DOX-induced cell death in tumor cells (**Figures 7D,E**). These results indicate that miR-499-5p is not involved in the process of DOX-induced tumor cell apoptosis.

### DISCUSSION

Doxorubicin is a powerful drug in the clinical fight against cancer. However, its cardiotoxicity is a major challenge and limits its application. A total of 20% of patients who receive DOX treatment develop heart failure, and DOX accumulation reaches 500 mg/m<sup>2</sup> in these patients' heart (Chatterjee et al., 2010). An understanding of the molecular mechanisms will ameliorate the cardiotoxicity and increase the clinical efficacy of DOX. miRNAs play important roles in DOX-induced cardiotoxicity. miR-146a

the means ± SD, n = 3 except in panel; <sup>∗</sup>P < 0.05.

is involved in DOX-induced cardiomyocyte cell death via the targeting of ErbB4 (Horie et al., 2010). miR-30 targets GATA to participate in DOX-induced cardiomyocyte apoptosis (Roca-Alonso et al., 2015). MicroRNA-208 also contributes to DOXinduced cardiomyocyte cell death via the targeting of GATA4 (Tony et al., 2015). Our previous work found that miR-532-3p was involved in DOX-induced cardiotoxicity via the repression of ARC expression (Wang et al., 2015b). miR-140-5p directly targets Nrf2 and Sirt2 and increases DOX-induced oxidative damage via altering FOXO3a expression levels (Zhao et al., 2018). We reported here that miR-499-5p efficiently prevented DOX cardiotoxicity via attenuating mitochondrial fission and cell apoptosis and targeting of p21 (**Figure 6E**).

Two other types of non-coding RNAs, lncRNAs and circRNAs, garnered great attention recently, and these non-coding RNAs are also involved in DOX cardiotoxicity and myocardial damage. lincRNA-p21 participates in DOX-related cardiac cell senescence via regulation of the Wnt/beta-catenin signaling pathway, and silencing of lincRNA-p21 effectively protects against DOX cardiotoxicity (Xie et al., 2018). Qki (Quaking), which is a RNA binding protein, inhibits DOX cardiotoxicity via regulation of cardiac circular RNAs (Gupta et al., 2018). Knockdown

of MiRt1 improves cardiac function, reduces apoptosis of cardiomyocytes and attenuates inflammatory cell infiltration in vivo (Li et al., 2017). lncRNA-cardiac autophagy inhibitory factor (CAIF) inhibits autophagy and attenuates myocardial infarction via the targeting of p53-mediated transcription of myocardin (Liu C.Y. et al., 2018). Microarray analysis found a total of 63 differentially expressed circRNAs, including 29 upregulated and 34 downregulated circRNAs, during myocardial infarction (Wu et al., 2016). Mitochondrial fission and apoptosisrelated circRNA (MFACR) regulates mitochondrial fission and apoptosis in the heart via direct targeting of miR-652-3p (Li et al., 2018). In conclusion, lncRNAs and circRNAs are potential targets for the prevention of DOX cardiotoxicity. Further study is needed to elucidate the functional role of these two non-coding RNAs.

The role of miR-499-5p in certain types of tumors is controversial. Several studies demonstrated that miR-499-5p suppressed cancer development and progression. Overexpression of miR-499-5p downregulated ETS1 expression, and it inhibited the migration and infiltration of HepG-2 cells (Wei et al., 2012). miR-499-5p could also inhibit the growth of tumor cells and improves the effectiveness of cancer treatment (Ando et al., 2014). miR-499-5p may also act as a tumor suppressor gene via the targeting of VAV3 (Li M. et al., 2016). However, other studies revealed that miR-499-5p was abnormally expressed in non-small cell lung cancer tissues, and it may be used as a biomarker for early diagnosis and evaluation of patient prognosis (Li et al., 2014). Our study found that miR-499-5p produced no effect on tumor cells, including gastric cancer, non-small cell lung cancer, colon cancer or liver cancer, which suggests that it increases the therapeutic potential of the tumor and protects the heart without promoting tumor development. Therefore, miR-499-5p may be an effective factor in the prevention of DOX cardiotoxicity.

Single nucleotide polymorphisms (SNPs) are the most frequent type of variation in the genome. Numerous study demonstrated that functional SNPs in miRNA genes affect different signaling pathways via altering the expression levels

of target genes (Zhi et al., 2012; Chen et al., 2014; Liu et al., 2017). Increasing evidence suggests that miRNA polymorphisms are associated with the susceptibility to heart diseases, such as myocardial infarction and coronary heart disease. Previous studies reported that miR-196a2 rs11614913 (Xu et al., 2009; Zhi et al., 2012), miR-146a rs2910164 (Chen et al., 2014), and miR-149 rs71428439 (Ding et al., 2013) were associated with the risk of heart disease. miR-499-5p is a heart-rich miRNA under physiological conditions, and some studies demonstrated that miR-499-5p levels were downregulated under pathological conditions (van Rooij et al., 2009; Wang et al., 2011). The plasma level of miR-499-5p may help distinguish acute myocardial

infarction and heart disease in patients (Olivieri et al., 2013). Our previous study found that the SNP rs3746444 of the miR-499 precursor affected the expression level and anti-apoptotic function of miR-499-5p (Ding et al., 2018). Whether SNPs in miRNA-499-5p are related to the susceptibility to cardiotoxicity need to be further investigated, which is of great significance for guiding medication and overcoming drug cardiotoxicity.

Well-balanced mitochondrial dynamics play an important role in cell life (Archer, 2014). Abnormal mitochondrial fission leads to the onset of cardiomyocyte apoptosis and the development of heart failure (Lee et al., 2004; Ong and Hausenloy, 2010). DOX-induced mitochondrial dysfunction is currently the major cause of cardiotoxicity (Ichikawa et al., 2014), and it was also documented in heart failure patients (Lebrecht et al., 2005). Our present study revealed the pivotal role of p21 in the promotion of mitochondrial fission caused by DOX treatment. Drp1-mediated mitochondrial fission is an important component of cell apoptosis (Martinou and Youle, 2006), and p21 participates in the Drp1-mediated mitochondrial fission that promotes HCC cell proliferation (Zhan et al., 2016). p21 also participated in the p53-mediated mitochondrial apoptosis program in nickel (II) induced nasal epithelial cytotoxicity (Lee et al., 2016). However, whether p21 regulates DOX-induced mitochondrial fission via Drp1 or p53 need to be further examined.

Taken together, we report that the downregulation of miR-499-5p in cardiomyocytes exposed to DOX is involved in DOX cardiotoxicity. There is emerging evidence for the involvement of p21 in the promotion of mitochondrial fission and cell apoptosis in cardiomyocyte exposed to DOX. miR-499-5p prevented mitochondrial fission and cell apoptosis in cardiomyocytes exposed to DOX via the targeting of p21 (**Figure 6E**). Therefore, the development of new therapeutic strategies based on the miR-499-5p-p21 axis is promising for the overcoming of DOX cardiotoxicity in cancer treatment.

#### ETHICS STATEMENT

fgene-09-00734 January 17, 2019 Time: 18:39 # 13

This study was carried out in accordance with the recommendations of Institutional Animal Care and Use Committee of Qingdao University Medical College. The protocol was approved by Institutional Animal Care and Use Committee of Qingdao University Medical College.

#### AUTHOR CONTRIBUTIONS

JW and TX designed the research. QW and WD performed the cellular experiments. XZ and TX performed

#### REFERENCES


the animal experiments. TY and XJ constructed the reporter construct and adenoviruses. WY and ZL analyzed the cardiac function. QW and JW wrote the manuscript. All authors approved the final version of the manuscript.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (81622005 and 81770232 to JW) and the Natural Science Foundation of Shandong Province (JQ201815 to JW).

functional maturation of human embryonic stem cell-derived cardiomyocytes. PLoS One 6:e27417. doi: 10.1371/journal.pone.0027417


mitochondrial damage. Environ. Toxicol. Pharmacol. 42, 76–84. doi: 10.1016/j.etap.2016.01.005


fgene-09-00734 January 17, 2019 Time: 18:39 # 14


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

Copyright © 2019 Wan, Xu, Ding, Zhang, Ji, Yu, Yu, Lin 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) 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.

# Non-Coding RNAs in Retinoblastoma

*Meropi Plousiou and Ivan Vannini\**

Biosciences Laboratory, Istituto Scientifico Romagnolo per lo Studio e la Cura dei Tumori (IRST) IRCCS, Meldola, Italy

Retinoblastoma (Rb) is the most common ocular pediatric malignancy that arises from the retina and is caused by a mutation of the two alleles of the tumor suppressor gene, RB1. Although early detection provides the opportunity of controlling the primary tumor with effective therapies, metastatic activity is fatal. Non-coding RNAs (ncRNAs) have emerged as important modifiers of a plethora of biological mechanisms including those involved in cancer. They are classified into short and long ncRNAs according to their length. Deregulation of all these molecules has also been shown to play a critical role in Rb pathogenesis and progression. It is believed that ncRNAs can provide new insights into novel regulatory mechanisms associated with clinical pathological characteristics, facilitating the development of therapeutic alternatives for the treatment of Rb. In this review, we describe a variety of ncRNAs, which capable of regulating the most likely candidate genes involved in the tumorigenesis of Rb, could prove useful in analyzing different aspects of this cancer.

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Feng Wang, Emory University School of Medicine, United States Alessio Naccarati, Italian Institute for Genomic Medicine, Italy

> \*Correspondence: Ivan Vannini ivan.vannini@irst.emr.it

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 05 October 2018 Accepted: 22 October 2019 Published: 14 November 2019

#### Citation:

Plousiou M and Vannini I (2019) Non-Coding RNAs in Retinoblastoma. Front. Genet. 10:1155. doi: 10.3389/fgene.2019.01155 Keywords: cancer, microRNAs, long ncRNAs, retinoblastoma, RB1

### INTRODUCTION

Retinoblastoma (Rb) is the most common primary ocular paediatric malignancy and the first to be identified as having hereditary features. Inactivation of the *RB1* gene is thought to be the main reason for the development of the tumor (Ma et al., 2014). The impact of Rb is substantial, with an incidence of nearly one in 15,000–20,000 births, resulting in 9,000 new cases every year (Kivelä, 2009; Thériault et al., 2014). Although it is possible to control the primary tumor with effective therapies, metastatic disease is still fatal. The survival rate is lower in under-developed countries because the disease is often diagnosed at later stages, and the mortality rate among children is 50%– 70% (Jabbour et al., 2012).

Rb can be either non-hereditary or hereditary and can affect only one eye (unilateral) or both (bilateral). In particular, the bilateral forms of heritable Rb show a germline mutation of the *RB1* gene which is followed by a second somatic inactivation of the other allele. Conversely, somatic inactivation of both *RB1* alleles in retinal cells is responsible for non-heritable cases (Thériault et al., 2014). The *RB1* gene is considered to be a central regulator of the cell cycle mechanism and is inactivated in a wide range of cancers. Its tumor suppressor function is mainly known due to the inhibition of the E2F1 transcription factor, which in the absence of *RB1* pushes cells from G1 to S phase of the cell cycle. However, *RB1* appears to function in various ways, controlling more than four types of protein interactions and taking on the role of transcriptional co-factor and adaptor protein. For example, it represses *E2F1* gene target transcription through the recruitment of chromatin remodelling complexes including HDAC, DNMT1, HP1A, and SUV39H1 to promoters. With regard to non-E2F1 transcriptional factors such as MYOD, HIF1a, and SP1, *RB1* appears to act as a transcriptional co-factor (Burkhart and Sage, 2008).

Another important gene involved in Rb progression is the enhancer of zeste 2 polycomb repressive complex 2 subunit (*EZH2*), which was the first overexpressed epigenetic enzyme identified in Rb samples. *EZH2* is a histone methyltransferase expressed only in Rb and is essential for tumor development because of its capacity to silence tumor suppressors such as p19/ p14ARF and p16/INK4a. Targeting EZH2 could be the basis for developing an epigenetic therapeutic approach in ocular oncology (Khan et al., 2015). However, new molecules that are involved in Rb are studied to improve the therapy.

Non-coding RNAs (ncRNAs) are transcripts that do not encode proteins. They are diffused in the human genome and abnormally dysregulated in cancer cells. Given that ncRNAs are often located in fragile sites (FRA), common breakpoint sites and in regions with loss of heterozygosity, they represent a new category of genes that participate in tumorigenesis (Calin et al., 2004; Calin et al., 2007). Some ncRNAs have an oncogenic function while others act as tumor suppressors (Sanchez Calle et al., 2018; Vannini et al., 2018a). ncRNAs are classified into two groups on the basis of the length of their sequence: short ncRNAs have a maximum length of 200 nucleotides and long ncRNAs (lncRNAs) are transcripts with sequences of over 200 nucleotides.

microRNAs (miRNAs), the most widely studied class of ncRNAs, are small molecules containing around 22 nucleotides. They regulate the expression of more than 60% of genes. miRNAs are included in the RNA-induced silencing complex (RISC), an miRNA effector machine. This complex binds the 3' untranslated region or, less frequently, the 5' untranslated region of mRNA target, determining the protein downregulation by mRNA degradation or translational repression. miRNAs also upregulate gene expression (Vasudevan et al., 2007). They are located in intergenic regions with independent promoters (Ambros et al., 2003) but can be transcribed by introns with the same promoter as that of the host gene (Lin et al., 2006).

lncRNAs are usually transcripts with 5' terminal methylguanosine cap, frequently polyadenylated, and alternatively spliced (Spizzo et al., 2012; Ulitsky and Bartel, 2013). They have a thermodynamically stable secondary structure with bulges and hairpin loops (Mercer and Mattick, 2013) that enables them to interact with proteins, mRNAs, ncRNAs, and DNA. lncRNAs regulate gene expression at varied levels, from mRNA translation to cytoplasmatic and nuclear epigenetic processes such as miRNA sponging (Vannini et al., 2018b).

The present review summarizes the most important dysregulated ncRNAs in Rb, the interaction with some of their target molecules, and the mechanisms involved in tumor progression.

#### miRNAs AS DIAGNOSTIC AND PROGNOSTIC BIOMARKERS IN Rb

The formation and progression of numerous cancer types is frequently correlated with an altered miRNA expression profile.

A comparison by Beta et al. between miRNA profiles in primary Rb tissues and miRNAs detected in the serum of

children with Rb revealed eight downregulated miRNAs (miR-216a, miR-217, let-7a, let-7i, let-7f, miR-9, miR-92a, miR-99b) and 25 upregulated (miR-103, miR-142-5b, miR-106b, miR-143, miR-148b, miR-17, miR-16, miR-183, miR-182, miR-19a, miR-18a, miR-29a, miR-29b, miR-29c, miR-20a, miR-30b, miR-30d, miR-34a, miR-494, miR-378, miR-513, miR-513-1, miR-513-2, miR-518c, miR-96) miRNAs. It would thus seem that these 33 RNA molecules are Rb-specific and could potentially influence tumorigenesis and tumor progression in the disease (Beta et al., 2013). Another study identified a group of 24 differentially expressed miRNAs (nine upregulated and 15 downregulated) in healthy retinal tissues and Rb tissues. Among these, 14 miRNAs including miR-20a, miR-373, miR-125b, let7a, let-7b, let-7c, miR-25, and miR-18a proved capable of distinguishing between Rb samples and healthy tissues, thus identifying potential biomarkers of Rb (Yang and Mei, 2015). Likewise, Liu et al. demonstrated that miR-320, let-7e, and miR-21 were disregulated in plasma of Rb patients and can thus be hypothesized as novel diagnostic biomarkers for the disease (Liu et al., 2014).

Despite the tumor heterogeneity of Rb, Castro-Magdonel et al. identified a common miRNA expression profile, highlighting miR-3613 as an interesting candidate for therapy. It was highly expressed in all of the examined samples and was also observed to have more than 36 tumor suppressor gene targets (Castro-Magdonel et al., 2017). The microenvironment was recently identified as one of the main factors influencing the background of many types of cancer. Hypoxia is considered as one of the first conditions of stress present in the tumor microenvironment. Interestingly, studies have shown that a hypoxic tumor microenvironment plays a crucial role in controlling treatment outcomes in Rb. It has, in fact, been associated with treatment failure given that it is capable of regulating various pathways including growth factor signalling, glycolysis, genetic instability, metastasis, and angiogenesis. Sudhakar et al. studied the expression of hypoxia-related proteins such as HIF-1A and survivin to understand whether hypoxia is present in Rb, observing that increased expression of these proteins induces resistance to cytotoxic therapy (Sudhakar et al., 2013). Various studies have shown that hypoxic conditions can modulate the expression of a group of miRNAs called hypoxiaregulated microRNAs (HRMs). More precisely, microarray analysis identified miR181b, miR30c-2, miR125a3p, miR497, and miR491-3p as the most important HRMs in Rb cells (Xu et al., 2011). By *in silico* and *in vitro* approaches, Venkatesan et al. identified two key miRNAs (miR486-3p, miR-532) that are downregulated in Rb. Their overexpression using mimic miRNA strategy on Rb cells led to apoptotic cell death (Venkatesan et al., 2015).

### miRNA PATHWAYS IN Rb

Although the critical role played by miRNAs in cancer has been demonstrated, further research is needed to clarify the link between cancer-related miRNAs and their target genes and to identify their correlation with multiple pathways associated with tumorigenesis. Through *in silico* and *in vitro* analysis of different cancer types it has been possible to identify and validate miRNAs that directly regulate RB1 gene (**Table 1**).

Lyu et al. observed low expression levels of miR-485 in Rb tissue and cell lines through reverse transcription-quantitative polymerase chain reaction (RT-qPCR). They also confirmed that miR-485 has a tumor suppressor function targeting Wnt family member 3A (Wnt3A) which activates the canonical Wnt signaling pathway, leading to decreased Rb proliferation, invasion, and migration (Lyu et al., 2019b).

Zhao et al. analyzed miR-361-3p expression levels by qRT-PCR in serum and tissue of Rb patients, in serum samples and normal retinal tissue from healthy controls, and in human Rb cell lines. The authors demonstrated that miR-361-3p was downregulated in Rb serum, Rb tissue and Rb cell lines compared with normal serum and normal retinal tissue. They also observed that miR-361-3p decreased Rb cell proliferation *via* targeting of GLI family zinc finger 1 and 3 (GLI 1/3) (Zhao and Cui, 2018).

Various studies have demonstrated that miR-183 is dysregulated in a great number of cancer types (Li et al., 2010; Lowery et al., 2010; Zhao et al., 2012). Interestingly, Wang et al. observed that miR-183 was downregulated in Rb cell lines and tissues with respect to healthy retinal tissues and that its forced overexpression inhibited the migration, proliferation and invasion capacity of Rb cells. Specifically, the mechanism underlying these results included lowdensity lipoprotein receptor-related protein (*LRP6*), a candidate target gene of miR-183 known to counteract the apoptotic effects of this miRNA. The authors showed that miR-183 directly targeted LRP6, downregulating its expression, and thus controlling the progression and development of Rb (Wang et al., 2014).

Another miRNA that has been shown to be involved in various cancer types, including Rb, is miR-204. Wu et al. reported that miR-204 is downregulated in Rb tissues and cell lines. The authors identified *cyclin D2* and matrix metalloproteinase *MMP-9* as its putative gene targets and focused on elucidating the mechanisms underlying this interaction (Wu et al., 2015).

MMP-9 is a protease that plays a strategic role in extracellular matrix (ECM) remodelling under normal conditions but also in the degradation of ECM by cancer cells during the metastasis process. Cyclin D2 is involved in the phosphorylation of RB1 and numerous studies have demonstrated that its expression levels are high in various cancers. It has emerged that cyclin D2, MMP-9, and miR-204 expression are inversely correlated and that miR-204 targets cyclin D2 and MMP-9, inhibiting tumor growth in Rb (Wu et al., 2015; Golabchi et al., 2017).

miR-29a expression is inversely correlated with cyclin D1 and matrix metalloproteinase (MMP-2) being part of signal transducer and an activator of transcription 3 (*STAT3*) downstream genes. Liu et al. demonstrated that miR-29a expression is very low in Rb (cells and tissues) and that STAT3 is a direct target of this miRNA (Liu et al., 2018).

Wang et al. discovered that miR-504 was decreased in Rb cell lines and tissue, reporting that miR-504 overexpression suppressed Rb cell proliferation and invasion. This suggests that miR-504 explicates its tumor suppressor function by directly targeting the astrocyte elevated gene 1 (AEG-1) known to be involved in the aggressiveness of Rb (Wang et al., 2019).

Guo et al. demonstrated that miR-98 suppressed Rb invasion, migration, and cell growth through the regulation of insulin-like growth factor1 receptor (IGF1R). IGF1R is a tyrosine kinase receptor that regulates IGF1-induced signaling events and plays an important role in cellular processes including differentiation, proliferation, and migration. The authors reported that miR-98 inhibited the k-Ras/Raf/MEK/ ERK signaling pathway *via* targeting of IGF1R in RB. In fact, the restoration of IGF1R reverted the effects of miR-98 on Rb migration, invasion, and cell viability (Guo L, et al., 2019).

Sun et al. demonstrated that miR-492 was upregulated in Rb cell lines and tissue. Downregulating this miRNA, they confirmed its oncogene function that leads to a decrease in Rb proliferation and invasion. Using bioinformatic analysis and luciferase reporter assay, the authors observed that miR-492 targeted large tumor-suppressor kinase 2 (LATS 2), a serine/threonine protein kinase (Sun et al., 2019).



Another potential therapeutic axis was proposed by Zhang et al. for miR-125a-5p/TAZ and EGFR. The authors demonstrated that miR-125a-5p was markedly downregulated in Rb and that its overexpression inhibited cell proliferation. Transcriptional co-activator with PDZ-binding motif (*TAZ*) is a novel oncogene that promotes tumor progression and is one of the elements of the Hippo tumor suppressor pathway. TAZ is overexpressed in Rb and has been shown to stimulate tumor progression *via* the *EGFR* pathway. miR-125a-5p, in targeting TAZ, suppresses EGFR pathway and consequently Rb progression (Zhang et al., 2016). Whilst miR-125a-5p is downregulated in Rb, miR-125b is reported to be upregulated in both tissue and Rb cells. It has been shown to suppress cell apoptosis and promote cancer cell proliferation *via* interaction with one of its putative target genes, DNA damage regulated autophagy modulator 2 (*DRAM2*).

DRAM2 plays a crucial role in TP53-mediated apoptosis and induces cell autophagy. Studies have confirmed that DRAM2 is downregulated in various cancer types, indicating that it may be part of the tumor signalling process. *DRAM2* gene is linked to Rb because of its involvement in the renewal and recycling mechanism of photoreceptor cells located in the retina, an essential procedure for visual function preservation.

Bai et al. showed that miR-125b exerts its oncogene biological function by directly targeting and downregulating *DRAM2* gene expression, consequently suppressing cell apoptosis (Bai et al., 2016).

The inactivation of both *TP53* and *RB1* pathways is an essential characteristic of tumorigenesis in the majority of cancer types. However, the classic *TP53* mutation does not occur in Rb. When *RB1* is inactivated, cells generally respond by activating *TP53* which induces cell cycle arrest and apoptosis. TP53 is regulated by nuclear-localized E3 ubiquitin ligase (MDM2), which targets TP53 for proteasomal degradation. In cells with a loss of *RB1*, E2F1 is activated and induces cyclin-dependent kinase inhibitor 2A (p14ARF), inhibiting MDM2 and leading to TP53 stabilization and TP53 target gene expression (Sherr, 2006). To et al. demonstrated that miR-24 represses p14ARF in Rb cells without *RB1* causing a block of activation of TP53 tumor surveillance (To et al., 2012) (**Figure 1**). This might explain the lack of mutation of the *TP53* pathway in Rb (Laurie et al., 2006; Xu et al., 2009; Ksander, 2010; Turhan, 2014).

miR-21 is reported to be one of the most dysregulated miRNAs in Rb. Shen et al. focused on this cancer-related miRNA and its potential gene targets in order to shed light on the complicated regulatory network. One of the identified gene targets is programmed cell death 4 (*PDCD4*), a well-known tumor suppressor gene that promotes cell apoptosis and inhibits cell migration and proliferation. Downregulation of PDCD4 expression has also been assessed in other cancer types including breast and lung cancer, hepatocellular carcinoma, and colorectal and squamous cell carcinoma. Shen et al. observed that miR-21 and PDCD4 were inversely correlated and that miR-21 directly targeted PDCD4 (Shen et al., 2014).

A regulator of Wnt signalling is Disheveled-Axin domain containing 1 (DIXDC1). Che et al. showed that DIXDC1 was significantly upregulated in Rb. They also observed that the expression of DIXDC1 and miR-186 was inversely correlated.

retinoblastoma (Rb) cone cell. In Rb cone cell without RB1, miR-24 blocks p14ARF activation. Consequently, MDM2 level is increased and leads to TP53 pathway block and inactivation of TP53- mediated surveillance.

miR-186 overexpression caused a downregulation of DIXDC1 and inhibited the proliferation and invasion in Rb cells (Che et al., 2018).

Wang et al. demonstrated that miR-138-5p has a tumor suppressor function, downregulating miR-138-5p and upregulating pyruvate dehydrogenase kinase 1 (PDK1) in Rb cells. PDK1 caused the phosphorylation of the pyruvate dehydrogenase enzyme. miR-138-5p overexpression led to a downregulation of PDK1 and consequently a decrease in migration, cell viability, invasion and induced apoptosis (Wang et al., 2017).

Montoya et al. elucidated the regulation of Rb proliferation by miR-31 and miR-200a, both of which were downregulated in Rb cell lines. Their induced overexpression by mimic transfection restricted the proliferative capacity of the Y-79 Rb cell line (Montoya et al., 2015).

miR-106b and its target gene Runt-related transcription factor 3 (Runx3) have also been identified as capable of modulating cancer progression. Runx3 is a tumor suppressor in Rb with an important role in mammalian development. Its loss is linked to a variety of cancers such as gastric cancer, glioblastoma, and bladder tumors. Yang et al. demonstrated that miR-106b binds to the Runx3 and downregulate its expression, which consequently leads to the development of the malignant phenotype. Inhibition of miR-106b and thus upregulation of Runx3 could represent be a possible therapy for Rb. (Yang et al., 2017).

Moreover, given the important role of the miR-17-92 cluster in the development of Rb, it is opportune to mention the advantage of the aptamer method, elucidated by Subramanian et al. An RNA aptamer directed against the primary miR-17-92 transcript was selected by the systemic evolution of ligands by exponential enrichment (SELEX) to inhibit the biogenesis of the miRNA cluster. The authors confirmed the inhibition of cluster biogenesis and a decrease in Rb cell proliferation (Subramanian et al., 2015).

However, the described investigations focused mainly on a single element (miRNA or gene). Li et al. studied genes, miRNAs, and transcription factors as elements of the regulatory networks, analyzing their relationship in Rb. The authors focused on the interactions between genes, which regulate miRNAs, and host genes including miRNAs and miRNAs targeting genes. The authors thus constructed three regulatory networks based on the relationships called the related network, differentially expressed network and global network. Data on Rb-related genes for the related network were obtained from GeneCards database and pertinent literature, whereas data on differently expressed genes in Rb for the differentially expressed network were obtained from the Cancer Genetics Web. Expressed miRNAs were extracted from literatures and mir2Disease, which is a database about differentially expressed miRNAs in various human diseases. Rb-related miRNAs were collected manually from permanent literatures.

The global network was constructed from all of the interactions that have been experimentally validated, making it too complex to be used or obtaining information. In this way, the authors were able to identify the pathways of the related elements and differentially expressed elements inserted in the other two networks (Li et al., 2014).

All of the above miRNAs are summarized in **Table 2**.

#### lncRNAs PATHWAYS IN Rb

In the same way as miRNAs, lncRNAs are responsible for various cellular functions and play an important role in many cancer types as biomarkers, tumor regulators, and predictors of prognosis. Evidence has emerged of interesting pathways


Frontiers in Genetics | www.frontiersin.org

involving lncRNAs and their putative gene targets in different tumors. Musahl et al. demonstrated that the depletion of the ncRNA-RB1, an lncRNA expressed by the RB1 promoter, reduced the expression of calreticulin (CARL). CARL is an endoplasmatic reticulum protein that, in pre-apoptosis, translocates to the cell surface and serves as a signal to phagocytic cells. As a result of ncRNA-RB1 depletion, tumor cell uptake by macrophages is inhibited (Musahl et al., 2015). E2F1-regulated inhibitor of cell death (ERIC) is also capable of controlling DNA damage response and subsequent apoptotic cell death, interacting with proliferation regulators such as E2F1. Specifically, Feldstein et al. showed that ERIC is regulated at transcriptional level by E2F1 and responds to DNA damage in osteosarcoma and lung cancer cell lines. The authors observed an upregulation of ERIC after DNA damage that regulated apoptosis. Consequently, the inhibition of ERIC expression led to increased apoptosis (Feldstein et al., 2013).

lncRNAs are involved in Rb progression. Li et al. observed that lncRNA 00152 (LINC00152) was upregulated in Rb cell lines and tissues. The authors also reported that Rb cells silenced for LINC00152 showed an inhibition of cell proliferation, migration, invasion, and colony formation, and increased apoptosis. Moreover, LINC00152 knockdown caused the activation of caspase-3 and caspase-8 *in vitro* and suppressed tumorigenesis in nude mouse models (Li et al., 2018).

Dong et al. reported that HOTAIR lncRNA is involved in Rb progression through the Notch signalling pathway. HOTAIR is an oncogenic lncRNA correlated with metastasis, invasion, tumorigenesis and drug resistance. It is significantly upregulated in human Rb tissues and the authors showed that lncRNA knockout impeded the proliferation of Y-79 cell line. Notch also plays an important role in tumor development processes. In particular, Dong et al. assessed the expression levels of Notch1 and Jagged 1, the most common ligand and receptor, respectively, in the Notch pathway. They found that Jagged 1 and Notch 1 expression decreased after HOTAIR knockdown, indicating that HOTAIR regulates tumor progression in Rb through the activation of the *Notch 1* pathway (Dong et al., 2016). Another mechanism of action of HOTAIR is that of a miRNA sponge. Yang et al. demonstrated that HOTAIR sponged miR-613, which subsequently did not trigger tyrosine protein kinase met (c-met), its direct target gene. *C-met* is a protoncogene and its upregulation through HOTAIR causes the progression of Rb (Yang et al., 2018).

The testis-associated highly conserved oncogenic lncRNA (THOR) has been identified as another Rb promoter. THOR is widely expressed in various cancer types including Rb, but is also restrictively expressed in healthy testis tissues. Shang et al. showed that THOR promoted the malignant phenotype of Rb by interacting with insulin-like growth factor 2 mRNA binding protein 2 (IGF2BP1) and by controlling the mRNA stability of the c-myc oncogene. C-myc must be associated with IGF2BP1 to prevent its degradation. Thus in Rb cells, upregulated THOR promotes the association of c-myc and IGF2BP1, leading to its stabilization and consequently enhancing the malignant phenotype (Shang, 2018).

Wang et al. discovered that the lncRNA differentiation antagonizing non-protein coding RNA (DANCR) is upregulated in Rb cell lines and tissues and also overexpressed in Rb patients, leading to poor overall and disease-free survival. The authors observed that DANCR sponged miR-34c and miR-613, two miRNAs with tumor suppression function that target matrix metallopeptidase 9 (MMP9), an important protein for the breakdown of ECM. When DANCR regulation was activated, MMP9 was upregulated, leading to tumor progression (Wang et al., 2018).

A further connection between lncRNA and miRNA in Rb was demonstrated by Zhang et al. in their investigation of the association between lncRNA CCAT1 and miR-218-5p, known to occur in other cancer types. In fact, Lu et al. had previously shown that the negative regulation of miR-218-5p by CCAT1 promoted tumor progression in lung cancer (Lu et al., 2016). In the same way, Zhang et al. proved that CCAT1 was upregulated in Rb. Confirming the same mechanism of negative interaction with miR-218-5p, the authors showed that there was a reduction in apoptosis and an increase in cell proliferation and migration capacity (Zhang et al., 2017).

Actin filament-associated protein 1-antisense RNA 1 (AFAP1-AS1) is yet another lncRNA that has been hypothesized as having an oncogenic function in Rb. This lncRNA is associated with cancer progression and has been found to be overexpressed in various cell lines and tumor types such as lung cancer, ovarian cancer, oesophageal cancer, gastric cancer, hepatocellular carcinoma, nasopharyngeal carcinoma, colorectal cancer, biliary tract cancer, and pancreatic ductal adenocarcinoma (Zhang et al., 2018b). The role of this lncRNA in Rb was unknown up until recently. Hao et al. compared normal retina cell lines with Rb cells in knockdown experiments, observing that AFAP1-AS1 downregulation inhibited cell cycle progression, invasion, and migration. The authors also confirmed the oncogenic function of AFAP1-AS1 through its upregulation in Rb, reporting that it caused larger tumor size and optic nerve and choroidal invasion (Hao et al., 2018).

BDNF antisense RNA (BDNF-AS) is an lncRNAs transcribed by RNA polymerase II and has proved to be reverse regulator of BDNF. BDNF is a member of the neurotrophin family of growth factors whose role is to facilitate neuron survival and support the differentiation and growth of new synapses and neurons. Shang et al. evaluated the expression of BDNF-AS in Rb cell lines and found that it was downregulated in both Rb cell lines and tissues. Conversely, they also demonstrated that forced expression of BDNF-AS diminished cancer proliferation and metastatic potential, arresting cells in Go/G1 phase and consequently downregulating the cell cycle-associated proteins cyclin E and CDC42 (Shang et al., 2017).

Another lncRNA associated with tumorigenesis is promoter of CDKN1A antisense DNA damage activated RNA (PANDAR), and there is evidence of its upregulation in several types of cancers (Peng and Fan, 2015; Jiang et al., 2017). Sheng et al. investigated the potential clinical role of PANDAR in Rb, observing that it was overexpressed in Rb cells and tissues and that it inhibited cell apoptosis by affecting the *Bcl-2/caspase-3*

pathway. Moreover, using online databases, the authors predicted the specificity protein 1 (SP1) as a potential transcriptional factor capable of binding directly to the PANDAR promoter region and of triggering its transcription. They subsequently confirmed *in vitro* this binding ability of SP1 to the PANDAR promoter region.

The new connection discovered between PANDAR and SP1 could represent an alternative therapeutic target for Rb (Sheng et al., 2018).

In addition to the other lncRNAs, BRAF-activated non-coding RNA (BANCR) also plays an important role in the progression of various cancers, including Rb. Su et al. demonstrated that BANCR was overexpressed in Rb cell lines and tissues and that it was associated with tumor development. In fact, knocked down BANCR limited tumor invasion, metastatic capacity, and proliferation, indicating its potential usefulness as a therapeutic target for Rb (Su et al., 2015).

lncRNA H19 was among the first lncRNAs to be discovered and has different functions in different cancers. Zhang et al. hypothesized that H19 may also be involved in Rb. Initially, the authors assessed H19 expression in Rb, observing that it was substantially downregulated in Rb tissues and cell lines. They also showed that H19 has seven binding sites where it directly binds the miR-17-92 cluster in a competitive way. It has been already seen that the miR-17-92 cluster suppresses p21, an important regulator of the cell cycle, leading to STAT3 activation. Zhang et al. showed that H19 inhibited the suppressor role of the miR-17-92 cluster in p21 gene, decreasing STAT3 activation (Zhang et al., 2018a).

Cheng et al. observed that lncRNAs X inactive specific transcript (XIST) had an oncogene function in Rb. The authors discovered that XIST binds miR-101, regulating the repression of its targets, E-box binding homeobox 1 and 2 (ZEB1, ZEB2). ZEB1 and ZEB2 are transcription factors that are responsible for the malignant features of different type of cancers. The binding of miR-101 to XIST and the subsequent oncogenic activity of ZEB1 and ZEB2 led to increased proliferation, invasion and migration of Rb cells (Cheng et al., 2019).

A similar mechanism was uncovered by Liu et al. who found that the lncRNA metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) was upregulated in Rb. The authors observed that MALAT1 silencing in cells caused the inhibition of cell viability, invasion and migration, activated apoptosis and upregulated miR-124.

They also demonstrated that MALAT1 sponged miR-124 which, consequently, did not target E-cadherin transcription repressor (Slug), its target gene. Slug activates ERK/MAPK and Wnt/β-catenin pathways and, when downregulated by miR-124, inhibits Rb progression (Liu S, et al., 2017).

The above-mentioned lncRNAs are summarized in **Table 3**.

### CIRCULAR RNAs PATHWAYS IN Rb

Circular RNAs (circ-RNAs) are an emerging group of ncRNAs present inside cells. Considering their form, they are less susceptible to RNAs degradation and are the result of a backsplicing of 5'splice position with the 3'splice position or exon skipping mechanism. Recent evidence suggests that various disorders such as nervous diseases, vascular inflammation and several types of cancer may be the result of their deregulation (Guo N, et al., 2019).

hsa\_circ\_0001649 is a new cancer associated circ-RNA and is reported to be the transcription product of Snf2 Histone Linker Phd Ring Helicase (SHPRH) tumor suppressor gene. Xing et al. showed *via* q-RT PCR analysis that hsa\_circ\_0001649 was downregulated in Rb tissue samples compared with normal tissue and that low expression of this circ-RNA was associated with aggressive phenotypes in Rb patients. To further confirm its involvement in tumorigenesis, the authors assessed hsa\_ circ\_0001649 enhanced expression in xenografts, reporting that tumors formed from hsa\_circ\_0001649-transfected cells

TABLE 3 | Long ncRNAs (lncRNAs) involved in retinoblastoma (Rb) and some of their putative target genes.


were smaller than those of the control group. In the same study, the authors investigated the molecular mechanisms of hsa\_ circ\_0001649 which are responsible for changes in Rb tumor cell proliferation. The Akt/mTOR apoptosis-related signalling pathway is known to be related to Rb progression, but there is still no evidence of the involvement of circ-RNA. Xing et al. suggested that the Akt/mTOR signalling pathway is regulated by hsa\_circ\_0001649. More precisely, after transfection of Rb cell lines with hsa\_circ\_0001649, the authors observed that p-AKT and p-mTOR were negatively correlated with hsa\_circ\_0001649 (Xing et al., 2018).

In another study, Lyu et al. analyzed the expression profile of circ-RNA in human Rb tissue and in corresponding normal retina, observing a general reduction in circ-RNA expression levels in Rb. This may have been a result of compromised back-splice machinery in the circ-RNA production or by an excessive consumption of circ-RNA, necessary for cell proliferation. In particular, TET1-hsa\_circ\_0093996 was significantly downregulated in Rb tissue. The authors also observed a downregulation of tumor suppressor PDCD4. Based on the *in silico* analysis, which revealed that miR-183 targeted PDCD4, the authors hypothesized that TET1-hsa\_circ\_0093996 sponged miR-183. In an attempt to create a regulatory axis, they assumed that TET1-hsa\_ circ\_0093996 downregulation increased unbound miR-183, which consequently targeted PDCD4 causing enhanced cell proliferation (Lyu et al., 2019a).

### CONCLUSIONS

ncRNAs are molecules physiologically present in humans where they regulate gene expression after transcription and subsequently control important mechanisms such as cell proliferation, development, and apoptosis. ncRNAs are deregulated in many types of cancer, suggesting that they are involved in carcinogenesis. In the present review, we explored short and long ncRNAs, which are deregulated in Rb, demonstrating that, in addition to *RB1,* there is a great number of other molecules involved in the development of the malignant phenotype of this type of cancer.

The evolution of genomic and genetic technologies together with the generation and development of bioinformatics have made it possible to manage the enormous quantity of accumulated data generated by large-scale high throughput analyses and basic research.

Profiling analysis is the first step to discover new mechanisms of ncRNAs involved in Rb development.

miRNA profiles in Rb have been discussed at length in this review (Beta et al., 2013; Yang and Mei, 2015). The studies in question identified miRNAs that are upregulated and downregulated in Rb with an oncogene or tumor suppressor function, respectively.

The function of miRNAs is explicated through the regulation of target genes involved in cell proliferation, migration, invasion, cell viability, and apoptosis, as demonstrated by Wang et al. (2017). Through a comparison of miRNAs and gene profiling it is possible to identify the pathways that determine Rb progression. These pathways are carefully validated through *in vitro* and *in vivo* experiments. miRNAs directly or indirectly regulate important gene such as cyclins (Wu et al., 2015; Golabchi et al., 2017) and TP53 (To et al., 2012; Bai et al., 2016), respectively. In turn, lncRNAs regulate miRNAs through sponge mechanism (Liu S, et al., 2017; Wang et al., 2018; Yang et al., 2018; Zhang et al., 2018a). Studies show the complex ncRNA mechanisms and the prevailing pathways that determine Rb progression.

Further research into the complex Rb pathways will help to identify novel ncRNA-based therapeutic approaches to counteract the aberrations of the ncRNA that are responsible for the development of Rb. Several studies described in this review analyzed the therapeutic potential of ncRNAs, but although they highlighted the therapeutic prospects of these molecules, their clinical implementation remains a challenge. A non-toxic delivery system is needed to selectively transport ncRNA-based therapeutics to the tumor site, e.g., antisense oligonucleotide or inhibitor against an oncogene (Adams et al., 2017). Moreover, the fact that a single miRNA binds more than 100 target genes makes target specificity a problem. The recent discovery that ncRNAs contained in exosomes interact with the tumor microenvironment and affect cancer growth and metastatic potential has opened up a new chapter on the intercellular crosstalk for cancer biology (Vannini et al., 2018a). Thus, new strategies to impair the exosome-mediated ncRNA transfer affecting cancer growth and dissemination can be hypothesized. The characterization of regulatory mechanisms of lncRNAs is also a critical aspect to complement the deficiency of precision medicine.

A better understanding of the role played by ncRNAs in chemoresistance would enable patients to be spared from nonbeneficial treatments but would also help to overcome the problem by modulating the expression of the ncRNAs involved in the resistance mechanisms.

Future studies will serve to clarify the role of ncRNAs as tumor suppressors or oncogenes and to design new ncRNAbased therapeutic approaches. The era of an ncRNA-based therapy for rb is fast approaching and will provide oncologists with a powerful tool for improving patients' odds against this often deadly tumor.

### AUTHOR CONTRIBUTIONS

MP found the articles that describe ncRNAs in Retinoblastoma and she wrote the manuscript. IV wrote and corrected the manuscript.

### ACKNOWLEDGMENTS

We would like to thank Francesco Mazza for helping with figure design.

## REFERENCES


retinoblastoma tumor suppressor. *Biochem. Biophys. Res. Commun*. 406, 518– 523. doi: 10.1016/j.bbrc.2011.02.065


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

*Copyright © 2019 Plousiou and Vannini. 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.*

# Transcriptional Repression of CYP3A4 by Increased miR-200a-3p and miR-150-5p Promotes Steatosis in vitro

Zhijun Huang1,2, Mengyao Wang1,3, Li Liu<sup>1</sup> , Jinfu Peng1,2, Chengxian Guo<sup>1</sup> , Xiaoping Chen<sup>4</sup> , Lu Huang1,2, Jieqiong Tan<sup>5</sup> and Guoping Yang1,2 \*

<sup>1</sup> Center for Clinical Pharmacology, The Third Xiangya Hospital, Central South University, Changsha, China, <sup>2</sup> Department of Pharmacy, The Third Xiangya Hospital, Central South University, Changsha, China, <sup>3</sup> Xiangya School of Pharmaceutical Sciences, Central South University, Changsha, China, <sup>4</sup> Institute of Clinical Pharmacology, Central South University, Changsha, China, <sup>5</sup> Center for Medical Genetics, Life Science School, Central South University, Changsha, China

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Mahshid Malakootian, Iran University of Medical Sciences, Iran Jun-An Chen, Academia Sinica, Taiwan

> \*Correspondence: Guoping Yang ygp9880@126.com

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 03 October 2018 Accepted: 06 May 2019 Published: 28 May 2019

#### Citation:

Huang Z, Wang M, Liu L, Peng J, Guo C, Chen X, Huang L, Tan J and Yang G (2019) Transcriptional Repression of CYP3A4 by Increased miR-200a-3p and miR-150-5p Promotes Steatosis in vitro. Front. Genet. 10:484. doi: 10.3389/fgene.2019.00484 Hepatic cytochrome P450 enzyme activities correlate with non-alcoholic fatty liver disease (NAFLD) and hepatic steatosis. The decreased activity of CYP3A4, an important drug-metabolizing enzyme, is associated with the progression of NAFLD. CYP3A4 is predicted as a target gene of miR-200a-3p and miR-150-5p by MicroInspector and TargetScan algorithms analyses. Here, we found decreased CYP3A4 and increased miR-200a-3p and miR-150-5p in LO2 cells with free fatty acid (FFA)-induced steatosis. Dual-luciferase assay confirmed that both miR-200a-3p and miR-150-5p targeted the 3 0 -untranslated region (3<sup>0</sup> -UTR) of CYP3A4 and that such interaction was abolished by miRNA binding site mutations in 3<sup>0</sup> -UTR of CYP3A4. Using miR-200a-3p and miR-150- 5p mimics and inhibitors, we further confirmed that endogenous CYP3A4 was regulated posttranscriptionally by miR-200a-3p or miR-150-5p. Moreover, miR-200a-3p and miR-150-5p inhibitors attenuated FFA-induced steatosis in LO2 cells, and such effect was dependent on CYP3Y4 expression. These results suggest that miR-200a-3p and miR-150-5p, through directly targeting 3<sup>0</sup> -UTR of CYP3A4, contribute to the development of FFA-induced steatosis.

Keywords: non-alcoholic fatty liver disease, CYP3A4, miR-200a-3p, miR-150-5p, LO2 cell line

### INTRODUCTION

Non-alcoholic fatty liver disease (NAFLD) has been the most common chronic liver disease in the world for the past 30 years (Younossi et al., 2018) with an estimated prevalence of 10– 40% (Younossi et al., 2016). NAFLD includes a spectrum of disease from simple liver steatosis to non-alcoholic steatohepatitis (NASH). NASH is considered as a leading indication for liver transplants in the near future, as it increases the risk of hepatocarcinoma (HCC) (Sattar et al., 2014; Diehl and Day, 2017).

Several genetic defects and environmental factors have been implicated in the pathogenesis of NAFLD. Specifically, a few studies have reported that NAFLD is associated with decreased expression and function of CYP3A4, a major member of the hepatic cytochrome P450 superfamily contributing to the metabolism of 45–60% of all drugs used in the clinical setting

(Wojnowski and Kamdem, 2006; Anglicheau et al., 2007). Donato et al. (2006) observed a significantly reduced CYP3A4 activity in both human liver tissue with steatosis and fatoverloaded hepatocytes cultured in vitro. Similar results were shown in nutritionally obese mice (Yoshinari et al., 2006; Maximos et al., 2017). Compared with healthy controls, NAFLD patients showed higher plasma concentration of CYP3A4 substrate, indicating impaired CYP3A4 function in NAFLD patients (Woolsey et al., 2015). Decreased protein expression and activity of CYP3A4 were observed with NAFLD development (Fisher et al., 2009). A recent study further confirmed reduced protein level and activity of CYP3A4 in liver tissues of NAFLD patients as compared to those of controls (Jamwal et al., 2018). However, despite all these observations, the underlying mechanism regulating the expression and function of CYP3A4 in NAFLD remains unclear.

MicroRNAs (miRNAs/miRs) are small non-coding RNAs that negatively regulate gene expression through binding to the 3 0 -untranslated region (3<sup>0</sup> -UTR) of mRNAs, thus altering the expression and function of various genes, including CYP3A4 (Pan et al., 2009; Swathy et al., 2017; Yan et al., 2017). miRNA-34a, miRNA-122, and miRNA-192 are considered as potential biomarkers of NAFLD staging (Liu et al., 2018). Importantly, high-throughput sequencing revealed that expressions of miR-150-5p and miR-200a-3p were significantly higher in NAFLD with fibrosis than in NAFLD without fibrosis (Leti et al., 2015). Therefore, we hypothesize that significant change of hepatic miRNAs in NAFLD could regulate CYP3A4 expression posttranscriptionally. Here, we report that miR-150-5p and miR-200a-3p directly regulate CYP3A4 and are involved in free fatty acid (FFA)-induced steatosis.

### MATERIALS AND METHODS

#### Reagents

Sodium salts of palmitic acid (PA) (P9767) and oleic acid (OA) (O7501), fatty acid (FA) free bovine serum albumin (BSA) (A8806), and BODIPY 493/503 (790389) were purchased from Sigma-Aldrich (MO, United States), and RPMI 1640 (11875-093) medium and fetal bovine serum (FBS) (26140079) were from GIBCO (Invitrogen, CA, United States). LipofectamineTM 2000 (11668019) was from Invitrogen (CA, United States). Oligonucleotide primers for CYP3A4 were synthesized by Sangon Biotech (Shanghai, China). All miRNA mimics, miRNA mimic negative controls, miRNA inhibitors, miRNA inhibitor negative controls, and primers for miRNA RT-qPCR were purchased from RiboBio (Guangzhou, China). All other chemicals and solvents were of the highest commercial grades.

### Cell Culture Model of Hepatic Steatosis in Vitro

LO2 cells were provided by Stem Cell Bank, Chinese Academy of Sciences and cultured in RPMI 1640 medium supplemented with 10% FBS at 37◦C/5% CO2. Steatosis was induced as previously described (Feldstein et al., 2004; Ricchi et al., 2009). PA and OA were codissolved in 10% FA-free BSA prepared in H2O. In accordance with previous studies (Wang et al., 2014), LO2 cells were exposed to a mixture of 1 mM OA and PA (final ratio 2:1) for 24 h. After 24 h of incubation, lipid droplets were stained by BODIPY according to a previously reported protocol (Qiu and Simon, 2016).

### In silico Identification of Putative miRNA Binding Sites

The 3<sup>0</sup> -UTR sequences of human CYP3A4 (GenBank sequence NM\_017460) were searched for the antisense matches to individual miRNAs using MicroInspector (Rusinov et al., 2005) and Target Scan (Lewis et al., 2005).

#### Transfection

All transfections were performed by LipofectamineTM 2000 according to the manufacturer's instructions.

### Real-Time PCR

Small RNA was extracted with an E.Z.N.A miRNA kit (Omega BIO-TEK, GA, United States) and reverse transcribed using PrimeScript RT Reagent Kit (Roche, Basel, Switzerland) with primers for specified miRNAs. The oligonucleotide sequences of STEM-LOOP RT primers were designed (Kramer, 2011) as follows: miR-200a-3p, 5<sup>0</sup> - GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGAT ACGACacatcgtt-3<sup>0</sup> ; miR-150-5p, 5<sup>0</sup> -GTCGTATCCAGTGCAGG GTCCGAGGTATTCGCACTGGATACGACctgtcccc-3<sup>0</sup> ; U6, 5<sup>0</sup> - GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGAT ACGACaaaaatat-3<sup>0</sup> . Real-time PCR was performed with SYBR Green PCR Master Mix (Roche, Basel, Switzerland) using the following conditions: 95◦C for 10 min followed by 40 cycles of amplification at 95◦C for 10 s and 59◦C for 30 s. Mature miR-200a-3p and miR-150-5p levels were normalized with U6. The oligonucleotide sequences of qPCR primers were as follows: miR-200a-3p, 5<sup>0</sup> -CACGCAtaacactgtctggtaa-3 0 ; miR-150-5p, 5<sup>0</sup> -CACGCActggtacagggcctgg-3<sup>0</sup> ; U6, 5 0 -CACGCAgcaaggatgacacgcaa-3<sup>0</sup> ; general reverse primer, 5 0 -CACGCATGGAAGGACGGG-3<sup>0</sup> .

To inhibit or induce miR-150-5p or miR-200a-3p, transient transfection of miRNA inhibitors or mimics (100 nM) was performed in LO2 cells using LipofectamineTM 2000. Specific miRNA-150 or miRNA-200a inhibitors or mimics were commercially purchased from RiboBio (Guangzhou, China), including anti-miRNA-150 (target sequence 5<sup>0</sup> -CUGGUA CAGGCCUGGGGGACAG-3<sup>0</sup> ), anti-miRNA-200a (target sequence 5<sup>0</sup> -UAACACUGUCUGGUAACGAUGU-3<sup>0</sup> ), synmiRNA-150 (target sequence 5<sup>0</sup> -CUGGUACAGGCCUGGGGG ACAG-3<sup>0</sup> ), and syn-miRNA-200a (target sequence 5 0 -UAACACUGUCUGGUAACGAUGU-3<sup>0</sup> ). miScript inhibitor negative control (100 nM) (RiboBio, Guangzhou, China) was used as internal reference for normalization.

Twenty-four hours after transfection, cells were treated with a mixture of 1 mM OA and PA (final ratio 2:1) for

24 h. To confirm the effect of miR-150-5p or miR-200a-3p inhibition or induction on CYP3A4 mRNA, total RNAs were extracted from LO2 cells with TRIzol (Life Technologies, CA, United States). Total mRNAs were reverse transcribed into cDNAs by the PrimeScript RT Reagent kit. Real-time PCR was performed by the SYBR Green PCR Master Mix using the following conditions: 95◦C for 10 min followed by 40 cycles of amplification at 95◦C for 10 s and 59◦C for 30 s. GAPDH (forward primer: AGAAGGCTGGGGCTCATTTG; reverse primer: AGGGGCCATCCACAGTCTTC) was used as an internal control to normalize CYP3A4 expression (forward primer: CCCGTTGTTCTAAAGGTTGA; reverse primer: TCTGGTGTTCTCAGGCACAG). qPCR was quantified using the formula 2−11CT and plotted as x-fold to the control.

### Western Blot

Cells in six-well plates were harvested posttransfection and treatment, and whole-cell lysates were prepared with RIPA lysis buffer (Beyotime, Beijing, China) supplemented with complete protease inhibitor and phenylmethanesulfonyl fluoride (Beyotime, Beijing, China). Protein concentrations were determined with the BCA Protein Assay Kit (Beyotime, Beijing, China). Whole-cell protein (20 µg) was separated on SDS-PAGE and electrophoretically transferred onto PVDF membrane (Millipore, CA, United States). The membrane was incubated with a selective rabbit antihuman CYP3A4 polyclonal antibody (Millipore, CA, United States) or mouse anti-human GAPDH antibodies (Zhongshan Inc., Guangzhou, China), and subsequently with the secondary antibody of HRP goat anti-rabbit IgG (Zhongshan Inc., Guangzhou, China) or rabbit anti-mouse IgG (Zhongshan Inc., Guangzhou, China). Images were acquired with GE Healthcare ImageQuant 350, and band densities were quantified with GeneTools (SynGene, Cambridge, United Kingdom).

#### Construction of Reporter Plasmids

The 3<sup>0</sup> -UTR of CYP3A4 gene corresponding to 1,620–2,792 nt (1,173 bp; accession no. NM\_001202855) was cloned into pmiR-RB-REPORTTM vector via XhoI and NotI restriction sites. The primers used for construction of wild-type CYP3A4 3 0 -UTR reporter plasmid were as follows: h-CYP3A4-F, 5<sup>0</sup> - CTTGACTCGAGATTTTCCTAAGGACTTCTGC-3<sup>0</sup> ; h-CYP 3A4-R, 5<sup>0</sup> -ATTGCGGCCGCAGGCTTATTGCTCAATC-3<sup>0</sup> . The sequence of the recombinant clones was confirmed by DNA sequencing and named as pmiR-RB-REPORTTM CYP3A4 3<sup>0</sup> -UTR WT. miRNA-200a binding site mutant (AGTGTTA changed to TGTGCCA) and miRNA-150 binding site double mutant (TTCCCAG changed to ATCCGAT and TGGGAGA changed to AGGCATA) were constructed using Q5 <sup>R</sup> Site-Directed Mutagenesis Kit (NEB, MA, United States) and confirmed by DNA sequencing. The firefly luciferase gene used 3<sup>0</sup> -UTR of CYP3A4 as the report luciferase, with Renilla luciferase gene as an internal control.

levels in steatosis cells were determined by real-time PCR and normalized with U6 snRNA. Values were expressed as mean ± SEM for three independent experiments. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, versus control.

### Dual-Luciferase Assay

The LO2 cells were seeded into 24-well plates. Firefly luciferase (0.1 µg) containing 3<sup>0</sup> -UTR of CYP3A4 in pmiR-RB-REPORTTM vector, along with miR-200a-3p or miR-150-5p mimic, was transfected into LO2 cells with Lipofectamine 2000. After 24 h of incubation, luciferase activities were measured with a luminometer (Tecan Infinite 200 Pro, Switzerland) using the Dual-Luciferase Reporter Assay System (Promega, Valencia, CA, United States). Firefly luciferase activity was normalized by Renilla luciferase activity and compared between different treatments.

#### Statistical Analysis

All values were expressed as the mean ± SEM. Comparisons of variables between groups were performed with an unpaired two-tailed Student's t-test. P < 0.05 was considered statistically significant.

miR-200a-3p (B) or miR-150-5p (C) mimics to CYP3A4 3<sup>0</sup> -UTR-Luc WT and miRNA binding site mutants (MUT). LO2 cells were cotransfected with a firefly luciferase reporter vector containing the indicated CYP3A4 3<sup>0</sup> -UTR constructs, a Renilla luciferase reporter as internal control, and treated with miR-200a-3p (B) or miR-150-5p (C) mimics. Values were mean ± SEM for three independent experiments. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, versus control; ns, not significant.

#### RESULTS

#### CYP3A4 Expression Is Decreased in FFA-Induced Steatosis Cells

CYP3A4 activity is reduced in human NAFLD as well as in mouse and in vitro cell models of the disease (Kolwankar et al., 2007; Woolsey et al., 2015). Here, we investigated CYP3A4 expression in FFA-induced steatosis cells in vitro. As shown in **Figures 1A,B**, CYP3A4 protein level was decreased in FFA-induced steatosis cells compared to control. Similarly, a significant decrease of CYP3A4 mRNA was observed in FFA-induced steatosis cells compared to control (**Figure 1C**).

### Increased Mature miR-200a-3p and miR-150-5p in FFA-Induced Steatosis Cells

To investigate whether decreased CYP3A4 in FFA-induced steatosis cells was due to miRNAs regulation, we used MicroInspector and TargetScan algorithms to screen antisense matches of CYP3A4 3<sup>0</sup> -UTR against human miRNAs. CYP3A4 was predicted to be a target gene of miR-200a-3p and miR-150- 5p **(Figure 2A**). Next, we determined the expression levels of mature miR-200a-3p and miR-150-5p in FFA-induced steatosis cells and found that both mature miR-200a-3p and miR-150- 5p were higher in FFA-induced steatosis cells than in control cells (**Figure 1C**).

#### miR-200a-3p and miR-150-5p Regulate CYP3A4 Expression by Targeting Its 3<sup>0</sup> -UTR

To investigate whether CYP3A4 can be directly regulated by mature miR-200a-3p and miR-150-5p, CYP3A4 3<sup>0</sup> -UTR WT and the two miRNA binding site mutants were cloned separately into pmiR-RB-REPORTTM vector for dual-luciferase assay. We showed that both miR-200a-3p and miR-150-5p interacted with CYP3A4 3<sup>0</sup> -UTR WT, but not the two miRNA binding site mutants (**Figures 2B,C**).

### miR-200a-3p and miR-150-5p Down-Regulate CYP3A4 Expression in FFA-Induced Steatosis Cells

To investigate the effect of miR-200a-3p and miR-150-5p on CYP3A4 expression, we examined changes of CYP3A4 mRNA and protein levels in response to inhibition or induction of miR-200a-3p or miR-150-5p. miR-200a-3p or miR-150-5p inhibitor increased CYP3A4 protein (**Figures 3A**–**D**) and mRNA (**Figure 3E**) levels. In addition, CYP3A4 protein and mRNA (**Figure 3F**) levels were decreased after miR-200a-3p or miR-150- 5p mimic treatment.

### miR-200a-3p or miR-150-5p Inhibitor Regulates FFA-Induced Steatosis via CYP3A4

We have demonstrated that miR-200a-3p or miR-150-5p negatively regulated CYP3A4 expression. Because of the importance of CYP3A4 activity in steatosis development (Donato et al., 2006; Hu et al., 2014), we then determined whether CYP3A4 mediated the regulatory effect of miR-200a-3p or miR-150-5p inhibitor on FFA-induced steatosis. miR-200a-3p or miR-150- 5p inhibitor increased CYP3A4 expression, which was abolished by CYP3A4 knockdown (**Figures 4A–D**). More importantly, miR-200a-3p or miR-150-5p inhibitor reduced FFA-induced steatosis assessed by the BODIPY 493/503 staining, and this effect was abrogated by CYP3A4 gene silencing (**Figures 4E,F**). This suggested that miR-200a-3p or miR-150-5p inhibitor, through up-regulating CYP3A4, protected against FFA-induced steatosis.

## DISCUSSION

In recent years, NAFLD has emerged as a major public health concern characterized by elevated serum FFAs and hepatocyte lipoapoptosis (Feldstein et al., 2003). Several reports have indicated the importance of both quantitative and qualitative (e.g., saturated versus unsaturated FAs) changes in dietary FAs as underlying causative mechanisms for NAFLD in both rodent

miR-150-5p inhibitors were transfected into LO2 cells with or without knockdown of CYP3A4. After 24 h, cells were exposed to 1 mM FFA or vehicle. The BODIPY 493/503 staining was performed to assess cellular steatosis. Green fluorescence indicated lipid droplets. The cell nucleus was stained by DAPI (Blue). <sup>∗</sup>p < 0.05, ∗∗p < 0.01, bar = 10 µm.

models and humans (Nehra et al., 2001; Musso et al., 2003; Wang et al., 2006). Increased supply of FAs or decreased lipid clearance in hepatocytes can set off esterification of FAs and formation of lipid droplets (Anderson and Borlak, 2008). In terms of mechanism, enzymes, important regulators of plasma and tissue FA composition, signaling pathways, and transcriptional factors controlling FA synthesis and gluconeogenesis have all been implicated in NAFLD (Malhi and Gores, 2008; Mota et al., 2016). Here, we demonstrated that miR-200a-3p and miR-150- 5p, through directly targeting 3<sup>0</sup> -UTR of CYP3A4, contributed to the development of FFA-induced steatosis in vitro.

Previous studies have evaluated the relationship between NAFLD and hepatic CYP3A activity. Fisher et al. (2009) explored the relationship between hepatic CYP3A4 and non-alcoholic hepatic steatosis in human liver samples. It was found that CYP3A4 protein expression and activity decreased with the progression of NAFLD. Other studies (Kolwankar et al., 2007; Jamwal et al., 2018) showed that liver samples with steatosis had significantly lower hepatic CYP3A4 activity than those without steatosis, which was consistent with in vitro findings in longchain FFA-induced steatosis. CYP3A4 mRNA and protein also decreased significantly in fatty mice induced by a high-fat diet (Yoshinari et al., 2006).

Although most of the CYPs are found within endoplasmic reticulum, CYP3A4 localizes in mitochondria and synthesizes arachidonic acid (AA)-derived epoxyeicosatrienoic acids (EETs), which promotes electron transport chain/respiration and mitochondrial function (Guo et al., 2017). It is well known that CYP enzyme activities are affected by steatosis, which poses major impact on drug metabolism and drug-induced hepatotoxicity (Gomez-Lechon et al., 2009). However, our study suggests that increased CYP3A4 expression prevents LO2 cells from FFA-induced steatosis. Our results demonstrate that CYP3A4 activity impairment is not only induced by hepatic steatosis but also a factor further promoting hepatic steatosis. This could explain the phenotype of cyp3a-null male mice,

#### REFERENCES


including increases in weight, liver triglycerides, and total lipids (Kumar et al., 2018).

It was previously shown that CYP3A4 could be regulated by miRNAs via direct and indirect targeting (Takagi et al., 2008; Pan et al., 2009; Swathy et al., 2017; Yan et al., 2017). Hepatic miRNA changes in NAFLD have been confirmed in several studies of dietary NASH in mice and rats (Pogribny et al., 2010; Alisi et al., 2011), as well as serum and liver tissue of NAFLD patients (Leti et al., 2015; Liu et al., 2018). In this study, we reported two new CYP3A4 regulators, including miR-200a-3p and miR-150-5p, in FFA-induced steatosis. Our results showed that expressions of miR-200a-3p and miR-150-5p increased significantly in hepatocytes with steatosis as compared to normal hepatocytes, which is consistent with previous studies reporting an increase of miR-200a-3p in NAFLD (Cheung et al., 2008; Alisi et al., 2011). More importantly, miR-200a-3p or miR-150-5p inhibitor can alleviate FFA-induced steatosis. Thus, our findings provide a potential therapeutic target for the use of miR-200a-3p and miR-150-5p inhibitors to protect against steatosis development.

### AUTHOR CONTRIBUTIONS

ZH and MW performed the experiments and wrote the manuscript. LL, JP, CG, XC, LH, and JT performed the experiments. ZH and GY designed the project and provided supervision and support for the project.

#### FUNDING

This work was supported by National Scientific Foundation of China (Nos. 81673520 and 81673519) and the New Xiangya Talent Project of the Third Xiangya Hospital of Central South University (No. 20180302).


with decreased CYP3A4 protein expression and activity in human liver. Mol. Pharm. 15, 2621–2632. doi: 10.1021/acs.molpharmaceut.8b00159


and apoptosis in cultured hepatocytes. J. Gastroenterol. Hepatol. 24, 830–840. doi: 10.1111/j.1440-1746.2008.05733.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 © 2019 Huang, Wang, Liu, Peng, Guo, Chen, Huang, Tan 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) 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.

# Autophagy-Modulating Long Non-coding RNAs (LncRNAs) and Their Molecular Events in Cancer

Md Zahirul Islam Khan, Shing Yau Tam and Helen Ka Wai Law\*

Department of Health Technology and Informatics, Faculty of Health and Social Sciences, The Hong Kong Polytechnic University, Hong Kong, Hong Kong

Cancer is a global threat of health. Cancer incidence and death is also increasing continuously because of poor understanding of diseases. Although, traditional treatments (surgery, radiotherapy, and chemotherapy) are effective against primary tumors, death rate is increasing because of metastasis development where traditional treatments have failed. Autophagy is a conserved regulatory process of eliminating proteins and damaged organelles. Numerous research revealed that autophagy has dual sword mechanisms including cancer progressions and suppressions. In most of the cases, it maintains homeostasis of cancer microenvironment by providing nutritional supplement under starvation and hypoxic conditions. Over the past few decades, stunning research evidence disclosed significant roles of long non-coding RNAs (lncRNAs) in the regulation of autophagy. LncRNAs are RNA containing more than 200 nucleotides, which have no protein-coding ability but they are found to be expressed in most of the cancers. It is also proved that, autophagy-modulating lncRNAs have significant impacts on pro-survival or pro-death roles in cancers. In this review, we highlighted the recently identified autophagy-modulating lncRNAs, their signaling transduction in cancer and mechanism in cancer. This review will explore newly emerging knowledge of cancer genetics and it may provide novel targets for cancer therapy.

Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Kaushlendra Tripathi, University of Alabama at Birmingham, United States Jai Prakash, University of Twente, Netherlands

> \*Correspondence: Helen Ka Wai Law hthelen@polyu.edu.hk

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 02 October 2018 Accepted: 31 December 2018 Published: 14 January 2019

#### Citation:

Islam Khan MZ, Tam SY and Law HKW (2019) Autophagy-Modulating Long Non-coding RNAs (LncRNAs) and Their Molecular Events in Cancer. Front. Genet. 9:750. doi: 10.3389/fgene.2018.00750 Keywords: autophagy, long non-coding RNAs, cancer, therapy, biomarkers

**Abbreviations:** ATGs, autophagy-related genes; BANCR, BRAF-activated long non-coding RNA; BDC, bladder cancer; BRC, breast cancer; BT, brain tumor; CASC2, cancer susceptibility candidate gene 2; CRC, colorectal cancer; CTA, LncRNA CTA; DFS, disease-free survival; EMT, epithelial–mesenchymal transition; ENDC, endometrial cancer; GAS5, growth arrestspecific 5; GC, gastric cancer; GMEC, glioma microvascular endothelial cancer; HCC, hepatocellular carcinoma; HMGB1, high-mobility group box 1 protein; HNC, head and neck cancer; HNF 1A-AS1, a natural antisense transcript of HNF 1A; HOTAIR, Hox transcript antisense RNA; HULC, highly up-regulated in liver cancer; LC, lung cancer; LCPAT1, lung cancer progression associated transcript 1; LEUK, leukemia; lncRNAs, long non-coding RNAs; LNM, lymphatic node metastasis; LVC, liver cancer; MALAT1, metastasis-associated lung adenocarcinoma transcript 1; MEG3, maternally expressed gene 3; MM, multiple myeloma; MPM, malignant pleural mesothelioma; ncRNAs, non-coding RNAs; NGS, next-generation sequencing; NSCLC, non-small-cell lung cancer; OS, overall survival; OSTS, osteosarcoma; OVC, ovarian cancer; PCA3, prostate cancer antigen 3; PDA, pancreatic ductal adenocarcinoma; PKM2, pyruvate kinase isozymes 2; PNC, pancreatic cancer; PPTC, papillary thyroid carcinoma; PRKD3, protein kinase D3; PTC, prostate cancer; PTENP1, pseudogene of tumor suppressor gene PTEN; PVT1, plasmacytoma variant translocation 1; ROR, regulator of reprogramming; RTB, retinoblastoma; SB, sleeping beauty; sirt1, silent information regulator 1; STAT3, signal transducer and activator of transcription 3; TNM, tumor node metastasis.

Cancer is a global threat of public health because of its high mortality rate. More than 14 million of new cancer cases and ∼60% of death were accounted in 2012 by GLOBOCAN data (**Figure 1A**). It is also assumed that more than 20 million new cancer cases will be counted in the year 2025 (Ferlay et al., 2015). Cancer death is increasing day by day, 8.8 million death were counted in 2015 where abundant death was reduced in low to moderate income countries (**Figure 1B**) (GBD 2015 Risk Factors Collaborators, 2016; WHO, 2018). The cancer cases in 10 Asian countries are also estimated to increase from in 6.1 million to 10.7 million in the year 2008 to 2030 respectively. The consecutive death was estimated to raise from 4.1 million to 7.5 million (Sankaranarayanan et al., 2014).

Metastatic tumors are the leading cause of death and pose a great challenge for cancer treatments. Although the molecular basis of carcinogenesis is different in different cancers, the metastasis developmental process is almost similar in all solid tumor cells (Bogenrieder and Herlyn, 2003). The aim of traditional treatments of cancer is either removing or destroying cancerous cells by surgery, radiation therapy, chemotherapy and sometimes personalized treatment. Theoretically, these traditional treatments are very promising but practically none of these are able to cure effectively and efficiently because of poor understating of mechanisms and development of metastasis (Wang Z. et al., 2016).

Autophagy is a highly conserved and critical regulatory process for cells to maintain homeostasis by lysosomal degradation of various proteins and damaged organelles. Dynamic roles of autophagy have been identified in cancers where it participates in cancer progression, prevention, as well as, drug resistance mechanisms (Santana-Codina et al., 2017). There are three types of autophagy: macroautophagy, microautophagy, and chaperone-mediated autophagy. Macroautophagy is sometimes also referred to as autophagy which is the major autophagic pathways and most extensively studied compared to microautophagy and chaperone-mediated autophagy. In macroautophagy, phagophore is initially formed and matured to autophagosome. Subsequently, autophagosome fused with a lysosome to degrade the internal materials in autolysosome (**Figure 2**) (Mizushima and Komatsu, 2011). Although autophagy may suppress tumors (Kung et al., 2011) in most cases, the induction of autophagy promotes tumorigenesis by providing survival capability of tumor under microenvironmental stress (Kung et al., 2011; Avalos et al., 2014; White, 2015). Autophagy promotes cancer by inhibiting tumor suppressor protein p53 and controlling the metabolism of cells (Amaravadi et al., 2016). Cellular metabolism and homeostasis are encoded by more than 30 ATGs, their translational products and transduction of signals (**Figure 2**) (Kim and Lee, 2014; Cicchini et al., 2015; Ktistakis and Tooze, 2016). Tumorigenesis of both benign and malignant tumors are controlled by either single or group of ATG genes (Blessing et al., 2017; Mowers et al., 2017). Thus, in most of the cancers, autophagy is key therapeutic choices in clinical trials (Mowers et al., 2016).

There is increasing evidence that a large number of ncRNAs are actively transcribed from the human genome, controlling diverse cellular metabolic process in growth and development of cancers through regulating various gene expressions (Lin and He, 2017). Most diverse ncRNAs, specifically lncRNAs, are largely involved in cancer initiation, maturation, metastasis and resistance against chemotherapy (Lin and He, 2017; Hu et al., 2018). LncRNAs are >200 nucleotides in length and are mostly transcribed by RNA polymerase II. According to GENCODE (version 18), 13,562 lncRNAs have been identified (Huang X. et al., 2017). LncRNAs have independent expressions in different tissues and different cancers and are responsible for various cancers including BRC, CRC, GC, glioma, HCC, leukemia, LVC, LC, OVC, PTC, and retinoblastoma (He et al., 2016; Huang J. et al., 2017; Huang X. et al., 2017; Li et al., 2017b,d,e; Luo et al., 2017; Ma et al., 2017d; Jiang et al., 2018; Li H. et al., 2018).

Over the past decades, advance research evidence has shown that lncRNAs regulate most of the cancers by means of controlling the autophagy process and modulating the transcriptional and post-transcriptional ATGs (Mowers et al., 2017; Zhang J. et al., 2017). In this review, we aim to provide an overview of recently identified autophagy-modulating lncRNAs autophagy in cancers, their mechanisms and future directions on therapeutic intervention.

The field of targeted therapy is rapidly developing with advanced genetic study and their successful applications. Previously considered junk molecules such as lncRNAs are now extensively studied to be established as novel diagnostic biomarkers and therapeutic targets. The biological and physiological roles of autophagy-modulating lncRNAs in carcinogenesis are being unveiled recently. The expression of lncRNAs greatly impacts on the extent of autophagy at different carcinogenic stages, mostly in advanced metastatic stages. A number of research articles suggested that lncRNAs induce or suppress autophagy through ATGs and their signaling pathways (**Table 1**). The complex process of autophagy modulation by expressions of lncRNAs may suppress or promote carcinogenesis under diverse physiological conditions (**Figure 3**). Here we described below the recently characterized lncRNAs and their mechanisms through inducing or inhibiting autophagy in different cancers. We also summarized some autophagymodulating lncRNA which worth further investigation in the field of cancer.

### AUTOPHAGY-INDUCING LncRNAs IN CANCER

Majority of the autophagy-modulating lncRNAs have a positive relationship with the induction of autophagy. Hence, increase expression of these lncRNA in tumor induced autophagy and decrease expression of these lncRNA suppressed autophagy. Some examples are listed as follow.

#### HOX Transcript Antisense RNA

HOX transcript antisense RNA contains 2158 nucleotides and was discovered by Rinn et al. (2007). It is located on intergenic

FIGURE 1 | Global cancer impact in (A) 2012 and (B) 2015. In 2012, the death toll 8.2 million from 14 million new cases. It raised gradually to 8.8 million in 2015. The growth of cancer death was more than 7% from 2012 to 2015. The most common cancers were Lung cancer, liver cancer, stomach cancer, colorectal cancer, esophageal cancer, and breast cancer was a major threat of women cancer death worldwide. These figures were based on data published by WHO and GLOBOCAN (Ferlay et al., 2015; GBD 2015 Risk Factors Collaborators, 2016; WHO, 2018).

space of HOXC11 and HOXC12 in chromosome 12q13.13 (Rinn et al., 2007). Abnormal expression of HOTAIR has been noticed for most of the cancers including, BT, BRC, CRC, GC, LVC, NSCLC, OVC, and PNC (Loewen et al., 2014; Zhou et al., 2014; Miao et al., 2016; Gerardo et al., 2017). Several research groups have reported HOTAIR association in different cancers evolutionary processes including, EMT, TNM, prognosis, drug resistance, metastasis, DFS, OS, and tumor development (Loewen et al., 2014; Zhou et al., 2014). Recently, a growing number of studies have revealed HOTAIR contexts in the

#### TABLE 1 | List of autophagy-modulating lncRNAs and their roles in various cancer.


BDC, bladder cancer; BRC, breast cancer; CRC, colorectal cancer; ENDC, endometrial cancer; GC, gastric cancer; GMEC, glioma microvascular endothelial cancer; HCC, hepatocellular carcinoma; LC, lung cancer; MM, multiple myeloma; NSCLC, non-small cell lung cancer; OSTS, osteosarcoma; OVC, ovarian cancer; PNC, pancreatic cancer; PPTC, papillary thyroid cancer; PTC, prostate cancer; TNM, tumor node metastasis.

regulation of autophagy, cancer progression, and drug resistance (Yang et al., 2016; Bao et al., 2017; Sun et al., 2017). Liu's research group proved that, upregulated HOTAIR in HCC cells and tissues induce autophagy by promoting two major ATG3 and ATG7 (Yang et al., 2016). It is also proved that, HOTAIR expression increase along with STAT3 and ATG12 (key of autophagosome formation) through suppressing cancer suppressing micro RNA miR-454-3p in chondrosarcoma (Bao et al., 2017). Sun et al. (2017) revealed that HOTAIR abundancy in ENDC cells significantly induce autophagy which controls the development of chemo-resistance toward cisplatin through the expression of Beclin-1 and P-glycoprotein.

### Metastasis-Associated Lung Adenocarcinoma Transcript 1

Metastasis-associated lung adenocarcinoma transcript 1 is located on chromosome 11q13, containing over 8.7 kb nucleotides and was first identified in NSCLC since 2003 (Ji et al., 2003). It has been proven that MALAT1 plays significant roles in the development, proliferation, invasion, and metastasis of BDC, BRC, CRC, HCC, LC, NSCLC, and osteosarcoma (OSTS) (Ji et al., 2003; Gutschner et al., 2013; Hou et al., 2017; Li et al., 2017c; Zuo et al., 2017; Xiong et al., 2018). Up-regulated MALAT1 promote proliferation and metastasis of PTC cells and tissues (Li L. et al., 2016) via inducing autophagy. To facilitate the process, MALAT1 interacts with RNA binding protein HuR to activate autophagy via controlling post-transcriptional effects of cytotoxic granule-associated RNA binding protein TIA1 (Li L. et al., 2016). It has also been postulated that, aberrant expression of MALAT1 modulate autophagy in various cancers including glioma, GC, HCC, and RTB by controlling micro RNAs miR-216b, miR-101, miR-124, and miR-23b-3p (Yuan et al., 2016; Fu et al., 2017; Huang X. et al., 2017; YiRen et al., 2017). To maintain homeostasis of the cancer micro-environments, up-regulated MALAT1 induce conserved autophagy process directly or indirectly to take part in the progression of chemo-resistance and multi-drug resistance (Yuan et al., 2016; YiRen et al., 2017). More recently, YiRen et al. (2017) revealed that MALAT1 regulate GC progression and autophagy-mediated chemo-resistance via controlling micro RNA miR-23b-3p (YiRen et al., 2017). Gao D. et al. (2017) demonstrated that MALAT1 is highly expressed in MM along with HMGB1 to promote carcinogenesis by significantly expressing two key autophagy regulatory proteins LC3B and Beclin 1 (The mammalian ortholog of yeast ATG6). Their in vivo investigation suggests that knockdown of MALAT1 would be an effective target of MM growth inhibition by autophagy suppression (Gao D. et al., 2017).

#### Plasmacytoma Variant Translocation 1

Plasmacytoma variant translocation 1 was first identified in murine leukemia virus-mediated T lymphomas. It contains 1716

nucleotides and is located on chromosome 8q24.21 (Zeidler et al., 1994). After its discovery, the roles of PVT1 have been identified in various cancers, including BDC, BRC, LC, MPM, and NSCLC (Cui et al., 2016; Li et al., 2017d; Lu et al., 2017; Zhang X.W. et al., 2017; Zhou et al., 2017; Guo et al., 2018; Li H. et al., 2018). Ma and coworkers proved that PVT1 is significantly upregulate in GMEC and promote Atg7 and Beclin-1 expression. They reported that excessive endothelial cell proliferation and migration is mediated by PTV1/Atg7/Beclin-1 (Ma et al., 2017c). Huang and co-workers revealed that expression of PVT1 directly activates ULK1, an autophagy activating protein, in PDA) cells, patients sample, and in vivo xenograft model. PVT1 promotes pathogenesis by regulating miR-20a-5p (Huang et al., 2018). Thus, PVT1/ULK1/autophagy/miR-20a-5p may lead to being a novel therapeutic target of PDA.

#### H19

Maternally expressed non-protein coding transcript H19 is lying on the imprinted region of chromosome 11p15.5 and it is 2.3 kb in length (Cui et al., 2002). H19 is transcribed by RNA polymerase II and dysregulation of H19 is associated with BRC, CRC, GC, Glioblastoma, HCC, HNC, LC and NSCLC (Hu et al., 2016; Matouk et al., 2016; Chen et al., 2017b; Luo et al., 2017). Brannan et al. (1990) first discovered H19 as a riboregulator. To date, numerous evidence has been established for H19's association in various human cancers through distinctive molecular pathways (Chen T. et al., 2016;Chen et al., 2017b). Expression of H19 was found to be increased in both PPTC cells and tissues along with estrogen receptor β which may trigger autophagy through regulating ROS and ERK1/2 pathways. Higher expression of H19 promotes PPTC pathogenesis where further investigation may lead to better understanding of PPTC carcinogenesis through H19/autophagy regulation (Li M. et al., 2018).

### Others

Increasing number of autophagy-modulating lncRNAs are being identified but some of them attracted less attention. Wang et al. (2014) described, BANCR activate autophagy and contribute to proliferation and apoptosis of both PPTC cells and tissues. The overexpression of BANCR promote conversion of LC3- II/LC3-I, activated autophagy promote cells growth and reduce apoptosis in G1 phase (Wang et al., 2014). LncRNA-p21 is a hypoxia-responsive intergenic non-coding RNA which is highly expressed in hepatoma and glioma (I¸sın et al., 2015; Shen et al., 2017). Overexpression of lncRNA-p21 is associated with autophagy induction in hepatoma and glioma cells through HIF-1/Akt/mTOR/P70S6K pathways, resulting in excessive proliferation, motility, reduced apoptosis, and reduced radiosensitivity. Therefore, knockdown of lncRNA-p21 is a new target of radiotherapy as its knockdown potentially alter the molecular events and increase radiosensitivity of hypoxic tumor cells (Shen et al., 2017). HNF 1A-AS1 is located on chromosome 12. It is associated with larger tumor size, advanced TNM stage, excessive growth and apoptosis process of HCC cells and tissues through sponging tumor suppressor miR-30b-5p and inducing autophagy (Liu et al., 2016). Yu et al. (2018), recently noticed LCPAT1 lncRNA in LC which is directly regulated with autophagic flux. The overexpression of LCPAT1 and LC3 was found to be in both LC cells and tumor samples which accelerate the autophagic flux formation to promote carcinogenesis. Whereas, knockdown of LCPAT1 can significantly reduce in vivo tumor size by reducing LC3, ATG3, ATG5, ATG7, ATG12, ATG14, and Beclin1expression (Yu et al., 2018). Chen and coworkers shown that PTENP1 is a lncRNA which is downregulated in HCC and SB based hybrid baculovirus vectors mediated insertion of PTENP1 potentially work as targeted anti-tumor agent in HCC cells by reducing proliferation and migration by activating autophagy in PI3K/AKT pathways (Chen et al., 2015).

### AUTOPHAGY-INHIBITING LncRNAs IN CANCER

Some lncRNA has an inverse relationship with autophagy and one example that has been studied extensively is MEG3. MEG3 is an imprinted gene which was first identified in 2000 (Miyoshi et al., 2000). It contains ∼1600 nucleotides and is found in 14q32.3 chromosomal position (Ma et al., 2018). Extensive research demonstrated that MEG3 expression is significantly reduced in cancer and it affects the proliferation, migration, and metastasis of most cancers including BRC, CRC, GC, glioma, HCC, LC, NSCLC, and PNC (Gong and Huang, 2017; He et al., 2017; Wang P. et al., 2017; Wei and Wang, 2017; Zhang C.Y. et al., 2017; Ma et al., 2018). Ying et al. (2013) speculated that MEG3 inversely regulate cellular autophagy process via the p53 pathways, and reduced MEG3 induce autophagy to promote BDC proliferation and progression. Down-regulation of MEG3 promote tumorigenesis and progression of epithelial OVC cells proliferation and colony formation through inhibiting autophagy process (Xiu et al., 2017). On the other hand, up-regulation of MEG3 inhibits the expression of autophagy-related proteins LC3, ATG3, and LAMP1 (Xiu et al., 2017). These findings led to the development of MEG3 as a potential biomarker of early diagnosis and treatment of OVC. More recently, Ma et al. (2017a) also proved that MEG3 is association with cisplatin-induced glioma cells death by regulating autophagy.

#### Others

Jiang et al. (2018) explored a novel relationship of CASC2 in temozolomide (chemotherapy drug) resistance of glioma. CASC2 is negatively downregulated with miR-193a-5p in temozolomide resistant glioma tissues and induce autophagy by controlling mTOR expression to promote drug resistance (Jiang et al., 2018). Ma et al. (2017b) have described lncRNA AC023115.3 up-regulated in glioma cells after cisplatin treatment and induce cisplatin-mediated apoptosis by inhibiting autophagy process via miR-26a/GSK3β axis. Wang and colleagues showed that lncRNA CTA (CTA) is significantly downregulated in OSTS cells and cancer tissues in contrast with the adjacent normal tissues. CTA downregulated expression is also associated with the advanced TNM stage, larger tumor size and reduced chemosensitivity of doxorubicin through autophagy process (Wang Z. et al., 2017). Micro-RNA miR-210 is negatively regulated with CTA in OSTS and promote apoptosis of OSTS cells, whilst overexpression of

CTA inhibit autophagy and sensitize to doxorubicin subsequently in both in vitro and in vivo (Wang Z. et al., 2017). PCA3 is a newly identified lncRNA, located on chromosome 9q21- 22 and highly specific for PTC (Popa et al., 2007). He et al. (2016) reported that PCA3 is overexpressed in PTC to promote proliferation, migration, and invasion by sponging miR-1261 through inhibiting PRKD3 and blocking protective autophagy. On the other hand, silencing of PCA3 is able to induce protective autophagy and lessen the PTC progression which could be a novel target of personalized treatment (He et al., 2016). Shan's team established that, silencing of lncRNA POU3F3 significantly reduce CRC cells proliferation, migration, and activate autophagy process by enhancing the expression of autophagy-related genes and proteins Beclin-1, ATG5, ATG7, and LC3 II which is novel therapeutic target of CRC (Shan et al., 2016).

### AUTOPHAGY-MODULATING LncRNAs EITHER INDUCE OR INHIBIT AUTOPHAGY IN CANCER

In the literature, we found some reports describing the same lncRNA but with opposite relationship with autophagy. Here, we include three examples which have been studied extensively.

#### Highly Up-Regulated in Liver Cancer

Highly up-regulated in liver cancer was an lncRNA originally characterized in HCC as a significantly overexpressed lncRNA (Panzitt et al., 2007). HULC contains two exons, 1.6k nucleotides in length, located on chromosome 6p24.3. It is significantly dysregulated in most of the cancers including CRC, GC, HCC, OSTS and PNC (Panzitt et al., 2007; Chen et al., 2017c; Li et al., 2017g; Ma et al., 2017d; Shaker et al., 2017; Yu et al., 2017). A number of groups addressed HULC dysregulation and its molecular mechanisms in various cancer cells proliferation, migration, apoptosis, and metastasis but limited reports have focused on autophagy. Zhao et al. (2014) established that overexpression of HULC is clinically correlated with the developmental process of GC by promoted proliferation, migration, invasion, and reduced cellular apoptosis by inducing autophagy. Xiong et al. (2017) found that HULC overexpression induces autophagy and resulting in reduced chemosensitivity of potent chemo drugs 5-fluorouracil, oxaliplatin, and pirarubicin in HCC cells. Moreover, inhibition of protective autophagy by silencing of HULC sensitize these three drugs activity through controlling sirt1 protein in HCC (Xiong et al., 2017). In the contrary, Chen et al. (2017a) reported that HULC suppresses in vitro apoptosis and in vivo tumor development through autophagy blockage in epithelial ovarian carcinoma. Up-regulated HULC inhibits expression of ATG7, LC3-II, and LAMP1 while activates SQSTM1/p62 to promote carcinogenic events (Chen et al., 2017a).

#### Growth Arrest-Specific 5

LncRNA GAS5 was first identified in 1988. It is a tumor suppressor lncRNA which contains 630 nucleotides and encoded at chromosome 1q25 (Schneider et al., 1988). So far, it is wellestablished that GAS5 plays key roles in diverse molecular functions in cancers (Pickard and Williams, 2015; Ma et al., 2016; Gao Q. et al., 2017). Meta-analysis of GAS5 shown that, it is associated with DFS, OS, LNM and tumor stages (I, II, III, IV) (Gao Q. et al., 2017). GAS5 is well-known for the negative regulation of most cancer cells survival (Song et al., 2014). Zhang and co-workers have demonstrated that down-regulating GAS5 would inhibit autophagy in NSCLC and facilitate drug resistance. Overexpression of GAS5 through vector mediated transfection induced autophagy and promote chemotherapy (cisplatin) response in NSCLC cells (Zhang et al., 2016). Gu et al. (2018) reported that GAS5 expression and autophagy were both downregulated in BRC cells and patients sample GAS5 expression is negatively correlated with depth tumor size, advanced TNM and poor prognosis of diseases. Interestingly, vector-mediated overexpression of GAS5 triggers autophagy and increases LC3, ATG3, and p62 expressions through sponging miR-23a. These findings may be developed into a targeted therapy for BRC through GAS5/miR-23a/ATG3 axis (Gu et al., 2018). However, another experiment conducted by Huo's group also demonstrated reduced GAS5 expression in cisplatin-resistant glioma cell lines. Further investigations on the mechanisms have shown that GAS5 down-regulated glioma cells become resistant to cisplatin by increasing autophagosomes formation (Huo and Chen, 2018).

#### Regulator of Reprogramming

The lncRNA ROR was first identified in pluripotent stem cells. It contains four exons, totally 2.6 kb in length and is located on chromosome 18q21.31 (Loewer et al., 2010; Zhan et al., 2016). ROR has been reported to be involved in isolated cellular processes, including growth, proliferation, migration, apoptosis, autophagy and metastasis of BRC, CRC, GC, HCC, NPC, and PNC cancers (Takahashi et al., 2014; Pan et al., 2016; Wang S. et al., 2016; Zhan et al., 2016; Li et al., 2017a; Peng et al., 2017; Wang Y. et al., 2017). Chen Y.M. et al. (2016) proved that ROR suppress autophagy and gemcitabine-induced cell death (apoptosis) in BRC cells by regulating miR-34a. In another study, however, Li's team showed that ROR is up-regulated in PNC to promote basal autophagy which suppresses PKM2 and reduce chemo-sensitivity (gemcitabine) of cells (Li C. et al., 2016). Li et al. (2017f) also identified ROR overexpression reduce autophagy to increase proliferation, invasion, migration, and tamoxifen resistance in BRC cells and tissues. On the other hand, silencing of ROR effectively increases the sensitivity of tamoxifen, proliferation, and migration by inducing autophagy (Li et al., 2017f).

### DISCUSSION

Advanced technologies have improved our understanding of the role of lncRNAs in cancer. At this stage, it is hard to consider lncRNAs as cancer biomarker because the sensitivity and specificity are still not at the desired level. However, researchers may consider using the autophagy-modulating lncRNAs for the

development of biomarkers and targeted cancer therapy. From the growing research based knowledge on lncRNAs, autophagy and cancers, we may declare that most lncRNAs are involved in tumorigenesis through inducing or inhibiting the autophagy pathway.

Autophagy is a conserved regulatory process which is essential for both normal and cancer microenvironments under stress and hypoxic conditions (Xu et al., 2017). Advances in NGS technologies identified more than 3500 putative lncRNAs and the epigenetic roles of lncRNAs have been identified in many diseases including cancers. The discovery of lncRNAs and their molecular signals in cancers through autophagy process has drawn keen attention by scientists as biomarkers and targets for cancer therapy of lncRNAs/autophagy modulated diseases (Xu et al., 2017). However, autophagy and lncRNAs research is still in its fancy compared to its quantity, limited research methods of relationship establishment studies and inadequate therapeutic and clinical trials. Therefore, a major part is still under investigations. The effects of autophagy-modulating lncRNAs are very controversial from its discovery. Both autophagy and lncRNAs may accelerate the carcinogenesis or suppress the cancers (Wang et al., 2015; Liu et al., 2016; Lu et al., 2017; Ma et al., 2017b; Xu et al., 2017). In addition, the dual roles playing autophagy and lncRNAs made of difficult to comprehend the regulatory mechanisms in cancers (Zhang J. et al., 2017).

The regulatory process of autophagy composed of initiation, phagophore nucleation, autophagosome structure formation, and autolysosomal fusion to degrade the unwanted proteins or cytoplasmic components from the body (Mizushima, 2007). It is now well-established that lncRNAs participate in cellular regulatory autophagy pathways (Xu et al., 2017; Yang et al., 2017; Sun, 2018), for example, H19, BRCA1, MEG3, PTENP1, and MALAT1 are involved in initiation; ROR, loc146880, and AC023115.3 are involved in nucleation; TGFB2-OT1, GAS5, HNF1A-AS1, PCGEM1, and HULC are involved in elongation; Chast and MALAT1 are involved in lysosomal fusion process (Yang et al., 2017). With the perspective of ATGs genes, lncRNAs provided a new paradigm of gene expression in the autophagy pathway. An increasing number of lncRNAs have been identified in cancer which triggers autophagy to either promotion or suppression of carcinogenesis (Xu et al., 2017; Sun, 2018). It is now a precious research question that, does autophagy impact on the expression of lncRNAs to promote or suppress cancer? So far, PVT1 is the only reported lncRNA which is regulated by autophagy in diabetes (Li Z. et al., 2016). To address this question in cancer we are currently working on autophagy regulated lncRNAs and their impact on CRC pathogenesis.

#### REFERENCES


### CONCLUSION AND INSIGHTS

Cancer is the second leading cause of death worldwide. Incidences and deaths are increasing because of poor understanding of the disease, diagnostic techniques, and proper treatments. Progression of treatments remains dissatisfactory because most of the cancer patients were diagnosed after development of metastasis. Recently circulating lncRNAs has been considered for diagnostic and therapeutic purpose even though their mechanisms remain unclear (Qi et al., 2016; Sun, 2017). In this review, we have described the roles autophagyregulated lncRNAs on cancer (**Table 1**), and their impact on autophagy regulation through the distinct network (**Figure 3**). At present, the research on the lncRNAs involvement on autophagy pathways is still in its primary stage. Although every day new lncRNAs are reporting on various cancer type, most of the researchs mainly focused on their impact on cancer pathogenesis. By considering the lncRNAs impact on autophagy mechanisms (Yang et al., 2017), more critical functions underlying autophagy are waiting to be demonstrated. Some autophagy-modulated lncRNAs described in this review are very specific with identical tissue types, for example HULC, PCA3, PVT1 may serve as a potential biomarkers and target therapy in identical cancer types (He et al., 2016; Ma et al., 2017c; Xiong et al., 2017). Therefore, extensive studies are needed to address the interaction between lncRNAs and the complex regulatory autophagy process with selective target genes and signaling transductions for the discovery of new targets, prognostic and diagnostic biomarkers and target therapy.

#### AUTHOR CONTRIBUTIONS

All authors have participated sufficiently in the work to take public responsibility for the content. We have worked together in the conception, drafting and revising the articles to provide an intellectual content of critical importance to the work described and final approval of the version to be published.

### FUNDING

This work was supported by Departmental Start-up and Seed Funding for HL, and Departmental Postgraduate Funds for MK and ST, The Hong Kong Polytechnic University.





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

Copyright © 2019 Islam Khan, Tam and Law. 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.

# miRNA–mRNA Associated With Survival in Endometrial Cancer

*Xiaofeng Xu1†, Tao Liu1,2†, Yijin Wang1,3, Jian Fu4, Qian Yang5, Jun Wu1\* and Huaijun Zhou1\**

*1 Department of Gynecology, The Affiliated Drum Tower Hospital of Nanjing University Medical School, Nanjing, China, 2 Medical College, Nanjing University, Nanjing, China, 3 Medical College, Southeast University, Nanjing, China, 4 Department of Gynecology, Suqian People's Hospital of Nanjing Drum Tower Hospital Group, Suqian, China, 5 Department of Gynecology and Obstetrics, The Pukou Hospital of Nanjing, The Fourth Affiliated Hospital of Nanjing Medical University, Nanjing, China*

Although various factors may contribute to its initiation and progression, the etiology and prognostic factors of endometrial carcinoma (EC) remains not fully understood. We sought to understand the role of changes in transcriptome during the progress of EC by exploring public datasets. The aberrant expression characteristics of EC based on RNA-Seq and miRNAseq data from The Cancer Genome Atlas (TCGA) were analyzed. Kaplan–Meier analysis was performed to assess the relationship between differently expressed genes (DEGs) and patient survival. As a result, 320 out of 4,613 differently expressed mRNAs (DE mRNAs) and 68 out of 531 differently expressed miRNAs (DE miRNAs) with a significantly poorer survival were determined. We predicted eight paired DE miRNAs and DE mRNAs through TargetScan. Patients with three out of the eight paired low rate of miRNA/mRNA (miR-497/EMX1, miR-23c/DMBX1, and miR-670/KCNS1) expression had a significantly poorer survival. Furthermore, the simultaneous presence of these selected low miRNA/mRNA pairs occurred in most patients and resulted in a significantly poorer survival rate. Luciferase reporter assay identified that EMX1 was a direct target of miR-497. Both low expression of miR-497 and overexpression of EMX1 were significantly associated with more advanced clinicopathologic characteristics (stage, tumor status, grade, and histology) besides survival (all P values < 0.05). Multivariate analysis also demonstrated that miR-497 remained an independent prognostic variable for overall survival. In summary, we identified that a series of DE mRNAs and miRNAs, including eight paired DE miRNAs and mRNAs, were associated with survival in EC. Clinical evaluation of downregulated miR-497 and paired upregulated EMX1 confirmed the value of our prediction analysis. The simultaneous presence of low rate of these selected low miRNA/ mRNA pairs (miR-497/EMX1, miR-23c/DMBX1, and miR-670/KCNS1) might have a better prediction value. Therefore, further studies are required to validate these findings.

Keywords: endometrial carcinoma, miR-497, EMX1, survival, TCGA

#### INTRODUCTION

Endometrial cancer (EC) is one of the most prevalent gynecological malignancies in the word (Siegel et al., 2018). The incidence has increased by 1.2% per year from 2005 to 2014, along with the mortality rates, which similarly increased during this period (Smith et al., 2018). Importantly, EC is one of only two common cancers that defy the general trend of improvement in morbidity and mortality, with a worse survival rate today than in the 1970s (Siegel et al., 2018). The American

#### *Edited by:*

*Ge Shan, University of Science and Technology of China, China*

#### *Reviewed by:*

*Andrés Fernando Muro, International Centre for Genetic Engineering and Biotechnology, Italy Feng Wang, Emory University School of Medicine, United States*

*\*Correspondence:*

*Huaijun Zhou zhouhj2007@126.com Jun Wu iamwujun2008@163.com*

*†These authors share first authorship.*

#### *Specialty section:*

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

*Received: 29 September 2018 Accepted: 16 July 2019 Published: 20 August 2019*

#### *Citation:*

*Xu X, Liu T, Wang Y, Fu J, Yang Q, Wu J and Zhou H (2019) miRNA– mRNA Associated With Survival in Endometrial Cancer. Front. Genet. 10:743. doi: 10.3389/fgene.2019.00743*

**213**

Cancer Society (ACS) estimates that 63,230 women will be diagnosed with EC and that more than 11,000 women will die from this disease in 2018 (Siegel et al., 2018). Clearly, the burden of EC is increasing in the USA and worldwide, hence increasing the need to investigate its causes and improve prevention, early diagnosis, and treatment (Binder and Mutch, 2014). Although various endocrine, genetic, and external factors may contribute to its initiation and progression, the etiology and prognostic factors of endometrial carcinoma remain not fully understood.

Previous analysis of the human genome revealed that while ~85% is transcribed, only ~1% is protein-coding mRNA (Dunham et al., 2012; Djebali et al., 2012; Dykes and Emanueli, 2017). With time, more and more attention was being paid to the research of non-coding RNAs, including microRNAs (miRNAs), which have important regulatory roles in cancer cellular biology. Studies demonstrated that mature miRNAs can regulate the expression of their target genes by imprecise complementation to the 3′-untranslated regions (UTRs), 5′-UTRs, and even coding sequences of the mRNAs to repress their translation (Ambros, 2004; Duursma et al., 2008; Sacco and Masotti, 2012). Srivastava et al. reviewed the amount of miRNAs differentially expressed in EC versus normal endometrial tissue, including the increased expression of miR-9, miR-92a, miR-141, miR-182, miR-183, miR-186, miR-200a, miR-205a, miR-222, miR-223, miR-410, miR-429, miR-449, and miR-1228 and downregulation of miR-99b,miR-143, miR-145, miR-193b, and miR-204 (Srivastava et al., 2017). Numerous miRNAs could regulate EC cells by silencing their target genes (Zhou et al., 2012; Chen et al., 2016). Though expression patterns of several miRNAs were found to be associated with the International Federation of Gynaecology and Obstetrics (FIGO) stage, grade, relapse, and nodal metastases in EC (Torres et al., 2013; Canlorbe et al., 2016; Srivastava et al., 2017), few studies focused on the relationship between both miRNAs and their target genes with patient survival.

The Cancer Genome Atlas (TCGA) studies have defined the molecular genetic landscape of EC and highlighted the molecular genetic diversity of both endometrioid and non-endometrioid cancers (Burk et al., 2017). Recently, accompanied by the advent era of sharing information, more and more cancer researches including EC were carried out based on TCGA database (Dellinger et al., 2016; Reyes et al., 2017; Wang et al., 2018; Wu and Zhang, 2018). However, the previous studies mainly focused on the single factor, such as differently expressed mRNAs (DE mRNAs), differently expressed miRNAs (DE miRNAs), or differently expressed long non-coding RNAs (DE lncRNAs) to find the potential etiology and prognostic factor of tumor.

In this study, considering the silence effect to their target genes of miRNAs, we performed survival analysis on both DE mRNAs and DE miRNAs comparing EC and normal samples from TCGA database. As a result, we revealed a group of DE mRNAs and DE miRNAs associated with survival. Furthermore, we identified that low expression of miR-497 and overexpression of its potential target gene Empty Spiracles Homeobox 1 (EMX1) were both related to more advanced clinicopathologic characteristics.

### METHODS

### Data Collection

Expression profiles of RNA-Seq and miRNA-seq for TCGA-UCEC project were downloaded from TCGA official website (https://cancergenome.nih.gov) in July 2018. RNA expression data of 543 EC cases and 35 normal cases were downloaded from the database, while miRNA expression data of 539 EC and 33 normal samples were included. The corresponding clinical information was downloaded from http://www.cbioportal.org. Data were collated and extracted for analysis.

#### Identification of Differently Expressed Genes (DEGs)

We applied the expression profiles to the edgeR package in R language and calculated for differential expression between tumor and normal group samples after normalization and filtration. The adjusted *P* values (adj. *P*) were applied to correct the false-positive results. Then the significant DEGs (adj. *P* < 0.05 and fold-change value larger than 2) were selected out for the next step analysis.

### Survival Analysis

In this study, survival analysis refers to the overall survival (OS) Kaplan–Meier estimate. We performed Kaplan–Meier analysis (R package "survival") to assess survival and relapse difference across cases with DEGs. For each gene, patients were divided into high-expression group and low-expression group according to the median expression level. Based on the survival curves of each group, these upregulated and downregulated DEGs with a poorer survival rate were within our consideration. *P* value < 0.05 was set as the cutoff point.

### Functional and Pathway Enrichment Analysis of DEGs

The Database for Annotation, Visualization and Integrated Discovery (DAVID, https://david.ncifcrf.gov/) is a biological database regularly used to facilitate functional annotation and pathway analysis. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis aims to identify and visualize significantly enriched pathways of molecular interactions, reactions, and relations. Gene Ontology (GO) analysis uses hypergeometric tests to perform enrichment analysis on gene sets (Falcon and Gentleman, 2007). We uploaded selected DEGs associated with survival to DAVID to perform KEGG pathway and GO enrichment analysis. The human genome was selected as the background list parameter, and *P* value < 0.05 was set as the cutoff point.

Gene Set Enrichment Analysis (GSEA) is a computational method that determines whether a predefined set of genes shows significant differences between two biological phenotypes under study (Subramanian et al., 2005). Gene set permutations were performed 1,000 times for each analysis. The nominal *P* value (NOM *P* value) and normalized enrichment score (NES) were used to sort the pathways enriched in each phenotype.

#### Prediction of DE miRNA-Targeted DE mRNAs Associated With OS

In order to improve the validity of our search results, we further excavated the relationship between the selected DE mRNAs and DE miRNAs through TargetScan (http://www.targetscan.org/ vert\_72/). TargetScan predicts biological targets of miRNAs by searching for the presence of conserved sites that match the seed region of each miRNA (Fromm et al., 2015). In brief, we not only searched for the presence of conserved sites among these DE mRNA that matched the seed region of DE miRNA but also searched for DE miRNA containing matched seed region for DE mRNA. These poorly conserved sites were excluded. Only these enrolled miRNA–mRNA pairs with opposite expression levels in EC samples as compared with normal tissues both presented in our dataset were adapted for further study.

#### Cell Culture and Luciferase Reporter Assay

Cell culture and Luciferase reporter assay were very similar with those in our previous study (Zhou et al., 2012). Briefly, 293T cell and human endometrial cancer Ishikawa cell were cultured in Dulbecco's modified Eagle's medium (DMEM; Gibco, USA) supplemented with 10% fetal bovine serum (FBS; Gibco) and penicillin (100 U/ml). The 300-nt-long 3′-UTR (wild type or mutant) of EMX1 containing the predicted conserved binding sites for miR-497 was cloned into H306 pMIR-LUC vectors purchased from Obio Technology (Shanghai, China). Cells were plated at a density of 1 × 105 in 12-well plate. After 24 h, the pMIR-LUC reporters were co-transfected with either miR-497 mimics or control using Lipofectamine 3000. Luciferase activity normalized to the Renilla Luciferase was measured by the Dual-Luciferase assay (Promega, USA) according to the manufacturer's instructions after 48 h on the Luminometer (Promega, USA). The assay was repeated three times.

### RNA Extraction and Quantitative RT-PCR

Total RNA was extracted from cells with isolator reagent (Vazyme, China). After measurement of the RNA concentration, cDNAs were generated from reverse transcription with the HiScript II 1st Strand cDNA Synthesis Kit (Vazyme, China). The expression level of miR-497 was measured according to the instructions of the ChamQTM Universal SYBR qPCR Master Mix kit (Vazyme, China) using the ABI-7300 Real-Time PCR Detection System (Applied Biosystems, USA). The bulgeloopTM miRNA Primer Sets (one RT primer and a pair of qPCR primers) specific for miR-497 were purchased from RiboBio (Guangzhou, China). The level of miR-497 was normalized to U6. Fold changes were calculated using the 2−ΔΔCt method. Each plate was run in triplicate.

#### Statistical Analysis

Logistic regression and *t* test were used to evaluate the relationship between gene expression and clinical–pathologic features. For gene expression level, the cutoff value was determined by its absolute median expression level (high- and low-expression groups). For the analysis of survival for patients with paired miRNA/mRNA, the median rate of miRNA/mRNA expression was used to divide groups for low and high rates. In logistic regression, the absolute gene expression level (high- and lowexpression group) was used as categorical dependent variable, while clinical–pathologic feature was used as independent variable. Among all the clinical–pathologic features, age and BMI were calculated as continuous variables and the rest as categorical variables. Kaplan–Meier method and Cox regression were used to analyze clinicopathologic characteristics associated with OS. The influence of gene expression on survival along with other clinical characteristics was determined by multivariate Cox analysis. *P* value < 0.05 was set as the cutoff point. \**P* < 0.05, \*\**P* < 0.01, and \*\*\**P* < 0.001.

### RESULTS

#### Identification of DE mRNAs and DE miRNAs With a Poorer Survival in EC

The RNA-Seq profile data of 543 EC and 35 normal cases were downloaded from TCGA database along with the miRNA-seq data of 539 EC and 33 normal samples. Among them, 533 patients shared two expression profiles (**Figure 1**). Under the threshold of adj. *P* < 0.05 and fold-change value larger than 2, a total of 4,613 DE mRNAs (3,221 upregulated and 1,392 downregulated) and 531 DE miRNAs (374 upregulated and 157 downregulated) were identified in EC compared with normal samples. The heatmaps and volcano plots are shown in **Supplementary Figure 1**.

To have a better understanding of the relationship between DEGs and patient survival, we conducted the "survival" package in R language to draw Kaplan–Meier curves according to the median expression level of DEGs among 533 patients. These upregulated and downregulated genes with a significantly poorer survival (*P* < 0.05), respectively, were determined. A total of 320 (280 upregulated and 40 downregulated) out of 4,613 DE mRNAs and 68 (43 upregulated and 25 downregulated) out of 531 DE miRNAs were chosen for further study (**Supplementary Table 1**). Kaplan– Meier curves of the partial genes between the high-expression group and low-expression group are shown in **Figure 2**.

#### Functional and Pathway Enrichment Analysis

The selected 320 DE mRNAs associated with survival were uploaded to DAVID to perform KEGG pathways and GO enrichment analysis. KEGG analysis revealed that these DEGs were mostly enriched in cell cycle, neuroactive ligand–receptor interaction, microRNAs in cancer, oocyte meiosis, and serotonergic synapse signaling pathways (**Figure 3A**). The top five enriched GO biological process (BP) terms included positive regulation of transcription from RNA polymerase II promoter, cell division, negative regulation of transcription from RNA polymerase II promoter, cell proliferation, and mitotic nuclear division. Nucleus, cytoplasm, nucleoplasm, microtubule, and neuronal cell body were the five most enriched GO terms for cellular component (CC). The top five enriched GO molecular function (MF) terms were protein binding, DNA binding, ATP binding, transcription factor activity,

FIGURE 2 | A total of 320 out of 4,613 DE mRNAs and 68 out of 531 DE miRNAs associated with a poorer survival rate were chosen. The eight plots show Kaplan–Meier curves of the partial DE miRNAs and DE mRNAs between cases of high-expression group and low-expression group divided according to the median expression level. (A and B) Upregulated miRNA-93 and miRNA-224. (C and D) Downregulated miRNA-4670 and miRNA-770. (E and F) Upregulated TPX2 and NEIL3. (G and H) Downregulated CTSK and KIAA1755.

and sequence-specific DNA binding (**Figures 3B**–**D**). The most significantly enriched pathways and enrichment terms are also shown in **Supplementary Table 2**.

#### Prediction Roles of Identified DE miRNAs and Their Potential Targeted mRNAs Associated With Survival

In order to better understand the role of DE miRNAs and their potential targeted DE mRNAs related to survival, we made target predictions through TargetScan. These genes containing conserved target sites, which matched the seed regions of miRNAs, were in our consideration. Besides, we only accepted these DE miRNAs and DE mRNAs with opposite expression levels in view of the silence effect of miRNA targeting. As a result, we predicted that 5 out of 68 DE miRNAs could interact with 8 out of 320 DE mRNAs (**Figure 4**), with one upregulated miRNA paired with two downregulated mRNAs and four downregulated miRNAs paired with six upregulated mRNAs (**Table 1**). The survival curves are also shown in **Figure 5**.

We further tried to assess the differences between high and low rates of expression levels for the eight miRNA/mRNA pairs under the cutoff of median rate. As shown in **Figure 6A**, patients with low rate of three paired miRNA/mRNA (miR-497/ EMX1, miR-23c/DMBX1, and miR-670/KCNS1) expression had a significantly poorer survival, while the rest seemed no different (data were not shown). Furthermore, under each cutoff of median rate, there were 156 patients carrying low rates of both miR-497/EMX1 and miR-23c/DMBX1 expression, 135 patients carrying low rates of both miR-497/EMX1 and miR-670/ KCNS1 expression, 160 patients carrying low rates of both miR-670/KCNS1 and miR-23c/DMBX1 expression, and 92 patients carrying low rates of all the three miRNA/mRNA pairs (**Table 2**). Surprisingly, these patients with the simultaneous presence of these two or three low miRNA/mRNA pairs had a significantly poorer survival rate (**Figure 6B**). These results strongly verified the validity and prediction capacity of our analysis.

## Clinical Evaluation of miR-497 and EMX1 Expression in EC

#### Patient Characteristics

We downloaded the clinical data of 533 primary EC from http://www.cbioportal.org. Patient characteristics are shown in **Table 3**. Median age at diagnosis of our study cohort was 64 year old (range 31–90 years old); 72.6% of the study group were white, with the rest scattered among other races. Most of the patients were endometrioid endometrial adenocarcinoma (EA; *n* = 402, 75.4%), 20.5% (*n* = 109) were serous, and 22 (4.1%) were mixed serous and endometrioid. The histopathologic distribution of EC was well differentiated (18.2%), moderately differentiated (22.5%), and poorly differentiated (59.3%). The cancer status included 419 cases without tumor (84.5%) and 77 cases with tumor (15.5%). In all patients, 32.7% had deep myometrial invasion, and 14.2% had positive peritoneal washings. About 16.8% of all patients had positive pelvic lymph nodes, while 10.3% had positive para-aortic lymph nodes. Stages I, II, III, and IV comprised 62.9%, 9.0%, 22.8%, and 5.3%, respectively. Median


FIGURE 4 | Eight predicted consequential pairings of DE miRNAs and target regions of DE mRNAs associated with survival. These enrolled miRNA–mRNA pairs not only contained conserved binding sites but also had opposite expression levels in EC samples as compared with normal tissues.


*\*P value < 0.05 represented a significantly different OS rate between patients grouped according to the median level of gene expression.*

*The bolded miR-497/EMX1 was selected as a representative of eight miRNA/mRNA pairs for subsequent study.*

follow-up for subjects alive at last contact was 23.5 months (range 0–190 months).

#### miR-497, EMX1 Expression, and Association With Clinicopathologic Variables

Univariate analysis using logistic regression revealed that miR-497 expression as a categorical dependent variable (based on median expression value) was associated with adverse prognostic clinical pathological features (**Table 4**). Decreased miR-497 expression in EC was significantly associated with high grade (OR = 0.228 for G1, G2 vs. G3), stage (OR = 0.470 for I vs. II, III, and IV), histology (OR = 0.175 for EA vs. non-EA), tumor status (OR = 0.391 for tumor free vs. with tumor), pelvic lymph node metastasis (OR = 0.294 for positive vs. negative), para-aortic lymph node metastasis (OR = 0.232 for positive vs. negative) and peritoneal washings (OR = 0.510 for positive vs. negative) (all *P* values < 0.05, **Figures 7A**–**D**).

On the contrary, high EMX1 expression was associated with poor prognostic clinicopathologic characteristics (**Table 5**). Overexpressed EMX1 in EC was associated with clinically advanced stage (OR = 1.668 for I vs. II, III, and IV), high grade (OR = 2.325 for G1, G2 vs. G3), histology (OR = 3.060 for EA vs. non-EA), tumor status (OR = 2.342 for tumor free vs. with tumor), and para-aortic lymph node metastasis (OR = 2.547 for positive vs. negative) (all *P* values < 0.05, **Figures 7E**–**H**).

These results suggested that EC patients with low expression of miR-497 and high expression of EMX1 were prone to progress to a more advanced stage than those with high expression of miR-497 and low expression of EMX1.

#### Survival Outcomes and Multivariate Analysis

Kaplan–Meier survival analysis showed that EC with low expression of miR-497 had a worse prognosis than that with high expression (**Figure 5G**). The univariate analysis revealed that reduced miR-497 correlated significantly with a poor OS (HR: 0.536; 95% CI: 0.345–0.831; *P* = 0.005).

Meanwhile, Kaplan–Meier survival analysis suggested that patients with high EMX1 expression also had a poorer prognosis than those with low EMX1 expression (**Figure 5I**). And, univariate analysis revealed that high EMX1 expression correlated significantly with a poorer OS rate (HR: 1.563, 95% CI: 1.019–2.396; *P* = 0.041).

presence of rates for more than one miRNA/mRNA pair among these three selected pairs. The median rate of miRNA/mRNA expression was used to divide groups

TABLE 2 | Patients with the simultaneous presence of low rates of miRNA/

mRNA pairs.

into low and high rates.


*\*P value represented a significantly different OS rate between patients with low and high rates of miRNA/mRNA expression.*

Other clinicopathologic factors associated with poor survival include high grade, histology, advanced stage, myometrial invasion, peritoneal washings, and lymph node metastasis (**Table 6A**).

In multivariate analysis (**Table 6b**), miR-497 remained an independent prognostic variable for OS, with an HR of 0.587 (CI: 0.363–0.951, *P* = 0.030), along with myometrial invasion.

#### EMX1 Was a Direct Target of miR-497

To explore whether EMX1 was a direct target of miR-497, we first detected the miR-497 levels in 293T and Ishikawa cells. As shown in **Figure 8A**, the miR-497 level in Ishikawa cell was very low, while it was much higher in 293T cell. We next performed a Dual-Luciferase reporter assay. The binding sites for miR-497 with EMX1 wildtype 3′-UTR (EMX1 3′-UTR) and EMX1 mutant 3′-UTR (EMX1 mu3′-UTR) are shown in **Figure 8B**. There was no loss of luciferase activity in Ishikawa cells with co-transfection of miR-497 mimics and mutated 3′-UTR of EMX1 plasmid, while a significant luciferase activity decrease was observed in cells co-transfected with miR-497 mimics and EMX1 3′-UTR plasmid (**Figure 8C**, \*\*\* *P* = 0.0001). We also performed the luciferase assay in 293T cells, which showed an increased relative luciferase activity of the construct having the mutated binding site compared with that of the wild-type site (**Figure 8D**, \*\* *P* < 0.01). These data indicated that EMX1 was a direct target of miR-497.

### GSEA Identified EMX1-Related Oncogenic Signaling Pathways

The GSEA method is used to deal with continuous data and identify gene sets that are enriched at the top (overexpressed vs. control) or bottom (underexpressed) of a ranked gene list (Subramanian et al., 2005). We performed GSEA with the ordered list of preliminary screening DEGs according to their correlation with EMX1 expression. MSigDB Collection (c6.all.v6.2.symbols.gmt), which represents signatures of cellular pathways often dysregulated in cancer, was applied to our GSEA analysis at the phenotype of EMX1 expression level. Twenty-one signaling pathways were significantly enriched based on their NES (NOM *P* value < 0.05, FDR *q*-value < 0.25, **Table 7**). **Supplementary Figure 2** shows that part of those oncogenic gene-involved cellular pathways, including PIGF, SRC, JNK KRAS.AMP.LUNG, PTEN, and E2F3, were differentially enriched in EMX1 high-expression phenotype.

### DISCUSSION

In the recent years, more and more evidence has revealed that miRNAs play important roles in the development and progression of tumors. They may serve as molecular indicators of prognosis



TABLE 4 | Univariate miR-497 expression\* association with clinicopathologic characteristics (logistic regression).


*\*Categorical dependent variable, greater or less than the median expression level.*

and targets for oncotherapy (Jonas and Izaurralde, 2015). In endometrial cancer, a number of studies have looked at the miRNA profiles using tumor tissues or blood samples with a goal to identify disease-specific biomarkers (Srivastava et al., 2017). However, previous studies usually focused on one or two specific miRNAs and their target genes based on their potential to regulate diverse biological processes. Few researchers carried out studies on the relationship between whole miRNA profile and their potential target mRNAs based on their impact on patient survival.

A miRNA-based research data source of which was also a TCGA-UCEC project was carried out not long ago (Wang et al., 2018). It identified miRNAs that were correlated with the occurrence and progression of EC and established a six-miRNA (miR-15a. MIMAT0000068, miR-142. MIMAT0000433, miR-142. MIMAT0000434, miR-3170. MIMAT0015045, miR-1976. MIMAT0009451, and miR-146a. MIMAT0000449) expression signature as a predictor for the OS of patients with EC. In this study, the six-miRNA model was based on these 144 DE miRNAs resulting only from 15 patients with corresponding miRNA expression of the paired adjacent tissues. Another miRNA-based study on TCGA-UCEC project focused on the difference of miRNA profile between metastatic and nonmetastatic ECs (Zhu et al., 2018). It revealed that four miRNAs (miR-1247 miR-3200, miR-150, and miR-301b) were differently expressed between two groups, miR-1247 was associated with metastasis of EC to the lung, and miR-3200 is associated with the clinical stage of EC. The researchers also performed functional enrichment analysis with these predicted potential target genes of the four miRNAs, which showed that they might be involved in multiple pathways of cancer, including the Wnt, NOTCH, and TGF-β signaling pathways and signaling pathways regulating pluripotency of stem cells. However, miRNAs can target multiple genes, and a single gene can be targeted by multiple miRNAs (Ambros, 2004). Unilateral research is not enough to illustrate the problem.

In this study, for the first time, we performed Kaplan–Meier analysis for all the DE mRNAs and DE miRNAs in a larger number of EC cases from TCGA project with the help of bioinformation technology. As a result, 320 out of 4,613 DE mRNAs and 68 out of 531 DE miRNAs with a significantly poorer survival were determined. For the selected DE mRNAs associated with survival, KEGG pathway and GO enrichment analysis revealed that these selected DE mRNAs were significantly enriched in several functional pathways, including microRNAs in cancer. We further made target predictions through TargetScan and found that 5 out of 68 DE miRNAs could interact with eight DE mRNAs from 320 DE mRNAs. However, since the cutoff value for dividing groups (high expression and low expression) was based on the absolute median, most *P* values for different OS rates of the selected paired genes were of borderline statistical significance. Xie et al. (2018) used median mRNA expression level of ALKBH1 as a separation for high- and low-expression groups based on another TCGA project of glioblastoma (*n* = 488), which also resulted in a minor difference (*P* = 0.0386) and provided evidence for ALKBH1 to be a potential therapeutically targetable node. Another way to perform analysis was by using the computing cutoff expression

FIGURE 7 | Association with DEG expression and clinicopathologic characteristics, including has-miR-497 expression associated with (A) stage, (B) tumor status, (C) grade, and (D) histology and EMX1 expression associated with (E) stage, (F) tumor status, (G) grade, and (H) histology. *P* values were calculated using *t* test.

TABLE 5 | Univariate EMX1 expression\* association with clinicopathologic characteristics (logistic regression).


*\*Categorical dependent variable, greater or less than the median expression level.*

level for the best separation with the smallest *P* value on survival and different numbers in two groups for each gene, which was also our initial design and resulted in a much larger number of DE mRNAs and miRNAs related to survival. To narrow the results of genes associated with survival, we finally chose the median truncation value, though some genes with potential predictability may be lost in our study.

Since each group of patients with a single miRNA or mRNA was made independently and their composition might be different, it would be interesting to determine differential expression of more than one miRNA or mRNA simultaneously in the patients. We tried to assess the differences between high and low rates of expression levels for the eight miRNA/ mRNA pairs. As a result, patients with a low rate of three paired miRNA/mRNA (miR-497/EMX1, miR-23c/DMBX1, and miR-670/KCNS1) had a significantly poorer survival, while the rest seemed no different. As the presence of a single miRNA/mRNA TABLE 6 | a. Overall survival and associations with clinicopathologic characteristics using Cox regression. b. Multivariate survival model after variable selection.


pair is clearly not sufficient to predict the outcome of a patient having EC, we further verified the simultaneous presence of low rate of more than one miRNA/mRNA pair among these three selected pairs in the same group of patients. Surprisingly, the

TABLE 7 | Gene Set Enrichment Analysis in EMX1 phenotype.


*NES, normalized enrichment score; NOM, nominal; FDR, false discovery rate. Gene sets with NOM P val < 0.05 and FDR q-val < 0.25 are considered significant.*

simultaneous presence of these selected low miRNA/mRNA pairs occurred in most EC patients and resulted in a significantly poorer survival rate, which strongly verified the validity and prediction capacity of our analysis. The experimental luciferase reporter assay confirmed that EMX1 was a direct target of miR-497. Clinical evaluation was assessed on miR-497 and EMX1 expression levels. Interestingly, they had an opposite significant relationship with several clinicopathologic characteristics besides survival. Multivariate analysis also demonstrated that miR-497 remained independent prognostic variables for OS. These data also provided evidence for the value of our prediction analysis in EC.

Among 68 DE miRNAs related to a poorer survival, 43 miRNAs were upregulated and 25 were downregulated. In 2013, Torres et al. conducted a study that aimed to reveal the relationship between DE miRNAs and EC (Torres et al., 2013). They found that upregulated miR-1228 was significantly associated with a poorer survival, which was consistent with our research result. Another research revealed that lower expression of miR-101, miR-10b, miR-139-5p, miR-152, miR-29b, and miR-455-5p was significantly correlated with poor OS and that decreased expression of miR-152 was a statistically independent risk factor for OS in EC (Hiroki et al., 2010). Mitamura et al. found that miR-31 was significantly upregulated in the EC patients with a high risk of recurrence compared with that observed in the lowrisk patients, and this higher expression correlated with a poorer progression-free survival (Mitamura et al., 2014).

miRNAs mainly function and result in silence effect at posttranscriptional level by base pairing with the 3′-UTR of their target mRNAs completely or incompletely (Moran et al., 2017). In the present study, we predicted eight potential target DE mRNAs for the five DE miRNAs, including miR-211, miR-23c, miR-670, miR-497, and miR-4770. To our knowledge, there is no basic experimental evidence for the eight pairs of miRNAs and mRNAs predicted. One of the mostly studied miRNAs, miR-497, was recognized as a tumor suppressor in many cancers (Zhao et al., 2013; Du et al., 2015; Zhao et al., 2015; Yang et al., 2016), which was also consistent with our result in EC. During our manuscript preparation, a meta-analysis (Feng et al., 2018) on the prognostic role of miR-497 in different cancer patients revealed that high-expression levels of miR-497 are less possible to have lymph node metastasis and have better overall survival, which indicated that miR-497 might be a potential biomarker and could be used to predict the better prognosis of different cancer types. A recent study showed that miR-497 negatively regulated glioma cells by targeting oncogene Wnt3α and that reduced expression of miR-497 was associated with poor disease-free and overall survival rates (Lu et al., 2018). It could also suppress clear cell renal cell carcinoma by targeting PD-L1, which was an immune-related oncogene (Qu et al., 2018). miR-497 could target SERPINE-1 and induce reversion of epithelial-to-mesenchymal transition in cutaneous squamous cell carcinoma (Mizrahi et al., 2018). However, these targets of miR-497 did not appear as a DE mRNA associated with survival in the present study. EMX1, a potential target of miR-497 predicted through TargetScan, also had a very close relationship with EC clinicopathologies and patient survival, which was in contrast to the miR-497 in our study. Asada et al. revealed that a high quartile of EMX1 methylation level had a significant univariate HR and a multivariate-adjusted HR of developing authentic metachronous gastric cancers (Asada et al., 2015). Its methylation level was also differently expressed in hepatocellular carcinoma (HCC), which showed a potential role in the development of HCC (Sun et al., 2018). However, most of the research on EMX1 was focused on brain development (Lim et al., 2015; Kobeissy et al., 2016). The relationship between miR-497/EMX1 and cancer needs further study.

Testis-expressed 19 (TEX19) was another potential target gene of miR-497 in our study. Zhong. et al. revealed that TEX19 exhibited increased expression in high-grade tumors and might represent a novel cancer-testis gene related to the progression of bladder cancer (Zhong et al., 2016). TEX19 was also required to drive cell proliferation in a range of cancer cell types, and its expression was linked to a poor prognosis for breast cancer, kidney cancer, prostate cancer, and glioma cancer (Planells-Palop et al., 2017). However, further study on its role in EC is needed.

Bu Y. et al. identified that miR-211 directly targets to Bmal1 and Clock in Burkitt's lymphoma, thereby suppressing both circadian oscillation and ongoing protein synthesis to facilitate tumor progression (Bu et al., 2018). miR-23c, as a target of lncRNA MALAT1, could directly repress its target ELAVL1 and inhibit hyperglycemia-induced cell pyroptosis (Li et al., 2017). Shi C. et al. showed evidence that miR-670 could induce cell proliferation in hepatocellular carcinoma by targeting PROX1 (Shi and Xu, 2016), while our study predicted that it might play an important role in EC progression *via* targeting Potassium Voltage-Gated Channel Modifier Subfamily S Member 1 (KCNS1), which was also a DE mRNA associated with survival in our study. Previous studies on KCNS1 mainly focused on pain (Tsantoulas et al., 2018; Wonkam et al., 2018). Besides the downregulated expression level of KCNS1 in metastatic breast carcinoma (Savci-Heijink et al., 2016), little research has been carried out on the relationship between KCNS1 and cancer.

DNA methyltransferase 3B (DNMT3B), one of the eight paired genes in our study, was a methyltransferase responsible for *de novo* DNA methylation. A previous study also showed that it was significantly upregulated in both grade I and grade III ECs as compared with normal controls (Jin et al., 2005). DNMT3B-involved divergent DNA methylation pathways and protein synthesis required for posttranscriptional regulation may be implicated in the development of type I and type II ECs (Xiong et al., 2005a; Xiong et al., 2005b). However, little is known about the relationships between EC and the rest of the potential miRNA-targeted DE mRNAs, including growth differentiation factor 7 (GDF7), sarcoglycan zeta (SGCZ), diencephalon/ mesencephalon homeobox 1 (DMBX1), and fatty acid desaturase 6 (FADS6). Further research is also needed.

There were emerging researches on EC using TCGA data. Studies revealed that L1CAM (Dellinger et al., 2016) and TFAP2B (Wu and Zhang, 2018) were the two DEGs associated with advanced clinicopathologic characteristics and independent predictors for survival in EC based on TCGA-UCEC data. However, the massive narrowing down process for the DEG screening was not shown. Another study focused on the mutation-expression profile and looked for guidance for EC drug discovery (Wong et al., 2016). Thus, while these studies and ours were all based on TCGA data for EC, the different focus and data analysis approaches led to the discovery of different aspects of the disease.

In this study, we identified a series of DE mRNAs and DE miRNAs associated with survival in EC. Furthermore, we predicted eight pairs of DE miRNAs and their potential target DE mRNAs related to survival. Clinical evaluation of downregulated miR-497 and paired upregulated EMX1 confirmed the value of our prediction analysis in EC. The simultaneous presence of low rate of these selected low miRNA/mRNA pairs (miR-497/

EMX1, miR-23c/DMBX1, and miR-670/KCNS1) might have a better prediction value. Therefore, further studies are required to validate these findings.

#### AUTHOR CONTRIBUTIONS

XX, JW, and HZ designed the study, checked the data, and prepared the manuscript. TL, YW, JF, and QY performed data collection and statistical analysis. TL searched the literature and took part in the manuscript preparation. JW and HZ conceived and supervised the project.

#### FUNDING

Funding was provided by the Natural Science Foundation of Jiangsu Province (BK20151096) and the Key Projects of National

#### REFERENCES


Health and Family Planning Commission of Nanjing City (ZKX17015) to HZ and the Fundamental Research Funds for the Central Universities (3332018179) to JW.

#### ACKNOWLEDGMENTS

We thank Juan Chen from CCN Department of ZTE for her computer technology support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene.2019.00743/ full#supplementary-material


cell proliferation and cancer prognosis. *Mol. Cancer* 16, 84. doi: 10.1186/ s12943-017-0653-4


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

*Copyright © 2019 Xu, Liu, Wang, Fu, Yang, Wu and Zhou. 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.*

# Long Non-coding RNA DLEU1 Promotes Proliferation and Invasion by Interacting With miR-381 and Enhancing HOXA13 Expression in Cervical Cancer

#### Chang Liu<sup>1</sup> \*, Xing Tian<sup>2</sup> , Jing Zhang<sup>3</sup> and Lifeng Jiang<sup>4</sup>

1 Intensive Care Unit, Cancer Hospital Affiliated to Zhengzhou University, Zhengzhou, China, <sup>2</sup> Department of Nephrology, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, China, <sup>3</sup> Department Gynecologic Tumor, Cancer Hospital Affiliated to Zhengzhou University, Zhengzhou, China, <sup>4</sup> Department of Chinese and Western Medicine, Cancer Hospital Affiliated to Zhengzhou University, Zhengzhou, China

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Young-Ho Ahn, Ewha Womans University, South Korea Sandeep Kumar, Emory University, United States

> \*Correspondence: Chang Liu liuchang677@sina.com

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 29 September 2018 Accepted: 23 November 2018 Published: 07 December 2018

#### Citation:

Liu C, Tian X, Zhang J and Jiang L (2018) Long Non-coding RNA DLEU1 Promotes Proliferation and Invasion by Interacting With miR-381 and Enhancing HOXA13 Expression in Cervical Cancer. Front. Genet. 9:629. doi: 10.3389/fgene.2018.00629 Although growing evidence has demonstrated that the long non-coding RNA DLEU1 is involved in the progression of various cancers, its functional role and underlying mechanisms have not been explored in cervical cancer (CC). In this study, we found that DLEU1 was up-regulated in both CC tissues and CC cell lines, and overexpression of DLEU1 was significantly correlated with shorter patient survival. Knockdown of DLEU1 suppressed CC cell proliferation and invasion, whereas overexpression of DLEU1 promoted the proliferation and invasion of CC cells. Bioinformatics analysis was used to elucidate the potential correlation between DLEU1 and miR-381. Moreover, qRT-PCR analysis, luciferase reporter assay and RNA immunoprecipitation assay confirmed that DLEU1 inhibited the expression of miR-381, and revealed a direct interaction between DLEU1 and miR-381. In addition, we demonstrated that miR-381 directly targeted HOXA13 in CC cells. The restoration of HOXA13 expression reversed DLEU1 knockdown or miR-381 overexpression-mediated suppression of cell proliferation and invasion. These results suggested that DLEU1 can promote CC cell proliferation and invasion via the miR-381/HOXA13 axis.

#### Keywords: DLEU1, long non-coding RNA, cervical cancer, miR-381, HOXA13

### INTRODUCTION

Cervical cancer (CC) is the second most common cause of death among women with various cancers, with over 500,000 new patients diagnosed and approximately 280,000 deaths each year (Torre et al., 2015). Although CC is curable if diagnosed at an early stage, many still present with an advanced or metastatic disease and have a worse prognosis. Metastasis is responsible for as much as 90% of cancer-induced mortality (Chaffer and Weinberg, 2011). Consequently, a better understanding of the molecular mechanism in metastatic CC is essential for the development of effective therapeutic strategies against CC.

MicroRNAs (miRNAs) and long non-coding RNAs (lncRNAs) have been shown to have crucial roles in carcinogenesis, metastasis and drug resistance (Han Li and Chen, 2015). MiRNAs

post-transcriptionally regulate the expression of oncogenes and tumor suppressor genes, thus modulating the biological behaviors of tumor cells (Fabian et al., 2010). LncRNAs are known to function as scaffolds or guides to regulate interactions between protein and genes, and as decoys to bind proteins or miRNAs to modulate the expression of their target genes (Sanchez Calle et al., 2018). Importantly, lncRNAs act as competing endogenous RNAs (ceRNAs) and participate in a miRNA-dependent crosstalk by competitively binding miRNAs, representing a critical mechanism of tumor development and metastasis (de Giorgio et al., 2013; Gao et al., 2017).

LncRNA DLEU1, located on chromosome 13q14.3, has been reported to be dysregulated in chronic lymphocytic leukemia, multiple myeloma, breast cancer, gastric cancer and ovarian cancer (Garding et al., 2013; Dowd et al., 2015; Wu et al., 2015; Wang et al., 2017; Li et al., 2018). High expression of DLEU1 was associated with poor prognosis of gastric cancer and contributes to cell proliferation (Li et al., 2018). DLEU1 was also shown to promote ovarian cancer cell proliferation and invasion by interacting with miR-490-3p (Wang et al., 2017). However, little is known about the role and the upstream regulatory mechanism of DLEU1 in CC.

In our current study, we found that DLEU1 was highly expressed in CC tissues and CC cell lines. Moreover, silencing DLEU1 expression obviously inhibited the proliferative and invasive ability of CC cells. Mechanistic analysis demonstrated that DLEU1 served as a ceRNA by sponging miR-381 and upregulating HOXA13 expression, thus promoting CC cell proliferation and invasion. The DLEU1/miR-381/HOXA13 axis might be a promising therapeutic target for CC.

### MATERIALS AND METHODS

#### Cell Cultures and Transfection

Human CC cell lines (SiHa cells and HeLa cells) and human normal cervical cells (Academia Sinica Cell Bank, Shanghai, China) were conserved in DMEM/F12 (GIBCO-BRL) mediums with ten percent of fetal bovine serum (FBS) under a wettish condition at 37◦C with 5% of CO2. Two different DLEU1 siRNAs (Genepharma, Shanghai, China) were used for knockdown of DLEU1 expression. Scrambled siRNA was used as a negative control (Genepharma, Shanghai, China). miR-381 mimic, control mimic, miR-381 inhibitor and control inhibitor were purchased from Genepharma (Shanghai, China). The DLEU1 expression plasmid pcDNA3.1-DLEU1 and the HOXA13 expression plasmid pcDNA3.1-HOXA13 were constructed by Genepharma (Shanghai, China). The empty vector pcDNA3.1 was used as negative control. Lipofectamine 2000 (Invitrogen, Carlsbad, CA, United States) was used for cell transfection following the guidance of the manufacturer's instructions.

#### Quantitative Real-Time PCR (qRT-PCR)

The expression level of DLEU1 was detected by real-time RT-PCR. In brief, total RNA from CC cells was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, United States) according to the manufacturer's instructions. We reversely transcribed 1 µg of RNA into cDNA in virtue of a Reverse Transcription Kit (Takara, Dalian, China). Real-time PCR analyses were performed using SYBR-Green-quantitative realtime PCR Master Mix kit (Toyobo Co., Osaka, Japan). The mirVanaTM qRT-PCR microRNA Detection Kit (Ambion Inc., Austin, TX, United States) was used for miR-381 detection according to the manufacturer's instructions. A specific stemloop RT primer was used for miR-381 detection. The primer sequences used have been reported (Lee et al., 2017). MiR-381 was normalized to U6. DLEU1 expression data were normalized to GAPDH.

#### Western Blot Analysis

The entire protein lysates were subjected to 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and then transferred to a PVDF membrane (Millipore, Bedford, MA, United States). Then the above PVDF membranes were incubated with the corresponding primary antibody HOXA13 (1:2000, Abcam, Cambridge, United Kingdom) and GAPDH (1:1000, Santa Cruz Biotech, Santa Cruz, CA, United States) overnight, followed by incubation with HRP-conjugated secondary antibodies (Santa Cruz, CA, United States). Protein bands were detected using an ECL western blotting kit (Amersham Biosciences, Buckinghamshire, United Kingdom). GAPDH was used as the loading control.

#### CCK-8 Assay

Cervical cancer cells (5 × 10<sup>3</sup> cells per well in 96-well plates) were transfected as described above. Cell proliferation was measured 72 h after transfection using a CCK-8 (Beyotime Institute of Biotechnology, Jiangsu, China) according to the manufacturer's instructions. The absorbance was measured at 450 nm by a microplate reader (Bio-Rad, Hercules, CA, United States).

#### In vitro Invasion Assay

The invasive ability of the cells was measured using transwell chambers (Corning, New York, United States), as described previously (Dong P. et al., 2017). In brief, cells (5 × 10<sup>4</sup> ) suspended in serum-free medium were transferred to the upper chamber. The medium containing 10% FBS was added as chemokine in the lower chamber. After 24 h, the invaded cells on the membrane lower surface were fixed with 75% methanol, and stained with crystal violet. Evaluation of invasive capacity was performed by counting invading cells under a microscope, and five random fields of view were analyzed for each chamber. All experiments were performed in triplicate.

#### Luciferase Reporter Assay

The wild-type DLEU1 (DLEU1-WT), mutant DLEU1 (DLEU1- MUT), wild-type HOXA13 3<sup>0</sup> -UTR (HOXA13-WT), and mutant HOXA13 3<sup>0</sup> -UTR (HOXA13-MUT) were synthesized and cloned into pMIR-GLOTM Luciferase vectors (Promega, Madison, WI, United States). For the luciferase reporter assay, CC cells were co-transfected with the above luciferase reporter

vectors containing DLEU1 (WT or MUT) or HOXA13 3 0 -UTR (WT or MUT) and miR-381 mimic, miR-381 inhibitor or their respective controls using Lipofectamine 2000 (Invitrogen). Luciferase activity was measured after 48 h. The relative luciferase activity was measured with the Dual-Luciferase Reporter Assay System (Promega, China). Firefly luciferase activity was normalized to that of Renilla luciferase.

#### RNA Immunoprecipitation Assay (RIP)

To verify the interaction between DLEU1 and miR-381, RNA immunoprecipitation assay was conducted using the Magna RIPTM RNA-Binding Protein Immunoprecipitation Kit (Millipore). Briefly, CC cells at 80% confluency were harvested and lysed in complete RIP lysis buffer. Then, the whole cell extract was co-immunoprecipitated with RIP buffer containing magnetic beads conjugated with anti-Argonaute2 (Ago2) antibody (Millipore, Bedford, MA, United States) or normal mouse IgG (Millipore) as a negative control. Samples were digested with proteinase K, and RNAs were isolated from the immunoprecipitation products were subjected to qRT-PCR analysis of DLEU1 and miR-381 expression.

#### Statistical Analysis

All statistical analyses were performed using SPSS 17.0 statistical software (IBM, Armonk, NY, United States). Data are presented as the mean ± standard deviation (SD) from at least three experiments. The significant differences were analyzed using Student's t-test or one-way ANOVA. P-values were based on the two-sided statistical analysis, and a P-value < 0.05 from a two-tailed test was considered significant.

#### Data Availability

All data generated during this study are available from the corresponding author on reasonable request.

#### RESULTS

#### DLEU1 Is Upregulated in CC Tissues and CC Cell Lines and Significantly Correlated With Shorter Patient Survival

To investigate the role of lncRNA in CRC metastasis, we first analyzed the expression of DLEU1 in CC tissues and normal cervical tissues using the TCGA database (Chandrashekar et al., 2017). We found that the level of DLEU1 was markedly increased in CC tissues in comparison with that in the normal samples (**Figure 1A**). Subsequently, we investigated the expression of DLEU1 in CC cell lines (SiHa cells and HeLa cells) and human normal cervical cells. Our qRT-PCR analysis suggested that the level of DLEU1 was significantly higher in CC cells compared with the normal cervical cells (**Figure 1B**), indicating that DLEU1 might have an oncogenic role in CC progression. To determine the association between DLEU1 expression and patient survival, we examined the expression of DLEU1 in TCGA dataset via the UALCAN web server. For the cohort of 291 patients, overexpression of DLEU1 was associated with poor survival (**Figure 1E**). The difference in survival is significant (P = 0.008). These data indicate that DLEU1

expression may be an important prognostic factor for patients with CC.

### Overexpression of DLEU1 Promotes CC Cell Proliferation and Invasion

To test the biological function of DLEU1 in CC cells, two different DLEU1-specific siRNAs were used to silence DLEU1 expression in SiHa cells, which exhibits the high level of DLEU1. The qRT-PCR analysis confirmed downregulation of DLEU1 levels in SiHa cells. Both siRNA-1 and siRNA-2 resulted in a significant down-regulation of DLEU1 expression (**Figure 2A**). The transfection with siRNA-1 was more effective than the transfection with siRNA-2 in terms of downregulating the DLEU1 level. We therefore chose to use DLEU1-siRNA-1 for all subsequent experiments.

Furthermore, HeLa cells with low expression of DLEU1 were selected in the following experiments. We stably overexpressed DLEU1 and found that DLEU1 level was significantly elevated after transfecting with pcDNA3.1- DLEU1 in HeLa cells (**Figure 2A**). To assess the influence of DLEU1 on cell proliferation and invasion, either lossof-function or gain-of-function assays were performed. Results of CCK-8 assay and invasion assay showed that knockdown of DLEU1 significantly weakened the proliferative and invasive ability of SiHa cells (**Figures 2B,C**). However, overexpression of DLEU1 increased the proliferation and invasion of HeLa cells, as measured using the CCK-8 assay and invasion assay (**Figures 2B,D**). These data indicated

FIGURE 2 | Silencing of DLEU1 inhibited, whereas overexpression of DLEU1 promoted proliferation and invasion of CC cells. (A) qRT-PCR analysis of DLEU1 expression in SiHa cells transfected with DLEU1 siRNA-1, DLEU1-siRNA-2 or control siRNA, and in HeLa cells transfected with pcDNA3.1-DLEU1 or empty vector pcDNA3.1. (B) CCK-8 assay was used to measure cell proliferation in SiHa cells transfected with DLEU1 siRNA-1 or control siRNA, and in HeLa cells transfected with pcDNA3.1-DLEU1 or empty vector pcDNA3.1. (C,D) Transwell invasion assay was performed to assess invasiveness in SiHa cells transfected with DLEU1 siRNA-1 or control siRNA (C), and in HeLa cells transfected with pcDNA3.1-DLEU1 or empty vector pcDNA3.1 (D). <sup>∗</sup>p < 0.05.

that overexpression of DLEU1 promoted CC cell growth and invasion.

## DLEU1 Acts as a ceRNA by Sponging miR-381 in CC Cells

To investigate whether DLEU1 functions as a molecular sponge of miRNA to liberate mRNA transcript targeted by miRNA, thereby contributing to CC progression, we used the public prediction algorithm StarBase V2.0 (Li et al., 2014) and identified miR-381 with complementary sequences to the DLEU1 transcript (**Figure 3A**). Some ceRNAs will degrade their miRNA binding partners (Liu et al., 2017).

To study the clinical relevance of miR-381 to human CC, we examined miR-381 expression in normal cells and CC cell lines. The level of miR-381 was significantly downregulated in SiHa and HeLa cells than the normal cells (**Figure 1C**). Using the BioExpress database<sup>1</sup> , we analyzed the TCGA

<sup>1</sup>https://hive.biochemistry.gwu.edu/bioxpress/about

data to evaluate miR-381 expression in human CC tissues and normal tissues. We found that miR-381 expression was markedly downregulated in CC tissues compared with normal tissues (**Figure 1F**). These results suggested a negative correlation between DLEU1 and miR-381 expression in CC (**Figure 1A**).

To examine whether DLEU1 has an impact on miR-381 expression, qRT-PCR was used to investigate the effect of DLEU1 knockdown or overexpression on miR-381 expression in CC cells. MiR-381 expression was up-regulated in SiHa cells with DLEU1 knockdown (**Figure 3B**). In contrast, overexpression of DLEU1 reduced the expression of miR-381 in HeLa cells (**Figure 3B**).

To further whether DLEU1 directly interacts with miR-381, we constructed luciferase reporter vectors containing wildtype or mutated DLEU1 and performed luciferase reporter assay. As shown in **Figure 3C**, ectopic expression of miR-381 resulted in a significant reduction in luciferase activity of wild-type DLEU1 in SiHa cells (**Figure 3C**), but had no

FIGURE 3 | DLEU1 acts as a ceRNA by sponging miR-381. (A) The predicted binding site of miR-381 to the DLEU1 sequence was shown. (B) qRT-PCR analysis of miR-381 in SiHa cells transfected with DLEU1 siRNA-1 or control siRNA, and in HeLa cells transfected with pcDNA3.1-DLEU1 or empty vector pcDNA3.1. (C) The relative luciferase activity in SiHa cells cotransfected with luciferase reporter vectors containing wild-type (WT) DLEU1 or mutant (MUT) DLEU1 and control mimic or miR-381 mimic, and in HeLa cells cotransfected with luciferase reporter vectors containing wild-type DLEU1 or mutated DLEU1 and miR-381 inhibitor or control inhibitor. (D) RIP assay was performed in SiHa cells. DLEU1 and miR-381 expression was detected using qRT-PCR. (E) Cell proliferation (left panel) and invasion (right panel) assessed in SiHa cells transfected with control siRNA + control inhibitor, DLEU1 siRNA-1 + control inhibitor, or DLEU1 siRNA-1 + miR-381 inhibitor. (F) Cell proliferation (left panel) and invasion (right panel) assessed in HeLa cells transfected with control vector + contro1 mimic, DLEU1 vector + contro1 mimic, or DLEU1 vector+ miR-381 mimic. <sup>∗</sup>p < 0.05.

evident inhibitory effect on mutant DLEU1. In addition, the transfection with miR-381 inhibitor led to a notable increase in the luciferase activity of wild-type DLEU1 in HeLa cells, while miR-381 did not affect the luciferase activity of mutated DLEU1 (**Figure 3C**).

In order to further verify the direct binding between miR-381 and DLEU1 at endogenous levels, RIP assay was performed to pull down endogenous miRNAs associated with DLEU1 in SiHa cells using the antibody against Ago2. We found that DLEU1 and miR-381 were specifically enriched in Ago2 pellets of SiHa cell extracts relative to the IgG control group (**Figure 3D**).

To determine whether DLEU1 exerted its function through miR-381 in CC cells, we performed the rescue experiments by transfecting DLEU1 siRNA in combination with miR-381 inhibitor into SiHa cells, or by transfecting DLEU1 vector in combination with miR-381 mimic into HeLa cells. The knockdown of DLEU1 significantly impeded cell proliferation and invasion of SiHa cells, while the inhibition of miR-381 significantly abrogated these effects (**Figure 3E**). Conversely, the induction in proliferative capacity and invasion ability caused by overexpressing DLEU1 could be largely reversed by the restoration of miR-381 in HeLa cells (**Figure 3F**). Taken together, our results indicated that DLEU1 could serve as a ceRNA by binding miR-381 in CC cells.

#### miR-381 Targeted HOXA13 in CC Cells

To further explore the potential targets of miR-381 in CC cells, we performed the bioinformatic-based target prediction analysis using TargetScan<sup>2</sup> . Among the potential target genes, HOXA13 was predicted to contain the binding sequence of miR-381 (**Figure 4A**), and its upregulation was shown to enhance cancer cell proliferation and invasion (He et al., 2017; Yu et al., 2018). Next, we explored the protein expression of HOXA13 in CC cells and normal cells. As demonstrated in **Figure 1D**, in comparison with that in normal cells, the expression of HOXA13 in CC cells was much higher, as measured using Western blot analysis, indicating that miR-381 directly targets HOXA13. We examined HOXA13 expression in TCGA CC datasets using the publicly available tool MethHC<sup>3</sup> , and found higher HOXA13 mRNA levels in CC tissues compared with normal tissues (**Figure 1G**), suggesting a positive correlation between DLEU1 in CC (**Figure 1A**).

Dual-luciferase reporter assay further showed that HOXA13 was the direct target of miR-381 (**Figure 4B**). As shown in **Figure 4C**, HOXA13 expression was decreased after transfecting with miR-381 mimic, but was increased after transfecting with miR-381 inhibitor.

#### Restoration of HOXA13 Reversed the Effects of DLEU1 Knockdown or miR-381 Overexpression in CC Cells

Then, we conducted the rescue assays to further affirm the regulatory role of DLEU1 silencing and miR-381 inhibition on

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

<sup>3</sup>http://methhc.mbc.nctu.edu.tw/php/index.php

FIGURE 4 | DLEU1 indirectly induced HOXA13 expression via inhibiting miR-381 expression. (A) Predicted miR-381 binding site in the 3<sup>0</sup> -UTR of HOXA13. (B) The relative luciferase activities of reporter vectors in SiHa cells (left panel) cotransfected with luciferase reporter vectors containing wild-type (WT) HOXA13 or mutated (MUT) HOXA13 and control mimic or miR-381 mimic, and in HeLa cells (right panel) cotransfected with luciferase reporter vectors containing wild-type HOXA13 or mutated HOXA13 and miR-381 inhibitor or control inhibitor. (C) The protein level of HOXA13 in SiHa cells transfected with miR-381 mimic or control mimic, and in HeLa cells transfected with miR-381 inhibitor or control inhibitor. (D) The protein expression of HOXA13 in SiHa cells transfected with contro1 siRNA + control inhibitor, DLEU1 siRNA-1 + control inhibitor, or DLEU1 siRNA-1 + miR-381 inhibitor, and in HeLa cells transfected with contro1 vector pcDNA3.1 + control mimic, pcDNA3.1-DLEU1 + control mimic, or pcDNA3.1-DLEU1 + miR-381 mimic. <sup>∗</sup>p < 0.05.

the protein expression of HOXA13 using Western blot analysis. The knockdown of DLEU1 reduced the protein level of HOXA13 in SiHa cells, while the transfection with miR-381 inhibitor apparently abolished this effect (**Figure 4D**). The induction of HOXA13 expression caused by DLEU1 overexpression could be largely suppressed by the transfection of miR-381 mimic in HeLa cells (**Figure 4D**), suggesting that DLEU1 liberated HOXA13 expression by competitively binding to miR-381.

Furthermore, CCK-8 and invasion assay showed that knockdown of DLEU1 or miR-381 overexpression significantly suppressed SiHa cell proliferation and invasion, while forced HOXA13 expression could partially rescue the anti-proliferation and anti-invasion effects mediated by the knockdown of DLEU1 or miR-381 overexpression (**Figure 5**). These data revealed that DLEU1 might promote CC cell growth and invasion by upregulating HOXA13 via competitively interacting with miR-381.

### DISCUSSION

fgene-09-00629 December 5, 2018 Time: 15:47 # 7

Emerging studies have revealed that lncRNA plays an important role in cancer progression, and they are aberrantly expressed in a variety of tumors including CC (Dong J. et al., 2017). For instance, lncRNA NORAD (Huo et al., 2018), TPT1- AS1 (Jiang et al., 2018) and NEAT1 (Guo et al., 2018) could promote cell growth and metastasis in CC. HOXD-AS1 motivates doxorubicin resistance in CC (Hou et al., 2018). HOTAIR and CCHE1 were overexpressed in CC tissues, and elevated expression of HOTAIR and CCHE1 were negative prognostic factors in CC (Kim et al., 2015; Chen et al., 2017). Up to now, studies have shown that DLEU1 functioned as an oncogenic lncRNA in several tumors (Garding et al., 2013; Dowd et al., 2015; Wu et al., 2015; Wang et al., 2017; Li et al., 2018), but the contribution of DLEU1 to CC cell proliferation and invasion and the underlying mechanisms still remained to be elucidated. In this study, we found that DLEU1 level was significantly increased in the CC tissues and CC cell lines. By performing loss-of-function and gain-of-function assays, our

results revealed the oncogenic role of DLEU1 in promoting cellular proliferation and invasiveness. Thus, our study provided a new insight into the molecular mechanism of DLEU1 in CC progression.

Although lncRNAs are involved in many pathological processes, the underlying molecular mechanisms remain largely unknown. Recently, a novel mechanism was proposed in which lncRNAs could serve as ceRNAs for miRNAs in cancer (Salmena et al., 2011). For example, DLEU1 functions as a ceRNA for miR-490-3p, thereby up-regulating the expression of CDK1, CCND1 and SMARCD1 and subsequently promoting the development and progression of CC (Wang et al., 2017). Here, we show that DLEU1 acts as an oncogene in vitro through binding miR-381. Our luciferase activity assays and RIP assays further confirmed that DLEU1 served as a molecular sponge for miR-381 to upregulate the expression of its target HOXA13, thus promoting CC pathogenesis.

Previously, miR-381 was reported to be dysregulated in many cancers, including gastric cancer (Zhang et al., 2017), oral squamous cell carcinoma (Yang et al., 2017), breast cancer (Xue et al., 2017), colorectal cancer (He et al., 2016) and ovarian cancer (Xia et al., 2016). Moreover, miR-381 acts as a tumor suppressor and was known to widely participate in tumor cell proliferation, invasion, metastasis and chemoresistance (He et al., 2016; Xia et al., 2016; Cao et al., 2017; Xue et al., 2017; Yang et al., 2017; Zhang et al., 2017; Huang et al., 2018). However, little was known about its function in CC. In the present study, we reported that miR-381 could significantly suppress CC cell growth and invasion. Furthermore, we showed the endogenous interaction between DLEU1 and miR-381 by RIP assays in CC cells, thus our results also indicated that the effects of miR-381 on proliferation and invasion are possibly due to its direct interaction with DLEU1.

HOXA13 has been reported to be upregulated and has been demonstrated to play an oncogenic role in tumorigenesis and progression in various tumors (Dong Y. et al., 2017 He et al., 2017). Although HOXA13 was highly expressed in CC tissues compared to normal cervical tissues (Saha et al., 2017), its role in the proliferation and invasion of CC cells has not been reported. Through the bioinformatic analysis, we predicted the potential miRNA binding site in HOXA13. Importantly, we found that miR-381 directly interacted with HOXA13 and downregulated its expression in CC cells. The ectopic HOXA13 expression could partially rescued the suppressive effects of miR-381 mimic or DLEU1 knockdown on cell proliferation and invasion. These experiments supported that HOXA13 acts as an oncogene in CC, and DLEU1 exerted its function via miR-381/HOXA13 axis in CC cells.

### CONCLUSION

Taken together, these results revealed that DLEU1 might facilitate the progression of CC partially through the miR-381/HOXA13

pcDNA3.1, or miR-381 mimic + pcDNA3.1-HOXA13. <sup>∗</sup>p < 0.05.

axis. The DLEU1/miR-381/HOXA13 axis should be considered as a potential therapeutic target against CC.

#### AUTHOR CONTRIBUTIONS

fgene-09-00629 December 5, 2018 Time: 15:47 # 8

CL provided the direction. CL, XT, and JZ performed the experiments. CL wrote the manuscript. LJ made significant

#### REFERENCES


revisions to the manuscript. All authors read and approved the final manuscript.

#### FUNDING

This work was supported by a grant from Cancer Hospital Affiliated to Zhengzhou University.



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

## The Functions of Non-coding RNAs in rRNA Regulation

Qi Yan<sup>1</sup> , Chengming Zhu<sup>1</sup> , Shouhong Guang1,2 \* and Xuezhu Feng<sup>1</sup> \*

<sup>1</sup> Hefei National Laboratory for Physical Sciences at the Microscale, School of Life Sciences, University of Science and Technology of China, Hefei, China, <sup>2</sup> CAS Center for Excellence in Molecular Cell Science, Chinese Academy of Sciences, Hefei, China

Ribosomes are ribonucleoprotein machines that decode the genetic information embedded in mRNAs into polypeptides. Ribosome biogenesis is tightly coordinated and controlled from the transcription of pre-rRNAs to the assembly of ribosomes. Defects or disorders in rRNA production result in a number of human ribosomopathy diseases. During the processes of rRNA synthesis, non-coding RNAs, especially snoRNAs, play important roles in pre-rRNA transcription, processing, and maturation. Recent research has started to reveal that other long and short non-coding RNAs, including risiRNA, LoNA, and SLERT (among others), are also involved in pre-rRNA transcription and rRNA production. Here, we summarize the current understanding of the mechanisms of non-coding RNA-mediated rRNA generation and regulation and their biological roles.

Keywords: rRNA, non-coding RNA, PAPAS, SLERT, 5S-OT, risiRNA, LoNA

#### Edited by:

Thomas S. Wingo, Emory University, United States

#### Reviewed by:

Fabrizio Loreni, University of Rome Tor Vergata, Italy Bing Yao, Emory University, United States

\*Correspondence:

Shouhong Guang sguang@ustc.edu.cn Xuezhu Feng fengxz@ustc.edu.cn

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 29 September 2018 Accepted: 18 March 2019 Published: 05 April 2019

#### Citation:

Yan Q, Zhu C, Guang S and Feng X (2019) The Functions of Non-coding RNAs in rRNA Regulation. Front. Genet. 10:290. doi: 10.3389/fgene.2019.00290

## INTRODUCTION

In humans, there are approximately 300–400 copies of rDNA genes per haploid genome that are distributed over five chromosomes (Henderson et al., 1973; Boisvert et al., 2007). Each rDNA unit is ∼43 kb long. rDNAs are transcribed by RNA polymerase I (Pol I) to generate prerRNAs that subsequently undergo multiple modifications and processing steps to remove the external transcribed spacers (ETSs) and internal transcribed spacers (ITSs) to produce mature 18S, 5.8S, and 28S rRNAs (McStay, 2016). Pol I activity is a key determinant for ribosome abundance and is essential for cell growth and proliferation (McStay, 2006; Srivastava et al., 2016). Interestingly, only some rDNA units are transcriptionally active. Uncontrolled rRNA synthesis by dysregulated Pol I is associated with aberrant cell proliferation and oncogenesis (Peltonen et al., 2014; Nguyen le et al., 2015).

Non-coding RNAs play essential roles in rRNA regulation. The small nucleolar RNA (snoRNA) is widely known to guide the nucleotide modifications and processing (Cech and Steitz, 2014; Sloan et al., 2017). Recently, increasingly more studies have started to reveal the roles of other classes of non-coding RNAs in regulating rRNA transcription and nucleolar function (Srivastava et al., 2016; Bazin et al., 2017). In this review, we will focus on recent work investigating how several long noncoding RNAs (lncRNAs) and antisense ribosomal siRNA (risiRNA) regulate rRNA expression and their potential biological roles in anti-stress reactions (**Table 1**).

### LncRNAs REGULATE rRNA TRANSCRIPTION IN THE NUCLEUS

Long non-coding RNAs comprise a fast-growing classes of RNA molecules with sizes greater than 200 nt. Most lncRNAs are first transcribed by polymerase II (Pol II), then capped, polyadenylated, and spliced after transcription (Cech and Steitz, 2014). lncRNAs localize in distinct subcellular compartments, including the nucleus, nucleolus and cytoplasm. Nuclear-localized lncRNAs, such

as MANTIS and Xist, may function as transcriptional or posttranscriptional regulators or structural scaffolds for nuclear domains (Sun et al., 2017). The nucleolar localization of lncRNAs suggests that they may modulate rRNA transcription and maturation (Cech and Steitz, 2014; Chen, 2016). Several studies revealed that lncRNAs regulate rDNA transcription by altering rDNA epigenetic status or by acting as "decoys" to inhibit transcription factor activity. Interestingly, some lncRNAs might contain short open reading frames that can be translated (Ji et al., 2015).

### Binding of pRNAs to TIP5 Induces Heterochromatin Formation of rDNA Genes

There are several clusters of tandemly arrayed rDNA genes exist in each mammalian genome, yet not all of these repeats are transcribed. rDNA exists in two types of chromatin – a euchromatic conformation that is actively transcribed and a heterochromatic conformation that is transcriptionally inactive. Silent rDNA repeats are marked by heterochromatic histone modifications and CpG methylation at the rDNA promoter (Schmitz et al., 2010). Silencing of rDNA depends on NoRC, a chromatin-remodeling complex that directs heterochromatin formation. NoRC function requires RNA that is complementary to the rDNA promoter, which is termed as promoter-associated RNAs (pRNA). pRNAs are 150∼300 nt long and are produced from rDNA promoters. TIP5 (TIF interacting protein 5), the large subunit of NoRC, binds to pRNAs. pRNA interacts with regulatory elements in the rDNA promoter, forms a DNA:RNA triplex, and is recognized by the DNA methyltransferase DNMT3b (**Figure 1**; Mayer et al., 2008; Bierhoff et al., 2010). Thus, the binding of NoRC to the rDNA promoter represses rDNA transcription through recruitment of histone modifying and DNA methylating enzymes (Santoro et al., 2002; Mayer et al., 2006; Guetg et al., 2010).

Interestingly, by using mature pRNAs to tether heterochromatin at nucleoli in embryonic stem cells, Savic´ et al. (2014) found that localized heterochromatin condensation of rDNA genes initiates establishment of highly condensed chromatin structures outside of the nucleolus. Meanwhile, the formation of such highly condensed, transcriptionally inactive heterochromatin promotes transcriptional activation of differentiation genes and loss of pluripotency of embryonic stem cells. NoRC safeguards genome stability by triggering heterochromatin formation at telomeres and centromeres (Postepska-Igielska et al., 2013). Whether and how pRNA and NoRC function together to maintain rDNA stability requires further investigation.

### LoNA Modulates rRNA to Promote Learning and Memory

When mice are trained with a Morris water maze, both rRNA and pre-rRNA levels are significantly elevated (Li et al., 2018). The trained mice exhibit decreased expression of the lncRNA LoNA in the hippocampus. LoNA is synthesized by Pol II and specifically enriched in nucleolus, and it can suppress the transcription of pre-rRNAs (**Figure 2**). The 5<sup>0</sup> portion of LoNA interacts with nucleolin (NCL), while its 3<sup>0</sup> portion contains a snoRNA that binds to fibrillarin (FBL). NCL can remodel rDNA loci, and can, therefore, modulate pre-rRNA transcription (Abdelmohsen and Gorospe, 2012; Durut and Saez-Vasquez, 2015). LoNA binds to NCL and inactivates the chromatin status of rDNA region. FBL is a component of C/D box small nucleolar ribonucleoproteins (snoRNPs), which direct 2<sup>0</sup> -O-methylation of rRNAs and participate in rRNA processing (Newton et al., 2003). LoNA competes with snoRNAs to bind to FBL, thereby altering the methylation status of rRNAs. By binding to both NCL and FBL proteins, LoNA suppresses rRNA production and alters ribosome heterogeneity (Li et al., 2018).

Nucleolar stress is accompanied by decreased rRNA synthesis and failures in ribosome biogenesis and functions, which are considered to be cellular stress events associated with aging and neurodegenerative diseases (Boulon et al., 2010; Wu et al., 2018). In patients with Alzheimer disease (AD), rRNA production decreases (Dönmez-Altunta¸s et al., 2005). In an AD animal model, LoNA expression is significantly increased in the mouse brain, which is accompanied by decreased rRNA levels (Li et al., 2018). Mice lacking LoNA are more efficient in locating the hidden platform in Morris water maze tests than are the control animals. LoNA-deficient AD mice show rescue of the learning and memory defects compared to the control animals in Morris water mazes and in object context discrimination behavioral tests. These results suggest that hippocampal LoNA is involved in learning and memory and may represent a potential therapeutic target for AD treatment.

### SLERT Regulates DDX21 Rings Associated With Pol I Transcription

SLERT is a box H/ACA snoRNA-ended lncRNA (Xing et al., 2017). SLERT contains 694 nt and is highly expressed in many human cell lines, especially in human embryonic stem cells and ovarian carcinoma cells. SLERT mainly accumulates in the nucleolus, and its localization depends on its box H/ACA snoRNA ends. SLERT depletion results in decreased levels of the 18S and 28S rRNAs, indicating that SLERT promotes rRNA production.

Mass spectrometry (MS) data of SLERT-associated proteins identified DDX21, a DEAD-box RNA helicase that is involved in multiple steps of ribosome biogenesis (Holmström et al., 2008; Calo et al., 2015; Sloan et al., 2015). SLERT depletion enhances the interaction between DDX21 and Pol I by tightening the DDX21 rings surrounding Pol I complexes, thereby suppressing rDNA transcription (Xing et al., 2017). Dysregulated rRNA synthesis by Pol I is associated with uncontrolled cancer cell proliferation (Nguyen le et al., 2015). The interaction between SLERT and DDX21, therefore, represents a potential therapeutic target for future anti-cancer drug discovery (Peltonen et al., 2014).

### PAPAS Responds to Environment Stresses to Maintain rRNA Suppression

Upon stress, cells utilize various strategies to suppress rDNA transcription to promote survival, for example, by inactivating


TABLE 1 | Non-coding RNAs regulate rRNA production.

certain transcription factors and inducing chromatin remodeling (Bierhoff et al., 2014; Holland et al., 2016). Furthermore, a class of lncRNAs, PAPAS, is expressed to inhibit prerRNA transcription (**Figure 3**). PAPAS is transcribed by RNA polymerase II from a fraction of the rDNA units in an antisense orientation, and, therefore, it is called "promoter and prerRNA antisense" (PAPAS) (Bierhoff et al., 2010, 2014). PAPAS comprises a heterogeneous population of 12–16 kb lncRNAs that are complementary to both the pre-rRNA coding region and the rDNA promoter. PAPAS responds to distinct stresses and modulates pre-rRNA synthesis accordingly.

In density-arrested or serum-deprived cells, pre-rRNA synthesis is suppressed and rDNA is enriched with H4K20me3 marks, while PAPAS is upregulated. After starved cells are refed with serum, the levels of H4K20me3 modifications and PAPAS decrease. RNA immunoprecipitation (RIP) experiments revealed that quiescence-induced PAPAS recruits Suv4-20h2

FIGURE 1 | Promoter-associated RNA (pRNA) targets NoRC to nucleolus to promote heterochromatin formation and rDNA silencing. pRNA is complimentary to rRNA genes and folds into a stem-loop structure. TIP5, the core factor of NoRC, recognizes pRNA, facilitates formation of heterochromatin at rRNA genes and promotes transcriptional gene silencing.

to transcription-competent rRNA genes to trigger H4K20me3 modification and chromatin compaction. Furthermore, siRNAmediated knockdown of endogenous PAPAS decreased the level of H4K20me3 modifications, but not the level of H3K9me3 modifications (Bierhoff et al., 2014).

Hypo-osmotic stress also upregulates PAPAS and inhibits rDNA transcription. However, unlike serum deprivation, hypoosmotic shock does not increase the Suv4-20h2 occupancy and H4K20me3 abundance at rDNA loci, but rather induces the degradation of Suv4-20h2 (Zhao et al., 2016a). The Mi-2/nucleosome remodeling and deacetylase factor (NuRD) is a multisubunit protein complex containing the HDAC1 histone deacetylase and the ATP-dependent remodeling enzyme CHD4 (Xue et al., 1998; Torchy et al., 2015). Upon hypo-osmotic stress, the elevated PAPAS associates with CHD4/NuRD and recruits them to rDNA regions where they deacetylate histone H4, remodel the promoter-bound nucleosomes, and reinforce transcriptional repression (Zhao et al., 2016a).

Similar to hypotonicity, heat shock also increases PAPAS expression, induces the degradation of Suv4-20h2, recruits NuRD to rDNA, and turns off transcription of pre-rRNA (Zhao et al., 2016b). Recent studies revealed the molecular mechanism of how PAPAS recruits CHD4/NuRD to rDNA. CHD4 is an RNAbinding protein that associates with both DNA and RNA via its N-terminal PHD and chromo-domains. Heat-shock elicits CHD4 dephosphorylation to facilitate its association with PAPAS (Zhao et al., 2018). PAPAS binds to the adjacent rDNA sequence via the formation of a DNA-RNA triplex, thereby directing CHD4/NuRD to rDNA, where it remodels the chromatin into a transcription refractory state (Zhao et al., 2016b, 2018).

### 5S-OT Plays a Cis Role in Regulating the Transcription of 5S rRNA and a Trans Role in Alternative Splicing of mRNAs

Unlike other Pol I-transcribed rRNAs, 5S rRNAs are transcribed by Pol III. 5S rRNA genes are clustered as tandem repeats with intergenic sequences, and they are located on distinct chromosomes (Ciganda and Williams, 2011). A number of

FIGURE 2 | LoNA modulates rRNA and promotes learning and memory. LoNA is synthesized by Pol II. LoNA interacts with nucleolin (NCL) and inactivates the chromatin status of rDNA region via reducing the loading of UBF and Pol I to rDNA loci. Meanwhile, LoNA competes with snoRNAs to bind to FBL, thereby altering the methylation status of rRNAs. Therefore, by binding to both NCL and FBL proteins, LoNA suppresses rRNA production and alters ribosome heterogeneity.

studies have revealed Pol II binding sites adjacent to Pol IIItranscribed genes, including the 5S rRNA genes (Oler et al., 2010; Hu et al., 2012). These cryptic Pol II transcripts may therefore modulate the transcription of neighboring 5S rRNAs.

Hu et al. (2016) identified a lncRNA, 5S rRNA overlapped transcripts (5S-OT), that is transcribed by RNA polymerase II and is complementary to the 5S rRNA. 5<sup>0</sup> and 3<sup>0</sup> rapid amplification of cDNA ends (RACE) experiments demonstrated that this transcript contains 847 and 354 nt in mice and humans, respectively. Chromatin immunoprecipitation (ChIP) experiments indicated that Pol II binds to the promoter at the 5S-OT transcription start site in both mouse and human cells.

Inhibition of Pol II with α-amanitin results in a decreased level of 5S-OT transcripts, which further leads to a reduction of nascent 5S rRNAs. Consistently, knocking down 5S-OT by siRNAs also inhibits the production of nascent 5S rRNAs. It was suggested that in mammalian cells, the lncRNA 5S-OT associates with 5S rDNA clusters where it promotes the transcription of 5S rRNAs, thus providing a mechanism to couple Pol II and Pol III transcription.

Furthermore, human 5S-OT contains an antisense Alu element at its 3<sup>0</sup> end (Hu et al., 2016). Alu is a primate-specific transposable element. The Alu element in the human 5S-OT gene belongs to the AluY subfamily (Batzer and Deininger, 2002). In human cells, 5S-OT regulates alternative splicing of multiple genes in trans via Alu/anti-Alu pairing with targeted genes and by interacting with the splicing factor U2AF65.

Therefore, the lncRNA 5S-OT modifies 5S rRNA and mRNAs via cis and trans mechanisms, respectively. Since 5S-OT is relatively conserved in eukaryotes from fission yeast to humans, it will be interesting to examine whether similar mechanisms are applicable in other organisms.

### SMALL REGULATORY RNAs INHIBIT PRE-rRNA VIA THE NUCLEAR RNAi PATHWAY

The gene silencing capacity of small interfering RNAs (siRNAs) was first described in Caenorhabditis elegans two decades ago (Fire et al., 1991). siRNAs silence complementary nucleic acids in both the cytoplasm and nucleus. Previous research has focused on the mechanism of siRNA-dependent regulation of mRNAs. In the cytoplasm, siRNAs can direct the degradation of targeted RNAs and inhibit protein translation (Ipsaro and Joshua-Tor, 2015). In the nucleus, siRNAs can guide heterochromatin formation and inhibit transcription elongation (Feng and Guang, 2013; Rechavi and Lev, 2017). Here, we will summarize our recent work that begins to illustrate the function of siRNAs in the regulation of ribosomal RNAs (**Figure 4**; Zhou et al., 2017a; Zhu et al., 2018).

Antisense ribosomal siRNAs (risiRNAs) are widely present in various organisms. In Schizosaccharomyces pombe lacking Cid14, rRNAs become substrates for the RNAi pathway, giving rise to siRNAs targeting rRNA (Buhler et al., 2008). In Neurospora crassa, DNA damage induces the expression of the Argonaute protein QDE-2 and a class of RNAs that interact with it (qiRNAs) from the ribosomal DNA locus (Lee et al., 2009).

In C. elegans, upon exposure to low temperature treatment or ultraviolet (UV) light, risiRNAs accumulate (Zhou et al., 2017b). risiRNAs are complementary to the 18S and 26S rRNAs, contain 22 nt, and start with a 5<sup>0</sup> guanosine. risiRNAs belong to a class of 22G-RNAs that are synthesized by the RNAdependent RNA polymerases (RdRPs) in C. elegans. risiRNAs associate with the Argonaute protein NRDE-3 and translocate to nucleolus, where they suppress pre-rRNA expression (Zhou et al., 2017b; Zhu et al., 2018).

Ribosomal siRNAs act to surveil erroneous rRNAs and maintain rRNA homeostasis. Misprocessed rRNAs are usually detected and degraded by multiple surveillance machineries, including the exosome and Trf4/Air2/Mtr4p

pathway to inhibit pre-rRNA expression.

polyadenylation (TRAMP) complexes (Schmidt and Butler, 2013). The exonuclease SUSI-1 (ceDIS3L2) is involved in the 3 <sup>0</sup>–5<sup>0</sup> degradation of oligouridylated rRNA fragments (Astuti et al., 2012; Zhou et al., 2017b). When the rRNA modification or processing steps are disrupted, or upon cold shock treatment or UV exposure, erroneous rRNAs are oligouridylated and recognized by RdRPs to generate risiRNAs via a poorly understood mechanism (Zhou et al., 2017b; Zhu et al., 2018). risiRNAs in turn silence rRNAs via the RNAi machinery to prohibit the accumulation of erroneous rRNAs.

Downregulation of rRNA transcription is one of the major strategies to preserve cellular homeostasis upon encountering stress conditions and to limit energy consumption under unfavorable conditions (Parlato and Liss, 2014). The risiRNA/RNAi-directed feedback loop, therefore, may compensate for dysfunctions in the exoribonuclease-engaged degradation of erroneous rRNAs. Consistently, when rRNA modification and processing steps are defective, the animals grow more slowly because of the presence of risiRNAs (Zhu et al., 2018).

### PERSPECTIVES

Ribosome biogenesis is tightly coordinated and controlled from the transcription of pre-rRNAs to the assembly of ribosomes, a process that is influenced by many developmental programs and environmental stress challenges (McStay, 2016; Sloan et al., 2017). Cells respond to these signals by modulating the transcription, processing, maturation of rRNAs and the assembly and usage of ribosomes. Defects or disorders in any of these steps lead to a number of human diseases. In addition to protein factors, small and long regulatory RNAs also play important roles in the regulation of pre-rRNA transcription and rRNA maturation. The regulatory RNAs may act to sense developmental signals and environmental stresses. For example, PAPAS can sense nutrition deprivation, heat-shock, and hypo-osmotic stresses. risiRNAs respond to cold shock and UV and further down regulate pre-rRNA transcription. risiRNAs also surveil the fidelity and precision of rRNA modifications and processing to avoid the accumulation of erroneous rRNAs expressed during the development of organisms. More interestingly, the lncRNA LoNA is involved in learning and memory by modulating rRNA transcription.

#### REFERENCES


Many questions still remain to be addressed to fully understand the mechanisms and roles of non-coding RNAs in anti-stress pathways and rRNA regulation. For example, how do these lncRNAs surveil distinct environmental stresses? Cold shock could induce untemplated addition of oligouridylation at the 3<sup>0</sup> ends of 26S rRNAs to elicit the production of risiRNAs (Zhou et al., 2017b). Heat-shock elicits CHD4 dephosphorylation to facilitate its association with PAPAS (Zhao et al., 2018). Were these surveillance mechanisms conserved among different organisms during evolution? Do poikilotherms and homeotherms use similar mechanisms to sense temperature alterations in the environment? Beside nuclear non-coding RNAs, cytoplasmic lncRNAs are frequently bound to and degraded at ribosomes (Carlevaro-Fita et al., 2016). Whether these cytoplasmic lncRNAs can in turn regulate rRNAs need further investigation. In particular, what are the biological roles of non-coding RNAs in regulating rRNAs during developmental processes? With new emerging technologies, many novel discoveries will help to answer these important questions.

### AUTHOR CONTRIBUTIONS

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

### FUNDING

This work was supported by grants from the National Key R&D Program of China (2018YFC1004500 and 2017YFA0102903), the National Natural Science Foundation of China (Nos. 31671346, 91640110, 31870812, and 31871300), the Major/Innovative Program of Development Foundation of Hefei Center for Physical Science and Technology (2017FXZY005), and CAS Interdisciplinary Innovation Team.

#### ACKNOWLEDGMENTS

We are grateful to the members of the SG's lab for their comments.

Natl. Acad. Sci. U.S.A. 114, E10018–E10027. doi: 10.1073/pnas.1708 433114


and histone deacetylase activities. Mol. Cell 2, 851–861. doi: 10.1016/S1097- 2765(00)80299-3


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

Copyright © 2019 Yan, Zhu, Guang and Feng. 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.

# Systematic Analysis of Intestinal MicroRNAs Expression in HCC: Identification of Suitable Reference Genes in Fecal Samples

*Hui Wang1, Yuan Lv1, Cao Wang1, Dongjing Leng1, Yan Yan1, Moyondafoluwa Blessing Fasae1, Syeda Madiha Zahra1, Yanan Jiang1,2, Zhiguo Wang1, Baofeng Yang1,2\* and Yunlong Bai1,2\**

#### *Edited by:*

*Thomas S. Wingo, Emory University, United States*

#### *Reviewed by:*

*Chenghua Li, Ningbo University, China Venugopal Thayanithy, Medical School, University of Minnesota, United States*

#### *\*Correspondence:*

*Baofeng Yang yangbf@ems.hrbmu.edu.cn Yunlong Bai baiyunlong@ems.hrbmu.edu.cn*

#### *Specialty section:*

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

*Received: 03 September 2018 Accepted: 01 July 2019 Published: 13 August 2019*

#### *Citation:*

*Wang H, Lv Y, Wang C, Leng D, Yan Y, Blessing Fasae M, Madiha Zahra S, Jiang Y, Wang Z, Yang B and Bai Y (2019) Systematic Analysis of Intestinal MicroRNAs Expression in HCC: Identification of Suitable Reference Genes in Fecal Samples. Front. Genet. 10:687. doi: 10.3389/fgene.2019.00687*

*1 Department of Pharmacology (State-Province Key Laboratories of Biomedicine–Pharmaceutics of China, Key Laboratory of Cardiovascular Medicine Research, Ministry of Education), College of Pharmacy, Harbin Medical University, Harbin, China, 2 Chronic Disease Research Institute, Translational Medicine Research and Cooperation Center of Northern China, Heilongjiang Academy of Medical Sciences, Harbin, China*

Hepatocellular carcinoma (HCC) is an extremely fatal malignancy. Intestinal microRNAs, which can be detected in fecal samples in humans may be involved in the pathological process of HCC. Therefore, screening for functional intestinal microRNAs in fecal samples and investigating their potential roles in the molecular progression of HCC are necessary. Quantitative real-time PCR (qRT-PCR) has been widely used in microRNA expression studies. However, few genes have been reported as reference genes for intestinal microRNAs in fecal samples. In order to obtain a more accurately analyzed intestinal microRNAs expression, we first searched for reliable reference genes for intestinal microRNAs expression normalization during qRT-PCR, using three software packages (GeNorm, NormFinder, and Bestkeeper). Next we screened and predicted the target genes of the differentially intestinal microRNAs of control and HCC mice through quantitative RT-PCR or miRtarBase. Finally, we also analyzed the mRNA targets for enrichment of Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways using the DAVID Bioinformatic Resources database. This study has successfully screened relatively suitable reference genes and we have discovered that the differential intestinal microRNAs play significant roles in the development of HCC. The top reference genes identified in this study could provide a theoretical foundation for the reasonable selection of a suitable reference gene. Furthermore, the detection of intestinal microRNAs expression may serve as a promising therapeutic target for the diagnosis and treatment of HCC.

Keywords: intestinal microRNAs, quantitative real-time PCR assays, reference genes, hepatocellular carcinoma, feces

### INTRODUCTION

Hepatocellular carcinoma (HCC) is one of the most common cancers in the world, and the overall 5-year survival rate for HCC is less than 12% (Torre et al., 2015). It is usually diagnosed at the late stage and is characterized by a low resection rate, high postoperative recurrence, and poor drug treatment results (Galun et al., 2015). At present, the only drug approved for the treatment of liver cancer is sorafenib. Oxaliplatin has been shown to be effective in the treatment of advanced liver cancer, but the drug resistance potentially limits its efficacy (Petrelli et al., 2014). Finding sensitive and specific early diagnostic biomarkers, and understanding the mechanisms involved in the development and progression of HCC are a major issue in need of urgent resolution.

MicroRNAs are small non-coding RNAs, with transcripts that are just 18-25 nucleotides in length, and they bind mainly to the 3′ untranslated regions (3′ UTRs) of target RNAs through the microRNAs' seed sequences (microRNA nucleotides 2–7) (Parodi et al., 2016). They are highly conservative and tissuespecific and exists widely in nematodes, drosophila, and plants, as well as in humans (Gillan et al., 2017; You et al., 2018). MicroRNAs regulate the expression of protein-coding gene expression in a sequence-specific manner through cleavage or translational repression in the genomes (Kagiya, 2016; Xiao et al., 2018). The abnormal expression of microRNAs is related to many diseases, including cancer, and miRNAs can also be used as tumor-suppressor genes or oncogenes. During the development and progression of human cancers, microRNAs have been observed to regulate cell proliferation, survival, differentiation, invasion and metastasis (Davis-Dusenbery and Hata, 2010). In patients with HCC caused by a hepatitis B virus infection, mir-29c could control cell proliferation and leads to apoptosis by targeting TNFAIP3 (Wang, Wang et al., 2011); mir-101 produces a pro-apoptotic effect by targeting Mcl-1 (Su et al., 2009). Since the discovery of their essential roles in the occurrence and development of diseases, microRNAs have been intensely studied as prognostic and diagnostic biomarkers and predictors of drug responses (Li et al., 2014).

Past research has focused mainly on post-transcriptional and translational controls regulated by non-coding RNAs (Ranjha and Paul, 2013). MicroRNAs could efficiently interrupt the synthesis of proteins excluding further transcriptional activation and any subsequent mRNA processing steps. Therefore, it provides cells with a more accurate, direct and energy-efficient way to manipulate protein expression (Xi et al., 2007; Ge et al., 2015; Connerty et al., 2016). Liu et al. (2016) have illustrated that microRNAs in feces, which originate from intestinal cells, can enter into bacteria and regulate their gene expression and growth. These special microRNAs are called intestinal microRNAs. They specifically show that mir-515-5p and mir-1226-5p can promote the growth of *Fn* and *E. coli*, respectively. Thus, knowing the differential expressions of intestinal microRNAs in fecal samples could be critical to understand physiological and pathological conditions. They could even be used as a potential marker for a new therapeutic strategy. However, the expression profiles and functions of the intestinal microRNAs present in the fecal samples of patients with HCC remains unclear.

Due to the instability of microRNAs, their low expression level and minor differences in sequence, highly sophisticated analytical methods are required. Quantitative real-time PCR (qRT-PCR) plays an important role in studying the biological functions of microRNA with strong specificity and high sensitivity (Daud and Scott, 2008).With the development of qRT-PCR, the selection of reference genes is crucial for the accuracy of the relative quantification of gene expressions (Liu et al., 2018; Zhou et al., 2018). However, only a few studies have been published that systematically evaluated the normalization targets in intestinal microRNAs qRT-PCR assays. Successfully screening for suitable reference genes first will provide reliable evidence for the quantitative analysis of intestinal microRNAs in fecal samples. Once the expression profile of intestinal microRNAs in HCC is confirmed by qRT-PCR, the functions of differentiated intestinal microRNAs could be investigated during the biological process.

This study aimed to find relatively stable reference genes and to clarify whether intestinal microRNAs from fecal samples play an important role in indicating HCC development. RNA was extracted from mouse fecal samples and microRNA expressions were detected by qRT-PCR. Four conventional housekeeping genes and 18 microRNAs were selected as candidate genes. Their expression stability was evaluated using GeNorm, NormFinder, and Bestkeeper software. These statistical analysis software programs were primarily used to identify appropriate reference genes in qRT-PCR experiments. We identified three genes (*mir-23a*, *GAPDH* and *let-7i*) which may be used as suitable reference genes for intestinal microRNAs. Next, we detected the expression levels of some intestinal microRNAs, which had already been observed in humans and mice, between the control and HCC group. Finally, we analyzed the mRNA targets of the differentiated intestinal microRNAs. Then GO terms and KEGG pathways were built using these databases. In this study, relatively suitable reference genes of intestinal microRNAs in fecal samples are first screened for, with a potential diagnosis and treatment for patients with HCC is then proposed.

#### MATERIALS AND METHODS

#### Samples Preparation and the Ethics Approval

Professor Feng Hai from Harbin Medical University supplied fecal samples from five control and five HCC-suffering mice. The C57BL/6 mice (14th postnatal days) were injected DEN at a single intraperitoneal dose of 25 mg/kg and the animals were observed until 10 months. At 10 months of age, mice injected with DEN developed striking liver HCCs with tumors that could be clearly observed. The mice were allowed to defecate normally, and the first three fecal pellets of each animal were collected into an empty 1.5 ml tube with a sterile toothpick. A new toothpick was used for each mouse. The tubes were closed immediately and placed into liquid nitrogen, finally transferred into a −80°C refrigerator for storage. The methods were performed in accordance with the National Guidelines for Experimental Animal Welfare (the Ministry of Science and Technology, People's Republic of China, 2006). All experimental protocols were pre-approved by the Experimental Animal Ethics Committee of Harbin Medical University, China.

### Total RNA Extraction

One ml of Trizol isolation reagent (Invitrogen, Carlsbad, CA) was added per 40–60 mg of frozen fecal samples and total RNAs was extracted according to the manufacturer's protocol. The only difference from the usual RNA extraction protocol used in fecal samples was the need for repeated extraction using chloroform. The concentrations and purity of the total RNA extracted were measured by ultraviolet spectrophotometer NanoDro 2000 (Thermo Scientific, Waltham, USA). Purity requirements were considered to be met if the extracted RNA had an A260/A280 value between 1.80 and 2.00.

#### Quantitative Real-Time PCR Assays

For reverse transcription, 1 μg of total RNA was reverse transcribed using a ReverTra Ace® qPCR RT Kit (Toyobo, Osaka, Japan). The reverse transcription used a 20 μl reaction with random primers or gene-specific stem-loop primers, designed as described previously (Chen et al., 2005). All microRNA RT primer sequences are shown in **Table S1**. At the beginning of the experiment, the amplification efficiencies of the primers were first tested. According to the relative standard curve, the amplification efficiency of primers was about 93–116%. The cDNA was amplified by real-time PCR using SYBR® Green Realtime PCR Master Mix (Toyobo, Osaka, Japan) on the ABI 7500 fast Real Time PCR system (Applied Biosystems, Carlsbad, CA, USA). The primer sequences of the primers (Invitrogen, Shanghai, China) are listed in **Table S1**. The quantitative RT-PCR reaction mixture was run in a 20 μl volume reaction. The mixture consisted of 10 μl SYBR Green PCR Master Mix, 1.0 μl of each specific primer, 2.0 μl cDNA template, and 6.0 μl RNase-free water. The reactions performed were as follows: 1 min at 95°C for pre-denaturation, 40 cycles of 15 s at 95°C for denaturation, 15 s at 60°C for annealing, 45 s at 72°C for extension.

### Data Preprocessing

Bestkeeper uses the cycle threshold (Ct) value directly from the qRT-PCR to analyze the stability of the gene, whereas the Ct value needs to be converted to a Q value before analysis by GeNormer and NormFinder. The formula for calculating the Q value is: *Q = 2^-(Ct/sample-Ct/min)*. The calculated results were then imported into Excel according to the rules of GeNormer and NormFinder as they only identify a specific format (Vandesompele et al., 2002, Andersen et al., 2004).

#### MicroRNA Target Prediction and Functional Classification

The miRTarBase database (http://miRTarBase.mbc.nctu.edu. tw/) is a commonly used database for microRNA-target interactions. Using miRTarBase, all the validated mRNA targets of the differentially expressed intestinal microRNAs of control and HCC mice were listed. To create more accurate and reliable predicted results, only target genes validated by at least one experimental method were chosen. The DAVID Bioinformatic Resources database (https://david.ncifcrf.gov/) was then used to analyze the mRNA targets for enrichment of GO terms and KEGG pathways. A P value of <0.05 was used as the significance cut-off.

#### Statistical Analysis

All measurement data were presented as mean ± SEM and analyzed by GraphPad Prism 5. Differences between groups were analyzed by Student's t-test, where P< 0.05 was considered as indicative of a statistically significant difference.

### RESULTS

#### A Schematic Diagram for the Experimental Process

In order to identify the suitable reference genes for intestinal microRNAs in fecal samples and to systematically analyze intestinal microRNAs expression in HCC, a schematic flow diagram of sample processing and microRNA analysis was first designed (**Figure 1**). Due to the novelty and scarcity of data in this area of study, 17 intestinal microRNAs, commonly expressed both in human and mouse fecal samples, were selected as candidate genes (Liu et al., 2016). Generally, *GAPDH*, *U6 snRNA*, *16S rRNA* and *5S rRNA* were also selected as they are usually used as reference genes.

#### Comparison of Ct Values Between Candidate Reference Genes

Even in different tissues or different growth stages of the same organ, there may be differences in the expression of the same gene. The Ct value of the gene is inversely proportional to the amount of target nucleic acid present. The lower the Ct value, the higher the amount of the gene present. According to its Ct value from real-time PCR, we analyzed the expression of each candidate reference gene (**Figure 2A**). The Ct values of 22 candidate reference genes ranged between 11 and 37. Large differences and a wide distribution range were found between the expressions of the candidate genes. The traditional reference genes *5S rRNA* had the highest expression level, where the Ct value was between 11.67-14.94; the expressions of *U6 snRNA*, *mir-192*, *mir-574*, *mir-194*,*let-7i*, and *mir-1224* were relatively high; *Mir-15a* and *mir-155* showed a larger Ct value with low expressions. Additionally, the box plots illustrated that the expression range of each candidate reference gene varied significantly (**Figures 2B**, **C**). The Ct value of *mir-155*, *mir-23a*, *mir-378*, *GAPDH*, *mir-574*, *let-7i* and *mir-1224* had fluctuating ranges that were comparatively small with a more concentrated distribution. Preliminary Ct values showed that the expression of *mir-23a*, *GAPDH*, *mir-574*, *let-7i*, and *mir-1224* was higher and had few differences between the samples.

### Expression Stability of Candidate Reference Genes

Three software packages (GeNorm, NormFinder, and Bestkeeper) were used for analyzing the average expression stabilities of the candidate reference genes. GeNorm and NormFinder software was specifically used for selecting reference genes. Before the analysis, it was necessary to convert the qRT-PCR result (Ct value) to a Q value. GeNorm software was written by Vandesompele in 2002 for screening reference genes where two or more reference genes can be chosen (Vandesompele et al., 2002). For the screening and selection of reference genes, an M value is usually displayed according to the output file of the GeNorm program. The default critical M value of the software is 1.5, where an M value of less than 1.5 indicates that the gene expression is relatively stable. The smaller the M value, the more stable the gene is. Through GeNorm software, we analyzed the M values of 22 candidate reference genes (**Figure 3A**). The data showed that the M value of *mir-200b* was the largest, whereas *GAPDH* and *mir-23a* were the smallest, indicating that *GAPDH* and *mir-23a* were the most stable genes. However, the traditional reference genes *U6 snRNA* and *5S rRNA* remained rather unstable. Another program, NormFinder, was written in 2004 by Claus, and was used to screen the stability of qualitative reference genes (Andersen et al., 2004). The calculation principle of this program is similar to that of GeNorm and the stability value used is also the M value. The results showed that *mir-23a*, *mir-378*, and *GAPDH* were relatively more constant (**Figure 3B**). *Mir-200b* and *mir-15a* were discarded in the results since they had M values that were too large. These results were similar to those obtained from the GeNorm software. The Bestkeeper software was also used to evaluate the stability of the candidate genes (Pfaffl et al., 2004). The Ct values of 22 candidate reference genes were calculated, along with the standard deviation (SD) and coefficient of variation (CV). Our results showed that higher gene stability is related to a smaller SD and CV. The data showed that *let-7i* had the lowest SD and CV values of 0.06 and 0.36, respectively, followed by *mir-23a*, *mir-574,* and *GAPDH* (**Figure 3C**, **Table S2**). The results from Bestkeeper were somewhat similar to those from GeNorm and NormFinder.

#### Comprehensive Evaluation of Candidate Reference Genes in the Three Software Programs

When comparing the results of the three software programs, we found that despite the similar results obtained, there were still differences among them. A geometric mean method was thus used to evaluate the stability of each candidate reference gene. The candidate genes' stabilities were ranked in each of the software programs and then the rankings were averaged across the three software programs (**Table 1**). The top three genes were *mir-23a*, *GAPDH* and *let-7i*. These three genes were found to be the most suitable reference genes for intestinal microRNAs. Considering that the expression and distribution of *let-7i* was the most concentrated and the highest, it was selected as a reference gene for the subsequent analysis. To further identify the stability of *let-7i* as a reference gene, a validated experiment was carried out in different animal models. We collected the fecal samples from some disease models whose incidence is higher, such as HCC, Type 2 diabetes (T2DM) and Myocardial Infarction (MI). The Ct value dispersion of the *let-7i* 

by geNorm software. (B) Average expression stability values analysis of candidate reference genes by NormFinder software. (C) Average expression CV values analysis of candidate reference genes by Bestkeeper software.


was detected by qRT-PCR. The result showed that there was no significant difference between different disease animal models (**Figure S1A**). Additionally, the boxplot also indicated that the distribution of *let-7i* was concentrated within groups (**Figure S1B**). These assays demonstrated that *let-7i* displayed higher stability which could be used as a reference gene for microRNA qRT-PCR in fecal samples.

#### Altered Intestinal MicroRNAs Expression Profiles in HCC Mice

In order to explore the functional intestinal microRNAs in HCC, we detected the expression levels of microRNAs which had been observed both in humans and mice. According to the comprehensive evaluation of candidate reference genes in the three software programs and their expression levels (Ct value), we chose *let-7i* as an internal reference gene. We demonstrated the presence of 11 miRNAs that displayed a decrease (*miR-155*, *miR-378*, *mir-23a*, *mir-26b*, *mir-29b, mir-194, mir-192, mir-15a, let-7a, mir-200a,*and *mir-200b*)*,* in the HCC group compared with the control group, while *mir-200c, mir-1224*, *mir-574, mir-141, let-7b,* and *let-7g* showed no significant differences between the groups (**Figure 4**).

#### Enrichment of Predicted mRNA Targets for Differentiated Intestinal MicroRNAs

To clarify downstream target gene networks for the intestinal microRNAs with HCC, we first listed all the potential mRNA targets of the differentiated intestinal microRNAs. These were predicted using miRTarBase. In order to improve the accuracy of the predicted mRNA targets, we only chose targets with strong evidence which had been supported by the reporter assay, western blotting or qPCR. The GO terms and KEGG pathways were then examined using the DAVID database. The results obtained from GO terms showed that the major processes in the pathology of HCC includes apoptosis, cell proliferation and the Wnt signaling pathway (**Figure 5A**). The KEGG pathway analysis explained that mRNA targets were mainly concentrated in the Jak-STAT pathway, apoptosis, and the p53 signaling pathway (**Figure 5B**). These results suggested that the differential intestinal microRNAs are likely to indicate the occurrence of HCC through predicted targets and pathways.

#### DISCUSSION

MicroRNAs are expected to become potential biomarkers for cancer diagnosis and new targets for disease treatments as aberrant expressions of microRNAs may be related to specific diseases. In recent years, quantitative RT-PCR assays have become the main method of studying the relative expression of microRNA. However, the quality of the results may be affected by some nonspecific variables, such as the quality of RNA, efficiency of PCR amplification and the selection of reference genes (Yang et al., 2017). Appropriate normalization is a key part of quantitative gene expression analysis which is often ignored. The reference gene is also known as a housekeeping gene and is usually used to correct experimental errors when testing the gene expression levels. The ideal internal reference gene needs to satisfy the following conditions: the gene's expression should be stable in different tissues, experimental conditions and growth stages (Lv et al., 2017, Molina et al., 2018). By using stable reference genes, the data in quantitative RT-PCR assays can be standardized. However, no reference genes have been found to be stable under all experimental conditions. Therefore, when studying specific gene expressions, it is necessary to select the most stable gene as the reference gene to correct the data and to ensure the accuracy of the results.

Interestingly, a new study suggests that fecal microRNAs could be identified as potential markers for intestinal malignancy (Ahmed et al., 2009). These fecal microRNAs exist in the gut lumen and feces and are called intestinal microRNAs. Some specific microRNAs that have been found in HCC cells and tissues may participate in tumor development, or could be a notable biological feature (Nelson and Weiss, 2008). The expression of microRNA in feces is therefore likely to become a new diagnostic marker in HCC. To achieve this potential, relatively accurate expression profiles of intestinal microRNA in HCC is a critical requirement. Because the reference gene of intestinal microRNA has not been identified yet, our first process step was to seek out a stable normalizer for the subsequent microRNA qRT-PCR experiments. Beginning with screening potentially stable reference genes, four traditional reference genes (*5S rRNA*, *U6 snRNA*, *GAPDH and 16S rRNA*) and 22 intestinal microRNAs co-expressed in humans and mice were used as candidate genes. Considering that normal samples are necessary in all experiments and conditions, we initiated our studies using control mice. First of all, the dispersion of the candidate genes' Ct values was detected by qRT- PCR. There was a particularly large range of Ct values from 11 to 37 (**Figure 2A**). The boxplot indicated that Ct value distribution of *let-7i* was the most concentrated whereas *mir-23a*, *mir-378* and *GAPDH* had slightly dispersed Ct values (**Figures 2B**, **C**). Preliminarily assays demonstrated that these four genes with little dispersion also displayed higher stability.

However, the Ct value distribution map could only roughly assess the stability of the candidate reference genes. In order to determine the stability of a reference gene more accurately, it is necessary to further evaluate it using reference gene analysis software. There are a number of software programs available for analyzing reference genes, including GeNorm, NormFinder and Bestkeeper. They are three different statistical models that are widely used to select the most appropriate reference gene (Shen et al., 2011). Both GeNorm and NormFinder evaluates the stability of the gene by measuring the M value. When the maximum expression of the gene's stability is achieved, it will be accompanied by a low M value. Bestkeeper is a comprehensive evaluation software program which depends on the SD and CV. The principle of determination is that the smaller the SD and CV is, the better the stability of the reference gene is. When SD is greater than one, it directly indicates that the expression of the reference gene is unstable. In our study, the stability of reference genes was analyzed by all three of these software programs. According to the qRT-PCR results, NormFinder showed that *GAPDH* was the most stable reference gene with the smallest M value (**Figure 3B**). This result was consistent with GeNorm software, with only one difference, GeNorm had the two best candidate reference genes and the other one was *mir-23a* (**Figure 3A**). This is in accordance with previous studies where *mir-23a* was considered to be a novel microRNA normalizer in profiling studies of cervical tissues (Shen et al., 2011). Nevertheless, the results of Bestkeeper were slightly different from those of GeNorm and NormFinder. The results showed that *let-7i* was one of the best reference genes with the smallest SD and CV (**Figure 3C**, **Table S1**), and was thus considered the most stable gene by the Bestkeeper software. Since different software programs have different algorithms, some differences are expected between the results obtained from each program. Therefore, to obtain relatively stable reference genes, we comprehensively analyzed the results from the three software programs. As presented in **Table 1**, it can be seen that the stabilities of the expression levels of the 22 candidate reference genes from the highest to the lowest were *mir-23a* > *GAPDH* > *let-7i* > *mir-378* > *mir-1224* > *mir-155* > *mir-574* > *16S Rrna* > *mir-29b* > *let-7b* > *mir-26b* > *mir-141 mir-200c* > *mir-192* > *let-7g* > *mir-194*  > *let-7a* > *mir-200a* > *U6 snRNA*  > *mir-200b* > *5s Rrna* > *mir-15a*. Commonly used conventional reference genes in our microRNA qRT-PCR assays, like *5S rRNA* and *U6 snRNA* were particularly unstable. This phenomenon indicates that traditional reference genes are not applicable to all samples. Therefore, gene stability should always be evaluated and verified before use. Conclusively, *mir-23a*, *GAPDH* and *let-7i* are the optimal reference genes for intestinal microRNA in normal mouse feces. This information could provide recommendations for intestinal microRNA qRT-PCR studies with diseases statuses or other conditions.

MicroRNAs could modulate multiple target mRNA degradations or translations in the same pathway. Therefore, it has been proposed to identify miRNA regulatory profiles through predicted target genes and established miRNA-mRNA expression modules (Fu et al., 2012). In order to explore the possible mechanism of intestinal microRNAs regulating HCC, we constructed the intestinal miRNAmRNA expression modules. Intestinal microRNA expression levels with HCC were first detected by quantitative RT-PCR. *Let-7i* was chosen as the reference gene since it showed a higher expression (Ct < 15) than *mir-23a* and *GAPDH* (**Figure 2A**). Results showed that there were differential intestinal microRNAs in the HCC group compared to the control group (**Figure 4**). Unexpectedly, *mir-23a* displayed a significant change in the HCC group, even though it has never been reported that *mir-23a* is a novel microRNA normalizer for relative quantification in human uterine cervical tissues. Indeed, this also clarified that the expression of reference genes is not always stable, and an appropriate reference gene under one experimental condition may not be suitable for all experimental conditions.

To better understand the biological functions of dysregulation intestinal microRNAs in HCC, a functional analysis of miRNAmRNA was performed. Thus, an analysis of the differential microRNAs for GO terms and KEGG pathways were conducted. The targets of the intestinal microRNAs were significantly enriched in cell proliferation, migration and differentiation, which is closely related to the TGF-beta, Wnt and VEGF signaling pathway that is involved in tumorigenesis, metastasis and therapy (**Figure 5A**). *miR-29b* plays a protective role in cardiac remodeling by targeting the TGF-beta/Smad3 pathway (Zhang et al., 2014). mTOR is the key point of various important signaling pathways in cells, as well as being activated in a subset of HCCs (Fikret et al., 2004). The enriched KEGG pathway indicated that differential intestinal microRNA targets were likely to be involved in hepatocellular carcinogenesis through the mTOR pathway (**Figure 5B**). Previous research has clarified that *mir-155* enhances mTOR activity in the progression of breast carcinomas (Martin et al., 2014). It is evident that mRNA targets of differential intestinal microRNAs were significantly enriched in the apoptotic process and apoptosis pathway (**Figure 5**), where apoptosis was indicated to be involved in liver cancer development. Low *mir-15a* expression has shown negative regulation by targeting *Bcl-2* in human leukemia that is directly or indirectly involved in the biological process of apoptosis (Calin et al., 2008). The above evidence supports that intestinal miRNA could be a potential oncogene and antioncogene that is involved in HCC development.

In conclusion, the top three genes that were comprehensively ranked (*mir-23a*, *GAPDH* and *let-7i*), provide recommendations for intestinal microRNA qRT-PCR studies with disease statuses or other conditions. Additionally, differential intestinal microRNAs in fecal samples were predicted to be a potential oncogene or antioncogene involved in HCC development and may provide a novel strategy for diagnosing HCC and other related diseases.

### AUTHOR CONTRIBUTIONS

HW and YL conceptualized and designed the experiments in this study. CW, DL, and YY performed experimental procedures related to the study. MB, SM, and YJ proofread and discussed articles relating to this research. ZW provides great help for revise article. YB and BY provided fnancial support and reviewed experimental directions. All authors contributed to the generation of experimental data for this study.

### ACKNOWLEDGMENTS

We would like to extend our gratitude to Howard L. Weiner and Shirong Liu from Brigham and Women's Hospital, Harvard Medical School for providing help in intestinal microRNA isolation. Also, to Professor Feng Hai from Harbin Medical University for supplying fecal samples from HCC mice. We also thank Chiam Kai Wen Rachel for modifying the language used in this article. This project was supported by the National Natural Science Foundation of China (Grant No. 81674326 and Grant No. 81730012).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene.2019.00687/ full#supplementary-material

FIGURE S1 | The expression level of let-7i by real-time PCR. (A) Cycle threshold (Ct) values for let-7i, n=5. (B) Distribution of Ct values of let-7i. Box representative the range of the Ct value distribution.

#### REFERENCES


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

*Copyright © 2019 Wang, Lv, Wang, Leng, Yan, Blessing Fasae, Madiha Zahra, Jiang, Wang, Yang and Bai. 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.*

## RNA Expression Profile and Potential Biomarkers in Patients With Spinocerebellar Ataxia Type 3 From Mainland China

Tianjiao Li<sup>1</sup> , Xiaocan Hou<sup>1</sup> , Zhao Chen<sup>1</sup> , Yun Peng<sup>1</sup> , Puzhi Wang<sup>1</sup> , Yue Xie<sup>1</sup> , Lang He<sup>1</sup> , Hongyu Yuan<sup>1</sup> , Huirong Peng<sup>1</sup> , Rong Qiu<sup>2</sup> , Kun Xia<sup>3</sup> , Beisha Tang1,3,4,5 and Hong Jiang1,3,4,5 \*

<sup>1</sup> Department of Neurology, Xiangya Hospital, Central South University, Changsha, China, <sup>2</sup> School of Information Science and Engineering, Central South University, Changsha, China, <sup>3</sup> Medical Genetics Research Center, Central South University, Changsha, China, <sup>4</sup> National Clinical Research Center for Geriatric Diseases, Xiangya Hospital, Central South University, Changsha, China, <sup>5</sup> Key Laboratory of Hunan Province in Neurodegenerative Disorders, Central South University, Changsha, China

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Matteo Chiara, University of Milan, Italy Venugopal Thayanithy, University of Minnesota, United States

> \*Correspondence: Hong Jiang jianghong73868@126.com

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 01 September 2018 Accepted: 29 May 2019 Published: 13 June 2019

#### Citation:

Li T, Hou X, Chen Z, Peng Y, Wang P, Xie Y, He L, Yuan H, Peng H, Qiu R, Xia K, Tang B and Jiang H (2019) RNA Expression Profile and Potential Biomarkers in Patients With Spinocerebellar Ataxia Type 3 From Mainland China. Front. Genet. 10:566. doi: 10.3389/fgene.2019.00566 Long non-coding RNAs (lncRNAs) play an important role in growth, development, and reproduction and undoubtedly contribute to the pathogenesis and progression of diseases. Emerging evidence suggests the involvement of lncRNAs as regulatory factors in pathological conditions, including some neurodegenerative diseases. Spinocerebellar Ataxia Type 3/Machado–Joseph Disease (SCA3/MJD) has a prominent prevalence in China. Because the role of lncRNAs in SCA3/MJD pathogenesis has not yet been investigated, we conducted a pilot study to investigate the expression profile of lncRNAs by high-throughput sequencing in 12 patients and 12 healthy individuals. The sequencing analysis detected 5,540 known and 2,759 novel lncRNAs. Six lncRNAs were confirmed to be differentially expressed in peripheral blood mononuclear cells between SCA3/MJD patients and healthy individuals and were further validated in cerebellar tissue. Based on these results, NONHSAT022144.2 and NONHSAT165686.1 may be involved in the pathogenesis of SCA3/MJD and may be potential biomarkers for SCA3/MJD. Together with NONHSAT022144.2 and NONHSAT165686.1, the other four novel lncRNAs increase our understanding of lncRNA expression profile.

Keywords: Spinocerebellar Ataxia Type 3, Machado–Joseph disease, long non-coding RNAs, expression profile, biomarker

### INTRODUCTION

Long non-coding RNAs (lncRNAs) play an essential role in regulating the expression of genes involved in almost all metabolic pathways and other biological macromolecules at both the transcriptional and post-transcriptional levels. lncRNAs also play a role in the regulation of growth, development, and reproduction (Wapinski and Chang, 2011; Khorkova et al., 2015; Lardenoije et al., 2015; Engreitz et al., 2016). A growing body of evidence suggests that lncRNAs are involved in the pathogenesis and progression of some neurodegenerative diseases, such as Alzheimer's Disease (AD) and Huntington's disease (HD) (Lin et al., 2014; Tan et al., 2014; Salta and De Strooper, 2017; Sunwoo et al., 2017; Zhou et al., 2018).

Alteration in the expression levels of some lncRNAs contributes to the pathogenesis of neurodegenerative diseases (Lin et al., 2014; Roberts et al., 2014; Tan et al., 2014; Zhang et al., 2016; Salta and De Strooper, 2017; Sunwoo et al., 2017; Gagliardi et al., 2018). For example, lnc-sca7 (ATXN7L3B) is highly conserved in the central nervous system of human and adult mice and transcriptionally regulates the expression of ATXN7, which is the pathogenic gene of Spinocerebellar Ataxia Type 7 (SCA7). Indeed, knockout of lnc-sca7 in N2a cells leads to a significant decrease in ATXN7 expression at the translational level. Accordingly, overexpression of lnc-sca7 significantly enhances transcription of ATXN7 (Tan et al., 2014; Salta and De Strooper, 2017). A more detailed study showed that the lncRNA TUNA has elevated expression in the thalamus and striatum. Gene expression analysis in brain tissue of 44 HD patients and 36 healthy individuals confirmed that TUNA expression in the caudate nucleus might be involved in the pathophysiology of HD (Lin et al., 2014; Salta and De Strooper, 2017). Additionally, the lncRNA NEAT1 is associated with the damage mechanism of HD (Sunwoo et al., 2017).

Polyglutamine disease (PolyQ disease) is a type of disease caused by dynamic mutation of a trinucleotide sequence; PolyQ diseases include spinocerebellar ataxia (SCA) types 1, 2, 3, 6, 7, and 17; dentatorubral–pallidoluysian atrophy (DRPLA); spinal and bulbar muscular atrophy (SBMA), and HD (Matos et al., 2011). Spinocerebellar Ataxia Type 3/Machado–Joseph Disease, one of the PolyQ diseases similar to SCA7 and HD (Wang et al., 2015), is the most common SCA subtype in mainland China (Chen et al., 2018). It is mainly caused by the abnormal expansion of the trinucleotide sequence CAG in the ATXN3 gene and has a broad age of onset ranging from 4- to over 70-years-old (Peng et al., 2014; Wang et al., 2015; Chen Z. et al., 2016; Chen et al., 2017). Although various pathogenic hypotheses have been proposed and supported experimentally for SCA3/MJD, no effective strategies are available for surveillance, delay progression, or treatment of the disease (Costa Mdo and Paulson, 2012). Given that some lncRNAs are involved in SCA7 and HD pathogenesis (Lin et al., 2014; Roberts et al., 2014; Tan et al., 2014; Zhang et al., 2016; Salta and De Strooper, 2017; Sunwoo et al., 2017; Gagliardi et al., 2018) and that we previously detected abnormal expression of three lncRNAs in the SCA3/MJD mouse model (Vidal et al., 2000), we speculated that some lncRNAs also contribute to SCA3/MJD pathogenesis. Largely because of their accessibility and ease of operation, peripheral blood mononuclear cells (PBMCs) are widely used to establish suitable biomarkers for neurodegenerative and other diseases (Coppola et al., 2011; Zhang et al., 2016; Gagliardi et al., 2018). To investigate the possible mechanisms of lncRNAs in SCA3/MJD pathogenesis and to explore potential SCA3/MJD biomarkers, we isolated PBMCs from patients and healthy individuals for high-throughput sequencing. Through bioinformatics analysis, we identified 5,540 known and 2,759 novel lncRNAs in patients and normal individuals. Validation including expression analysis in human brains enabled us to target six lncRNAs. This provides useful information for understanding the pathogenesis of SCA3/MJD and enabling new treatment strategy options.

### MATERIALS AND METHODS

### Sample Collection and PBMC Isolation

All participants, including the 12 patients and 12 healthy individuals from Xiangya Hospital of Central South University, were required to sign consent forms prior to collection of their blood samples. The subjects included six males and six females ranging in age from 18- to 62-years-old. All of the patients met the clinical manifestations of SCA3/MJD, had been diagnosed with SCA3/MJD by neurologists and geneticists, and were confirmed to not have any other known brain or neurological disorder. By matching age and gender, we found suitable healthy individuals from the physical examination center at Xiangya Hospital. Ten milliliters of peripheral blood was collected for isolation of PBMCs. The blood samples collected from participants were processed within 2 h to isolate PBMCs using lymphocyte solution (Ficoll–Hypaque) based on the standard protocol instructions.

### RNA Extraction and High-Throughput Sequencing

TRIzolTM reagent (Invitrogen, Carlsbad, CA, United States) was added to isolated PBMCs for total RNA extraction, and genomic DNA was removed with the RNeasy kit (Qiagen, Hilden, Germany). Ribosomal RNA removal was conducted using a biotin-labeled specific probe from the Ribo-ZeroTM rRNA Removal Kit (Epicentre) <sup>R</sup> prior to RNA library preparation. The first strand cDNA was then synthesized using random primers and reverse transcriptase in the TruSeq <sup>R</sup> Stranded kit (Illumina) followed by synthesis of the double-stranded cDNA using DNA polymerase I and RNaseH. The adaptor-ligated cDNA fragments were amplified to generate the final cDNA library followed by library purification. High-throughput sequencing was carried out by BGI Genomics with the Illumina Hi-seq X-Ten platform.

High-throughput sequencing obtained raw reads as data. Subsequently, we used the short reads alignment tool SOAP<sup>1</sup> to compare reads to the ribosomal database and filtered data by removing the ribosomal reads on the alignment, which allows for up to five mismatches. Then, reads containing only adapter dimers or reads with 10% or more N or low-quality bases were filtered out. After that, we obtained the clean reads. Clean reads were aligned to the reference genome with the HISAT software<sup>2</sup> and assembled with StringTie<sup>3</sup> . The Hg19 genome assembly version was used for alignment and assembly. By aligning all reads to the genome and constructing all possible transcripts in each alignment region, only the transcripts with fragments per kilobase of transcript per million reads mapped (FPKM) values ≥ 0.5, coverage values > 1, and length values > 200 bp were retained for further quantitative analysis. To compare the expression levels between samples, it was necessary to standardize the expression level of the gene. Therefore, we aligned the clean reads to the reference sequence with Bowtie2<sup>4</sup> , and expression

<sup>1</sup>https://github.com/BGI-flexlab/SOAPnuke

<sup>2</sup>http://www.ccb.jhu.edu/software/hisat

<sup>3</sup>http://ccb.jhu.edu/software/stringtie

<sup>4</sup>http://bowtie-bio.sourceforge.net/bowtie2/index.shtml

#### TABLE 1 | The classification of overlap.

fgene-10-00566 June 12, 2019 Time: 17:26 # 3


Lnc-Overlap-mRNA means lncRNA on the same strand as mRNA, and lncRNA has an overlap with mRNA.

Lnc-AntiOverlap-mRNA means that lncRNA and mRNA are on different strands, and there is an overlap between lncRNA and mRNA.

Lnc-CompleteIn-mRNAExon means that lncRNA and mRNA are on the same strand, and the lncRNA completely falls in the Exon region of mRNA.

Lnc-AntiCompleteIn-mRNAExon means lncRNA and mRNA are on different strands, and the lncRNA completely falls in the Exon region of mRNA.

mRNA-CompleteIn-LncExon means that lncRNA and mRNA are on the same strand, and the mRNA completely falls in the Exon region of lncRNA.

mRNA-AntiCompleteIn-LncExon means that lncRNA and mRNA are on different strands, and mRNA completely falls in the Exon region of lncRNA.

Lnc-CompleteIn-mRNAIntron means that lncRNA and mRNA are on the same strand, and the lncRNA completely falls in the Intron region of mRNA.

Lnc-AntiCompleteIn-mRNAIntron means lncRNA and mRNA are on different strands, and the lncRNA completely falls in the Intron region of mRNA.

mRNA-CompleteIn-LncIntron means that lncRNA and mRNA are on the same strand, and mRNA completely falls in the Intron region of lncRNA.

mRNA-AntiCompleteIn-LncIntron means that lncRNA and mRNA are on different strands, and mRNA completely falls in the Intron region of lncRNA.

levels of genes and transcripts were calculated by using RSEM<sup>5</sup> . The standardized method of RSEM is FPKM. Its calculation method is FPKM(A) = 106C/(NL/10<sup>3</sup> ), where FPKM(A) is the expression level of gene A, C is the number of fragments that are uniquely aligned to gene A, N is the total fragment number that is aligned to the reference gene, and L is the number of bases of the gene A coding region. We needed to predict whether transcripts had coding ability to determine if they were lncRNAs. Three prediction programs – CPC<sup>6</sup> , txCdsPredict<sup>7</sup> , and

<sup>5</sup>http://deweylab.biostat.wisc.edu/rsem

<sup>6</sup>http://CPC.cbi.pku.edu.cn

<sup>7</sup>http://hgdownload.soe.ucsc.edu/admin/jksrc.zip

TABLE 2 | The classification of overlap of DEGs.

CNCI<sup>8</sup> – were used to score the transcripts' coding ability to distinguish mRNA and lncRNA by different score ranges. The thresholds in the CPC and CNCI software were set to 0, and mRNAs had scores greater than 0, while lncRNAs had scores less than 0. The txCdspredict threshold was set to 500, and mRNA transcripts had scores over 500, lncRNAs had scores while less than 500. Additionally, transcripts were compared to the protein database Pfam<sup>9</sup> to classify coding RNAs (mRNAs) and lncRNAs. We considered transcripts lncRNAs only if they met the criteria of at least three of the four software and databases, and then we analyzed their predicted target gene functional annotation. Moreover, we considered a fold change ≥ 2.00 and false discovery rate ≤ 0.001 significant in DEGseq (Wang et al., 2010) to compare lncRNA expression differences the SCA3/MJD and control group All data alignments, assembly and analysis were conducted by BGI Genomics.

#### Target Gene Prediction

We determined the Spearman and Pearson correlation coefficients to determine lncRNA targets; only mRNAs with a Spearman correlation ≥ 0.6 and Pearson correlation ≥ 0.6 were considered target genes of a given lncRNA. When the lncRNA overlapped with the target gene, we classified the lncRNAs and their target mRNAs based on their positional relationship.

#### Gene Function Annotation

We analyzed differentially expressed genes (DEGs) and their target genes with the GO database<sup>10</sup> to explore possible functional enrichment analyses of DEGs and identify functional modules enriched in DEGs. To determine the pathways that DEGs and their target genes are concentrated in, we performed enrichment analysis using the KEGG database<sup>11</sup> .

### Quantitative Real-Time PCR Verification in PBMCs and Human Brain Tissue

Reverse transcription was carried out with the GoldenstarTM RT6 cDNA Synthesis Kit (Beijing TsingKe Biotech Co. Ltd., China). Based on the analysis of the high-throughput sequencing, 20 lncRNAs were selected for quantitative real-time PCR (qRT-PCR) to confirm their differential expression between the case

https://github.com/www-bioinfo-org/CNCI http://pfam.xfam.org/ http://www.geneontology.org/ https://www.kegg.jp/


The details of overlap class included in Table 1, and NA means the lncRNA and their target gene are on different chromosomes.

and control groups. Further validation was performed in 28 SCA3/MJD patients and 28 healthy individuals as mentioned in the section "Materials and Methods." Primers for RT-PCR listed in **Supplementary Table S1** were designed by Primer 5, and the qRT-PCR was performed per the instructions of the 2 × T5 Fast qPCR Mix (SYBR Green I) Kit. We used GAPDH as an internal control and the 2−11Ct method to calculate the quantitative expression of lncRNAs. The Wilcoxon rank-sum test was used to compare statistical significance in the expression levels of DEGs between groups, and p-values < 0.05 were considered statistically significant. For further validation, statistically significant lncRNAs were validated by qRT-PCR in cerebellar tissue from a SCA3/MJD patient and a healthy individual.

#### RESULTS

### Expression Profiles of lncRNAs and mRNAs

A total of 124,394 transcripts was detected by high-throughput sequencing, including 15,926 novel lncRNAs, 13,651 novel mRNAs, 61,335 known lncRNAs, and 33,482 known mRNAs. After data filtering and DEG analysis with DEGseq, we identified 5,540 known lncRNAs and 2,759 novel lncRNAs with differential expression. Additionally, we identified 4,701 known mRNAs and 2,517 novel mRNAs.

#### Summary of the Cis- and Trans-Regulation and Classification of the Overlap

Through target gene prediction analyses, we predicted the possible regulatory patterns of lncRNAs. All of the positional information and the complementarity between mRNAs and lncRNAs are shown in **Supplementary Material II**, and we indicate how lncRNAs may act on their target genes. The overlap is divided into 10 classes (**Table 1**), including Lnc-OverlapmRNA with 812 pairs, Lnc-AntiOverlap-mRNA with 204 pairs, Lnc-Complete In-mRNAExon with 153 pairs, Lnc-AntiComplete In-mRNAExon with eight pairs, mRNA-CompleteIn-LncExon with 47 pairs, and mRNA-AntiCompleteIn-LncExon with 2 pairs. Lnc-CompleteIn-mRNA Intron with 235 pairs, Lnc-AntiComplete In-mRNA Intron with 64 pairs, mRNA-Complete

In-LncIntron with 19 pairs, and mRNA-AntiComplete In-LncIntron with 18 pairs. As listed in **Table 2**, we predicted the target genes for 5 of 20 tested lncRNAs in this study.

### Gene Function Annotation

fgene-10-00566 June 12, 2019 Time: 17:26 # 5

Gene Ontology (GO) analysis showed that DEGs were significantly enriched in certain functional terms, and the top 20 GO terms are listed in **Figure 1**. Similarly, we performed GO analysis of target genes of DEGs and listed significant enriched GO terms in **Figure 2**. KEGG analysis indicated that DEGs and their target genes were enriched in different pathways. We listed the top 20 terms in **Figures 3**, **4**.

### Validation of lncRNAs

Among the 20 lncRNAs we selected, we verified that 6 lncRNAs are statistically different in PBMCs. The NONHSAT165686.1, LTCONS\_00051791, LTCONS\_00175021, and LTCONS\_00175040 lncRNAs were up-regulated by 5.6, 4.8, 2.0, and 2.4-fold, respectively. There were also two downregulated lncRNAs. The expression of NONHSAT022144.2 in the SCA3/MJD group was approximately one-quarter of the control group, while the expression of LTCONS\_00176188 in the SCA3/MJD group was approximately two-thirds of the control group (**Figure 5**). All six lncRNAs were verified in cerebellum tissue (**Figure 6**). The dysregulation of these four lncRNAs, NONHSAT165686.1, LTCONS\_00051791, NONHSAT022144.2, and LTCONS\_00176188, in cerebellum tissue was consistent with the results in PBMCs. In contrast, the expression trends for LTCONS\_00175021 and LTCONS\_00175040 in cerebellum tissue differed from those in PBMCs.

### DISCUSSION

To explore possible biomarkers for SCA3/MJD, we performed high-throughput sequencing of 12 SCA3/MJD patients and

binding (GO:0031726), regulation of fever generation (GO:0031620), positive regulation of fever generation (GO:0031622), positive regulation of heat generation (GO:0031652), positive regulation of natural killer cell chemotaxis (GO:2000503), response to external stimulus (GO:0009605), regulation of heat generation (GO:0031650), regulation of lymphocyte chemotaxis (GO:1901623), regulation of natural killer cell chemotaxis (GO:2000501), cytokine receptor binding (GO:0005126).

mellitus (ko04940), Basal cell carcinoma (ko05217), Legionellosis (ko05134), HTLV-I infection (ko05166), Prostate cancer (ko05215), Epstein-Barr virus infection (ko05169), Wnt signaling pathway (ko04310), Cytokine-cytokine receptor interaction (ko04060), Hematopoietic cell lineage (ko04640), Allograft rejection (ko05330).

12 healthy individuals. We found 3,812 known lncRNAs and 2,300 novel lncRNAs that were up-regulated in the SCA3/MJD group compared to the control group, while the remainder of the lncRNAs was down-regulated. We analyzed 20 lncRNAs with qRT- PCR, including 8 known lncRNAs and 12 novel lncRNAs. The results suggested that six of them were significantly differentially expressed in the two groups, including two known and four novel lncRNAs.

According to the NONCODE database<sup>12</sup>, NONHSAT022144.2 is located on chr11: 65499058–65506444 from database hg38, is most highly expressed in the heart (FPKM = 202.808) and second most in the brain (FPKM = 152.041). According to the NCBI database<sup>13</sup>, NONHSAT022144.2 aligns to the three transcripts of MALAT1, which are NR\_002819.4, NR\_144567.1, and NR\_144568. The mature MALAT1 transcript is not stabilized by a poly(A) tail but instead has a 3<sup>0</sup> -triple helical structure. The MALAT1 gene is associated with cancer metastasis, cell migration, and cell cycle regulation. Studies have confirmed that MALAT1 inhibits miR-101 expression (Li et al., 2017), and miR-101 regulates apoptosis, cellular stress, metastasis, autophagy, and tumor growth (Assali et al., 2018). In the SAMP8 mice which are an AD animal model, miR-101 is an important node that regulates the expression of gene networks in the brain (Cheng et al., 2013). Additionally, MALAT1 plays an important role in the differentiation of N2a cells, and knockdown of MALAT1 may lead to neurite outgrowth and cell death through the ERK/MAPK signaling pathway (Chen L. et al., 2016). This evidence indicates that MALAT1 is important in the development, growth, differentiation, and function of the nervous system, and SCA3/MJD is closely related to the functional regression of the nervous system. Notably, MALAT1 is also known as NEAT2 (nuclear paraspeckle assembly transcript 2), and its family member NEAT1 resists neuronal damage and contributes to neuroprotection in patients with HD. This suggests that NEAT1 is a potential target for therapeutic treatment (Sunwoo et al., 2017). It is well-known that the pathogenesis of HD and SCA3/MJD, which are both PolyQ diseases, is similar. This may be evidence that NONHSAT022144.2 could be a

<sup>12</sup>https://www.bioinfo.org/NONCODE2016/

<sup>13</sup>https://www.ncbi.nlm.nih.gov/gene/378938

FIGURE 4 | The top 20 significantly enriched pathway terms for the target gene of DEGs. The top 20 KEGG pathway terms are Legionellosis (ko05134), MAPK signaling pathway (ko04010), NF-kappa B signaling pathway (ko04064), Leishmaniasis (ko05140), Prion diseases (ko05020), Protein processing in endoplasmic reticulum (ko04141), Rheumatoid arthritis (ko05323), Toll-like receptor signaling pathway (ko04620), Influenza A (ko05164), Estrogen signaling pathway (ko04915), Systemic lupus erythematosus (ko05322), Cytokine-cytokine receptor interaction (ko04060), Salmonella infection (ko05132), Measles (ko05162), TNF signaling pathway (ko04668), Antigen processing and presentation (ko04612), Spliceosome (ko03040)Toxoplasmosis (ko05145), Type I diabetes mellitus (ko04940), Epstein-Barr virus infection (ko05169).

potential therapeutic molecule for SCA3/MJD. Additionally, we found that the expression level of this lncRNA was significantly decreased in SCA3/MJD patients relative to healthy individuals, suggesting its potential as a biomarker and therapeutic molecule.

NONHSAT165686.1 is located on Chr13:48233221–48261860, as described in the hg38 and NONCODE database<sup>14</sup>, and it aligns to the ITM2B, NM\_021999.4 transcript. To our knowledge, mutation of this gene leads to two autosomal dominant neurodegenerative diseases: familial British dementia (FBD) and familial Danish dementia (FDD). FBD is characterized by progressive dementia, cerebellar ataxia, and spasticity and is partially similar to SCA3/MJD (Vidal et al., 1999). Coincidentally, FDD shares similar symptoms such as progressive ataxia (Vidal et al., 2000). FBD also shares certain similarities with SCA3/MJD pathogenesis. Mitochondrial dysfunction has been implicated in the pathogenesis of cerebral amyloidosis such as FBD (Todd et al., 2014). Similarly, mitochondrial dysfunction is involved in the pathogenesis of SCA3/MJD, and ataxin-3 proteolysis produces toxic fragments, leading to mitochondrial defects (Harmuth et al., 2018). Expression of ITM2B induces apoptosis and is associated with loss of mitochondrial membrane potential, release of cytochrome c, and induction of apoptosis by a caspase-dependent mitochondrial pathway (Fleischer et al., 2002). Our results show that the expression of NONHSAT165686.1 is 5.6-fold higher in SCA3/MJD than the control group. This suggests that NONHSAT165686.1 might be associated with the pathogenesis and clinical features of SCA3/MJD.

Although only two human cerebellar tissues were used for validation, it seems that these lncRNAs are differentially expressed in cerebellar tissue between SCA3/MJD patients and healthy individuals. More importantly, we detected a significant decrease in LTCONS\_00175040 expression in cerebellar tissue of SCA3/MJD patients relative to the healthy individual. We only used two human samples for validation due to the limitation of human brain sample collection. Even though we cannot exclude the possibility that the difference between these two human brain samples (the SCA3/MJD patients and the healthy individuals) was caused by individual variation, this result at least confirms the expression of these lncRNAs in the cerebellum. Thus,

<sup>14</sup>http://www.noncode.org/show\_rna.php?id=NONHSAT165686&version=1& utd=1#

and LTCONS\_00051791 are up-regulated.Compared with the control group, the expression level of NONHSAT165686.1 is about 5.6 times (p = 0.036), the expression of LTCONS\_00051791 is about 4.8 times (p = 0.018), and LTCONS\_00175021 and LTCONS\_00175040 are both up more than twice (p = 0.027 of LTCONS\_00175021 and p = 0.032 of LTCONS\_00175040). Contrary to this, both known lncRNA NONHSAT022144.2 and novel lncRNA LTCONS\_00176188 are down-regulated, their expression levels are 0.259 times (p = 0.000) and 0.640 times (p = 0.043) that of the control group, respectively.

further validation using cerebellar tissues from more patients and healthy individuals is indispensable for concluding that LTCONS\_00175040 is an important molecule. Additionally, more than 10 other lncRNAs have been identified near the LTCONS\_00175040 lncRNA locus, indicating that this is a lncRNA-enriched region. Furthermore, four novel lncRNAs contribute to the expansion and improvement of the lncRNA expression profile.

In summary, our study may open a new angle for dissecting SCA3/MJD pathogenesis based on lncRNA analysis. By further exploring potential functions and pathways in which the lncRNAs are involved, we anticipate that lncRNAs such as NONHSAT022144.2 and NONHSAT165686.1 will be biomarkers or even potential therapeutic targets for SCA3/MJD treatment.

#### ETHICS STATEMENT

fgene-10-00566 June 12, 2019 Time: 17:26 # 9

The study was approved by the Ethics Committee of Xiangya Hospital of Central South University in China (equivalent to an Institutional Review Board), and written informed consent was obtained from all of the patients.

#### AUTHOR CONTRIBUTIONS

All the authors contributed to the experimental design and sample collection. TL completed the statistical analysis and wrote the manuscript.

#### FUNDING

This work was supported by the National Key Research and Development Program of China (Nos. 2016YFC0905100

#### REFERENCES


and 2016YFC0901504 to HJ), the National Natural Science Foundation of China (No. 81771231 to HJ and No. 81600995 to Yuting Shi), Scientific Research Foundation of Health Commission of Hunan Province (No. B2019183 to HJ), the Key Research and Development Program of Hunan Province (No. 2018SK2092 to HJ), the Clinical and Rehabilitation Fund of Peking University Weiming Biotech Group (No. xywm2015I10 to HJ), Youth Foundation of Xiangya Hospital (No. 2017Q03

ACKNOWLEDGMENTS

(No. 1053320170177 to TL).

The authors sincerely thank all of the patients and healthy individuals who volunteered for and supported this study. The authors also thank those who helped conduct this study.

to ZC), and Independent Exploration and Innovation Project of Graduate Students of Central South University

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00566/full#supplementary-material

3498–3505. doi: 10.1002/1521-4141(200212)32:12<3498::aid-immu3498>3.0. co;2-c


Philos. Trans. R. Soc. Lond. B Biol. Sci. 369:20130507. doi: 10.1098/rstb.2013. 0507


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

The handling Editor declared a past collaboration with several of the authors HJ and BT.

Copyright © 2019 Li, Hou, Chen, Peng, Wang, Xie, He, Yuan, Peng, Qiu, Xia, Tang and Jiang. 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.

# Long Non-coding RNA NEAT1: A Novel Target for Diagnosis and Therapy in Human Tumors

Peixin Dong<sup>1</sup> \* † , Ying Xiong2†, Junming Yue3,4, Sharon J. B. Hanley <sup>1</sup> , Noriko Kobayashi <sup>1</sup> , Yukiharu Todo<sup>5</sup> and Hidemichi Watari <sup>1</sup> \*

*<sup>1</sup> Department of Obstetrics and Gynecology, Hokkaido University School of Medicine, Hokkaido University, Sapporo, Japan, <sup>2</sup> State Key Laboratory of Oncology in South China, Department of Gynecology, Sun Yat-sen University Cancer Center, Guangzhou, China, <sup>3</sup> Department of Pathology and Laboratory Medicine, University of Tennessee Health Science Center, Memphis, TN, United States, <sup>4</sup> Center for Cancer Research, University of Tennessee Health Science Center, Memphis, TN, United States, <sup>5</sup> Division of Gynecologic Oncology, National Hospital Organization, Hokkaido Cancer Center (NHO), Sapporo, Japan*

#### Edited by:

*Ge Shan, University of Science and Technology of China, China*

#### Reviewed by:

*Jernej Ule, University College London, United Kingdom Peter G. Zaphiropoulos, Karolinska Institutet (KI), Sweden*

#### \*Correspondence:

*Peixin Dong dpx1cn@gmail.com Hidemichi Watari watarih@med.hokudai.ac.jp*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to RNA, a section of the journal Frontiers in Genetics*

Received: *09 August 2018* Accepted: *24 September 2018* Published: *15 October 2018*

#### Citation:

*Dong P, Xiong Y, Yue J, Hanley SJB, Kobayashi N, Todo Y and Watari H (2018) Long Non-coding RNA NEAT1: A Novel Target for Diagnosis and Therapy in Human Tumors. Front. Genet. 9:471. doi: 10.3389/fgene.2018.00471* The nuclear paraspeckle assembly transcript 1 (NEAT1, a long non-coding RNA) is frequently overexpressed in human tumors, and higher NEAT1 expression is correlated with worse survival in cancer patients. NEAT1 drives tumor initiation and progression by modulating the expression of genes involved in the regulation of tumor cell growth, migration, invasion, metastasis, epithelial-to-mesenchymal transition, stem cell-like phenotype, chemoresistance and radioresistance, indicating the potential for NEAT1 to be a novel diagnostic biomarker and therapeutic target. Mechanistically, NEAT1 functions as a scaffold RNA molecule by interacting with EZH2 (a subunit of the polycomb repressive complex) to influence the expression of downstream effectors of EZH2, it also acts as a microRNA (miRNA) sponge to suppress the interactions between miRNAs and target mRNAs, and affects the expression of miR-129 by promoting the DNA methylation of the miR-129 promoter region. Knockdown of NEAT1 via small interfering RNA or short hairpin RNA inhibits the malignant behavior of tumor cells. In this review, we highlight the latest insights into the expression pattern, biological roles and mechanisms underlying the function and regulation of NEAT1 in tumors, and especially focus on its clinical implication as a new diagnostic biomarker and an attractive therapeutic target for cancers.

Keywords: NEAT1, nuclear paraspeckle assembly transcript 1, long non-coding RNA, cancer diagnosis, cancer treatment, EMT, microRNA

#### INTRODUCTION

Although most of the genome (around 80%) is actively transcribed into RNA, <2% is actually translated into functional proteins (ENCODE Project Consortium, 2012), suggesting that the dominant fraction of the transcriptome consists of non-coding RNAs (ncRNAs). NcRNAs can be classified into small ncRNAs (<200 nt) or long ncRNAs (>200 nt). One class of small ncRNAs, microRNAs (miRNAs), bind to the 3′ -untranslated region of the target gene mRNA, thereby repressing target mRNA translation or inducing mRNA degradation. In contrast to well-studied miRNAs, the function and mechanism of long ncRNAs (lncRNAs) in human cancers are poorly characterized.

### CATEGORIES OF LNCRNAS

According to genomic location, lncRNAs can be further subdivided into exon or intron sense-overlapping lncRNAs, intergenic lncRNAs, antisense lncRNAs, bidirectional lncRNAs and enhancer lncRNAs (Esteller, 2011; Thum and Condorelli, 2015; **Figure 1**).

#### FUNCTIONS AND MECHANISMS OF LNCRNAS

The function of lncRNA depends on their subcellular location (Chen L. L., 2016). Up to 30% of lncRNAs are found exclusively in the nucleus, ∼15% are found exclusively in the cytoplasm, while the remaining lncRNAs are present in both nucleus and cytoplasm (Kapranov et al., 2007). In the nucleus, lncRNAs are important transcriptional and epigenetic modulators of nuclear functions, whereas cytoplasmic lncRNAs target mRNA transcripts and modulate mRNA stability and translation (Mercer and Mattick, 2013).

LncRNAs could form complex secondary and tertiary structures, which are proposed to provide multiple binding sites for other molecules (Liu et al., 2017b). LncRNAs regulates the expression of target genes by interacting with other molecules such as protein, DNA and RNA (**Figure 2**). Some lncRNAs can guide site-specific recruitment of transcriptional activators or suppressors to genomic sites and regulate gene expression (Mercer et al., 2009; Tripathi et al., 2010; Kotake et al., 2011; Wang and Chang, 2011; Li Y. et al., 2016; Xie et al., 2016). A number of lncRNAs may act as scaffolding proteins by recruiting chromatin remodeling complexes, including the polycomb repressive complex 1 (PRC1) and PRC2, to silence target-specific genes (Mercer et al., 2009; Tripathi et al., 2010; Kotake et al., 2011; Wang and Chang, 2011; Li Y. et al., 2016; Xie et al., 2016). Furthermore, lncRNAs can scaffold HBXIP and LSD1 to form a complex that activates transcription of c-Myc target genes (Mercer et al., 2009; Tripathi et al., 2010; Kotake et al., 2011; Wang and Chang, 2011; Li Y. et al., 2016; Xie et al., 2016). lncRNAs can bind to transcription factors as decoys to sequester them away from their targets, thereby affecting gene transcription (Mercer et al., 2009; Tripathi et al., 2010; Kotake et al., 2011; Wang and Chang, 2011; Li Y. et al., 2016; Xie et al., 2016). LncRNAs may serve as molecular sponges by harboring binding sites for miRNAs and sequester them away from their mRNA targets (Mercer et al., 2009; Tripathi et al., 2010; Kotake et al., 2011; Wang and Chang, 2011; Li Y. et al., 2016; Xie et al., 2016). Certain LncRNAs can regulate RNA splicing either by interacting with splicing factors or by binding the splicing junctions of pre-mRNA (Mercer et al., 2009; Tripathi et al., 2010; Kotake et al., 2011; Wang and Chang, 2011; Li Y. et al., 2016; Xie et al., 2016).

#### THE ROLES OF LNCRNAS IN CANCER

Increasing number of studies have shown that lncRNAs are either up- or down-regulated in cancers. Some lncRNAs are expressed in a cell type-specific manner during differentiation and in certain cancers, whereas several lncRNAs such as MALAT1 are widely overexpressed in various cancers (Prensner and Chinnaiyan, 2011; Wang Y. et al., 2016). Their dysregulation is associated with tumorigenesis, cancer progression, metastasis and prognosis of various tumors (Schmitt and Chang, 2016), suggesting that lncRNAs may have roles as potential biomarkers for cancer. Actually, lncRNAs have been identified as prognostic and diagnostic biomarkers for lymph node metastasis and distant metastasis in early-stage cancer (Chen J. et al., 2016). Furthermore, extracellular lncRNAs can be stable and detectable in bodily fluids (such as blood and urine), thus the levels of circulating lncRNAs can be considered as promising noninvasive biomarkers over conventional biomarkers (Shi et al., 2016).

Through the mechanisms mentioned in **Figure 2**, lncRNAs exert their oncogenic or tumor suppressive functions in human tumors and they are key regulators of pathways involving all hallmarks of cancer (Prensner and Chinnaiyan, 2011; Schmitt and Chang, 2016; Wang Y. et al., 2016). Due to the massive involvement of lncRNAs in the mechanistic, functional and translational aspects of cancer biology, they have been explored as therapeutic targets. For example, aberrant expression of the lncRNA H19 occurs in ovarian cancer and other types of cancers (Yoshimura et al., 2018). The toxin vector DTA-H19 is a plasmid that expresses the diphtheria toxin A chain under the control of the H19 gene regulatory sequences (Mizrahi et al., 2009). The intra-tumoral injection of DTA-H19 caused significant inhibition of tumor growth in ovarian cancer xenograft models (Mizrahi et al., 2009). The safety, tolerability, and efficacy of DTA-H19 have been verified in a phase 1/2a clinical trial for the treatment of H19-overexpressing bladder cancer (Sidi et al., 2008).

Recent studies have demonstrated a complicated interplay and cross-regulation among different species of non-coding RNAs (Deng et al., 2016). LncRNAs and miRNAs can directly interact with and regulate each other through at least four mechanisms (Deng et al., 2016): (i) miRNA can decrease the abundance of lncRNA by reducing its stability (Yoon et al., 2012); (ii) lncRNAs can serve as sponges or decoys for miRNAs to decrease the available levels of miRNAs (Cesana et al., 2011); (iii) lncRNAs may compete with miRNAs for binding to mRNAs (Faghihi et al., 2010); and (iv) lncRNAs generate miRNAs from their exons and introns (Cesana et al., 2011).

### NEAT1, A NOVEL PLAYER IN TUMOR

Nuclear paraspeckle assembly transcript 1 (NEAT1, a lncRNA) is transcribed from the familial tumor syndrome multiple endocrine neoplasia (MEN) type 1 loci on chromosome 11q13.1 and encodes two transcriptional variants, NEAT1-1 (3756 bp) and NEAT1-2 (22,743 bp) (Bond and Fox, 2009; **Figure 3A**). NEAT1 is enriched in the nucleus but also found in the cytoplasm (van Heesch et al., 2014). NEAT1 appears to be dispensable for normal embryonic development and adult life, as mice lacking

NEAT1 develop normally (Nakagawa et al., 2011). However, another study reported that genetic ablation of NEAT1 resulted in aberrant mammary gland morphogenesis and lactation defects (Standaert et al., 2014). Whether the loss of NEAT1 is compatible with normal cell viability and normal development should be further evaluated.

NEAT1 displays typical characteristics of cancer drivers, because it is responsible for tumor initiation and progression, shown.

and its frequent dysregulation in cancers correlates with clinical features such as metastasis, recurrence rate and patient survival (Lanzós et al., 2017).

Increasing evidence suggested that NEAT1 is overexpressed in many solid tumors, including non-small cell lung cancer (Pan et al., 2015; Sun et al., 2016), ovarian cancer (Kim et al., 2010; Chen Z. J. et al., 2016), cervical cancer (Han et al., 2017), hepatocellular carcinoma (Guo et al., 2015; Liu Z. et al., 2017), colorectal cancer (Li et al., 2015; Peng et al., 2017), gastric cancer (Ma et al., 2016), esophageal squamous cell carcinoma (Chen et al., 2015), endometrial cancer (Li Z. et al., 2016; Wang J. et al., 2017), cholangiocarcinoma (Zhang C. et al., 2018), laryngeal squamous cell cancer (Wang P. et al., 2016), pancreatic cancer (Huang et al., 2017), thyroid carcinoma (Li J. H et al., 2017), oral squamous cell carcinoma (Huang et al., 2018), nasopharyngeal carcinoma (Cheng and Guo, 2017; Liu F. et al., 2018), osteosarcoma (Wang H. et al., 2017; Hu et al., 2018), breast cancer (Zhang et al., 2017; Zhao et al., 2017), glioma (Zhen et al., 2016) and renal cell carcinoma (Liu et al., 2017a; Ning et al., 2017). However, its expression was reduced in acute promyelocytic leukemia (Zeng et al., 2014), indicating that the role of NEAT1 may vary with cancer types.

The overexpression of NEAT1 was significantly associated with poor overall survival in non-small cell lung cancer (Pan et al., 2015), ovarian cancer (Chen Z. J. et al., 2016), cervical cancer (Han et al., 2017), colorectal cancer (Li et al., 2015; Peng et al., 2017), hepatocellular carcinoma (Liu Z. et al., 2017), esophageal squamous cell carcinoma (Chen et al., 2015), pancreatic cancer (Huang et al., 2017), oral squamous cell carcinoma (Huang et al., 2018), breast cancer (Zhao et al., 2017), gastric cancer (Fu et al., 2016), breast cancer (Zhang et al., 2017), and renal cell carcinoma (Ning et al., 2017). Moreover, higher expression levels of NEAT1 were positively correlated with cancer stage and metastasis in endometrial cancer (Li Z. et al., 2016), ovarian cancer (Chen Z. J. et al., 2016), non-small cell lung cancer (Pan et al., 2015; Sun et al., 2017), hepatocellular carcinoma (Sun et al., 2016), esophageal squamous cell carcinoma (Chen et al., 2015), laryngeal squamous cell cancer (Wang P. et al., 2016), breast cancer (Zhao et al., 2017), gastric cancer (Fu et al., 2016), and osteosarcoma (Hu et al., 2018).

### MECHANISMS OF NEAT1 DYSREGULATION IN CANCER

The expression of NEAT1 in cancer cells is controlled by the following mechanisms: genetic alterations (such as copy number gain and gene mutation), transcription factors, DNA methylation, miRNAs and RNA-binding protein (**Figure 3B**).

The Cancer Genome Atlas (TCGA, a large-scale cancer genomics project) has revealed molecular alterations such as gene mutations, copy-number changes, upregulation or downregulation of mRNA and lncRNA across diverse tumor types. A comprehensive study analyzing lncRNA alterations in the TCGA datasets covering 5,860 tumor samples from 13 cancer types revealed that on average 13.16% of lncRNAs underwent copy number gains and 13.53% of lncRNAs underwent copy number loss (Yan et al., 2015). Here, we used the TCGA cervical, endometrial and ovarian cancer datasets to annotate genetic alterations of NEAT1 in these gynecological cancers via the cBioPortal online application (http://www.cbioportal. org). Consistent with previous observations (Kim et al., 2010; Chen Z. J. et al., 2016; Li Z. et al., 2016; Han et al., 2017; Wang J. et al., 2017), we found that genetic alterations occur in 5–11% of cervical, endometrial and ovarian cancers, and the predominant alterations were amplification and RNA upregulation (**Figure 3C**). The presence of copy number gain in NEAT1 gene might provide a possible explanation for high NEAT1 expression in these cancers. In addition, deep sequencing analysis of breast cancers has indicated mutations in the NEAT1 promoter region (Rheinbay et al., 2017). Another group also found an elevation of mutations in the NEAT1 promoter in renal cell carcinoma, and these mutations were associated with increased NEAT1 expression and unfavorable patient survival (Li S. et al., 2017). Although the functional impact of these mutations is unknown, mutations that occur in the NEAT1 promoter might affect the binding of a protein to the NEAT1 promoter and alter its expression (Rheinbay et al., 2017).

Several studies reported that transcription factors such as hypoxia-inducible factor (HIF)-2 and RUNX1 bind to the locus of NEAT1 and induce its expression in breast cancer (Choudhry et al., 2015; Barutcu et al., 2016). STAT3 and NF-κB, two downstream effectors of EGFR signaling, could bind to and activate the NEAT1 promoter in glioblastoma (Chen et al., 2018). On the other hand, BRCA1 inhibits NEAT1 expression through binding to its genomic binding site upstream of the NEAT1 gene in breast cancer (Lo et al., 2016). At the post-transcriptional level, NEAT1 is physically associated with and stabilized by an RNA-binding protein HuR (Chai et al., 2016).

Epigenetic mechanisms such as aberrant DNA methylation and miRNA dysregulation account for the aberrant expression of lncRNAs in tumors (Choudhry et al., 2015; Yan et al., 2015). The expression of NEAT1 was increased by the treatment with 5-AZA in hepatocellular carcinoma cells, indicating that DNA methylation is an important determinant of NEAT1 expression (Fang et al., 2017). Numerous miRNAs that directly target lncRNAs have identified in tumor cells (Braconi et al., 2011), and miRNAs that directly interact with NEAT1 will be discussed below. How epigenetic mechanisms (such as histone modifications) contribute to the transcriptional control of NEAT1 expression warrants further investigation.

#### NEAT1 CONTROLS CANCER INITIATION AND PROGRESSION

NEAT1 inhibits cell cycle arrest and apoptosis, but promotes migration, invasion, metastasis, epithelial-to-mesenchymal transition (EMT), stem cell-like phenotype, chemoresistance and radioresistance, through at least three different molecular mechanisms (**Figure 4** and **Table 1**): (i) NEAT1 functions as a scaffold RNA molecule by interacting with EZH2 (a subunit of the polycomb repressive complex) to influence the expression of downstream effectors of EZH2, (ii) NEAT1 acts as miRNA sponges to antagonize the interactions between multiple tumor suppressor miRNAs and target mRNAs, and (iii) NEAT1 suppresses the expression of miR-129 by promoting the DNA methylation of the miR-129 promoter region.

In gastric cancer, laryngeal squamous cell cancer, pancreatic cancer, oral squamous cell carcinoma, nasopharyngeal carcinoma and breast cancer, knockdown of NEAT1 via small interfering RNA (siRNA) inhibited cell proliferation (Ma et al., 2016; Wang P. et al., 2016; Cheng and Guo, 2017; Huang et al., 2017, 2018; Qian et al., 2017). Similarly, the inhibition of NEAT1 via siRNA reduced cancer cell proliferation, migration and invasion in esophageal squamous cell carcinoma, endometrial cancer, glioma, hepatocellular carcinoma and renal cell carcinoma (Chen et al., 2015; Li Z. et al., 2016; Zhen et al., 2016; Ning et al., 2017; Wang Z. et al., 2017). Furthermore, short hairpin RNA (shRNA)-mediated silencing of NEAT1 significantly impaired cell proliferation, migration and invasion in cholangiocarcinoma (Zhang C. et al., 2018). Conversely, forced expression of NEAT1 enhanced cell growth, migration and invasion in endometrial cancer, papillary thyroid cancer, nasopharyngeal carcinoma, ovarian cancer, non-small cell lung cancer and breast cancer (Li Z. et al., 2016; Sun et al., 2016; Cheng and Guo, 2017; Ding et al., 2017; Wang J. et al., 2017; Zhao et al., 2017; Zhang H. et al., 2018).

EMT is a complex process in which epithelial cells acquire the characteristics of invasive mesenchymal cells and has been shown to contribute to tumorigenesis, invasion, metastasis, resistance to conventional chemotherapy, radiotherapy and small-moleculetargeted therapy (Dong et al., 2016; Nieto et al., 2016; Huo et al., 2017). Cancer stem cells (CSCs) are a class of pluripotent cells that possess a capacity for self-renewal and are resistant to chemotherapy and radiotherapy (Ayob and Ramasamy, 2018). Previous studies have established a link between EMT and CSC formation (Mani et al., 2008; Dong et al., 2014). NEAT1 overexpression promotes EMT and invasion in breast cancer, renal cell carcinoma, renal cell carcinoma, hepatoblastoma and nasopharyngeal carcinoma (Lu et al., 2016; Fu et al., 2017; Liu et al., 2017a; Li W. et al., 2017; Ning et al., 2017; Zhang et al., 2017; Zheng et al., 2018). Importantly, NEAT1 was found to be overexpressed in CD133<sup>+</sup> glioma stem cells and knockdown of NEAT1 by siRNA reduced the ability of these cells to form colonies in soft agar (Yang et al., 2017). Glioma stem cells transfected with NEAT1 shRNA exhibited weaker proliferation, migration and invasion than that of those cells transfected with control shRNA (Gong et al., 2016). In sphere-forming cells generated from certain non-small cell lung cancer cell lines, downregulation of NEAT1 resulted in a significant decrease in the expression of CSC markers (CD133, CD44, ABCG2, Sox2, Nanog, and Oct-4) (Jiang et al., 2018). Not surprisingly, NEAT1 silencing by siRNA sensitized tumor cells to anti-cancer drugs such as sorafenib (Liu et al., 2017a), cisplatin (Hu et al., 2018; Liu F. et al., 2018), dexamethasone (Wu and Wang, 2018), and paclitaxel (An et al., 2017). shRNA-mediated reduction of NEAT1 enhanced the in vitro radiosensitivity of cancer cell lines to radiation therapy (Lu et al., 2016; Han et al., 2017).

NEAT1-1 and NEAT1-2 appear to have distinct roles in regulating the phenotypes of cancer cells. For example, in colorectal cancer cell lines, knockdown of NEAT1-1 could inhibit cell invasion and proliferation, whereas knockdown of NEAT1- 2 promoted cell growth (Wu et al., 2015). The same study showed that expression of NEAT1-1 was significantly higher in liver metastatic lesions compared with adjacent normal colorectal tissues and primary colorectal cancer tissues (Wu et al., 2015). These findings suggested that NAT1-1 may act as a carcinogenic factor, while NEAT1-2 may be a tumor suppressor in colorectal cancer. It appears that the expression of NEAT1 isoforms is regulated in a cell type-specific manner. For example, in adult mouse tissues, NEAT1-1 is expressed in a broad range of cell types (Nakagawa et al., 2011). However, NEAT1-2 expression is largely restricted to the epithelial layers of digestive tissues (Nakagawa et al., 2011). The compositions of NEAT1 isoforms may vary in different cancer types. These dynamic changes in NEAT1 isoform expression may affect the main function of NEAT1 as an oncogene or a tumor suppressor. Therefore, elucidating the exact contribution of NEAT1 isoforms to the development of human tumors would be an exciting direction for future studies.

Mechanistically, NEAT1 recruits EZH2 to form the PRC2 complex and mediate the expression of EZH2 target genes, thereby promoting tumor cell growth and invasion in glioblastoma and cholangiocarcinoma (Chen et al., 2018; Zhang C. et al., 2018; **Figure 4** and **Table 1**).

Additionally, by sponging a set of miRNAs, such as let-7a (Liu F. et al., 2018), let-7e (Gong et al., 2016), miR-34a (Ding et al., 2017; Liu et al., 2017a), miR-34c (Hu et al., 2018), miR-101 (Qian et al., 2017; Wang Y. et al., 2017), miR-106b (Sun et al., 2018), miR-107 (Wang P. et al., 2016; Yang et al., 2017), miR-124 (Chai et al., 2016; Cheng and Guo, 2017; Liu X. et al., 2018), miR-193a (Wu and Wang, 2018), miR-193b (Han et al., 2017), miR-194 (An et al., 2017; Wang H. et al., 2017), miR-204 (Lu et al., 2016), miR-211 (Li X. et al., 2017), miR-214 (Li J. H et al., 2017; Wang J. et al., 2017), miR-218 (Zhao et al., 2017), miR-365 (Huang et al., 2018), miR-377 (Sun et al., 2016), miR-449 (Zhen et al., 2016), miR-506 (Huang et al., 2017), and miR-613 (Wang Z. et al., 2017), NEAT1 could abolish miRNAs-mediated suppression of their target genes, therefore promoting tumor cell growth, migration, invasion, metastasis, EMT, stem cell-like phenotype, chemoresistance and radioresistance (**Figure 4** and **Table 1**).

Changes in miRNA expression through altered DNA methylation affect tumor progression. NEAT1 was shown to silence the expression of miR-129 by increasing the DNA methylation level in its promoter region in breast cancer cells (Lo et al., 2016; **Table 1**).

Of note, NEAT1 overexpression induced the activity of the Wnt/β-catenin signaling, either by inhibiting the expression of miR-214 (Wang J. et al., 2017) or by interacting with EZH2 to

#### TABLE 1 | NEAT1 is a key regulator of cancer initiation and progression.



*NA, Undetermined.*

reduce the expression of negative regulators of WNT/β-catenin signaling such as ICAT, GSK3B and Axin2 (Chen et al., 2018 **Table 1**).

#### NEAT1 AS A POTENTIAL DIAGNOSTIC BIOMARKER OF CANCER

Cancer type- or subtype-specific expression patterns and the sensitive and inexpensive quantitative detection methods for ncRNAs make lncRNAs suitable as hopeful biomarkers for cancer diagnosis and prognosis (Brunner et al., 2012; Bijnsdorp et al., 2017). LncRNAs can exhibit different expression patterns between subtypes of the same cancer (Su et al., 2014). Several lncRNAs (such as PCA3, PCGEM1 and PCAT-1) are highly specific to prostate cancer (Crea et al., 2014) and have been used to identify primary tumors.

Although lncRNAs are commonly expressed at lower levels than protein-coding mRNAs, some lncRNAs (including NEAT1) exhibit moderate to high levels of expression in clinical samples. Moreover, lncRNAs can be detected in body fluids such as serum, plasma, urine and saliva of cancer patients (Chandra Gupta and Nandan Tripathi, 2017). The diagnostic value of NEAT1 was validated in a cohort of colorectal cancer patients, because the whole-blood NEAT1 expression was significantly increased in cancer patients than persons without cancer (Wang Y. et al.,

2017). Therefore, NEAT1 may be possibly used as a biomarker for the existence of primary cancer. The same study also reported that the whole-blood expression of NEAT1-1 was significantly higher in colorectal cancer patients with distant metastasis than in those without metastasis (Wang Y. et al., 2017). Thus, the whole-blood NEAT1 expression might serve as a biomarker for prediction of distant metastasis.

### NEAT1 AS A POTENTIAL THERAPEUTIC TARGET IN CANCER

Given the importance of the lncRNA-mediated networks that broadly affect cancer imitation and progression, lncRNAs are of interest for developing novel therapeutics. LncRNAs can be targeted by multiple approaches (Parasramka et al., 2016; Worku et al., 2017; Arun et al., 2018; **Figure 5**): (i) CRISPR/Cas-9 system can be used to knockout lncRNA; (ii) lncRNA transcript can be knocked down using lncRNA-specific siRNA and antisense oligonucleotide (ASO); (iii) small synthetic molecules/peptides/aptamers can be designed to block and antagonize the binding of lncRNAs with their binding partners (such as protein, DNA, RNA, or other interacting complexes); (iv) selective PRC1/2 inhibitors such as GSK343 (Ihira et al., 2017) can be used to inhibit the activity of lncRNA that function through chromatin remodeling; (v) miRNA regulates the expression of lncRNA, thus restoration or blocking of a specific miRNA using miRNA mimics or anti-miRNA inhibitor (such as locked nucleic acid and miRNA sponge) can indirectly alter the expression of lncRNA; (vi) synthetic lncRNA mimics could be used to restore the expression of tumor suppressor lncRNA; and (vii) approaches to target lncRNAs can be used in combination with other therapies such as chemotherapy and radiotherapy to enhance their effectiveness. For example, targeted depletion of NEAT1 by siRNA (Li J. H et al., 2017) or shRNA (Zhang C. et al., 2018; Zhang H. et al., 2018) displayed a significant therapeutic effect in vivo. In vivo studies also confirmed that knockdown of NEAT1 sensitized tumor cells to cisplatin/paclitaxel- or radio-induced tumor regression (An et al., 2017; Han et al., 2017; Hu et al., 2018).

Apart from these strategies, targeting the upstream signaling pathways of NEAT1 or its binding partners would be another method to modulate NEAT1 expression or function in malignant tumors. Despite their potential side effects, blocking of NFκB and STAT3 activity may represent a good approach to combat NEAT1-overexpressing tumors (Grivennikov and Karin, 2010; Liby et al., 2010). In addition, hypoxia is frequently observed in solid tumors and HIF-1/-2 has been potential targets for developing novel cancer therapeutics. A number of small molecule inhibitors of HIF-2 have been developed and might be used to downregulate the expression of NEAT1 (Yu et al., 2017). HuR is overexpressed in a wide variety of cancer types and stabilizes a large subset of mRNAs, which encode proteins implicated in tumor cell proliferation, survival, angiogenesis, invasion and metastasis (Kotta-Loizou et al., 2016). High-throughput screening methods have been established to identify small molecular molecules against HuR (Meisner et al., 2007). Thus, HuR inhibition may offer new conceptual routes to treat cancers expressing high levels of NEAT1.

#### PERSPECTIVES AND CHALLENGES

The current research on NEAT1 is still at a very early stage, but growing evidence has clearly identified NEAT1 as an attractive biomarker of cancer and an ideal candidate for lncRNA therapeutics. Although the broad implications of NEAT1 in cancers have been described, several key challenges associated with NEAT1 still exist: (i) The exact mechanisms for NEAT1 mediated carcinogenesis and metastasis remain elusive, although the interactions between PRC2 complex and miRNAs appears to be at least involved in these processes. High-throughput technologies such as ChIRP, PAR-CLIP and iCLIP have been used to reveal NEAT1-DNA or NEAT1-protein interactions (Hafner et al., 2010; Zhao et al., 2010; Chu et al., 2011; Tollervey et al., 2011; Yoon et al., 2014). (ii) The precise mechanisms by which the genetic and epigenetic factors contribute to NEAT1

#### REFERENCES


dysregulation, as well as the downstream signaling pathways of NEAT1, are still far from clear. (iii) The NEAT1 loci might be targeted for therapeutic purposes using CRISPR/Cas-9 genome editing technology. However, a challenge is the safety evaluation of CRISPR/Cas-9-based therapeutics, such as the risk of disease occurrence due to unwanted mutations, the immune response to the delivery system and other toxic side effects. Taken together, NEAT1 has the potential to provide a new biomarker for the diagnosis and monitoring of cancers and will need to be taken into consideration in the development of NEAT1-targeting therapeutics (possibly in combination with chemotherapy and radiotherapy).

#### AUTHOR CONTRIBUTIONS

PD and HW provided direction. PD, YX, and HW wrote the manuscript. JY, SH, NK, and YT made significant revisions to the manuscript. All authors read and approved the final manuscript.

#### FUNDING

This work was supported by a grant from JSPS Grant-in-Aid for Scientific Research (C) (16K11123 and 18K09278), the Science and Technology Planning Project of Guangdong Province, China (2014A020212124) and an NIH/NCI grant 1R21CA216585-01A1 to JY.


and modulating the Ras-MAPK pathway. Cancer Biol. Ther. 19, 534–542. doi: 10.1080/15384047.2018.1450119


in ovarian cancer. Cell. Physiol. Biochem. 44, 948–966. doi: 10.1159/0004 85395


metastasis of cholangiocarcinoma cells. Oncol. Res. 26, 879–888. doi: 10.3727/096504017X15024935181289


**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 Dong, Xiong, Yue, Hanley, Kobayashi, Todo and Watari. 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.

# Interpreting Non-coding Genetic Variation in Multiple Sclerosis Genome-Wide Associated Regions

Elvezia Maria Paraboschi<sup>1</sup> , Giulia Cardamone<sup>1</sup> , Giulia Soldà1,2, Stefano Duga1,2 and Rosanna Asselta1,2 \*

<sup>1</sup> Department of Biomedical Sciences, Humanitas University, Milan, Italy, <sup>2</sup> Humanitas Clinical and Research Center, Milan, Italy

Multiple sclerosis (MS) is the most common neurological disorder in young adults. Despite extensive studies, only a fraction of MS heritability has been explained, with association studies focusing primarily on protein-coding genes, essentially for the difficulty of interpreting non-coding features. However, non-coding RNAs (ncRNAs) and functional elements, such as super-enhancers (SE), are crucial regulators of many pathways and cellular mechanisms, and they have been implicated in a growing number of diseases. In this work, we searched for possible enrichments in noncoding elements at MS genome-wide associated loci, with the aim to highlight their possible involvement in the susceptibility to the disease. We first reconstructed the linkage disequilibrium (LD) structure of the Italian population using data of 727,478 single-nucleotide polymorphisms (SNPs) from 1,668 healthy individuals. The genomic coordinates of the obtained LD blocks were intersected with those of the top hits identified in previously published MS genome-wide association studies (GWAS). By a bootstrapping approach, we hence demonstrated a striking enrichment of non-coding elements, especially of circular RNAs (circRNAs) mapping in the 73 LD blocks harboring MS-associated SNPs. In particular, we found a total of 482 circRNAs (annotated in publicly available databases) vs. a mean of 194 ± 65 in the random sets of LD blocks, using 1,000 iterations. As a proof of concept of a possible functional relevance of this observation, we experimentally verified that the expression levels of a circRNA derived from an MS-associated locus, i.e., hsa\_circ\_0043813 from the STAT3 gene, can be modulated by the three genotypes at the disease-associated SNP. Finally, by evaluating RNA-seq data of two cell lines, SH-SY5Y and Jurkat cells, representing tissues relevant for MS, we identified 18 (two novel) circRNAs derived from MS-associated genes. In conclusion, this work showed for the first time that MS-GWAS top hits map in LD blocks enriched in circRNAs, suggesting circRNAs as possible novel contributors to the disease pathogenesis.

Keywords: multiple sclerosis, single-nucleotide polymorphism, association, long non-coding RNA, circular RNA, micro RNA, super-enhancer

#### Edited by:

Ge Shan, University of Science and Technology of China, China

#### Reviewed by:

Jun Yasuda, Miyagi Cancer Center, Japan Murray John Cairns, The University of Newcastle, Australia

> \*Correspondence: Rosanna Asselta rosanna.asselta@hunimed.eu

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 31 July 2018 Accepted: 30 November 2018 Published: 17 December 2018

#### Citation:

Paraboschi EM, Cardamone G, Soldà G, Duga S and Asselta R (2018) Interpreting Non-coding Genetic Variation in Multiple Sclerosis Genome-Wide Associated Regions. Front. Genet. 9:647. doi: 10.3389/fgene.2018.00647

Multiple sclerosis (MS) is a chronic autoimmune disease of the central nervous system, characterized by demyelination and progressive neurological impairment (Brownlee et al., 2017). Epidemiological studies showed both an important role of the environment in determining MS risk (Ramagopalan et al., 2010), and a strong contribution of genetic components (Belbasis et al., 2015). To date, besides the human leukocyte antigen (HLA) gene cluster (Patsopoulos et al., 2013), genome-wide association studies (GWAS) identified several common variants contributing to disease pathogenesis with mild effects on risk, many of which located within or close to genes displaying primarily immunologic functions (International Multiple Sclerosis Genetics Consortium [IMSGC] et al., 2007, 2011, 2013; Aulchenko et al., 2008; Comabella et al., 2008; Australia and New Zealand Multiple Sclerosis Genetics Consortium (ANZgene), 2009; Baranzini et al., 2009; de Jager et al., 2009; Jakkula et al., 2010; Nischwitz et al., 2010; Sanna et al., 2010; Patsopoulos et al., 2011; Martinelli-Boneschi et al., 2012; Matesanz et al., 2012). Despite these extensive efforts, the identified GWAS variants explain only 28% of the sibling recurrence risk (International Multiple Sclerosis Genetics Consortium [IMSGC] et al., 2013), thus implicating that the complete spectrum of MS genetic determinants is still far from being complete. These studies focused primarily on proteincoding genes, due to the difficulty of interpreting non-coding features. However, advances in the systematic annotation of non-coding genes and non-coding functional elements are revolutionizing genetic approaches and are paving the way to build a map that can help reveal "hidden" processes underlying disease associations (Ward and Kellis, 2012).

In this frame, non-coding RNAs (ncRNAs) have recently emerged as crucial regulators of many pathways and cellular mechanisms (Vidigal and Ventura, 2015; Barrett and Salzman, 2016; Quinn and Chang, 2016), and they have been implicated in a growing number of diseases (Mendell and Olson, 2012; Vuci ˇ cevi ´ c et al., 2014 ´ ). Many long ncRNAs (lncRNAs), for instance, were shown to contribute to the pathogenesis of neurological and psychiatric conditions in different ways, from regulation of transcription to modulation of RNA processing and translation (Vuci ˇ cevi ´ c et al., 2014 ´ ). In addition, microRNAs (miRNAs) dysregulation was associated with several disorders, such as different kinds of cancers and immune-related diseases (Mendell and Olson, 2012). Another group of ncRNAs with regulatory functions is represented by circular RNAs (circRNAs), a novel class of RNAs generated from the back-splicing of exons or introns (Jeck et al., 2013). By acting as miRNA sponges, or by binding to RNA-associated proteins, circRNAs regulate gene expression at the transcriptional or post-transcriptional level, although their exact mechanism of action still needs to be clarified (Greene et al., 2017). Moreover, they have been associated with human diseases such as ischemic heart disease, Alzheimer's disease, diabetes, cancer, as well as MS (Cardamone et al., 2017; Greene et al., 2017; Iparraguirre et al., 2017).

Among non-coding functional elements, also super-enhancers (SEs) have been described as key gene expression regulators (Pott and Lieb, 2015). SEs are genomic regions characterized by a strong enrichment in binding sites both for transcriptional coactivators, specifically the Mediator protein, and for factors generally associated with enhancer activity, such as RNA polymerase II and chromatin factors (Pott and Lieb, 2015). Very interestingly, many SE regions are significantly enriched in disease-associated single-nucleotide polymorphisms (SNPs), including those related to autoimmunity, and more specifically to MS (Hnisz et al., 2013; Farh et al., 2015). The enrichment in GWAS variants within enhancers suggests that they influence the disease risk by altering gene regulation. However, only a few disease-associated SNPs directly alter a transcription factor motif; many trait-associated SNPs instead modulate the enhancer activity by changing nearby nucleotides, resulting in slight but critical alterations of gene expression (Farh et al., 2015).

In this work, we aim at identifying ncRNAs and SEs mapping in proximity of MS GWAS-significant signals that could point to so-far unexplored mechanisms involved in the susceptibility to the disease.

### MATERIALS AND METHODS

#### Defining the Linkage Disequilibrium (LD) Structure of the Italian Population

The global LD structure of the Italian population was explored by using genome-wide genotyping data (727,478 quality-checked markers, genotyped with the Affymetrix 6.0 GeneChip platform; Affymetrix, Santa Clara, CA, United States) obtained from 1,668 healthy controls (for genotyping details see Myocardial Infarction Genetics Consortium et al., 2009). Haplotype blocks were estimated with the Plink program (Purcell et al., 2007) following the default procedure described for the Haploview software (Barrett et al., 2005). Pairwise LD was calculated for SNPs within 200 kb for autosomal chromosomes. Chromosome X was excluded from this analysis, leading the total number of SNPs used for LD studies to 699,676.

To verify whether the Italian LD structure was comparable to the European one, we analyzed the 1000 Genomes data on European subjects (phase 1 project) (1000 Genomes Project Consortium et al., 2015). This test was performed on chromosome 22 data, by selecting only those SNPs whose genetic information was available both in the Italian and 1000 Genome populations. These were used to calculate the European LD structure.

### Retrieving the Reference Files for ncRNAs and Regulatory Elements

Reference files for the analysis were retrieved for lncRNAs, miRNAs, circRNAs, and SEs. In particular: (1) The reference gene transfer format (GTF) file for lncRNAs was obtained from GENCODE (Harrow et al., 2012), selecting the comprehensive gene annotation of lncRNA genes on the reference chromosomes, version 25<sup>1</sup> . (2) The miRNA reference file was downloaded from miRBase<sup>2</sup> (Griffiths-Jones, 2004; Griffiths-Jones et al., 2006, 2008;

<sup>1</sup>https://www.gencodegenes.org/human/release\_25lift37.html <sup>2</sup>http://www.mirbase.org/

Kozomara and Griffiths-Jones, 2011, 2014) version 20. (3) The circRNA reference file was obtained from the circBase database<sup>3</sup> (Glažar et al., 2014), by downloading data from all the available studies on humans (Jeck et al., 2013; Memczak et al., 2013; Salzman et al., 2013; Zhang et al., 2013; Rybak-Wolf et al., 2015). (4) The SE reference file was downloaded from the SEA: Super-Enhancer Archive<sup>4</sup> (Wei et al., 2016) based on studies on humans.

In all cases, genome version hg19 was considered; databases were accessed on April 2016.

#### Defining Overlapping Regions Between LD Blocks, MS Genome-Wide Significant SNPs, and ncRNAs/SEs

Multiple sclerosis-associated SNPs, excluding those mapping in the highly complex HLA region, were retrieved from the literature (**Supplementary Table 1**) (International Multiple Sclerosis Genetics Consortium [IMSGC] et al., 2013). Their genomic coordinates were crossed with those of the LD blocks, to identify the blocks in which each single SNP resides.

The next step was searching for partial/total overlapping between LD blocks containing the genome-wide associated SNPs and the different classes of ncRNAs (lncRNAs, miRNAs and circRNAs) or SE elements. The overlaps were identified on the basis of the genomic coordinates of each LD block (using as borders the physical positions of the most 5<sup>0</sup> and 3<sup>0</sup> SNPs belonging to the block) and of each ncRNAs/SE elements (for these genomic features, coordinates were extracted from the reference files described in the previous section). The final list includes both the elements completely contained within the LD blocks and those showing only a partial overlap. Filtering for redundancy was used to eliminate multiple annotations referring to the same element. All procedures were performed using awk command line (described in the section **Supplementary Material**).

### Enrichment Analysis

To determine if the MS-related LD blocks are significantly enriched in ncRNA genes and SE elements, a bootstrapping strategy was adopted.

First, a set of random SNPs was extracted from the "Genome-Wide Human SNP array 6.0" manifest (copy number variants were excluded), which is one of the most used genotyping arrays in MS GWAS. The number of SNPs to be extracted was chosen in order to obtain either a number of LD blocks similar to the one of the MS-related analysis (Random set I) or an overall genomic region of equal length (i.e., 3.8 Mb; Random set II). Again, the HLA region and X chromosome were avoided. Moreover, since about half of the MS-associated SNPs are located in introns (**Supplementary Table 1**) (International Multiple Sclerosis Genetics Consortium [IMSGC] et al., 2013), the random SNP sets were constructed to mirror the proportion of intronic SNPs of the MS list. More in particular, to perform this step,

<sup>3</sup>http://www.circbase.org/

<sup>4</sup>https://sea.nebulagene.com/SEA/index.html

two complete lists of SNPs from the "Genome-Wide Human SNP array 6.0" manifest were generated: one containing only SNPs annotated as intronic in the manifest file, the second containing only extragenic SNPs. SNPs were chosen from both lists with a randomized procedure (using the gshuf Unix command), respecting the constraints above mentioned.

Then, the LD blocks in which the random SNPs reside were identified, and a search for overlapping regions between LD blocks and ncRNAs/SE regions was performed, as described above.

Finally, the results were filtered to avoid redundancy, and the total number of lncRNAs, miRNAs, circRNAs, and SEs was annotated. The entire procedure was repeated 1,000 times for random set I and II, and the outputs of each set averaged, in order to compare the resulting means with the result obtained with the MS SNP set. The comparison was based on the % of times in which the same (or a larger) number of lncRNAs, circRNAs, miRNAs, or SEs was obtained in the 1,000 iterations respect to the MS dataset. Enrichment p-values were calculated according to Davison and Hinkley method (Davison and Hinkley, 1997).

All analyses were performed using in-house developed Perl scripts (listed in the section **Supplementary Material**).

The entire procedure is schematized in **Supplementary Figure 1**.

#### Replication on an Unrelated Disease

To test the specificity of the analysis on MS, we repeated the entire workflow considering a disease with a completely different etiology, i.e., coronary artery disease (CAD).

The list of CAD-associated SNPs was derived from the literature (**Supplementary Table 2**) (Nikpay et al., 2015).

### Genotype-Dependent Analysis of circRNA Expression

DNA samples were extracted from whole blood of 35 healthy donors using an automated DNA extractor (Maxwell 16 System; Promega, Madison, WI, United States). All subjects gave written informed consent in accordance with the Declaration of Helsinki. To genotype the MS-associated SNP rs2293152, PCR amplifications (GoTaq; Promega) and Sanger sequencing, using the BigDye Terminator Cycle Sequencing Ready Reaction Kit v1.1 and an ABI-3500 Genetic Analyzer (Thermo Fisher Scientific, Waltham, MA, United States), were performed following standard protocols.

Peripheral blood mononuclear cells (PBMCs) of the same healthy donors were isolated by means of centrifugation on a Lympholyte Cell separation medium (Cederlane Laboratories Limited, Hornby, ON, Canada) gradient. RNA extraction was performed using the EuroGold Trifast kit (Euroclone, Wetherby, United Kingdom). RNA was reverse-transcribed using the Superscript-III Reverse Transcriptase (Thermo Fisher Scientific) and random hexamers (Promega), according to the manufacturers' instructions.

Semi-quantitative real-time RT-PCRs to detect the expression levels of circRNA hsa\_circ\_0043813 were performed by using divergent primers (5<sup>0</sup> -ACATTCTGGGCACAAACACA-3<sup>0</sup> and

5 0 -CCTCTGAGAGCTGCAACG-3<sup>0</sup> ), the FastStart SYBR Green Master mix (Roche, Basel, Switzerland), and a LightCycler 480 (Roche). HMBS (hydroxymethylbilane synthase) was used as housekeeping gene; reactions were performed in triplicate, and expression data were analyzed using the GeNorm software (Vandesompele et al., 2002).

#### CircRNA Analysis by RNA Sequencing

RNA was extracted using the Maxwell 16 LEV simplyRNA Cells Kit (Promega) from SH-SY5Y (human neuroblastoma) and Jurkat E6-1 (human T lymphocyte) cell lines. RNA quality was assessed by the LabChip GX Touch instrument (PerkinElmer, Waltham, MA, United States). RNA sequencing was performed using the TruSeq Stranded Total RNA Library Prep Kit (Illumina, San Diego, CA, United States), following the manufacturer's instructions and a paired-end sequencing strategy. SH-SY5Y and Jurkat samples underwent a high-coverage paired-end 75- and 150-bp strand-specific sequencing, respectively, using a NextSeq 500 platform (Illumina).

The circRNA analysis was then performed using the DCC software (Cheng et al., 2016). In detail, raw reads were first aligned to the hg19 version of the genome using STAR (Dobin et al., 2013), switching on the detection of chimeric alignments to detect reads containing backspliced products, as suggested by the DCC manual. In a first step, reads were mapped using both mates; subsequently, an additional separate mate mapping was performed. After mapping, DCC was used to analyze the chimeric reads to detect circRNAs. Only those circRNAs supported by at least five reads were considered for further analyses. CircRNAs mapping on mitochondrial DNA or in repetitive regions of the genome were filtered out.

#### Data Repository

Raw sequence files of SH-SY5Y have been deposited in NCBI Sequence Read Archive (SRA) under the following Bioproject ID: PRJNA483101, and with the accession number SRP155458; raw sequence files of Jurkat cells have been deposited in the GEO database (Accession No.: GSE110525).

### RESULTS

### The Global LD Structure of the Italian Population Is Not Different From the European One

The LD structure of the Italian population was built by using data on 699,676 genotyped SNPs on 1,668 healthy subjects. A total of 96,666 LD blocks (with an average length of 17.48 kb; range 0.01– 200 kb) were identified in autosomes, ranging from 1,421 blocks of chromosome 22 to 8,148 of chromosome 2 (on average: 4,394 blocks per chromosome).

The comparison of the LD structure of chromosome 22 of the Italian and European populations confirmed a substantial similarity in the blocks distribution (Pearson's χ <sup>2</sup> p = 0.79), thus suggesting that the Italian population is a good representation of the European LD structure, and confirming the data previously obtained by Mueller et al. (2005). The substantial overlap between the structures of the Italian and of the European populations was also confirmed by a principal component analysis performed using genotype data of our cohort and those from the 1000 Genome project (503 available individuals; **Supplementary Figure 2**).

#### MS-Associated Regions Are Significantly Enriched in circRNAs and SEs

With the aim of identifying ncRNAs and SE regulatory elements mapping in LD blocks harboring MS GWAS-significant signals, we selected through literature data mining all those SNPs that reached a genome-wide significant threshold in MS GWAS studies and meta-analyses (**Supplementary Table 1**). We excluded from this list all SNPs mapping in the HLA region as well as those located on the X chromosome. The list of 97 SNPs was intersected with that of LD blocks inferred for the Italian population. Our automated pipeline allowed the identification of 73 LD blocks, each harboring a single genome-wide significant SNP. For 24 out of the 97 SNPs used for the analysis, it was not possible to establish a precise LD block.

TABLE 1 | Results of the analysis performed on the MS-associated loci (upper part) and on 1,000 random sets (middle and lower part).


§Average length of the LD blocks. <sup>∗</sup>For the random sets analysis, the average values calculated on 1,000 iterations are indicated. ∗∗% of times in which the same or a larger number of lncRNAs, circRNAs, miRNAs, or SEs was obtained in the 1,000 iterations as compared to the MS dataset. p-Values were calculated as described (Davison and Hinkley, 1997); in detail: p-value = [1+sum(s > = s0)]/(N + 1), where s is the observed value in the random set, s0 is the value of the observed MS-specific result, and N is the number of bootstraps. n, number; SD, standard deviation.

The genomic coordinates of the 73 LD blocks were hence intersected with those of ncRNAs and SEs annotated in public genomic databases. This analysis evidenced the presence of 30 lncRNAs, 482 circRNAs, 7 miRNAs, and 23 SEs partially or totally overlapping the 73 identified blocks (**Table 1**). To test for possible significant enrichments in non-coding elements, the same workflow was applied to randomly selected SNPs (Random set I). To this aim, subsets of 94 SNPs were randomly selected from the genome, and the process was repeated 1,000 times. The pipeline identified, on average, 73.8 LD blocks (median value: 74) in which 22.6 lncRNAs (median: 22), 193.8 circRNAs (median: 185), 2.1 miRNAs (median: 2), and only 2.4 SEs (median: 2) were located (**Table 1**). Comparing the random set I results with the MS dataset, we obtained the same or a larger number of lncRNAs, circRNAs, and miRNAs in the 11, 0, and 3.6%, of the iterations, respectively, thus indicating a very strong enrichment in circRNAs in MS dataset. None of the randomly selected SNP subsets evidenced the same or a larger number of SE hits when compared to the MS list, thus confirming the enrichment in these regulatory elements previously reported by Farh et al. (2015) (observations that, however, were focused specifically on immune-related genes). Since the average length of the LD blocks in the Random set I was 22% lower than the one of MS LD blocks, we repeated the enrichment analysis on a genomic region spanning the same length as the MS dataset (3.8 Mb; Random set II). Also in this case, 1,000 iterations confirmed a strong enrichment in circRNAs and SEs in MS LD blocks (**Table 1**).

To test the specificity of the results, we applied the same pipeline on another dataset, composed of SNPs associated with CAD, a disease with a different etiology from MS. In this case, 55 SNPs were retrieved from the literature and used for the analysis, and 36 LD blocks were identified (**Table 2**). The workflow evidenced the presence of 19 lncRNAs, 122 circRNAs, 1 miRNA, and 2 SEs mapping within the corresponding blocks. When compared to the CAD dataset, the random bootstrapping strategy applied 1,000 times evidenced the same (or a larger) number of lncRNAs, circRNAs, miRNAs, and SEs in the 4.6, 12.4, 45.9, and 26.7% of the iterations, respectively (**Table 2**), thus suggesting only a slight enrichment in lncRNAs in the CAD dataset. No significant enrichment was evidenced in circRNAs and SEs.


<sup>∗</sup>For the random sets analysis, the average values calculated on 1,000 iterations are indicated. n, number; SD, standard deviation. ∗∗% of times in which the same or a larger number of lncRNAs, circRNAs, miRNAs, or SEs was obtained in the 1,000 iterations as compared to the CAD dataset. p-Values were calculated as described in the footnote of Table 1.

FIGURE 1 | Characterization of the hsa\_circ\_0043813 circRNA deriving from the STAT3 gene. (A) Schematic representation of the STAT3 genomic region spanning from exon 12 to 14. Exons are depicted as boxes (in scale), and introns as lines. The position of the SNP rs2293152 is shown by an arrow. (B) Schematic representation of the formation of the STAT3 circRNA hsa\_circ\_0043813 through a back-splicing event between exons 14 and 12. Exons are approximately drawn to scale; the curved arrow joins the 5<sup>0</sup> splice site of exon 14 to 3<sup>0</sup> splice site of exon 12. On the right, a schematic representation of the circRNA is depicted; arrows below exon 12 and 14 indicate the divergent primer couple used to detect the circRNA. Below the scheme, direct-sequencing electropherogram shows the head-to-tail splice junction, indicated by a black arrow, located between exons 14 and 12. (C) Boxplots showing expression levels of the hsa\_circ\_0043813 circRNA measured by semi-quantitative real-time RT-PCR in PBMCs of 35 healthy controls. Boxes define the interquartile range; the thick line refers to the median. Results were normalized to expression levels of the HMBS housekeeping gene, and for each sample three technical replicates were performed. The number of subjects belonging to each group is also indicated (n). The significance level of t-test analysis is shown. <sup>∗</sup>p < 0.05; ns, not significant.

The list of circRNAs/SE elements mapping within MSor CAD-specific blocks is given in the **Supplementary Table 3**.

is also shown. Yellow circles represent those circRNAs derived from MS-associated genes. (C) Venn diagrams representing the number of circRNAs that were already described in circBase for each cell line.

### SNP rs2293152 Genotype Influences STAT3 hsa\_circ\_0043813 Expression Levels

The newly observed enrichment in circRNAs in MS led us to test, as a proof of concept, whether the different genotypes of an MSassociated SNP could influence the expression levels of a circRNA mapping in the corresponding LD block. To this aim, we decided to better characterize a circRNA deriving from STAT3 (Signal Transducer and Activator of Transcription 3), a gene necessary for pro-inflammatory cytokines signaling (Adamson et al., 2009) and that it is required for differentiation and expansion of Th17 cells, key players of MS disease activity (Brucklacher-Waldert et al., 2009). In particular, four different SNPs mapping in STAT3 have been described as associated with MS in GWA studies: rs744166 (Jakkula et al., 2010), rs2293152 (Patsopoulos et al., 2011), rs9891119 (International Multiple Sclerosis Genetics Consortium [IMSGC] et al., 2011), and rs4796791 (International Multiple Sclerosis Genetics Consortium [IMSGC] et al., 2013). Among them, rs2293152 is in close proximity to a proteincoding exon, being located 50 nt upstream of STAT3 exon 14 (**Figure 1A**), and could in theory affect the expression levels of the circRNAs hsa\_circ\_0043813 (CircBase, chr17:40481427- 40481794, hg19), which is composed of exons 12, 13, and 14 (Refseq NM\_003150).

To confirm the existence/expression of hsa\_circ\_0043813, we first performed a RT-PCR assay with a divergent primer couple tagging exons 12 and 13 on RNA extracted from PBMCs of two healthy controls. Direct sequencing of the circRNA product confirmed the presence of the backspliced exons 12 and 14, joined by a head-to-tail splice junction (**Figure 1B**). The analysis of the expression levels of the hsa\_circ\_0043813 circRNA was hence performed on a total of 35 healthy subjects: 6 homozygous for the CC genotype, 16 heterozygous, and 13 homozygous GG (**Figure 1C**). Our data showed significant different expression levels upon genotype stratification, with the CC subjects showing the highest levels of expression (one-way ANOVA p = 0.023).

### CircRNA Landscape in SH-SY5Y and Jurkat T Cell Lines

Due to the striking enrichment in circRNAs mapping in the regions associated with MS, we looked at the circRNA landscape of two cell lines, SH-SY5Y and Jurkat cells, representing tissues relevant for MS, by analyzing high-coverage RNA-seq data already available in our lab. We obtained ∼186 and 197 million reads for SH-SY5Y and Jurkat cells, respectively (**Figure 2A**). The circRNA analysis detected the presence of 539 circRNAs supported by at least five reads in SH-SY5Y cells, and of 2,032 circRNAs in Jurkat cells. About half (52%) of circRNAs identified in SH-SY5Y cells were also present in the Jurkat sample (**Figure 2B**). Most of the detected circRNAs were already annotated in circBase (89% for SH-SY5Y, 68% for Jurkat cells; **Figure 2C**). In **Supplementary Table 3** we listed the 61 and 643


<sup>∗</sup>circRNA position is according to DCC output and referred to version hg19 of the genome. ∗∗circRNA identifier is referred to circBase nomenclature.

circRNAs that were newly identified in SH-SY5Y and Jurkat cells, respectively. Interestingly, we identified 18 (two novel) and 4 circRNAs derived from the MS-associated genes in Jurkat and SH-SY5Y cells, respectively (**Figure 2B** and **Table 3**).

Finally, UBAP2 and WHSC1, and MPP6 and ZNF124 were the genes giving origin to the highest number of different circRNAs species in Jurkat and SH-SY5Y, respectively (**Supplementary Figure 2**). These genes show completely different genomic structures, going from 29 exons distributed on a region of 127 kb in the case of the UBAP2 gene, to 4 exons spread over 16 kb for the ZNF124 gene. Instead, for all these genes we observed highest level of expression in cell lines of lymphoid origin and in SH-SY5Y when compared to other cell lines (source<sup>5</sup> ).

#### DISCUSSION

New classes of ncRNAs have been described over the last years; they all display regulatory functions, being part of a large RNA communication network that ultimately regulates the fundamental cellular functions (Adams et al., 2017). Many of them have in fact emerged as regulators of crucial mechanisms (Vidigal and Ventura, 2015; Barrett and Salzman, 2016; Quinn and Chang, 2016), and evidence suggests their implication in various diseases (Mendell and Olson, 2012; Vuci ˇ cevi ´ c et al., ´ 2014). Given this background, in this work we aimed at identifying possible enrichments in non-coding elements at MS genome-wide associated loci, that could point to their involvement in the susceptibility to the disease.

By taking advantage of the top hits identified in MS GWASs and of the LD structure of the Italian population, we demonstrated a striking enrichment of circRNAs in the LD blocks harboring MS-associated SNPs. This result suggests that this class of ncRNAs could play an important role in the disease predisposition and supports emerging evidence in the literature indicating that a dysregulation of the back-splicing process could be a signature of the disease. More specifically, our group identified in MS patients, for the first time, one dysregulated circRNA (Cardamone et al., 2017) derived from GSDMB, a gene associated with susceptibility to asthma and autoimmune diseases. Subsequently, Iparraguirre et al. (2017) performed a microarray analysis identifying 406 differentially expressed circRNAs and validating two of them (both deriving from the ANXA2 gene). As the biogenesis of circRNAs competes with premRNA splicing (Ashwal-Fluss et al., 2014), alterations in the back-splicing process may also interfere with alternative splicing (AS), a mechanism already demonstrated to be dysregulated in MS (Evsyukova et al., 2010; Paraboschi et al., 2015).

Considering that AS dysregulation has been described as a possible pathogenic mechanism underlying autoimmune diseases (Evsyukova et al., 2010; Paraboschi et al., 2015; Juan-Mateu et al., 2016), and given the tight interconnection between AS and back-splicing, we hypothesized that an enrichment in circRNAs could be a signature also for other autoimmune disorders. Recent findings showed that immune-mediated diseases have a complex network of shared genetic architecture, with ∼70% of the associated loci for each disease being shared with other autoimmune disorders (Farh et al., 2015). We hence investigated whether we could identify a ncRNA signature also in systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA), taking advantage of the GWAS top hits for these diseases (Okada et al., 2014; Bentham et al., 2015). By applying the same pipeline used for MS, we observed in SLE a circRNA enrichment in the LD blocks corresponding to GWAS signals (**Supplementary Tables 3, 4**). This result is in line with a growing body of evidence in the literature of an involvement of circRNAs in the

<sup>5</sup>https://www.proteinatlas.org/

disease: circHLA-C was in fact shown to be increased in patients affected by lupus nephritis (LN), a kidney disease caused by SLE. In addition, circHLA-C was correlated with clinical disease activities and was suggested to act as sponge for miR-150 (which, in turn, positively correlates with renal chronicity index in LN patients) (Luan et al., 2018). Regarding RA, we could not find any circRNA enrichment in the LD blocks corresponding to genomewide associated loci (**Supplementary Table 5**). This finding, however, may not be so surprising: systematic reviews on familiar clustering of autoimmune disorders found evidence of an inverse clustering of RA and MS, suggesting that these two pathologies might be less closely related than other autoimmune diseases (Richard-Miceli and Criswell, 2012).

In our work, by studying STAT3 hsa\_circ\_0043813, we also showed that the expression level of specific circRNAs may be influenced by the genotype of disease-associated SNPs (which might be defined as circ-eQTL). This observation could be very useful in understanding the functional impact of diseaseassociated SNPs, a task that still remains a key challenge of the post-GWAS era. Our hypothesis is that some variants associated with MS may impact on the biogenesis or on the sequence of circRNAs. This is in line with what has been reported for circANRIL, the only example of circRNA for which a link between disease associated SNPs and circRNA biogenesis has been demonstrated (Holdt et al., 2016). CircANRIL derives from the lncRNA ANRIL (Burd et al., 2010), transcribed from the CAD risk locus on chromosome 9p21. Holdt et al. (2016) demonstrated that carriers of the CAD-protective haplotype at this locus have significantly increased expression of circANRIL, and this is inversely correlated with the expression of linear ANRIL (linANRIL). Moreover, highest circANRIL:linANRIL ratios are found in CAD-free patients, thus implying an atheroprotective role of circANRIL. It is therefore likely that SNPs contained in the 9p21 haplotype are responsible for differential circANRIL formation, and that subtle genotypedirected gene expression differences may modulate the risk to develop the disease (Holdt et al., 2016). On the basis of this example, we can speculate that there might be other cases in which a disease-associated SNP exerts its functional effect by modulating the levels of specific circRNAs and, hence, modifying the ratio of the circular:linear isoforms. Of note, the RNA-seq analysis of the circRNA landscape in Jurkat cells highlighted

#### REFERENCES


the existence of 18 circRNAs deriving from seven MS-associated genes (∼8% of the total number of genes here considered; **Supplementary Table 1**). This group of circRNAs, together with their linear counterparts, could be a good starting point for an in-depth analysis of circular:linear isoform ratio in PBMCs of MS patients vs. controls, also in the perspective to find novel, simple, and reliable biomarkers for MS susceptibility and progression.

We are aware that our work has the potential limitation of comparing MS-associated loci, which are by definition nonrandom, with randomly sampled genomic regions. However, we think we have accounted for the main sources of bias by considering regions of equal length and exon density, and by performing a large number of iterations.

In conclusion, this work showed for the first time that MS-GWAS top hits map in LD blocks enriched in circRNAs, suggesting that this feature could be shared by other autoimmune diseases, and pointing to circRNAs as possible novel contributors to the disease pathogenesis.

#### ETHICS STATEMENT

The study was approved by the Ethics Committee of the Humanitas Research Hospital and conducted according to the Declaration of Helsinki. All subject signed an appropriate informed consent.

#### AUTHOR CONTRIBUTIONS

EP and RA conceived and designed the experiments. GC and EP performed the experiments. RA and EP analyzed the data. EP drafted the paper. GC, GS, SD, and RA critically revised the manuscript. RA supervised the entire study.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2018.00647/full#supplementary-material


maps. Bioinformatics 21, 263–265. doi: 10.1093/bioinformatics/ bth457


sclerosis. J. Neuroimmunol. 227, 162–166. doi: 10.1016/j.jneuroim.2010.06 .003


**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 Paraboschi, Cardamone, Soldà, Duga and Asselta. 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.

# Association of mir-196a-2 rs11614913 and mir-149 rs2292832 Polymorphisms With Risk of Cancer: An Updated Meta-Analysis

Jalal Choupani 1†, Ziba Nariman-Saleh-Fam2†, Zahra Saadatian<sup>3</sup> , Elaheh Ouladsahebmadarek <sup>2</sup> , Andrea Masotti <sup>4</sup> \* and Milad Bastami 5,6 \*

1 Immunology Research Center, Tabriz University of Medical Sciences, Tabriz, Iran, <sup>2</sup> Women's Reproductive Health Research Center, Tabriz University of Medical Sciences, Tabriz, Iran, <sup>3</sup> Department of Medical Genetics, Faculty of Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran, <sup>4</sup> Research Laboratories, Bambino Gesù Children's Hospital-IRCCS, Rome, Italy, <sup>5</sup> Drug Applied Research Center, Tabriz University of Medical Sciences, Tabriz, Iran, <sup>6</sup> Department of Medical Genetics, Faculty of Medicine, Tabriz University of Medical Sciences, Tabriz, Iran

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Peter Igaz, Semmelweis University, Hungary Graziella Curtale, The Scripps Research Institute, United States

#### \*Correspondence:

Andrea Masotti andrea.masotti@opbg.net Milad Bastami bastamim@tbzmed.ac.ir

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 29 July 2018 Accepted: 19 February 2019 Published: 15 March 2019

#### Citation:

Choupani J, Nariman-Saleh-Fam Z, Saadatian Z, Ouladsahebmadarek E, Masotti A and Bastami M (2019) Association of mir-196a-2 rs11614913 and mir-149 rs2292832 Polymorphisms With Risk of Cancer: An Updated Meta-Analysis. Front. Genet. 10:186. doi: 10.3389/fgene.2019.00186 Background: Accumulating evidence suggests that functional dysregulations of miRNAs, especially miR-196a-2 and miR-149, in cancers could be attributed to polymorphisms in miRNA sequences. This study was aimed at clarifying the association of mir-196a-2 rs11614913 and mir-149 rs2292832 with cancer risk by performing an updated meta-analysis of genetic association studies.

Methods: PubMed, Embase, Scopus, and ScienceDirect databases were searched until 9 April 2018 to identify eligible studies. Studies should meet the following criteria to be included in the meta-analysis: evaluation of genetic association between rs11614913 and/or rs2292832 and susceptibility to cancer; A case-control design; Written in English; Availability of sufficient data for estimating odds ratio (OR) and its 95% confidence interval (95%CI). Studies that met the following criteria were excluded: review articles, meta-analysis, abstracts or conference papers; duplicate publications; studies on animals or cell-lines; studies without a case-control design; studies that did not report genotype frequencies. Pooled ORs and 95% CIs were estimated using a total of 111 studies (41,673 cases and 49,570 controls) for mir-196a rs11614913 and 44 studies (15,954 cases and 19,594 controls) for mir-149 rs2292832. Stratified analysis according to quality scores, genotyping method, ethnicity, broad cancer category and cancer type was also performed.

Results: Mir-196a-2 rs11614913 T allele was associated with decreased cancer risk in overall population. The association was only significant in Asians but not Caucasians. In subgroup analysis, significant associations were found in high quality studies, gynecological cancers, ovarian, breast, and hepatocellular cancer. Mir-149 rs2292832 was not associated with cancer risk in overall population and there were no differences between Asians and Caucasians. However, the T allele was associated with a decrease risk of gastrointestinal tract cancers under the heterozygote model and an increased risk of colorectal cancer under the recessive model.

**286**

Conclusions: The present meta-analysis suggests that mir-196a-2 rs11614913 may contribute to the risk of cancer especially in Asians. Mir-149 rs2292832 may modulate the risk of gastrointestinal tract cancers especially colorectal cancer. This study had some limitations such as significant heterogeneity in most contrasts, limited number of studies enrolling Africans or Caucasians ancestry and lack of adjustment for covariates and environmental interactions.

Keywords: microRNA, polymorphism, meta-analysis, cancer, mir-196a-2, mir-149

#### INTRODUCTION

Despite remarkable recent progress in clinical management, diagnosis and treatment, cancer has remained one of the major causes of death worldwide. According to the recent World Health Organization (WHO) report, about one in six deaths were caused by cancer in 2015. It is predicted that cancer-related death will increase up to 13.2 million by 2030 worldwide (Ferlay et al., 2010; Bray et al., 2012). Complex genetic and environmental risk factors and also interactions between these components contribute to the etiopathology of different cancers. Until recent years, much effort has been devoted to link the alteration of protein coding genes to tumorigenesis. However, latest evidence has demonstrated the emerging role of noncoding RNAs in cancer development and, especially, introduced microRNAs (miRNAs) as new players in pathobiology of cancers (Peng and Croce, 2016). MiRNAs are short noncoding functional RNAs that are involved in the regulation of transcriptome (Ha and Kim, 2014). They modulate important cellular processes both in normal physiology and disease state and are involved in almost all cellular processes altered during tumorigenesis (Osada and Takahashi, 2007; Li et al., 2009). Human mir-196a (MIR196A2, HGNC:31568) and mir-149 (MIR149, HGNC: 31536) are wellstudied miRNAs that may function either as oncomiRs, by targeting tumor suppressor genes, or as tumor suppressors, by targeting oncogenes, in different conditions (Lu et al., 2016; He J. et al., 2018; Ow et al., 2018). It has been shown that single nucleotide polymorphism (SNP) in miRNA genes, such as hsa-mir-196a-2 rs11614913 and hsa-mir-149 rs2292832, may influence their functions through altering miRNA expression, maturation and/or efficiency of targeting and, thereby, contribute to the risk of cancer (Hu et al., 2008; Hoffman et al., 2009; Tu et al., 2012; Nariman-Saleh-Fam et al., 2016, 2017). Several association studies in a range of populations evaluated the contribution of mir-196a-2 rs11614913 and mir-149 rs2292832 to cancer risk; but results are inconclusive. Therefore, this study was aimed at clarifying the association of mir-196a-2 rs11614913 and mir-149 rs2292832 with cancer risk by performing an updated meta-analysis of genetic association studies.

## MATERIALS AND METHODS

#### Publication Search

To identify all potentially eligible publications, PubMed, Embase, Scopus and ScienceDirect databases were searched, with respect to specific search tips of each database, using following keywords. ("microRNA 196a2" OR "miRNA-196a2" OR "mir-196a2" OR "mir196a" OR "mir-196a-2" OR "pre-mir-196a" OR "premir196a" OR "196a" OR "rs11614913") OR ("microRNA 149" OR "miRNA-149" OR "mir-149" OR "mir149" OR "pre-mir-149" OR "pre-mir149" OR "rs2292832") AND ("single nucleotide polymorphism" OR "SNP" OR "variant" OR "variation" OR "polymorphism" OR "mutation" OR "locus") AND ("neoplasm" OR "cancer" OR "tumor" OR "carcinoma" OR "sarcoma" OR "lymphoma" OR "adenoma" OR "leukemia" OR "leucemia" OR "malignancy" OR "malignance" OR "malignant" OR "glioma"). Last search was performed on 9 April 2018. References of the relevant literature and review articles were also evaluated to identify all potentially eligible articles. This meta-analysis carried out in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Liberati et al., 2009).

#### Inclusion and Exclusion Criteria

Studies should meet the following criteria to be included in the meta-analysis: (1) evaluation of genetic association between rs11614913 and/or rs2292832 and susceptibility to cancer; (2) a case-control design; (3) Written in English; (4) Availability of sufficient data for estimating odds ratio (OR) and its 95% confidence interval (95%CI). Studies that met the following criteria were excluded: (1) review articles, meta-analysis, abstracts or conference papers; (2) duplicate publications; (3) studies on animals or cell-lines; (4) studies without a case-control design (5) studies that did not report genotype frequencies.

#### Data Extraction

Data was extracted from each eligible study and manually checked. Then, items were recorded for each eligible study: the first author, publication year, category of cancer, type of cancer, country, ethnicity, source of controls, genotyping method, number of subjects in the case and the control groups, genotype counts for each SNP in the case and the control groups. A broad cancer category was assigned for each study according to the following scheme: gastrointestinal tract cancers (GI, including gastric, esophageal, colorectal, bladder, pancreatic, or hepatocellular cancers), head and neck squamous cell

**Abbreviations:** BC, breast cancer; BlC, bladder cancer; CRC, colorectal cancer; ESCC, esophageal squamous cell carcinoma; GC, gastric cancer; GI, cancers of digestive system; GyC, Gynecologic cancers; HCC, Hepatocellular cancer; HM, Hematological malignancies; HNC, Head and neck carcinoma; HWE, Hardy-Weinberg equilibrium; HWD, Hardy-Weinberg deviation; LC, lung cancer; UG, Urogenital cancers; OC, Oral cancer; OR, Odds ratio; OvC, ovarian cancer; PC, prostate cancer.

carcinoma (HNC, including oral, non-oral, and nasopharyngeal cancers), gynecologic cancer (GyC, including endometrial, ovarian, and cervical cancers), hematological malignancies (HM, including leukemia and lymphomas), urogenital cancers (UG, including prostate, renal cell, and bladder cancers), or other cancers.

#### Quality Assessment

The quality of each study was assessed using quality assessment criteria for genetic association studies used elsewhere (Thakkinstian et al., 2011; Xue et al., 2015). This score is based on seven items including representativeness of cases, representativeness of controls, ascertainment of cancer case, control selection, genotyping examination, Hardy-Weinberg Equilibrium (HWE) status in controls, and total sample size. Quality scores ranged from 0 (lowest) to 15 (highest). Studies which were scored equal to or less than eight were regarded as low quality, while those with scores of greater than eight were regarded as high quality.

#### Statistical Analysis

The Meta package for R was used to perform meta-analysis (Schwarzer, 2007). Association of rs11614913 and rs2292832 with cancer was estimated by calculating pooled ORs and their 95% CIs assuming homozygote, heterozygote, dominant, recessive, and allelic models. Heterogeneity was assessed using the Chi-squared based Q test (Lau et al., 1997). In the presence of a significant heterogeneity (i.e., P-value of Qtest < 0.05 or I <sup>2</sup> > 50%), the random effect (RE) model (DerSimonian and Laird, 1986) was used to calculate pooled ORs and 95% CIs. Otherwise, the fixed effect (FE) model was used (Mantel and Haenszel, 1959). Significance of the pooled OR was determined by the Z-test (P < 0.05 was considered significant). In cases of remarkable heterogeneity (i.e., I 2 > 50%), the potential sources of heterogeneity across studies was explored using univariate meta-regression and stratified analysis. Moreover, subgroup analyses based on genotyping method, study quality, ethnicities, broad cancer categories, and cancer types were carried out. To assess consistency of results and influence of each study on the pooled OR, sensitivity analysis was done by omitting one study at a time and recalculating summary OR and 95% CI. Publication bias was evaluated by the Begg's rank correlation test of funnel plot asymmetry (Begg and Mazumdar, 1994) the "Trim and Fill" approach was used to correct for asymmetry in cases of significant rank correlation test (Duval and Tweedie, 2000a,b). All Pvalues were two-sided and P-value < 0.05 was considered statistically significant. All statistical analyses were performed in R (version 3.3.1).

### Dealing With HWD (Departure From Hardy-Weinberg Equilibrium)

Departure from Hardy-Weinberg equilibrium (HWE) may be caused by a range of factors, among which genotyping error is more importantly relevant to the association study context. Currently there is no consensus on the way of handling association studies with the controls not in HWE, but it has been recommended that such studies should not be excluded from meta-analysis (Minelli et al., 2008). However, sensitivity analysis should be performed to evaluate the possible effects of such studies on the pooled estimates (Attia et al., 2003; Thakkinstian et al., 2005; Zintzaras and Lau, 2008; Wang X. B. et al., 2014). In the present metaanalysis, the following approach with regards to HWE-deviated studies was followed. Departure of genotype distributions from HWE (i.e., HWD) in the control group of each study was evaluated using the Chi-squared or the exact goodness of fit test. Meta-analyses, including the overall and subgroup analyses, were performed considering all eligible studies including HWD studies. However, to evaluate possible impacts of HWE-deviated studies, HWD sensitivity analysis was performed by evaluating the influence of excluding these studies on point estimates and identifying the influenced genotype contrasts. In cases that excluding HWD studies altered the result of meta-analysis, ORs of such studies were adjusted for HWE deviation by means of incorporating the HWEexpected genotype counts in the control group as recommended (Trikalinos et al., 2006; Zintzaras et al., 2006; Zintzaras, 2008; Zintzaras and Lau, 2008; Srivastava and Srivastava, 2012) and the HWD-adjusted pooled ORs were calculated in genotype contrasts.

## RESULTS

### Study Characteristics

The process of selecting eligible studies is depicted in **Figure 1**. A total of 1,645 articles were found from different sources outlined in materials and methods and screened by reading titles and abstracts. A total of 1,509 articles were excluded in which 577 articles were duplicates, 114 articles were abstracts or conference meetings, 86 articles were meta-analysis, 404 were review articles, 7 articles were not written in English, 26 articles were related to other diseases, 36 articles were related to other genes, or polymorphisms and 259 more articles had either obvious irrelevant study design or irrelevant disease/gene. The full text of the remaining 136 articles were evaluated and 9 more articles were also excluded as they did not have either sufficient data to calculate ORs and 95%CIs (n: 4) or an association study design (n: 5). Finally, a total of 127 eligible articles remained (Horikawa et al., 2008; Yang et al., 2008, 2016; Hoffman et al., 2009; Hu et al., 2009, 2013; Tian et al., 2009; Catucci et al., 2010; Christensen et al., 2010; Dou et al., 2010; Kim et al., 2010, 2012; Kontorovich et al., 2010; Li et al., 2010, 2014; Liu et al., 2010, 2013, 2014; Okubo et al., 2010; Peng et al., 2010; Qi et al., 2010, 2014, 2015; Srivastava et al., 2010, 2017; Wang et al., 2010, 2013, 2016; Akkiz et al., 2011; George et al., 2011; Hong et al., 2011; Jedlinski et al., 2011; Mittal et al., 2011; Vinci et al., 2011, 2013; Zhan et al., 2011; Zhou et al., 2011, 2014; Alshatwi et al., 2012; Chen et al., 2012; Chu et al., 2012, 2014; Hezova et al., 2012; Linhares et al., 2012; Min et al., 2012; Tu et al., 2012; Zhang M. et al., 2012; Zhang M. W. et al., 2012; Zhu et al., 2012; Ahn et al., 2013; Han et al., 2013; Huang et al., 2013, 2017; Lv et al., 2013; Ma et al., 2013; Pavlakis et al., 2013; Umar et al., 2013; Wei et al., 2013, 2014; Zhang et al., 2013; Bansal et al., 2014;

Dikeakos et al., 2014; Du et al., 2014; Hao et al., 2014; Kou et al., 2014; Kupcinskas et al., 2014a,b; Omrani et al., 2014; Parlayan et al., 2014; Pu et al., 2014; Qu et al., 2014; Roy et al., 2014; Tong et al., 2014; Wang N. et al., 2014; Wang R. et al., 2014; Wang X. H. et al., 2014; Deng et al., 2015; Dikaiakos et al., 2015; Dong et al., 2015; He et al., 2015; Li T. et al., 2015; Liu, 2015; Li X. et al., 2015; Martin-Guerrero et al., 2015; Nikolic et al., ´ 2015; Pratedrat et al., 2015; Sodhi et al., 2015; Sushma et al., 2015; Yan et al., 2015; Yin et al., 2015, 2016, 2017; Dai et al., 2016; Gu and Tu, 2016; Hashemi et al., 2016; Jiang et al., 2016; Li H. et al., 2016; Li J. et al., 2016; Li M. et al., 2016; Miao et al., 2016; Morales et al., 2016; Ni and Huang, 2016; Peckham-Gregory et al., 2016; Shen et al., 2016; Song et al., 2016; Su et al., 2016; Sun et al., 2016; Toraih et al., 2016a,b; Zhang L. H. et al., 2016; Zhao et al., 2016; Afsharzadeh et al., 2017; Bodal et al., 2017; Cîmpeanu et al., 2017; Poltronieri-Oliveira et al., 2017; Rakmanee et al., 2017; Rogoveanu et al., 2017; Tandon et al., 2017; Zhang E. et al., 2017; Abdel-Hamid et al., 2018; Damodaran et al., 2018; Doulah et al., 2018; He J. et al., 2018; He Y. et al., 2018; Mashayekhi et al., 2018; Minh et al., 2018; Ranjbar TABLE 1 | Main characteristics of studies evaluating cancer risk associated with miR-196a-2 rs11614913 C/T which included in the current meta-analysis.





Genotype distributions are sorted as CC/CT/TT.

<sup>a</sup>Genotype distributions are sorted as CC/CT/TT. PB, population-based design; HB, hospital-based design; NC, Not clear; ALL, Acute myeloblastic leukemia; BC, Breast cancer; BlC, Bladder cancer; CLL, Chronic lymphoblastic leukemia; CRC, Colorectal cancer; CSCC, Cervical Squamous Cell Carcinoma; ESCC, Esophageal squamous cell carcinoma; GBC, Gallbladder carcinoma; GC, Gastric cancer; HCC, Hepatocellular carcinoma; HNSCC, Head and neck squamous cell carcinoma; LC, Lung cancer; NB, Neuroblastoma; NC, Nasopharyngeal carcinoma; NHL, non-Hodgkin lymphoma; OC, Oral cancer; OvC, Ovarian cancer; PC, Prostate cancer; PNC, Pancreatic cancer; RCC, Renal cell carcinoma.

et al., 2018). Characteristics of the included studies are tabulated in **Tables 1**, **2**.

A total of 110 articles which included 111 studies (41,673 cases and 49,570 controls) evaluated the association of miR-196a-2 rs11614913 and cancer risk (**Table 1**). The article by Catucci et al. included two studies on separate populations (Catucci et al., 2010). In a study on head and neck carcinoma the genotype frequencies were not reported in the paper and data were retrieved by contacting the authors (Christensen et al., 2010). The final meta-analysis of mir-196a-2 rs11614913 and cancer risk included 111 studies (including 41,673 patients and 49,570 controls), among which 93 were scored greater than eight in the quality assessment and regarded as high quality studies. In 93 out of 111 studies, the genotype distribution of rs11614913 in the control group was concordant with HWE. Mir-196a-2 rs11614913 were genotype using a range of techniques with the most common being PCR-RFLP (n = 53). With regard to the ethnicity, 85 studies were performed in Asians, 20 were performed in Caucasians and the remaining six studies were included either Africans or individuals from different ancestries (mixed ancestry). In four studies the patients were subgrouped into multiple cancer types. Namely, Liu evaluated both ovarian and endometrial cancers (Liu, 2015), Parlayan evaluated gastric, lung, colorectal, prostate, and acute leukemia (Parlayan et al., 2014), Toraih studied both GI and non-GI cancers in a study (Toraih et al., 2016b) and hepatic and renal cancers in another study (Toraih et al., 2016a). For these studies genotype distribution of all patients were used to calculate point estimates in the overall analysis. However, in subgroup analysis, as these studies were assigned to more than one subgroup, the genotype distribution of patients with the relevant cancer category/type was used for pooling data. When studies were subgrouped according to the broad cancer category, there were 49 gastrointestinal tract cancers (GI), nine head and neck squamous cell carcinoma (HNSCC), six gynecologic cancers (GyC), six hematological malignancies (HM), 12 urogenital cancers (UG), and 34 other cancers. When studies subgrouped according to cancer type, there were 21 breast cancer (BC), 18 hepatocellular carcinoma (HCC), 13 gastric cancer (GC), ten colorectal cancer (CRC), nine lung cancer (LC), four bladder cancer (BlC), five prostate cancer (PC), six oral carcinoma (OC), four ovarian cancer (OvC), six esophageal cancer (ESCC), and 22 other cancer types.

Moreover, 43 articles comprising 44 studies (15,954 cases and 19,594 controls) evaluated the association of mir-149 rs2292832 and cancer risk (**Table 2**), among which 39 studies were evaluated as being high quality (quality score > 8). The genotype distribution of rs2292832 in the control groups of 34 studies were in agreement with HWE. The main genotyping technique was PCR-RFLP (29 studies). Most studies (n: 39) were performed in Asian populations and only few studies (n: 5) had focused on Caucasians. According to the broad cancer category, there were 22 GI studies, seven HNC studies and 15 studies on other cancer types. When studies were subgrouped by cancer type, there were four studies on BC, nine studies on HCC, eight studies on GC, five studies on CRC, three studies on LC and 15 studies on other types of cancer.

#### Quantitative Synthesis

#### Association of mir-196a2 rs11614913 and Cancer Risk

Statistically significant associations between mir-196a2 rs11614913 and cancer risk were observed assuming the homozygote (TT vs. CC, ORRE [95% CI]: 0.88 [0.79–0.98], P: 0.027) and the recessive (TT vs. CC+CT, ORRE [95% TABLE 2 | Main characteristics of studies evaluating cancer risk associated with miR-149 rs2292832 C/T which included in the current meta-analysis.



Genotype distributions are sorted as CC/CT/TT.

<sup>a</sup>Genotype distributions are sorted as CC/CT/TT. PB, population-based design; HB, hospital-based design; BC, Breast cancer; CRC, Colorectal cancer; GC, Gastric cancer; HCC, Hepatocellular carcinoma; HNSCC, Head and neck squamous cell carcinoma; LC, Lung cancer; NB, Neuroblastoma; NC, Nasopharyngeal carcinoma; NSCLC, non-Small cell lung carcinoma; OC, Oral cancer; OvC, Ovarian cancer; PC, Prostate cancer; PTC, Papillary thyroid carcinoma; RCC, Renal cell carcinoma; KS, Kaposi sarcoma.

TABLE 3 | Summary of the results of meta-analysis of cancer risk associated with miR-196a-2 rs11614913 and miR-149 rs2292832.


<sup>a</sup>Number of studies in each contrast.

<sup>b</sup>Pooled OR and 95% CI (Random-effect model).

<sup>c</sup>P-value of the Z-test.

<sup>e</sup>P-value of the Begg's test.

<sup>f</sup> For these models (homozygote, recessive, and allelic comparisons of rs2292832) "trim and fill" adjusted results are shown (consult with Supplementary Figure S4 for forest plots and Supplementary Figure S11 for funnel plots of these models).

Significant associations are shown in boldface.

CI]: 0.89 [0.83–0.95], P: 0.001) models (**Table 3**). Mir-196a-2 rs11614913 was not associated with cancer risk in the heterozygote and dominant models and there was a nonsignificant borderline association in the allele contrast (**Table 3**). **Supplementary Figure S1** shows the forest plots of association of mir-196a-2 rs11614913 and cancer risk in five models. The results of subgroup analysis for mir-196a-2 rs11614913 are shown in **Table 4**. Decreased risk of cancer was found in high quality studies under the homozygote (TT vs. CC, ORRE [95% CI]: 0.87 [0.77–0.97], P: 0.017), the recessive (TT vs. CC+CT, ORRE [95% CI]: 0.88 [0.81–0.95], P: 0.001) and the allele contrasts (T vs. C, ORRE [95% CI]: 0.93 [0.88–0.99], P: 0.020). In subgroup analysis by genotyping method, the only significant association was observed under the recessive model for studies which used a method other than PCR-RFLP (TT vs. CC+CT, ORRE [95% CI]: 0.89 [0.82–0.96], P: 0.007). When sub-grouped by ethnicity, decreased risks of cancer under the homozygote (TT vs. CC, ORRE [95% CI]: 0.86 [0.77–0.97], P: 0.016), the recessive (TT vs. CC+CT, ORRE [95% CI]: 0.87 [0.81–0.93] P: 0.0004) and the allelic (T vs. C, ORRE [95% CI]: 0.93 [0.89–0.98], P: 0.015) models were found only in Asians but not in Caucasians or the African/mixed ancestry subgroups (**Table 4**, **Figure 2**). Subgrouping by broad cancer categories indicated that mir-196a-2 rs11614913 was associated by a decreased risk of gynecologic cancer (GyC) assuming the recessive model (TT vs. CC+CT, ORFE [95% CI]: 0.80 [0.68–0.95], P: 0.010) and the allelic contrast

<sup>d</sup>P-value of the Q-test.


11 March 2019 | Volume 10 | Article 186

Frontiers in Genetics | www.frontiersin.org

TABLE 4 |

Meta-analysis

 of miR-196a-2

 rs11614913

 and cancer risk.

**296**



 bPooledORsand95%confidenceintervals.

 cP-valueoftheheterogeneitytest.

 dMeta-analysisofallstudiesexcludingthosewiththecontrolgroup

Significant associations

 are shown in boldface.

 not in HWE. eThesesubgroupswerefoundtobeinfluencedbydeparturefromHWE.PleaseconsultwiththeSupplementaryTableS2fordetailsonHWDsensitivityandadjustments.

 GI, cancers of digestive system; HNC, Head and neck carcinoma; GyC, Gynecologic cancers; HM, Hematological malignancies; UG, Urogenital cancers; BC, breast cancer; HCC, Hepatocellular cancer; GC, gastric cancer; CRC,colorectal cancer; LC, lung cancer; BlC, bladder cancer; PC, prostate cancer; OC, Oral cancer; OvC, ovarian cancer; ESCC, esophageal squamous cell carcinoma.

(T vs. C, ORFE [95% CI]: 0.88 [0.79–0.98], P: 0.021) (**Figure 3**). No significant findings were observed for gastrointestinal, head and neck, hematological or urogenital cancers (**Table 4**). **Supplementary Figure S2** presents the forest plots for subgroups according to the broad cancer categories. Further subgrouping by cancer type revealed significant association of mir-196a-2 rs11614913 with hepatocellular carcinoma (**Figure 4**) under the homozygote model (TT vs. CC, ORRE [95% CI]: 0.73[0.57–0.94], P: 0.017), the recessive model (TT vs. CC+CT, ORRE [95% CI]: 0.79 [0.66–0.95], P: 0.017) and the allele contrast (T vs. C, ORFE [95% CI]: 0.88 [0.78–0.98], P: 0.030), and with ovarian cancer (**Figure 5**) under the recessive model (TT vs. CC+CT, ORFE [95% CI]: 0.73[0.60–0.90], P: 0.003). **Supplementary Figure S3** presents the forest plots for subgroups according to cancer type.

HWD sensitivity analysis (i.e., excluding studies with controls deviated from HWE) revealed stable results in the overall analysis under the homozygote, heterozygote, dominant, and recessive models (**Table 4**). However, excluding HWD studies made the borderline allele contrast statistically significant (**Table 4**). Moreover, excluding HWE violating studies had no dramatic effects on subgroup meta-analyses using quality of studies, genotyping methods, the ethnicity and the broad cancer category (**Supplementary Table S1**). In meta-analysis subgrouped by cancer type, the results were also stable for gastric, bladder, oral, ovarian, prostate, and esophageal cancer subgroups after excluding HWD studies (**Supplementary Table S1**). However, excluding such studies altered the results for the breast, hepatocellular, colorectal, and lung cancer subgroups. Therefore, for these subgroups, pooled ORs were estimated to account for departures from HWE (denoted as HWD-adjusted ORs) (**Supplementary Table S2**). When corrected for HWD, mir-196a-2 rs11614913 was found to be significantly associated with breast cancer under the homozygote (TT vs. CC, HWDadjusted ORRE [95% CI]: 0.75 [0.61–0.93], P: 0.011) and recessive (TT vs. CC+CT, HWD-adjusted ORRE [95% CI]: 0.84 [0.71–0.98], P: 0.030) models (**Figure 6**). The association with hepatocellular cancer under the homozygote and the recessive models was remained significant after adjustment for HWD (TT vs. CC, HWD-adjusted ORRE [95% CI]: 0.69 [0.53– 0.91], P: 0.011 and TT vs. CC+CT, HWD-adjusted ORRE [95% CI]: 0.72 [0.57–0.90], P: 0.008). Furthermore, adjustment for HWD confirmed that mir-196a-2 rs11614913 is not associated with colorectal or lung cancer assuming any genetic model (**Supplementary Table S2**).

#### Association of mir-149 rs2292832 and Cancer Risk

The overall analysis showed no significant association with cancer risk under any genetic model (**Table 3**). **Supplementary Figure S4** shows the forest plots for the association of mir-149 rs2292832 and cancer risk under different genetic models. However, in subgroup analyses (**Table 5**) significant association of rs2292832 with cancer risk was observed in studies which used a genotyping method other than PCR-RFLP (CT vs. CC, ORFE [95% CI]: 0.88 [0.79– 0.98], P: 0.025). When subgrouped by broad cancer category, a decreased risk of gastrointestinal tract cancers was found in the heterozygote model (**Figure 7A**, CT vs. CC, ORFE

[95% CI]: 0.87 [0.79–0.97], P: 0.011). Subgrouping by cancer type, however, revealed an increased risk of colorectal cancer for individuals carrying TT genotype compared to those who carry at least one C allele (**Figure 7B**, TT vs. CT+CC, ORFE [95% CI]: 1.21 [1.04–1.40], P: 0.011). No significant association was observed for other comparisons (**Table 5**). **Supplementary Figures S5**, **S6** show the forest plots for subgroup analysis according to the broad cancer category and cancer type, respectively. Sensitivity analysis revealed that HWD studies had no significant effect on point estimates in the overall meta-analysis of mir-149 rs2292832 and cancer risk, and still no significant association was observed in overall analysis (**Table 5**). Moreover, most subgroup analyses were also stable after removing HWD studies (**Supplementary Table S3**). However, removing studies with HWD controls influenced comparisons in three subgroups: (i) non-PCR-RFLP subgroup (heterozygote model); (ii) the breast cancer subgroup (recessive model); (iii) the colorectal cancer subgroup (recessive model). Therefore, for these subgroups, HWE-expected genotype distributions in controls were used for pooling ORs (denoted as HWD-adjusted OR) (**Supplementary Table S4**). Adjusting for HWD in these subgroups confirmed the results of original analyses and showed that rs2292832 is associated with cancer risk in non-RFLP subgroup under the heterozygote model (**Figure 8**, CT vs. CC, HWD-adjusted ORRE [95% CI]: 0.68 [0.48–0.98], P: 0.040) and with colorectal cancer risk under the recessive model (TT vs. CT+CC, HWD-adjusted ORFE [95% CI]: 1.29 [1.11–1.50], P: 0.0007). No association with breast cancer risk was identified after adjusting for HWD (**Supplementary Table S4**).

#### Heterogeneity, Meta-regression, and Sensitivity Analysis

Heterogeneity was evaluated for both polymorphisms in all genetic models (**Tables 3**, **4**, **5**). Significant between study heterogeneity was observed in the overall estimation under all genetic models for mir-196a-2 rs11614913 and consequently random effect model was used. Univariate meta-regression using cancer type, country, ethnicity, the quality of study (either high or low), genotyping method, source of controls (PB or HB) or HWE was performed to identify potential sources of heterogeneity. For mir-196a-2, meta-regression showed that at least a part of the observed between study heterogeneity in the heterozygote (R 2 : 24.23%, P: 0.007) and dominant (R 2 : 19.86%, P: 0.028) models could be attributed to the country moderator. However, there was still significant unaccounted heterogeneity even after correcting for the effect of country moderator (Heterozygote I 2 : 63.78 and dominant I 2 : 72.05, P for test of residual heterogeneity < 0.0001). Moreover, Galbraith plot analysis demonstrated three studies (Lv et al., 2013; Wang et al., 2013; Dikeakos et al., 2014) as the most extreme outliers in all genetic models that account for a considerable portion of the observed heterogeneities (**Supplementary Figure S7**). Excluding these studies led to a 11.6% reduction of I 2 in the homozygote model (from 77.1 to 65.5%), a 12.8% reduction in the heterozygote model (from 70.6 to 57.8%), a 12.3%

mode to represent 85 studies in the Asian subgroup. The result of meta-analysis is shown beneath the second column.

FIGURE 3 | Forest plot of cancer risk associated with mir-196a-2 rs11614913 in the gynecological cancers subgroup. From left to right: recessive (TT vs. CT+CC), allelic (T vs. C), and homozygote (TT vs. CC) contrast.

reduction in the dominant model (from 76.9 to 64.6%), an 9.1% reduction in the recessive model (from 68.8 to 59.7%) and a 11.3% reduction in the allelic model (from 79 to 67.7%). However, excluding these studies did not alter any genotypic contrast and results were comparable to the overall analyses (data not shown). Sensitivity analysis by omitting one study at a time revealed that no individual study significantly influenced the genotype contrasts (**Supplementary Figure S8**). In the allele contrast, omitting no single study dramatically influenced pooled OR or its 95%CI. However, given that the original 95%CI was borderline (0.90–1.00), omitting some studies lead the upper limit of 95%CI to fall slightly below one (**Supplementary Figure S8-e**).

Statistically significant heterogeneity was also observed in the overall analysis of miR-149 rs2292832 and cancer risk and, consequently, RE model was used to

estimate pooled OR (**Table 3**). Subgrouping by study level moderators led to a reduction in heterogeneity in some subgroups (**Table 5**). However, univariate metaregression showed no statistically significant source of heterogeneity (All P > 0.05). Sensitivity analysis by omitting one study at a time revealed no single influential study (**Supplementary Figure S9**).

### Publication Bias

Rank correlation test of the mir-196a-2 rs11614913 Begger's funnel plot asymmetry revealed no statistically significant evidence of publication bias in any contrast (**Table 3** and **Supplementary Figure S10**). However, rank correlation test for asymmetry of mir-149 rs2292832 funnel plots showed statistically significant results in the homozygote, recessive and the allelic contrasts (**Table 3** and **Supplementary Figure S11**). Consequently, the "trim and fill" approach (Duval and Tweedie, 2000a,b) employed to correct for funnel plot asymmetry arising from publication bias in these models. The results of overall analysis using original studies or trim and fill method under the three models were comparable (see **Supplementary Figure S4** to compare forest plots of the original studies vs. trim-and-fill method and **Supplementary Figure S11** for funnel plots). After excluding studies with controls deviating from HWE in sensitivity analysis, rank correlation test was still significant in the mentioned three contrasts.

## DISCUSSION

The possible contribution of miRNAs, especially mir-196a-2 and mir-149, to the risk of cancer has stimulated great attention in recent years. Many studies evaluated the functional alterations of these micro-regulators in a wide range of cancers. Accumulating evidence suggests that, at least a part of functional dysregulations of miRNAs in cancers could be attributed to polymorphisms in miRNA sequences (Hu et al., 2008; Hoffman et al., 2009; Tu et al., 2012; Ghaedi et al., 2015; Nariman-Saleh-Fam et al., 2016, 2017). Two mature miRNAs, miR-196a-5p and miR-196a-3p, are generated from the stem-loop structure of hsa-mir-196a-2 (Kozomara and Griffiths-Jones, 2014) with the studied polymorphism, rs11614913, residing in the 3 ′ arm (**Figure 9A**). This polymorphism, therefore, may potentially alter miRNA processing and also binding to related target mRNAs (Hoffman et al., 2009) (**Figure 9B**). Previous studies have shown that the expression level of mature miR-196a-3p was higher in CC carriers with lung cancer compared to CT and TT individuals (Hu et al., 2008). More evidences have been provided by Hoffman et al. (2009) who observed elevated expression of mature mir-196a-2 forms in MCF-7 cells transfected with pre-mir-196a-C vector compared with cells transfected with pre-mir-196a-T vector. The potential of rs11614913 in influencing targeting function of mir-196a-2 has also been documented by whole-genome expression microarrays which found different numbers of

dysregulated mRNAs after transfecting cells with pre-mir-196a-C or pre-mir-196a-T vector (Hoffman et al., 2009). Hsa-mir-149 also generates two mature miRNAs (miR-149-5p and miR-149-3p) and the studied polymorphism, rs2292832, does not reside in the mature sequence of neither miR-149-5p or miR-149-3p (**Figure 10A**). Therefore, it has been hypothesized that rs2292832 is not a structure-shifting polymorphism for pri-mir-149 or pre-mir-149 (Wei et al., 2014). However, Tu et al. reported that the T allele may disrupt the maturation process compared with the C allele and, consequently, decrease miR-149 expression (Tu et al., 2012) (**Figure 10B**) in head and neck squamous cell carcinoma patients.



polymorphisms and cancer risk with contradictory results merits the need for comprehensive systematic reviews and meta-analyses. Several meta-analyses have evaluated the risk of cancer associated with mir-196a-2 rs11614913 or mir-149 rs2292832 (Chu et al., 2011; Zhang H. et al., 2012; Feng et al., 2016; Yan et al., 2017; Liu et al., 2018). However, the conclusion of these studies with regards to the subgroup analysis and the significant genetic model varies due, at least in part, to differences in the number of studies included or in the methodology. Moreover, several recent genetic association studies have not been included in previously published metaanalyses. Therefore, it was necessary to perform an updated meta-analysis with larger number of studies to clarify the association of mir-196a-2 rs11614913 or mir-149 rs2292832 with cancer risk. Therefore, compared to the previous meta-analysis, we included more studies in the analyses. The present metaanalysis also evaluated and corrected for the possible influence of departure of the control group of association studies from HWE. Although checking for departure from HWE has been recommended, currently there is no consensus about how HWD studies should be handled in meta-analysis (Minelli et al., 2008). Result of simulations suggests no advantage for excluding these studies (Minelli et al., 2008). However, sensitivity analysis to detect any possible bias imposed by such studies and/or using HWE-expected counts instead of the observed genotype frequencies have been recommended and implemented in several studies (Attia et al., 2003; Thakkinstian et al., 2005; Trikalinos et al., 2006; Zintzaras et al., 2006; Zintzaras, 2008; Zintzaras and Lau, 2008; Srivastava and Srivastava, 2012; Wang X. B. et al., 2014). The current study noticed that most analyses, especially those with sufficiently large number of studies, were not influenced by excluding HWD studies. However, as it is rationally expected, excluding HWD studies from some subgroup with relatively small number of studies may influence the analysis, and therefore in such situations adjusted analyses were preferred.

Increasing number of association studies evaluating miRNA

For mir-196a-2 rs11614913, the previous largest metaanalysis, conducted by Liu et al. (2018), included 84 studies compared to 111 studies in the present meta-analysis. By including 41,673 patients and 49,570 control subjects, the present meta-analysis showed a decreased risk of cancer in the homozygote and the recessive models (**Table 3**). Although, the association was not significant in allele contrast, the OR and 95%CI of the allele contrast were borderline and influenced by excluding some individual studies. Excluding HWE-deviated or low quality studies yielded significant associations under allelic model. As there is, currently, no way to adjust allele frequencies for departure from HWE, the possibility that HWD studies may bias the allele contrast cannot be rolled out and a definite conclusion cannot be drawn under allelic model. Apart from the allele contrast, the results of other genetic models were statistically stable and not influenced by removing any single study, HWE deviated or low quality studies. The results also suggest that mir-196a-2 rs11614913 may pose an ethnic dependent effect on cancer risk as associations with cancer were only observed in Asians. However, it should be noted that most

dMeta-analysis

eThese subgroups were found to be influenced by departure from HWE. Please refer to the

fAlthough the point estimate of I

GI, Significant associations

Gastrointestinal

 tract cancers; HNC, Head and neck cancers; BC, Breast cancer; HCC, Hepatocellular

 are shown in boldface.

2 was zero for the allele contrast in the low quality subgroup, the random effect model was used based on the 95%CI of I

 of all studies excluding those with the control group not in HWE.

Supplementary

 Table S4 for details on HWD sensitivity and adjustment analysis for these subgroups.

 cancers; GC, Gastric cancer; CRC, Colorectal cancer; LC, Lung cancer.

2 (0–69.8%) and small number of studies.

studies enrolled Asian patients, mainly Chinese patients, and the number of studies involving other ethnicities were relatively small. Moreover, different minor allele frequencies (MAF) may partly contribute to the observed differences among ethnicities (Average MAF in Asians T: 0.501 ± 0.127, Caucasians T: 0.410 ± 0.1, Others T: 0.338 ± 0.066). In-line with previous studies, the current meta-analysis also confirmed that rs11614913 is associated with decreased risks of hepatocellular cancer under three genetic models (Liu et al., 2018) and that it may not modulate risk of urogenital cancers (Wang et al., 2017). The results of meta-analysis of all studies, subgroup analysis by ethnicity and hepatocellular cancer are in agreement with findings of the previous largest meta-analysis (Liu et al., 2018). However, the increase in the number of analyzed studies led to discrepancies with regards to conclusions in some subgroup analysis. (i) In contrast to the studies by Liu and Pan (Pan et al., 2017; Liu et al., 2018), the present meta-analysis did not find a significant association with head and neck carcinoma in any genetic model. This discrepancy may be attributed to the number of studies included in meta-analyses [nine studies in the present meta-analysis compared to four studies in the meta-analysis by Liu et al. (2018)]. Furthermore, differences in defining head and neck cancer may also explain different conclusions drawn from the present study and the study by Pan and colleagues (Pan et al., 2017). They included esophageal cancer as a type of head and neck cancer, whereas we considered it as a type of gastrointestinal tract cancers (according to the ICD-10-CM C15-C26). (ii) Additionally, the present meta-analysis found significant associations between mir-196a-2 rs11614913 and decreased risks of gynecologic cancers (especially ovarian cancer), which have not been reported in any previous metaanalysis. Interestingly, low heterogeneity was observed in the gynecological cancers subgroup assuming the two significant contrasts (i.e., recessive and allelic). (iii) Although previous metaanalyses (Yan et al., 2017; Zhang H. et al., 2017; Liu et al., 2018) failed to find a significant association between mir-196a-2 rs11614913 and breast cancer, the current study showed, by incorporating more association studies and performing

HWD sensitivity analysis, that adjusting for departure from HWE may reveal significant associations under the homozygote and recessive contrasts (**Figure 6** and **Supplementary Table S2**). (iv) Moreover, contradictory to previous meta-analyses, no association with gastric (Yan et al., 2017), colorectal (Xie et al., 2015; Yan et al., 2017) or lung cancer (Ren et al., 2016; Yan et al., 2017; Liu et al., 2018), was found. Apart from larger sample sizes and correcting for HWD, sometimes this discrepancy in results may also be related to methodological differences in the design, specifically the inclusion criteria, of meta-analyses. As a case in point, studies by Hu et al. (2008) and Yoon et al. (2012) did not meet the inclusion criteria of our study, as they deal with the survival or recurrence risk of lung cancer patients with approaches that differed from routine case-control genetic association studies. However, we noticed that these studies were included in a previous meta-analysis (Liu et al., 2018).

(i.e., He, 2018; Kontorovich et al., 2010; Pratedrat et al., 2015; Jiang et al., 2016).

The current meta-analysis also showed that mir-149 rs2292832 is not associated with risk of cancer in any genetic model and the results were statistically reliable, as summary effects were not influenced by excluding any single study, HWE-deviated or low quality studies. No differences in cancer risk was observed between ethnicities. Similar to miR-196a-2 polymorphism, the Asian subgroup comprised a large proportion of studies and relatively few studies with limited sample sizes were performed in Caucasians. Therefore, a definite conclusion cannot be drawn in Caucasians and more studies are needed further clarify the association of this SNP with cancer risk in Caucasians. The results of overall analysis were comparable to previous meta-analyses (Li L. et al., 2015; Feng et al., 2016). By pooling the results of 22 studies, this meta-analysis found a decreased risk of gastrointestinal tract (GI) cancers for individuals who carry the CT genotype compared to those with the CC genotype in a heterozygote model. Interestingly, there was no significant heterogeneity in the GI subgroup assuming the heterozygote model indicating the reliability of meta-analysis in this subgroup. Previous meta-analyses yielded different results with regard to GI cancers. A previous meta-analysis of seven studies on GI cancers suggested a marginally elevated risk under the recessive model (TT vs. CT+CC), while another pooling of 10 studies found a borderline decreased risk for the CT vs. TT contrast. Although no significant association was identified in the head and neck cancers subgroup, it should be noted that significant heterogeneity was present in all models except the recessive contrast and number of samples were relatively small. For colorectal cancer the association was in reverse direction

and an increased risk was observed in individuals who carry the TT genotype compared with subject who carry at-least one C allele (**Table 5**). A similar association based on three studies was previously reported (Rong et al., 2017), but not reproduced in other meta-analyses (Li et al., 2013; Feng et al., 2016). Taken together, the current results based on five studies suggest an increased risk for colorectal cancer that was stable after correcting for departure from HWE. Although no significant heterogeneity was detected in the colorectal cancer subgroup under any genetic model, it should be noted that the limited number of studies may influence heterogeneity evaluation and more definite conclusion may be drawn by analyzing larger sample sizes. In the case of breast cancer, a previous meta-analysis of three studies found a significant association (Feng et al., 2016). We found no significant association in the original and HWD-adjusted analysis. However, number of studies in the colorectal and breast cancer subgroups are relatively limited and results should be interpreted with caution. More studies with large sample sizes are needed for a definite conclusion.

However, the present study has some limitations. First, significant heterogeneity was present in most analyses especially for mir-196a-2 polymorphism. We, therefore, used random effect model and performed meta-regression; but no significant source of heterogeneity was observed for most analyses, suggesting that other unknown study level moderators may contribute to the heterogeneity. Second: The molecular mechanisms underlying association of these miRNAs-SNP with risk of cancer are complex

(−5p and −3p forms) expression.

and might be strongly affected by different genetic background as well as other masked variables. This, in turn, may limit the efficacy of the overall analysis especially in the case of miR-196a-2 rs11614913. Stratified analyses based on a specific cancer category or a cancer type may help to reduce this heterogeneity and, therefore, are considered to be more reliable. Third, this study was based on unadjusted ORs of the original studies and no adjustment for covariates like age and gender or interaction with environmental factors were done and this fact may also potentially contribute to the between study heterogeneity. Fourth, some limitations such as language restriction or lack of access to the genotype counts of mir-196a-2 rs11614913 in four studies with insufficiently reported data may bring in publication bias. The trim and fill method has been shown to reduce the bias in estimates in the presence of publication bias and heterogeneity (Peters et al., 2007). However, it has been recommended that this method should be considered as sensitivity analysis as we cannot be sure that asymmetry in funnel plot is truly caused by publication bias (Peters et al., 2007). Although rank correlation test of funnel plots of mir-149 rs2292832 was significant in three genetic models raising the possibility of publication bias, adjusting for such a bias using trim-and-fill method did not afford any change in analysis of overall studies in these models (**Supplementary Figure S4**). Fifth, number of studies in some subgroup analyses was limited and, consequently, results of such analysis should be interpreted with caution. Most studies were performed enrolling Asian patients and the number of studies on Caucasians or Africans was limited. Therefore, more association studies with larger sample sizes on Africans and Caucasians are needed to make precise estimations of cancer risk associated with the studied polymorphisms. Assigning ethnicity to each study population could be another limitation of meta-analysis of association studies as each ethnicity may regroup several sub-populations with somewhat different genetic background. Sixth, the control groups of association studies were not uniformly defined and non-differential misclassification bias may have occurred.

In conclusion, this meta-analysis showed that mir-196a-2 rs11614913 T allele is associated with decreased cancer risk in overall population, high quality studies and studies on Asian populations. It is also associated with a decreased risk of gynecological cancers, ovarian, breast and hepatocellular cancer. Mir-149 rs2292832 was not associated with cancer risk in overall population, high quality studies, Asians or Caucasians. However, the T allele was associated with a decrease risk of gastrointestinal tract cancers under the heterozygote model and an increased risk of colorectal cancer under the recessive model.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

MB and AM conceived the original idea and supervised the project. JC, ZN-S-F, and ZS contributed to the literature search and data management. MB, JC, and ZN-S-F wrote the manuscript with support from all authors. MB contributed to the data analysis, interpretation of results, and data visualization with inputs from AM. EO and ZS assisted with data visualization. All authors provided critical feedback, discussed the results, and contributed to the final manuscript.

#### ACKNOWLEDGMENTS

We would like to thank Dr. Karl Kelsey and Dr. Brock Christensen for providing the genotype counts of mir-196a-2 rs11614913 for their study (Christensen et al., 2010).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00186/full#supplementary-material

(2008-2030): a population-based study. Lancet Oncol. 13, 790–801. doi: 10.1016/S1470-2045(12)70211-5


and gastric cancer risk in japanese population. Helicobacter 15, 524–531. doi: 10.1111/j.1523-5378.2010.00806.x


and 499a variants with hepatic and renal cancer susceptibility a preliminary study. Asian Pac. J. Cancer Prevent. 17, 3369–3375.


MIR 196A2 Expression in Breast Cancer. DNA Cell Biol. 35, 81–87. doi: 10.1089/dna.2014.2785


haplotypes of methylenetetrahydrofolate reductase gene polymorphisms in acute lymphoblastic leukemia. Eur. J. Epidemiol. 21, 501–510. doi: 10.1007/s10654-006-9027-8

Zintzaras, E., and Lau, J. (2008). Synthesis of genetic association studies for pertinent gene-disease associations requires appropriate methodological and statistical approaches. J. Clin. Epidemiol. 61, 634–645. doi: 10.1016/j.jclinepi.2007.12.011

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

Copyright © 2019 Choupani, Nariman-Saleh-Fam, Saadatian, Ouladsahebmadarek, Masotti and Bastami. 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.

# miRNome Reveals New Insights Into the Molecular Biology of Field Cancerization in Gastric Cancer

Adenilson Pereira1,2† , Fabiano Moreira1,2† , Tatiana Vinasco-Sandoval<sup>2</sup> , Adenard Cunha<sup>2</sup> , Amanda Vidal<sup>2</sup> , André M. Ribeiro-dos-Santos<sup>1</sup> , Pablo Pinto<sup>1</sup> , Leandro Magalhães<sup>1</sup> , Mônica Assumpção<sup>2</sup> , Samia Demachki<sup>2</sup> , Sidney Santos1,2, Paulo Assumpção<sup>2</sup> and Ândrea Ribeiro-dos-Santos1,2 \*

<sup>1</sup> Laboratory of Human and Medical Genetics, Institute of Biological Sciences, Graduate Program of Genetics and Molecular Biology, Federal University of Pará, Belém, Brazil, <sup>2</sup> Research Center on Oncology, Graduate Program of Oncology

and Medical Science, Federal University of Pará, Belém, Brazil

#### Edited by:

Ge Shan, University of Science and Technology of China, China

#### Reviewed by:

Dominic C. Voon, Kanazawa University, Japan Mahshid Malakootian, Iran University of Medical Sciences, Iran

#### \*Correspondence:

Ândrea Ribeiro-dos-Santos akelyufpa@gmail.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 05 June 2018 Accepted: 04 June 2019 Published: 19 June 2019

#### Citation:

Pereira A, Moreira F, Vinasco-Sandoval T, Cunha A, Vidal A, Ribeiro-dos-Santos AM, Pinto P, Magalhães L, Assumpção M, Demachki S, Santos S, Assumpção P and Ribeiro-dos-Santos (2019) miRNome Reveals New Insights Into the Molecular Biology of Field Cancerization in Gastric Cancer. Front. Genet. 10:592. doi: 10.3389/fgene.2019.00592 Background: MicroRNAs (miRNAs) play an important role in gastric carcinogenesis and have been associated with gastric field cancerization; however, their role is not fully understood in this process. We performed the miRNome sequencing of non-cancerous, adjacent to tumor and gastric cancer samples to understand the involvement of these small RNAs in gastric field cancerization.

Methods: We analyzed samples of patients without cancer as control (non-cancerous gastric samples) and adjacent to cancer and gastric cancer paired samples, and considered miRNAs with |log2(fold change)| > 2 and Padj < 0.05 to be statistically significant. The identification of target genes, functional analysis and enrichment in KEGG pathways were realized in the TargetCompare, miRTargetLink, and DAVID tools. We also performed receiver operating characteristic (ROC) curves and miRNAs that had an AUC > 0.85 were considered to be potential biomarkers.

Results: We found 14 miRNAs exclusively deregulated in gastric cancer, of which six have potential diagnostic value for advanced disease. Nine miRNAs with known tumor suppressor activities (TS-miRs) were deregulated exclusively in adjacent tissue. Of these, five have potential diagnostic value for the early stages of gastric cancer. Functional analysis of these TS-miRs revealed that they regulate important cellular signaling pathways (PI3K-Akt, HIF-1, Ras, Rap1, ErbB, and MAPK signaling pathways), that are involved in gastric carcinogenesis. Seven miRNAs were differentially expressed in both gastric cancer and adjacent regarding to non-cancerous tissues; among them, hsa-miR-200a-3p and hsa-miR-873-5p have potential diagnostic value for early and advanced stages of the disease. Only hsa-miR-196a-5p was differentially expressed between adjacent to cancer and gastric cancer tissues. In addition, the other miRNAs identified in this study were not differentially expressed between adjacent to cancer and gastric cancer, suggesting that these tissues are very similar and that share these molecular changes.

Conclusion: Our results show that gastric cancer and adjacent tissues have a similar miRNA expression profile, indicating that studied miRNAs are intimately associated with field cancerization in gastric cancer. The overexpression of TS-miRs in adjacent tissues may be a barrier against tumorigenesis within these pre-cancerous conditions prior to the eventual formation or relapse of a tumor. Additionally, these miRNAs have a great accuracy in discriminating non-cancerous from adjacent to tumor and cancer tissues and can be potentially useful as biomarkers for gastric cancer.

Keywords: miRNome, miRNAs, gastric cancer, field cancerization, biomarkers

#### INTRODUCTION

Gastric cancer (GC) is an aggressive disease that is considered the third leading cause of cancer death worldwide (Ferlay et al., 2015). GC is usually diagnosed in late stages, due in part to the limited efficiency of existing biomarkers, and the 5-year survival rates of these patients do not exceed 30% (Correa, 2013; Yakirevich and Resnick, 2013). Studies have shown that microRNAs (miRNAs) have excellent sensitivity/specificity and a high discriminatory capacity, which make them potentially useful as molecular biomarkers for this type of cancer (Wu et al., 2014; Shin et al., 2015; Vidal et al., 2016).

MicroRNAs are small non-coding RNAs (∼18–30 nt) that regulate gene expression post-transcriptionally, disrupting the expression of target mRNAs (Bartel, 2004; Wu et al., 2014). These molecules play an important role in multiple pathways and the processes responsible for the maintenance of healthy tissue homeostasis (Schneider, 2012; Runtsch et al., 2014). In humans, studies suggest that miRNAs are part of the complex regulatory network of the healthy stomach (Ribeiro-dos-Santos et al., 2010; Moreira et al., 2014a). Therefore, when miRNAs are deregulated in the human stomach, they compromise important pathways that regulate the normal functions of this organ (Zhang et al., 2014), contributing to the onset and progression of gastric carcinogenesis (Wu et al., 2014).

In recent years, our research group joined efforts and used different technologies to understand the relationship between the deregulation of these small RNAs and gastric carcinogenesis (Assumpção et al., 2015; Darnet et al., 2015; Vidal et al., 2016; Magalhães et al., 2018; Pereira et al., 2019). More recently, we have demonstrated the deregulation of miRNAs during the evolution of Correa's cascade (Vidal et al., 2016) and in field cancerization in GC (Assumpção et al., 2015; Pereira et al., 2019).

Field cancerization assumes that the tissues adjacent to the tumor have molecular changes (e.g., genetic and/or epigenetic) that make them susceptible to the onset of tumors or recurrences (Slaughter et al., 1953; Chai and Brown, 2009; Curtius et al., 2018). The preexisting molecular changes in this tissue (which may be caused by an insult to the healthy epithelium) are precursors of carcinogenesis and serve as background for the establishment and progression of the tumor (Hattori and Ushijima, 2016; Padmanabhan et al., 2017; Curtius et al., 2018). In fact, studies have shown that adjacent to the tumor tissues have similar molecular changes to those found in the tumor as well as unique and exclusive changes, which distinguish them from non-cancerous tissues (Assumpção et al., 2015; Takeshima et al., 2015; Aran et al., 2017; Vidal et al., 2017; Yoshida et al., 2017; Pereira et al., 2019). Recently, miRNAs were related to field cancerization in GC (Assumpção et al., 2015; Pereira et al., 2019), however, little is known about the true role played by these small non-coding RNAs in this process. The use of robust large-scale sequencing technologies is an excellent strategy in both the discovery of new biomarkers and in providing an overview of the complex relationship between miRNAs and field cancerization.

In this study, we used deep sequencing to evaluate the overall expression profile of miRNAs in non-cancerous gastric tissues, adjacent to cancer and with cancer tissues in order to identify miRNAs involved in the field cancerization. In addition, we evaluated the discriminatory performance of miRNAs as biomarkers of gastric carcinogenesis. Our results show new deregulated miRNAs, which are potentially useful as biomarkers for this cancer and suggest a new molecular mechanism involved in the biology of field cancerization.

#### MATERIALS AND METHODS

### Biological Material

A total of 45 fresh samples of stomach antrum tissues were included in the present study. The non-cancerous control (NC) samples were collected from 15 patients without cancer (chronic gastritis; ± H. pylori; mean age = 59.2) during an upper digestive endoscopy (tissue fragments of approximately 4 millimeters). In addition, 15 tumor-adjacent tissues (histopathologically without cancer) and 15 gastric adenocarcinoma samples were collected from patients with gastric cancer. The tumor-adjacent (ADJ) and gastric adenocarcinoma (GC) samples were all paired (± H. pylori; mean age = 59.9).

The tissues were obtained from patients treated at the Hospital Universitário João de Barros Barreto (HUJBB), Belém, Pará, Brazil. Samples were collected prior to antibiotic, chemotherapeutic, and/or radiotherapeutic treatment. Immediately after collection, all samples were frozen and stored in liquid nitrogen until analysis.

Histopathological characterization of the samples, such as tumor subtype, degree of differentiation, depth of invasion, involvement of lymph nodes, and/or distant metastases were extracted from pathological reports performed by the HUJBB Department of Pathology. Histopathological analysis of the tumor fragments was performed according to Lauren's classification (Lauren, 1965).

#### Ethics Statement

fgene-10-00592 June 17, 2019 Time: 17:32 # 3

This study was reviewed and approved by the Ethics Committee of the Center of Oncology Research of the Federal University of Pará (Protocol No. 1.081.340). All study participants or their legal guardian provided informed written consent in accordance with the Helsinki Declaration of 1964, the Nuremberg Code, in compliance with the National Health Council's Research Guidelines Involving Human Beings (Res. CNS 466/12).

#### RNA Extraction, Small RNA Library Construction, and Sequencing

Total RNA was extracted using TRIzol <sup>R</sup> reagent (Thermo Fisher Scientific). After isolation, total RNA was stored at −80◦C until further analysis. The total RNA amount was determined using a Qubit <sup>R</sup> 2.0 (Life Technologies, Foster City, CA, United States), and an Agilent RNA ScreenTape assay and 2200 TapeStation Instrument (Agilent Technologies, United States) were used to detect RNA integrity. Samples with an RNA integrity number (RIN) ≥ 5 were sequenced.

For small RNA-seq, 1 µg of total RNA per sample was used for library preparation utilizing TruSeq Small RNA Sample Prep Kits (Illumina, San Diego, CA, United States). A DNA ScreenTape assay in a 2200 TapeStation Instrument (Agilent Technologies, United States) and real-time PCR with a KAPA Library Quantification Kit (KAPABIOSYSTEM, United States) were used to validate and quantify each library. A 4-nM library pool comprising all samples was sequenced using a MiSeq reagent kit v3 150 cycle on a MiSeq System (Illumina, San Diego, CA, United States).

The raw sequencing reads of all libraries have been deposited at EBI-ENA (PRJEB27213).

#### Bioinformatics Analysis

The resulting reads were pre-processed and quality filtered (qv > 25). We used STAR (Dobin et al., 2013) aligner to map the reads to the human genome reference (GRCH37). We quantified mature miRNA sequencing using miRBase human annotation (v20). Counting expression data was performed with HTSeq (Anders et al., 2015).

Differential expression analysis of all processed data was performed using the bioconductor-DESeq2 package (Love et al., 2014) in R software, with a detection threshold of 10 counts per miRNA (present at least 10 read counts in at least of the libraries). Comparison between (i) gastric cancer (GC) vs. noncancerous (NC) samples; (ii) adjacent to gastric cancer (ADJ) vs. NC samples; and (iii) GC vs. ADJ samples were made separately. Adjusted P-values ≤ 0.05 and |log2(fold change)| > 2 were considered statistically significant.

For graphical analysis of miRNAs, expression data was normalized to RPKM. Heatmaps were used for hierarchical clustering of differentially expressed miRNAs. The area under the curve (AUC > 0.85) from the receiver operating characteristic (ROC) curves was used to identify biomarkers with the best

sensitivity/specificity relation, and a discriminant analysis of principal components (DAPC) was constructed to infer the number of clusters of epigenetically related samples. All graphical analyses were performed using the R statistical platform.

### Identification of Differentially Expressed miRNA Target Genes

We used two online tools to identify the differentially expressed (D. E.) miRNA target genes: (i) TargetCompare (Moreira et al., 2014b) and (ii) miRTargetLink Human (Hamberg et al., 2016). TargetCompare database allows the user to filter miRNAs and its targets genes that are associated with determined diseases, such as GC. miRTargetLink Human is a tool that allows the search for interactions between target genes and miRNAs that have been experimentally validated by molecular biology techniques.

The identified target genes were submitted to functional annotation and enrichment in KEGG pathways using DAVID Bioinformatics Resources v.6.8 online tool (Huang et al., 2009a,b).

## RESULTS

After quality control (**Supplementary Figure S1**), alignment and transcript quantitation, several small non-coding RNAs (sncRNAs) and other transcript fragments were identified. From them, ∼30% (∼9.5 million reads) were recognized as microRNAs reads, identifying 1,144 mature miRNAs. Approximately 90% of the miRNA reads (∼8.5 million reads) were concentrated on the 35 most expressed miRNAs. The number and representativity of D. E. expressed miRNAs identified in this study is similar to other studies that performed miRNome analysis (Hou et al., 2011; Maltseva et al., 2014; Liu et al., 2015; Castro-Magdonel et al., 2017; Liang et al., 2017).

### Differentially Expressed miRNAs in GC vs. NC Analysis

We found 21 differentially expressed (D. E.) miRNAs when we compared gastric cancer samples (GC) with non-cancerous gastric samples (NC) (**Figure 1** and **Supplementary Table S1**),

of which eight were down-regulated and 13 were up-regulated in GC (**Supplementary Table S1**). The heatmap obtained from the normalized expression of the D. E. miRNAs perfectly clustered the GC and NC samples together (**Figure 2A**).

Fourteen miRNAs were exclusively D. E. in the GC vs. NC analysis: nine were up-regulated (six oncomiRs, one TS-miR and two with unknown functions in GC) and five were downregulated (all TS-miRs) in GC (**Figure 1** and **Table 1**). Six miRNAs, of which five were down-regulated (hsa-miR-141-3p, -miR-148a-3p, -miR-148a-5p, -miR-153-3p, and -miR-375) and one were up-regulated (hsa-miR-196a-5p), presented the best sensitivity/specificity relation (AUC > 0.85) and were considered as potential biomarkers to identify GC (**Figure 3**).

#### Differentially Expressed miRNAs in ADJ vs. NC Analysis

Comparing adjacent to gastric cancer samples (ADJ) with NC revealed 16 D. E. miRNAs (**Figure 1** and **Supplementary Table S2**), of which two were down-regulated and 14 were up-regulated (**Supplementary Table S2**). The heatmap obtained from the normalized expression of the D. E. miRNAs clustered ADJ and NC samples together (**Figure 2B**). Furthermore, nine miRNAs (eight TS-miRs and one with unknown function in GC) were exclusively D. E. in the ADJ vs. NC analysis: they were all up-regulated in ADJ (**Figure 1** and **Table 2**).

Five D. E. miRNAs (hsa-miR-99a, -miR-100-5p, -miR-125b-5p, -miR-145-3p, and -miR-145b-5p, all up-regulated in ADJ) presented the best sensitivity/specificity relation (AUC > 0.85) and were considered as potential early biomarkers for GC (**Figure 4**).

### Functional Analysis of the Up-Regulated TS-miRs in ADJ Tissue

To evaluate the biological and functional role of the eight upregulated TS-miRs and hsa-miR-99a-5p in ADJ tissue (**Table 2**),

TABLE 1 | Deregulated miRNAs only in gastric cancer.


GC, gastric cancer; ADJ, adjacent to cancer; NC, non-cancerous; Up, upregulated; Down, down-regulated; TS-miR, tumor suppressor miRNA; OncomiR, oncogenic miRNA. (-) Not differentially expressed. (<sup>∗</sup> ) Putative role in GC. (1) Wang et al., 2012; (2) Lu et al., 2018; (3) Shao et al., 2019; (4) Sun et al., 2012; (5) Pan et al., 2017; (6) Liao et al., 2012; (7) Deng et al., 2014; (8) He et al., 2017; (9) Wang et al., 2018; (10) Li B.S. et al., 2015; (11) LArki et al., 2018; (12) Zuo et al., 2015; (13) Zhou et al., 2019; (14) Gao et al., 2016; (15) Zhang et al., 2016; (16) Xie et al., 2019; (17) Zheng et al., 2011; (18) Wang and Liu, 2016; (19) Ouyang et al., 2018; (20) Ding et al., 2010; (21) Chen et al., 2017; (22) Hwang et al., 2018.

we performed enrichment and functional annotation of their experimentally validated common target genes. We used nine miRNAs as input in the online tools TargetCompare (Moreira et al., 2014b) and miRTargetLink Human (Hamberg et al., 2016), which output five miRNAs (hsa-miR-let7-c-5p, -miR-99a-5p, -miR-100-5p, -miR-133a-3p, and -miR-145-5p) that regulate 18 common target genes (**Figure 5A**). The enrichment and functional annotation analysis of the 18 target genes performed in DAVID v.6.8 (Huang et al., 2009a,b) revealed that 13 genes participate in 23 different biological pathways

TABLE 2 | Deregulated miRNAs only in adjacent to tumor tissue.


GC, gastric cancer; ADJ, adjacent to cancer; NC, non-cancerous; Up, upregulated; TS-miR, tumor suppressor miRNA; OncomiR, oncogenic miRNA. (-) Not differentially expressed. (<sup>∗</sup> ) Putative role in GC. (?) Unknown role in GC. (1) Tsai et al., 2015; (2) Qiu et al., 2014; (3) Zhang et al., 2018; (4) Yang et al., 2017; (5) Wu et al., 2013; (6) Lei et al., 2017; (7) Shi et al., 2015; (8) Zhao et al., 2017.

(**Supplementary Table S3**), and 11 genes are involved in nine biological pathways that are important for the development and progression of GC (**Figure 5B**).

### Differentially Expressed miRNAs Common to GC vs. ADJ vs. NC Analysis

hsa-miR-125b-5p, miR-125b-1-3p, -miR-200a-3p, miR-218-1-3p, miR-490-3p, -miR-493-5p and -miR-873-5p were D. E. in both the GC vs. NC (**Table 3** and **Supplementary Table S1**) and ADJ vs. NC (**Table 3** and **Supplementary Table S2**) analyses. These miRNAs were not D. E. in the GC vs. ADJ analysis (**Table 3**), suggesting that these two tissues are similar regarding the expression of these miRNAs. In addition, the DAPC plot (**Figure 6**) generated by all D. E. miRNAs indicates that GC and ADJ, despite generating distinct clusters, have much more similarity in its expression profiles when compared with NC tissue, which clustered apart. Thus, we assembled the GC and ADJ samples in a single group to compare to the NC samples during the ROC curve analysis. Two down-regulated miRNAs (hsa-miR-200a-3p and hsa-miR-873-5p) had the best sensitivity/specificity relation and were considered as potential biomarkers to identify gastric carcinogenesis (**Figures 7A,B**).

Comparing GC with ADJ, only one miRNA (hsa-miR-196a-5p) was differentially expressed in GC (**Figure 1** and **Table 1**). hsa-miR-196a-5p was significantly up-regulated in GC (P = 0.029) but had a low sensitivity/specificity relation in the ROC curve analysis (AUC < 0.85). hsa-miR-196a-5p was also up-regulated in the GC vs. NC comparison (**Table 1** and **Supplementary Table S1**). The difference in the number of miRNAs D. E. in GC vs. ADJ compared with the previous analysis corroborates to premise of that these two tissues have similar miRNA expression profiles.

### DISCUSSION

miRNA deregulation is closely related to gastric cancer development (Wu et al., 2014; Zhang et al., 2014) and its relationship to field cancerization becomes evident (Assumpção et al., 2015; Pereira et al., 2019). In the search for new biomarkers and a better understanding of epigenetic field cancerization in GC, we evaluated the global expression profile of miRNAs in NC, ADJ, and GC gastric samples. Our data showed that all three types of tissue share many differentially expressed miRNAs and present miRNAs that occur exclusively in either

ADJ or GC tissues, behaving as molecular signatures for those conditions (**Tables 1**, **2**).

Among the deregulated miRNAs only in GC (**Table 1**), seven up-regulated (hsa-miR-135b-5p, -miR-196a-5p, -miR-196b-5p, -miR-215-5p, -miR-224-5p, -miR-615-3p, and -miR-25-5p) and five down-regulated (hsa-miR-135a-5p, -miR-148a-3p, -miR-148a-5p, -miR-153-3p, and -miR-375) miRNAs are likely oncomiRs and TS-miRs, respectively. In fact, studies demonstrate that miR-135b (Wang et al., 2012; Lu et al., 2018; Shao et al., 2019), miR-196a (Pan et al., 2017), miR-196b (Liao et al., 2012), miR-215 (Deng et al., 2014), miR-224 (He et al., 2017), miR-615-3p (Wang et al., 2018), and miR-25 (Li B.S. et al., 2015; LArki et al., 2018) were reported as oncomiRs, while miR-135a (Zhang et al., 2016; Xie et al., 2019), miR-148a (Zheng et al., 2011), miR-153 (Wang and Liu, 2016; Ouyang et al., 2018) and miR-375 (Ding et al., 2010; Chen et al., 2017; Hwang et al., 2018) were reported as TS-miRs in GC. These results suggest that the joint deregulation of both oncomiRs and TS-miRs is required for the support and progression of GC.

hsa-miR-196a-5p was up-regulated in GC when comparing to ADJ and NC tissues; at the same time, it was not differentially expressed between the ADJ and NC tissues. Deregulation of this miRNA is important for GC progression because it promotes cell proliferation by down-regulating the expression of CDKN1B

TABLE 3 | Deregulated miRNAs considering both adjacent to tumor and gastric cancer tissues.


GC, gastric cancer; ADJ, adjacent to cancer; NC, non-cancerous; Up, upregulated; Down, down-regulated; TS-miR, tumor suppressor miRNA; OncomiR, oncogenic miRNA. (-) Not differentially expressed. (<sup>∗</sup> ) Putative role in GC. (1) Wu et al., 2015; (2) Zhang et al., 2017b; (3) Deng et al., 2017; (4) Tang et al., 2016; (5) Zhang et al., 2017a; (6) Wang et al., 2017; (7) Shen et al., 2015; (8) Qu et al., 2017; (9) Yu et al., 2019; (10) Cong et al., 2013; (11) Chen et al., 2016; (12) Cao et al., 2016; (13) Zhou et al., 2015.

(p27kip<sup>1</sup> ) tumor suppressor (Sun et al., 2012) and invasion and epithelial to mesenchymal transition of cancer stem cells by down-regulating the expression of SMAD4 in GC (Pan et al., 2017). In addition, hsa-miR-196a has been associated with metastasis in lymph nodes and the clinical stage of GC (Li H.L. et al., 2015).

Nine miRNAs were found to be up-regulated only in ADJ tissue (hsa-miR-let7-c-5p, -miR-99a-5p, -miR-100-5p, -miR-133a-3p, -miR-133b-3p, -miR-143-5p, -miR-145-3p, -miR-145-5p, and -miR-320b) and may act as TS-miRs in this tissue. Studies have shown that miR-let7-c (Tsai et al., 2015), miR-133a (Qiu et al., 2014; Zhang et al., 2018), miR-133b (Qiu et al., 2014; Yang et al., 2017), miR-143-5p (Wu et al., 2013; Lei et al., 2017), miR-145 (Qiu et al., 2014; Lei et al., 2017), miR-100-5p (Shi et al., 2015), and miR-320b (Zhao et al., 2017) are down-regulated in GC and play the role of TS-miRs because they inhibit proliferation, migration, invasion, and cell cycle progression.

Gene enrichment analysis in KEGG pathways revealed that these miRNAs regulate genes involved in important pathways that contribute to gastric carcinogenesis, such as the PI3K-Akt, HIF-1, Ras, Rap1, ErbB, and MAPK signaling pathways (**Figures 5A,B**). These pathways control important cell functions such as proliferation, migration, invasion, and progression of the cell cycle. The overexpression of TS-miRs in ADJ tissue may be a mechanism to compensate for preexisting molecular alterations in an attempt to contain the tumorigenesis process. However, we believe that during the progression of carcinogenesis in ADJ tissue, the up-regulated TS-miRs identified herein are inversely deregulated (become downregulated), contributing to the eventual onset, establishment and progression of GC. In a previous study, we reported the overexpression of the TS-miR hsa-miR-29c in adjacent to the gastric cancer tissues and its down-regulation in GC and NC tissues (Pereira et al., 2019).

The intense inflammatory process to which the adjacent tissue is subjected may be one of the causes of over representation of these TS-miRs, since inflammation can alter the local microenvironment by stimulating the expression of some miRNAs. Studies have shown that IL-6 and IL-17 pro-inflammatory interleukins activity may stimulate the expression of miRNAs with known oncogenic activity (miR-21 and miR-135b) in tissues submitted to the intense inflammatory process (Löffler et al., 2007; Iliopoulos et al., 2010; Matsuyama et al., 2011; Rozovski et al., 2013; Singh et al., 2015). Another possibility would be the existence of pre-cancerous lesions in adjacent tissues (e.g., atrophic and non-atrophic gastritis and intestinal metaplasia) and/or Helicobacter pylori infection, since the expression levels of many miRNAs (e.g., miR-21, miR-29c, miR-135b, miR-155, miR-204, and miR-223) may be change under these conditions (Link et al., 2015; Vidal et al., 2016; Pereira et al., 2019). Therefore, the local microenvironment and the histopathological characteristics of these tissues can directly influence the expression of many miRNAs, making them susceptible to eventual molecular alterations and carcinogenesis.

Our data (overexpression of nine TS-miRs only in adjacent to the gastric cancer tissues) corroborate with the findings of Aran et al. (2017), who used the transcriptomic profile of normal tissue adjacent to the tumor (NAT) to demonstrate that this tissue has an intermediate and unique expression profile when compared to truly normal and cancerous tissues. Therefore, despite the similarity between adjacent and tumor tissues, the former is not a malignant tissue, but it is not a molecularly normal tissue either (De Assumpção et al., 2016; Aran et al., 2017).

Many genetic and epigenetic changes identified in GC are shared by ADJ tissue (Assumpção et al., 2015; Takeshima et al., 2015; Aran et al., 2017; Vidal et al., 2017; Yoshida et al., 2017; Pereira et al., 2019). Probably some of these shared alterations were already present in the adjacent tissue before the onset of the tumor in situ and contributed to its establishment, making this tissue still susceptible to carcinogenesis even after surgical removal of the tumor.

We found that both ADJ and GC tissues share the upregulation of hsa-miR-125b-5p and hsa-miR-125b-1-3p oncomiRs and the down-regulation of hsa-miR-200a-3p and hsa-miR-873-5p TS-miRs, suggesting that the joint deregulation of both oncomiRs and TS-miRs is required for the progression of gastric carcinogenesis. In addition, seven miRNAs (hsamiR-125b-5p, -miR-125b-1-3p, -miR-200a-3p, -miR-218-1-3p, miR-490-3p, -miR-873-5p, and -miR-493-5p) are deregulated in both ADJ and GC tissues when compared to NC tissue; however, these miRNAs are not D. E. between ADJ and GC tissues. Our results suggest that these two tissues share molecular alterations and that ADJ is an epigenetically altered tissue. Five miRNAs (hsa-miR-125b-5p, -miR-125b-1-3p, -miR-493-5p, -miR-200a-3p, and -miR-873-5p) can promote tumor onset and progression, as studies have shown that the deregulation of miR-125b (Wu et al., 2015; Zhang et al., 2017b), miR-493-5p (Zhou et al., 2015), miR-200a (Cong et al., 2013), and miR-873 (Cao et al., 2016; Chen et al., 2016) contributes to cell proliferation, migration, invasion, and cell growth in GC.

Although we have not performed further experimental assays to confirm the deregulation and the functional role of the identified miRNAs in our study, the literature consistently

FIGURE 6 | Discriminant analysis of principal component (DAPC) plot shows that NC, ADJ, and GC samples form distinct clusters. The DAPC analysis showed great similarity between GC and ADJ tissues.

supports and corroborates our findings and hypothesis through studies that have applied safe, sensitive and reliable techniques and methods (e.g., RT-qPCR, Western Blot, Cell and/or Reporter Assays) that provides strong evidences of these miRNAs' deregulation in GC (Ding et al., 2010; Zheng et al., 2011; Liao et al., 2012; Sun et al., 2012; Wang et al., 2012, 2017, 2018; Cong et al., 2013; Wu et al., 2013, 2015; Deng et al., 2014, 2017; Qiu et al., 2014; Li B.S. et al., 2015; Shen et al., 2015; Shi et al., 2015; Tsai et al., 2015; Zhou et al., 2015, 2019; Zuo et al., 2015; Cao et al., 2016; Chen et al., 2016, 2017; Gao et al., 2016; Tang et al., 2016; Vidal et al., 2016; Wang and Liu, 2016; Zhang et al., 2016, 2017a,b, 2018; He et al., 2017; Lei et al., 2017; Pan et al., 2017; Qu et al., 2017; Yang et al., 2017; Zhao et al., 2017; Hwang et al., 2018; LArki et al., 2018; Lu et al., 2018; Magalhães et al., 2018; Ouyang et al., 2018; Pereira et al., 2019; Shao et al., 2019; Xie et al., 2019; Yu et al., 2019).

For the evaluation of the studied miRNAs as potential biomarkers, we selected three different groups of markers with AUC > 85%. Among the D. E. miRNAs only in GC, six miRNAs (hsa-miR-141-3p, -miR-148a-3p, -miR-148b-5p, -miR-153-3p, -miR-196a-5p, and -miR-375) are potentially useful in identifying patients with advanced disease (**Figure 3**). Among the deregulated miRNAs only in ADJ tissue, five (hsa-miR-99a-5p, -miR-100-5p, -miR-125b-5p, -miR-145-3p, and -miR-145-5p) are potentially useful in identifying patients susceptible to tumor development or the early stages of gastric cancer (**Figure 4**). Two miRNAs (hsa-miR-200a-3p and hsa-miR-873-5p) are potentially useful in identifying both patients with established disease and patients susceptible to developing it (**Figure 7**). Thus, these miRNAs may be a potential diagnostic alternative for this type of cancer because the currently available biomarkers do not have a good sensitivity/specificity relationship, which makes it difficult to diagnose the disease early and to start curative treatment.

Studies that analyzed the data from miRNA sequencing in gastric cancer demonstrated their deregulation in this type of tumor (Assumpção et al., 2015; Darnet et al., 2015; Liu et al., 2015; Liang et al., 2017). Among the miRNAs found in these studies, hsa-miR-99a, -miR-100, -miR-133a/b-3p, -miR-135b, -miR-141, -miR-143, -miR-145, -miR-148a, -miR-196a/b, -miR-200a, -miR-218-1, -miR-215, and -miR-490 were also observed in the present study. In addition to those miRNAs, we also found 16 novel differentially expressed miRNAs in the gastric field cancerization.

Liu et al. (2015) analyzed miRNome in gastric cancer downloaded from the TCGA; however, this database basically has samples from European and Asian populations. These authors identified 54 D. E. miRNAs (using the Padj < 0.05 and |log<sup>2</sup> (fold change)| > 3) and two miRNAs (hsa-miR-133a/b) were the most D. E. Liang et al. (2017) also analyzed miRNomes in GC using samples from the same database (TCGA) and identified 43 D. E. miRNAs (using the FDR < 0.001 and |log<sup>2</sup> (fold change)| > 1.5); of these, 5 miRNAs (hsa-miR-30a, -miR-135b, -miR-133b, -miR-143, and -miR-145) were associated with patient survival time. We emphasize that the "normal" or "healthy" tissues used in these two studies are of patients with the disease (adjacent to the gastric cancer tissues), since the TCGA does not have expression data of gastric tissues of patients without the history of GC. By using samples from the Brazilian population (that has genetic admixture), we identified 30 miRNAs differentially expressed when comparing patients with GC and individuals without the history of GC (we considered Padj < 0.05 and |log2(fold change)| > 2). In addition, we found that the hsamiR-196a-5p was D. E. between the GC and ADJ tissues. Many D. E. miRNAs identified by Liu et al. (2015) and Liang et al. (2017) were also identified in this study; however, there are differences in the number of D. E. miRNAs found among these studies (mainly in GC and ADJ analysis), which may be a consequence of: (i) the statistical criteria used, (ii) the number of samples used, and/or (iii) the ethnical and genetic characteristics of the studied populations. Our data are important because we analyzed the miRNA sequencing of a larger number of samples from the Brazilian population. This population has a genetic contribution from different parental populations, such as the European, African, Asian, and Amerindian ones (Santos et al., 2010; Andrade et al., 2018). The strong genetic substructure and admixture (Santos et al., 2010) of our population may interfere in the expression profile of some genes (Pinto et al., 2015; Dluzen et al., 2016). Therefore, this study provides important and relevant information about the expression profile of miRNAs associated to gastric field cancerization in populations with genetic substructure and admixture, such as the Brazilian one.

Overall, our study was able to demonstrate that the tissue adjacent to gastric cancer shares some epigenetic changes (miRNAs deregulation) present in the tumor and also has unique and exclusive alterations; therefore, it should not be used as a healthy and/or normal tissue as a benchmark for gastric cancer. Thus, we recommend the use of gastric samples from patients with no history of GC as a control to exclude any biases that the adjacent tissue may provide in the miRNAs' expression profile in GC.

In this study, we also observed that the tissues adjacent to GC have an over representation of microRNAs with known tumor suppressor activities, suggesting that these microRNAs may represent a barrier against tumorigenesis within these precancerous tissues prior to the eventual formation of a tumor. The excellent performance of the studied miRNAs in identifying with good sensitivity and specificity, both early and advanced stages of the disease, make them potentially useful as biomarkers and therapeutic targets for GC.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the National Health Council's Research Guidelines Involving Human Beings (Res. CNS 466/12), Research Ethics Committees and National Research Ethics Commission. The protocol was approved by the Ethics Committee of the Center of Oncology Research of the Federal University of Pará (Protocol No. 1.081.340). All subjects gave written informed consent in accordance with the Declaration of Helsinki.

#### AUTHOR CONTRIBUTIONS

AP and FM study design, formal analyses, interpretation of data, and drafting of manuscript. ÂRS, PA, and SS study concept and design and obtaining funding. PA, MA, and AC provided clinical materials. SD histopathological evaluation. TV-S, AV, PP, LM, and AMRS prepared nucleic acids, sequencing libraries, and performed the experiments. All authors reviewed the manuscript.

### FUNDING

This study was supported by Rede de Pesquisa em Genômica Populacional Humana (CAPES/Biologia Computacional: No. 3381/2013/CAPES) and PROPESP/UFPA. AP was supported by CAPES Ph.D. fellowship (PROEX – 0487) in Genetics and Molecular Biology (UFPA) and is supported actually by CAPES/BRASIL Post-Doctoral fellowship (Edital No. 51/2013/Biologia Computacional; Proc. No. 88887.313819/2019-00) in Oncology and Medical Sciences (UFPA). ÂRS was supported by CNPq/Productivity: 304413/2015-1 and SS was supported by CNPq/Productivity: 305 256/2013-3.

#### ACKNOWLEDGMENTS

fgene-10-00592 June 17, 2019 Time: 17:32 # 10

We thank to Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Conselho Nacional de Desenvolvimento

#### REFERENCES


Científico e Tecnológico (CNPq), and PROPESP/UFPA for the financial support and fellowships.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2019.00592/full#supplementary-material


by directly targeting transducer of ERBB2, 1 and correlates with poor survival. Oncogene 34, 2556–2565. doi: 10.1038/onc.2014.214


progression via targeting transcription factor Sp1 in gastric cancer. FEBS Lett. 588, 1168–1177. doi: 10.1016/j.febslet.2014.02.054



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

Copyright © 2019 Pereira, Moreira, Vinasco-Sandoval, Cunha, Vidal, Ribeiro-dos-Santos, Pinto, Magalhães, Assumpção, Demachki, Santos, Assumpção and Ribeirodos-Santos. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Epigenetics of Alzheimer's Disease: Factors and Therapeutic Implications

Xiaolei Liu1,2, Bin Jiao1,3,4 and Lu Shen1,3,4,5 \*

<sup>1</sup> Department of Neurology, Xiangya Hospital, Central South University, Changsha, China, <sup>2</sup> The Center of Gerontology and Geriatrics, West China Hospital, Sichuan University, Chengdu, China, <sup>3</sup> National Clinical Research Center for Geriatric Disorders, Central South University, Changsha, China, <sup>4</sup> Key Laboratory of Hunan Province in Neurodegenerative Disorders, Central South University, Changsha, China, <sup>5</sup> Key Laboratory of Organ Injury, Aging and Regenerative Medicine of Hunan Province, Changsha, China

Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Melanie Carless, Texas Biomedical Research Institute, United States Andrew Shafik, Emory University, United States Xingshun Xu, The Second Affiliated Hospital of Soochow University, China

> \*Correspondence: Lu Shen shenlu@csu.edu.cn; shenlu2505@126.com

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 04 June 2018 Accepted: 08 November 2018 Published: 30 November 2018

#### Citation:

Liu X, Jiao B and Shen L (2018) The Epigenetics of Alzheimer's Disease: Factors and Therapeutic Implications. Front. Genet. 9:579. doi: 10.3389/fgene.2018.00579 Alzheimer's disease (AD) is a well-known neurodegenerative disorder that imposes a great burden on the world. The mechanisms of AD are not yet fully understood. Current insight into the role of epigenetics in the mechanism of AD focuses on DNA methylation, remodeling of chromatin, histone modifications and non-coding RNA regulation. This review summarizes the current state of knowledge regarding the role of epigenetics in AD and the possibilities for epigenetically based therapeutics. The general conclusion is that epigenetic mechanisms play a variety of crucial roles in the development of AD, and there are a number of viable possibilities for treatments based on modulating these effects, but significant advances in knowledge and technology will be needed to move these treatments from the bench to the bedside.

Keywords: Alzheimer's disease (AD), epigenetics, non-coding RNA (ncRNA), DNA methylation, histone modifications and chromatin structure

### INTRODUCTION

Alzheimer's disease (AD), an irreversible and progressive neurological disease, is the most common form of neurodegenerative dementia (Mckhann et al., 2011; Rocca et al., 2011; Lansdall et al., 2017). The most prominent early symptom is short-term memory loss. As the disease progresses, other symptoms such as personality changes, apathy and language problems emerge. 95 percent of patients who are hospitalized with AD have the sporadic form, which is late-onset Alzheimer's disease (LOAD) (Diniz et al., 2017; Zhao et al., 2017). The pathology of LOAD is multi-factorial with biological, genetic and environmental factors interacting with each other to aggravate the process of AD. Many genetic risk factors for LOAD have been identified. Among them, the ε4 isoform of apolipoprotein E (ApoE4) is well-known as the strongest genetic risk factor for LOAD (Lane-Donovan and Herz, 2017). However, less than 1 percent of patients have autosomal dominant inherited Alzheimer's disease (DIAD) which has a much earlier age of onset, around 45 years old. Genetic mutations in the genes encoding APP, PS1, and PS2 that cause overproduction or formation of an aberrant form of Aβ are found in this group (Levy et al., 1990; Masters et al., 2015).

At the cellular level, AD can be characterized by the appearance of extracellular plaques from accumulations of insoluble amyloid beta (Aβ) filaments, intracellular neurofibrillary tangles of hyperphosphorylated tau and neuroinflammation (Selkoe, 2012). It appears that AD is not only

one or two types of diseases, but is rather a group of diseases with similar APP and Tau pathologies that are triggered by different mechanisms (Sery et al., 2013). Among them, the most popular theory is the amyloid cascade hypothesis and detection of the accumulation of Aβ is now available in MCI and AD patients using cerebrospinal fluid biomarkers and PET. However, it is still difficult to find stable biomarkers which could help to discover this disease much earlier, and the exact mechanisms of AD are still unknown. Now the number of studies examining epigenetic mechanisms playing in the etiology of AD has risen dramatically. With this increased focus, the epigenetic modification in AD has become a highly popular topic. This review summarizes the literature on the role of epigenetic mechanisms in the development of AD and the possibilities for treating AD by epigenetically based interventions.

### EPIGENETICS FACTORS IN AD

Epigenetic factors include DNA methylation, histone modifications, chromatin remodeling, and regulation by non-coding RNA (**Figure 1**) (Morange, 2002; Holliday, 2006; Santana et al., 2017). The functions of epigenetics have been studied in various areas of biology, cancer biology and so on (Jones and Laird, 1999; Champagne, 2013). Now, many researchers are focusing on investigating the potential roles of epigenetics in AD pathogenesis.

### DNA METHYLATION

DNA methylation modifies cytosine residues by adding methyl groups in cytosine/guanine-rich regions such as CpG islands (Mehler, 2008). The process is initiated by DNA methyltransferases (DNMT), including DNMT1, DNMT2, DNMT3a, and DNMT3b (Kemme et al., 2017). In this section we discuss the possibility that AD could be caused by aberrations in DNA methylation in some certain genes and the potential role of methylation in being a biomarker in AD.

For example, it was found that some cytosines, particularly those at −207 to approximately −182, in the promoter region of the APP gene are mostly methylated and that their demethylation with age may led to Aβ deposition in the aged brain (Tohgi et al., 1999a,b). Methylation of the microtubule-related protein tau (MAPT) gene could also suppress the MAPT expression, which could affect the levels of tau protein (Zhang et al., 2016). Mano et al. (2017) used postmortem brain samples from AD patients and found that the expression of BRCA1 was significantly upregulated, consisting of its hypomethylation. Additionally, BRCA1 protein levels were increased in response to Aβ and became mislocalized to the cytoplasm, in both in vitro cellular and in vivo mouse models (Mano et al., 2017). Recently, it was found that reduced DNA methylation at the triggering receptor expressed on myeloid cells 2 (TREM2) gene intron 1 caused higher TREM2 mRNA expression in the leukocytes of AD subjects than in the controls (Ozaki et al., 2017). Furthermore, DNA methylation in some other genes also plays a role in the mechanisms of AD and might be a potential biomarker. A study found that DNA methylation (CpG5) of the BDNF gene promoter and a tag SNP (rs6265) have a significant role in the etiology of amnestic mild cognitive impairment (aMCI) and its progression to AD (Xie et al., 2017a). In addition, a 5-year longitudinal study using multivariate Cox regression analysis revealed that elevated methylation of CpG5 of BDNF promoters was a significant independent predictor of AD conversion. This suggests that increased levels of peripheral BDNF promoter methylation may be an epigenetic biomarker indicating the transformation of aMCI to AD (Xie et al., 2017b). Kobayashi et al. (2016) examined the DNA methylation levels of the COASY and SPINT gene promoter regions and found that DNA methylation in the two regions was significantly increased in AD and aMCI as compared to controls. Di Francesco et al. (2015) found that global DNA methylation in the peripheral blood mononuclear cells of LOAD patients was higher compared to healthy controls, and higher DNA methylation levels were associated with the presence of APOE ε4 allele (p = 0.0043) and APOE ε3 carriers (p = 0.05) in the global population. This indicated global DNA methylation in peripheral samples is a useful marker for screening individuals at risk of developing AD (Di Francesco et al., 2015). Moreover, DNA methylation levels in the NCAPH2/LMF2 promoter region were found to be a useful biomarker for the diagnosis of AD and aMCI (Shinagawa et al., 2016). One study has found that APOE CGI holds lower DNA methylation levels in AD compared to control in the frontal lobe, mostly in the non-neuronal cells of the AD brain (Tulloch et al., 2018). Additionally, PICALM gene methylation was found to be associated with cognitive decline in blood cells of AD patients (Mercorio et al., 2018). Recently, it was discovered that elevated DNA methylation across a 48-kb region spanning the HOXA gene cluster is associated with AD neuropathology (Smith et al., 2018). Moreover, a differentially methylated region was identified within the alternative promoter of the PLD3 gene showing higher DNA methylation levels in the AD hippocampus compared to controls (Blanco-Luquin et al., 2018). All of these indicated that the methylation levels of some genes could be potential biomarkers in AD.

In some postmortem brains or neuronal cells of AD individuals, a lot of work has been done to find more specific evidence. In cortical neurons of a postmortem AD brain, immunoreactivity for 5-methylcytosine (5-mC) was decreased compared to the control (Mastroeni et al., 2010). Levels of 5mC were reduced in the hippocampus, entorhinal cortex and cerebellum of patients with AD (Chouliaras et al., 2013; Condliffe et al., 2014). In hippocampal tissue, Mastroeni et al. (2010) found weak staining for antibodies against DNA methylation maintenance factors in AD cases, contrasting with normal brains. More compellingly, in a comparison of cortical neurons of monozygotic twins, one of whom had AD and the other not, they found extensive co-localization of 5 mC, with three cell-specific markers in the non-AD twin yet an absence of co-localization in the AD twin (Mastroeni et al., 2009). Furthermore, a study identified that numerous AD-related genes, such as MCF2L, ANK1, MAP2, LRRC8B, STK32C, and S100B, have cell-type-specific methylation signatures and document

differential methylation dynamics through DNA methylation analysis on purified neurons and glia cells (Gasparoni et al., 2018). Differences were found in DNA methylation in differentiated human neurons with and without Aβ treatment, and that Aβ could change the DNA methylation status of some cell-fate genes (Taher et al., 2014).

### CHROMATIN REMODELING AND HISTONE MODIFICATIONS

Chromatin is an assemblage of genomic DNA, histone proteins and associated factors. Chromatin can be dynamically altered by various modifications of histones. Other possible modifications include nucleosome repositioning, chromatin remodeling, and nuclear compartmentalization (Qureshi and Mehler, 2010). Histones are basic proteins that form the building blocks of nucleosomes. Their tails are susceptible to various modifications, including lysine (K), arginine (R), and histidine (H) methylation (Lardenoije et al., 2015). These alterations together constitute the so-called histone code. The histone code is very complex and includes several types of covalent modifications, such as acetylation, methylation, phosphorylation, and ubiquitination, of at least 20 possible sites within the histone proteins (Jenuwein and Allis, 2001).

Among all the above modifications, acetylation is certainly the best characterized post-translational modification of core histones. Histone acetylation is catalyzed by histone acetyltransferase (HAT) while deacetylation is influenced by histone deacetylase (HDAC). It is known that acetylation of histones is associated with enhanced rates of gene transcription, while deacetylation represses gene expression by condensing the chromatin. Here, we want to discuss the role played by HDAC in the AD mechanisms.

Class I HDACs, such as HDAC2 and HDAC3, are expressed at much higher levels than the other HDACs in the memoryassociated regions of the brain (Datta et al., 2018). A recent study used a mouse model of AD, which deleted the HDAC1 and HDAC2 genes in microglia cells, leading to a decrease in amyloid load and improved cognitive impairment by enhancing microglial amyloid phagocytosis (Broide et al., 2007). In the CK-p25 AD mouse model, Graff et al. (2012) found that elevated HDAC2 levels epigenetically block the expression of neuroplasticity genes during neurodegeneration, and HDAC2 reduces the histone acetylation of genes important for learning and memory. In another AD mouse model, HDAC2 was found to be strongly expressed in the hippocampus and prefrontal cortex. Neuron-specific overexpression of HDAC2 led to a decrease of dendritic spine density, synapse number, synaptic plasticity and memory formation (Levenson et al., 2004; Guan et al., 2009). In

conditions of stress and injury, the level of HDAC2 increases, and this causes decreased expression of genes related to memory and cognition (Levenson et al., 2004; Graff et al., 2012). HDAC3 also plays a vital role in regulating long-term memory formation. Deletion of HDAC3 in the dorsal hippocampus leads to enhanced long-term memory for object location (Mcquown et al., 2011).

In addition to class I HDACs, there is some evidence that class II HDACs are involved to some degree in AD. Class II HDACs are subdivided into class IIa and class IIb (Haggarty and Tsai, 2011). HDAC6, a class IIb HDAC, is elevated by 52% and 91% in the cortex and hippocampus of patients with AD. HDAC6 co-localizes with tau protein in the AD hippocampus, and tau phosphorylation can be decreased by reducing HDAC6 levels (Ding et al., 2008). Govindarajan et al. (2013) using a mouse model of AD, found that decreasing HDAC6 levels could improve cognitive impairment. HDAC4, a member of class IIa, is reported to be associated with synaptic plasticity and memory formation, and loss of it is detrimental to learning and memory (Kim et al., 2012). D'Addario et al. (2017) found that in the peripheral blood cells of monozygotic twins discordant for AD, higher levels of gene expression of HDAC2 and HDAC9 were found in the AD twin. Recently, it was reported that HDAC2 dysregulation could contribute to cholinergic nucleus basalis of Meynert (nbM) neuronal dysfunction, NFT pathology, and cognitive decline during clinical progression of AD (Mahady et al., 2018).

The class III HDACs are called sirtuins, SIRT1-7 (Gray and Ekstrom, 2001). The level of SIRT1 is reduced in the AD cortex, and SIRT1 levels are negatively correlated with the accumulation of paired helical tau filaments (Julien et al., 2009). Tau acetylation can promote pathological tau aggregation, while SIRT1 causes tau deacetylation. There is extensive evidence that SIRT1 reduction is involved in AD tau pathology, and tau acetylation of lysine 28 inhibits tau function via impaired tau-microtubule interactions, promoting tau aggregation (Cohen et al., 2011). Moreover, a significant positive correlation between SIRT1 levels and dementia was found whereby dementia risk increases by a factor of 1.16 due to an increase in the SIRT1 level and a factor of 24.23 due to a decrease in the TLR4 level (Kilic et al., 2018).

Histone acetyltransferases are also involved in the regulation of AD neurogenesis. For example, Tip60, a type of histone acetyltransferase, epigenetically regulates genes enriched in neurons, and its function in axonal transport is mediated by APP. More importantly, Tip60 plays a neuroprotective role in APP-induced axonal transport and functional locomotion defects (Johnson et al., 2013).

In conclusion, a range of studies indicates that histone modifications play a vital role in the development of AD. HDACs can both promote and impair memory formation and cognition.

#### NON-CODING RNA REGULATION

Non-coding RNAs (ncRNAs) can cause changes in gene expression and the production of proteins and they are involved in the pathology of many neurodegenerative diseases. A recent study used microarray analysis to characterize the expression patterns of circular RNAs (circRNAs), miRNAs and mRNAs in hippocampal tissue from an Aβ1-42-induced AD rat model, and found that a total of 555 circRNAs, 183 miRNAs, and 319 mRNAs were significantly dysregulated (change ≥2-fold and p-value < 0.05) in the hippocampus (Wang et al., 2018). In this section, we discuss three aspects of the relationship between ncRNAs and AD.

### miRNAS IN AD

In AD, miRNAs are expressed differently in the postmortem brain, blood monocytes and cerebrospinal fluid (CSF). The expression of miRNAs in the hippocampus, medial frontal gyrus and CSF is regionally and stage-specifically altered in AD patient brains using a sensitive qRT-PCR platform, and miRNAs are involved in pathways related to amyloid processing, neurogenesis, insulin resistance, and innate immunity in AD pathogenesis (Cogswell et al., 2008). Levels of members of the miR-29 family, including miR-29a, miR-29b, and miR-29c, are decreased in the brains of patients with AD, and this is related to high BACE1 protein expression (Hebert et al., 2008). It has been reported that miRNA-7, miRNA-9-1, miRNA-23a/miRNA-27a, miRNA-34a, miRNA-125b-1, miRNA-146a, and miRNA-155 are significantly increased in abundance in AD-affected superior temporal lobe neocortexes (Brodmann area 22). These seven upregulated miRNAs are involved in the coordination of amyloid production and clearance (Pogue and Lukiw, 2018). Higaki et al. (2018) suggested that miR-200b/c reduces Aβ secretion and Aβ-induced cognitive impairment by promoting insulin signaling. Moreover, exosome miR-135a and miR-384 were found to be upregulated, while miR-193b was downregulated, in the serum of AD patients compared to that of normal controls (Yang et al., 2018). A study using Tg2576 AD transgenic mice and human AD brain samples found that miR-206 could regulate brain-derived neurotrophic factor (BDNF), which regulates synaptic plasticity and memory (Lee et al., 2012). Moreover, of 8098 individually measured miRNAs in blood cells, six of these miRNAs such as miR-107, miR-125b, miR-146a, miR-181c, miR-29b, and miR-342 were found to be significantly downregulated in individuals with AD compared to controls (Fransquet and Ryan, 2018). These findings indicate that miRNAs are potential biomarkers for AD, and that miRNAs might be involved in the pathophysiology of AD.

### LONG NON-CODING RNAs IN AD

Long non-coding RNAs (lncRNA) also participate in the pathology of AD. Using a microarray analysis, it was found that in the hippocampal tissue of a rat model of AD, a total of 315 lncRNAs and 311 mRNAs were significantly dysregulated (≥2 fold, p < 0.05) (Yang et al., 2017). A type of ncRNA called 51A has been found to regulate the SORL1 gene, which encodes a sorting receptor for the APP holoprotein (Rogaeva et al., 2007). Another

study concluded that 51A might increase AD susceptibility by increasing APP formation (Ciarlo et al., 2013). Other lncRNAs are also related to AD mechanisms. For example, RNA 17A was found to be increased in AD. This lncRNA regulates GABAB receptor alternative splicing and signaling in response to inflammatory stimuli (Massone et al., 2011). NDM29 RNA upregulation is accompanied by altered APP modulation and can promote the cleavage activities of BACE (Massone et al., 2012). BC200 RNA is reported to be highly increased in the AD brain in parallel with disease progression (Wu et al., 2013). BC1, another type of lncRNA, is found to induce APP mRNA translation via association with fragile X syndrome protein (FMRP) in mouse model AD brains (Zhang T. et al., 2017). It was recently reported that EBF3 knockdown can reverse Aβ25-35-induced apoptosis in SH-SY5Y cells, and that lncRNA EBF3-AS promotes neuron apoptosis in AD through regulation of EBF3 expression (Gu et al., 2018). Furthermore, the overexpression of Aβ and Aβ-42 in AD could increase BACE1 antisense transcript lncRNA levels, leading to amyloid protein aggregation (Faghihi et al., 2008).

### OTHER ncRNAs IN AD

In addition to miRNA and lncRNA, there are other classes of ncRNAs that play roles in AD that are not as well understood. Recently, it was reported that circRNA-associatedceRNA networks in an AD mouse model are involved in Aβ clearance and myelin function (Zhang S. et al., 2017). For example, circRNA CIRS-7 significantly decreases in brains with AD. Its effect is to reduce the protein levels of APP and BACE1 (Lukiw, 2013). Y RNA is reported to play an essential pathological role in the altered mRNA landscape of AD (Champagne et al., 2001). Another interesting ncRNA is small NRSE dsRNA, which is a 20bp double-stranded RNA. NRSE dsRNA has been found to bind to the REST complex and activate it. The REST complex protects neurons from oxidative stress and amyloid β-protein toxicity (Rougeulle et al., 1998). The U1 snRNP is reported to show very high expression in neurofibrillary tangles in sporadic AD and early-onset AD (Hales et al., 2014a,b). All these findings point to ways that ncRNAs may take part in the development of AD.

### APPLICATIONS OF EPIGENETICS TO AD THERAPEUTICS

### DNA Methylation-Based Therapy

It has been shown that hypomethylation of AD risk genes such as APP, PSEN1, and PSEN2 has effects on learning and memory. Studies have shown that increases in methyl donor S-adenosyl-L-methionine (SAM) can decrease APP and PSEN1 expression by promoter hypermethylation (Scarpa et al., 2003; Fuso et al., 2005). It may also be relevant that the impaired learning ability induced by lead can be improved significantly by SAM, perhaps by ameliorating the impairments in the long-term potentiation of excitatory postsynaptic potentials that are caused by lead (Cao et al., 2008). Moreover, increasing levels of B12, folate and other methionine sources in the diet can promote methionine bioavailability and reverse elevated expression of APP and PSEN1 (Chan and Shea, 2007; Fuso et al., 2008).

Several studies have focused on finding DNMT inhibitors that could modulate the methylation of AD risk genes. DNMT inhibitors such as azacitidine and decitabine have been approved by the FDA for used in tumors and leukemia (Momparler et al., 1997; Issa et al., 2004). This approach has also been used in some other neurodegenerative diseases, including Friedreich's ataxia and fragile X syndrome (Chiurazzi et al., 1998; Sandi et al., 2013).

It has also been found that DNA methylation is related to the state of NSC differentiation. As a result, the use of the DNA methylation inhibitor 5-aza-cytidine to treat NSCs could alter DNA methylation, thereby disrupting migration and differentiation (Singh et al., 2009). Loss of Dnmt3a, a type of DNA methyltransferase, was found to impair hematopoietic stem cell differentiation (Challen et al., 2011). Another study showed that intrinsic epigenetic mechanisms play critical roles in the regulation of adult NSC functions. Using methylated CpG binding protein 1 (Mbd1)-deficient mice, it was found that both acute knockdown of Mbd1 or overexpression of fibroblast growth factor 2 (Fgf-2) in adult neural stem and progenitor cells inhibited neuronal differentiation (Li et al., 2008).

### Histone Modification-Based Therapy

Histone deacetylase inhibitors, including valproic acid (VPA), sodium 4-phenylbutyrate (4-PBA), vorinostat (SAHA), trichostatin A (TSA) and nicotinamide, have been shown to produce good results in AD mouse models. We discuss these five inhibitors separately.

Valproic acid has been shown to inhibit Aβproduction in a transgenic AD mouse model (Su et al., 2004). In neuroblastoma and glioma cell lines, VPA could prevent amyloid-β aggregation by increasing clusterin expression (Nuutinen et al., 2010). VPA was also found to decrease Aβ production in transgenic mice, improving their behavioral deficits (Qing et al., 2008). In an AD mouse model it relieved contextual memory deficits via histone H4 acetylation (Kilgore et al., 2010).

4-PBA is reported to decrease β-amyloid production and reverse spatial learning and memory deficits. In the Tg2576 AD mouse model, administration of 4-PBA led to a decrease in tau phosphorylation by increasing GSK-3β (Ricobaraza et al., 2009) as well as increasing intraneuronal Aβ clearance and restoration of dendritic spine densities in hippocampal CA1 pyramidal neurons (Ricobaraza et al., 2012).

SAHA is found to increase the H4K12 acetylation level, thus restoring learning ability in a mouse model (Peleg et al., 2010). In an astrocyte cell line, it induced the expression of clusterin, which might prevent the progress of AD pathogenesis (Nuutinen et al., 2010). In the APPswe/PS1dE9 AD mouse model, SAHA was reported to restore contextual memory by inhibiting HDAC6 in the early stage of AD (Kilgore et al., 2010).

Trichostatin A is found to induce brain-derived neurotrophic factor (BDNF) exon I–IX mRNA in Neuro-2a cells (Ishimaru et al., 2010). Another study of hippocampal neurons also suggested that TSA treatment could be useful at promoter 1 by acetylating histones (H3AcK9/K14) (Tian et al., 2010).

Nicotinamide was found to restore cognitive deficits. Nicotinamide can also increase the acetylation of alpha-tubulin and MAP2c, thereby increasing microtubule stability. These observations suggest a possibility that nicotinamide could be useful in treating AD (Green et al., 2008).

#### ncRNA Regulation-Based Therapy

fgene-09-00579 November 29, 2018 Time: 10:16 # 6

Non-coding RNAs are widely expressed in the brain and show a range of differences between AD and healthy controls. Numerous studies have focused on finding potential treatments using ncRNAs. Studies have found that overexpression of miR-124 and miR-195 could down-regulate Aβ levels by controlling BACE1 gene expression (Fang et al., 2012; Zhu et al., 2012). It is reported that DNA antisense oligonucleotides can hasten the degradation of UBE3A-ATS, which might be an avenue of treatment for AD (Meng et al., 2015). In Tg-19959 mice, knockdown of BACE1 or BACE1-AS transcripts causes reductions in BACE1 protein and insoluble Aβ (Modarresi et al., 2011). In an in vitro preparation, MiR-16 was found to be an effective inhibitor of amyloid precursor protein (APP) and BACE1 expression, Aβ peptide production, and tau phosphorylation (Parsi et al., 2015).

miRNA regulators have also been evaluated for use in treating AD. For example, a type of neutralizing inhibitor of miR-206 called AM206 was injected into the cerebral ventricles of an AD mouse model and found to improve memory function (Lee et al., 2012) as well as improving hippocampal neurogenesis and synaptic density. It was reported that RNA interference-mediated knockdown of the long–form phosphodiesterase–4D (PDE4D) enzyme could reverse memory deficits caused by accumulation of amyloid-beta (42) (Zhang et al., 2014). Another study found that miR-34c levels were increased in the hippocampus of AD patients, leading to a suggestion that it might be a biomarker for AD. In an AD mouse model, targeting miR-34 was found to lead to the recovery of learning ability (Zovoilis et al., 2011).

### DISCUSSION AND FUTURE PERSPECTIVES

Alzheimer's disease is a progressive neurodegenerative disease which is already a substantial burden on the world. The mechanisms of AD are complex and have been the subject of many theories. We have gained insight into the role of epigenetic modulation, which comprises DNA methylation, histone modification and the effects of ncRNAs. While many differences in these three kinds of epigenetic regulation have been observed between AD patients or models and controls, much work will still be needed to understand this topic and to find potential targets for the treatment of AD.

One of the main challenges for epigenetic therapy is to find molecules that can pass through the blood-brain barrier (BBB). In general, only molecules less than 500 daltons, with fewer than 8 pairs of hydrogen bonds, can move through the BBB. There is a variety of ways of working around this limitation, including BBB disruption, intracerebral implantation and intracerebroventricular infusion (Pardridge, 2005). As discussed above, DNMT inhibitors and HDAC inhibitors are potential therapeutic drugs, and these two kinds of molecules are able to cross the BBB (Hockly et al., 2003; Rai et al., 2008; Kwa et al., 2011). However, several additional factors need to be considered, including efficacy, toxicity, delivery and patient factors. Even though DNMT inhibitors are widely used in treating tumors and leukemia, they are not yet suitable for continuous treatment of AD. Because of their genotoxicity and low stability, they remain at the preclinical stage of development (Cuadrado-Tejedor et al., 2013; Erdmann et al., 2015). However, the clinical side effects of HDAC inhibitors are well known (Subramanian et al., 2010). As an example, the HDAC inhibitor SAHA can induce reactive oxygen species and DNA damage in acute myeloid leukemia cells (Petruccelli et al., 2011). Therefore, more detailed studies are needed to develop safe and effective treatments using DNMT inhibitors and HDAC inhibitors.

Other approaches to ncRNA-based therapy has been examined in recent years. A study showed that a short peptide derived from rabies virus glycoprotein (RVG) assists the transvascular delivery of small interfering RNA (siRNA) to the brain, leading to specific gene silencing inside the brain. Intravenous treatment with RVG-9R-bound antiviral siRNA afforded robust protection against fatal viral encephalitis in mice (Kumar et al., 2007). Other studies have shown that nanoparticle carriers may be a useful method of moving drugs through the BBB (Kreuter, 2001). All of these are methods that may be useful for implementing small interfering RNA therapy.

## CONCLUSION

In conclusion, we have discussed various epigenetic approaches to the treatment of AD, and new methods have emerged to aid in validating epigenetic targets. However, careful characterization of chemical probes is essential to ensure accurate biological conclusions. A new generation of selective chemical probes needs to be developed to reduce side effects. Studies on epigenetics may lead researchers to new methods of AD diagnosis and therapy. Even though the transition from bench to bedside may be challenging in the near future, current research strongly suggests that epigenetic regulation will become an important tool in AD treatment.

### AUTHOR CONTRIBUTIONS

XL contributed to the central idea and wrote the initial draft of the manuscript. BJ and LS contributed to refining the ideas, carrying out additional modifications, and finalizing the manuscript. All authors contributed to writing and revising the manuscript.

## FUNDING

This study was supported by the National Natural Science Foundation of China (No. 81701134 to BJ) and Youth Research Foundation of Xiangya Hospital, Central South University (No. 2016Q01 to BJ).

### REFERENCES

fgene-09-00579 November 29, 2018 Time: 10:16 # 7


lobar degeneration syndromes. Brain 140, 1792–1807. doi: 10.1093/brain/ awx101


fgene-09-00579 November 29, 2018 Time: 10:16 # 8



**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 Liu, Jiao and Shen. 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.

# miR-205-5p Mediated Downregulation of PTEN Contributes to Cisplatin Resistance in C13K Human Ovarian Cancer Cells

Xiaoyan Shi1,3† , Lan Xiao<sup>4</sup>† , Xiaolu Mao<sup>5</sup>† , Jinrong He1,3, Yu Ding1,3, Jin Huang1,3 , Caixia Peng1,3 \* and Zihui Xu1,2 \*

<sup>1</sup> Key Laboratory for Molecular Diagnosis of Hubei Province, The Central Hospital of Wuhan, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, <sup>2</sup> Department of Endocrinology & Metabolism, Renmin Hospital of Wuhan University, Wuhan, China, <sup>3</sup> Central Laboratory, The Central Hospital of Wuhan, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, <sup>4</sup> Department of Obstetrics and Gynecology, The First Affiliated Hospital, An Hui Medical University, Hefei, China, <sup>5</sup> Department of Clinical Laboratory, The Central Hospital of Wuhan, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China

#### Edited by:

Yujing Li, Emory University, United States

#### Reviewed by:

Feng Wang, Emory University School of Medicine, United States Sandeep Kumar, Emory University, United States John P. Hagan, University of Texas Health Science Center at Houston, United States

#### \*Correspondence:

Caixia Peng echo\_by\_me@126.com Zihui Xu tj017@163.com †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to RNA, a section of the journal Frontiers in Genetics

Received: 02 May 2018 Accepted: 31 October 2018 Published: 19 November 2018

#### Citation:

Shi X, Xiao L, Mao X, He J, Ding Y, Huang J, Peng C and Xu Z (2018) miR-205-5p Mediated Downregulation of PTEN Contributes to Cisplatin Resistance in C13K Human Ovarian Cancer Cells. Front. Genet. 9:555. doi: 10.3389/fgene.2018.00555 Cisplatin resistance is a major cause of treatment failure in advanced ovarian cancer. The limited evidence shows the paradoxical regulation of miR-205 on chemotherapy resistance in cancer. Herein, we found that miR-205-5p was enormously increased in cisplatin-resistant C13K ovarian cancer cells compared with its cisplatin-sensitive OV2008 parental cells using miRNA microarrays, which was further verified by quantitative PCR. Furthermore, we confirmed that inhibition of miR-205-5p upregulated PTEN and subsequently attenuated its downstream target p-AKT, which inversed C13K cells from cisplatin resistance to sensitivity. Our data suggest that miR-205- 5p contributes to cisplatin resistance in C13K ovarian cancer cells may via targeting PTEN/AKT pathway.

Keywords: miR-205-5p, PTEN, cisplatin resistance, ovarian cancer, AKT

### INTRODUCTION

Cisplatin is an anti-cancer chemotherapy drug which used to treat ovarian cancer in women who have already received surgery and/or radiation treatment. However, either resistance to cisplatin or multi-drug resistance (MDR) to cisplatin-centered chemotherapy is a major cause of treatment failure in human ovarian cancer. As we know, four well-characterized mechanisms have been suggested to account for cisplatin-resistance in cancer: (1) pre-target resistance – involving steps preceding the binding of cisplatin to DNA, (2) on-target resistance – directly related to DNAcisplatin adducts, (3) post-target resistance – concerning the lethal signaling pathways elicited by cisplatin-mediated DNA damage, and (4) off-target resistance – affecting molecular circuitries that do not present obvious links with cisplatin-elicited signals (Galluzzi et al., 2012, 2014). Besides lots of genes involved, emerging evidences demonstrate that miRNAs such as miR-141 (Seki, 2011), miR-184 (Tung et al., 2016), miR-199a (Wang et al., 2013), miR-214 (Yang et al., 2008), and miR-421 (Ge et al., 2016) contribute to cisplatin-resistance in cancer. miR-205, a frequently silenced microRNA in cancer, has recently been implicated in chemotherapy resistance. De Cola et al. (2015) showed that upregulation of miR-205 leaded to lapatinib resistance in breast cancer stem cells via targeting ErbB/HER receptors. On the contrary, however, results from

Puhr et al. (2012), Bhatnagar et al. (2010), and Alla et al. (2012) suggested that downregulation of miR-205 contributed to docetaxel or cisplatin resistance in prostate cancer, and cisplatin resistance in melanoma cells. Therefore, more extensive and detailed studies are required to explore the role of miR-205 in drug resistance associated with cancer chemotherapy.

In present study, we measured the miRNA expression profiles of cisplatin-resistant C13K ovarian cancer cell line compared with its cisplatin-sensitive OV2008 parent cell line using miRNA microarrays, and found that miR-205-5p was enormously increased in cisplatin-resistant C13K ovarian cancer cells, which was further verified by quantitative real-time PCR (qPCR). Furthermore, we investigated the effects of miR-205- 5p on cisplatin resistance in ovarian cancer cells and the underlying mechanism, and confirmed that inhibition of miR-205-5p upregulated PTEN and subsequently attenuated its downstream target p-AKT, which inversed C13K cells from cisplatin resistance to sensitivity. Our data suggest that miR-205- 5p contributes to cisplatin resistance in C13K ovarian cancer cells via targeting PTEN/AKT pathway.

### MATERIALS AND METHODS

### Cell Culture and Transfection

The cisplatin-sensitive human ovarian cancer cell line OV2008 and its cisplatin-resistant clone C13K were supplied by Dr Wencheng Ding (Cancer Biology Research Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology) (Asselin et al., 2001; Su et al., 2011). OV2008 cells were maintained in complete RPMI-1640 medium supplemented with 10% fetal bovine serum at 37◦C in a humidified atmosphere containing 5% CO2. C13K cells were cultured in RPMI1640 supplemented with 10% fetal bovine-serum medium containing 1 µmol/L cisplatin to maintain resistance. The cells were transfected with 100 pmol hsa-miR-205-5p mimics or hsa-miR-205-5p inhibitor using Lipofectamine 2000 transfection reagent (Invitrogen, Carlsbad, CA, United States). The sequences for hsa-miR-205-5p are as follow: hsa-miR-205-5p mimics (5<sup>0</sup> -UCCUUCA UUCCACCGGAGUCUG-3<sup>0</sup> , 5<sup>0</sup> - GACUCCGGUGGAAUGA AGGAUU-3<sup>0</sup> ); hsa-miR-205-5p mimics negative control (NC) (5<sup>0</sup> - UUCUCCGAACGUGUCACGUTT-3<sup>0</sup> , 5<sup>0</sup> -ACGUGACACGU UCGGAGAATT-3<sup>0</sup> ); hsa-miR-205-5p inhibitor (5<sup>0</sup> -CAGA CUCCGGUGGAAUGAAGGA-3<sup>0</sup> ); mircoRNA inhibitor NC (50 -CAGUACUUUUGUGUAGUACAA -3<sup>0</sup> ).

#### Cytotoxicity Analysis

Cell cytotoxicity was analyzed with Cell Counting Kit-8 (CCK-8) assay kit (Sigma-Aldrich, St. Louis, MO, United States). Briefly, cells were seeded into flat-bottomed 96-well plates at a density of 8000 per well and incubated overnight. Cisplatin (Sigma-Aldrich, St. Louis, MO, United States) with concentrations of 0, 20, 40, 60, 80, 100 and 120 µM was added after adherence. Then cells were continuously cultured for 24 or 48 h followed by treatment with 10 ul of CCK-8 solution for additional 1h at 37◦C, and the absorbance (A) was measured at 450 nm by an Enspire microplate reader (Perkin Elmer, United States). Cell viability (%) = experimental group A value/control group A value × 100. IC50 values (50% inhibition of surviving fraction) were then estimated using the fitted dose-response curves for cell viability.

### RNA Extraction and miRNA Microarray Analysis

Total RNA was extracted by Trizol reagent (Invitrogen, Carlsbad, CA, United States). The RNA sample was purified with an RNeasy Mini Column (Qiagen, CA, United States) and the RNA quality was assessed by 1.0% agarose gel electrophoresis. miRNA-profiling was carried out by miRNA microarray using Affymetrix miRNA 4.0 and Affymetrix GeneChip (Affymetrix, United States). The random-variance model (RVM) F test was applied to filter differentially expressed genes for the two ovarian cancer cell lines. After the significance analysis and false discovery rate (FDR) analysis, differentially expressed genes were selected according to their p-value threshold.

### qPCR

miRNAs expression was measured using TaqMan miRNA reverse transcription kit and TaqMan miRNA assay kits (Applied Biosystems, United States). Primer sequences for hsa-miR-205-5p: (5<sup>0</sup> -CCTTCATTCCACCGGAGT-3<sup>0</sup> ; 50 -GTCCAGTTTT TTTTTTTTTTTCAGACT-3<sup>0</sup> ). PTEN mRNA expression was measured using TaqMan mRNA reverse transcription kit and SYBR Green Supermix (BioRad, United States), primer sequences for PTEN: (5<sup>0</sup> -TGGATTCGACTTAGACTTGACCT; 5 0 -TTTGGCGGTGTCATA ATGTCTT).

#### Cell Apoptosis Assay

Cell apoptosis was measured by Flow Cytometry. All samples were washed in phosphate-buffered saline and resuspended in 200 µl binding buffer. Next, 5 µl Annexin–V-fluorescein isothiocyanate and 10 µl propidium iodide (PI; 1 µg/ml) were added and the cell suspension was incubated in a dark chamber for 1h at room temperature. Cell apoptosis was then determined using a FACSCalibur flow cytometer (BD Biosciences, United States) and data were analyzed using CellQuest software (BD Biosciences, United States).

### Western Blotting Analysis

Cells were harvested with ice-cold PBS and lysed in lysis buffer containing a protease inhibitor cocktail. A total of 60 µg protein was separated by 10% SDS-PAGE and transferred to polyvinylidene fluoride membranes. Following blocking with Tris-buffered saline containing 5% skimmed milk for 1 h at room temperature, the membranes were incubated with the primary antibodies (anti-PTEN, anti-actin, anti-AKT, antiphospho-AKT, Santa Cruz or Cell Signaling, United States) in blocking buffer overnight at 4◦C. The membranes were washed three times in TBST and then incubated with horseradish peroxidase-conjugated anti-mouse/rabbit antibodies at a dilution of 1:3,000 for 1 h at room temperature. Signals were detected on X-ray film using an enhanced chemiluminescent detection system (Pierce Biotechnology, Inc., Rockford, IL, United States).

### Statistical Analysis

fgene-09-00555 November 19, 2018 Time: 15:31 # 3

All experiments were repeated at least three times and the data were expressed as the mean ± standard deviation (SD). Statistical analysis was performed with SPSS 18.0 for Windows (SPSS, Inc., Chicago, IL, United States). Statistical significance was determined by the Student t-test or one-way ANOVA. P < 0.05 was considered to indicate a statistically significant difference.

### RESULTS

### Differentially Expressed miRNAs in Cisplatin-Resistant Variant C13K Cells Versus Its Cisplatin-Sensitive OV2008 Parental Cells

C13K cells are cisplatin-resistant variant originating from its cisplatin-sensitive OV2008 parental cells (Su et al., 2011). We used miRNA microarray to determine miRNA expression profiles in both C13K and OV2008 cell lines. We have identified 113 miRNAs that were significantly differentially expressed (GEO accession number: GSE120256) in C13K cells, including 32 upregulated miRNAs (such as hsa-miR-205-5p, hsa-miR-200c-3p, hsa-miR-100-5p, hsa-miR-155-5p, and hsamiR-125b-5p) and 81 down-regulated miRNAs (such as hsamiR-214-3p, hsa-miR-199a-3p, hsa-miR-199b-3p, hsa-miR-199a-5p), compared to OV2008 cells (**Figure 1A**). Surprisingly, miR-205-5p was around 9000 fold changes among 32 upregulated miRNAs, which was further verified by qPCR (**Figure 1B**).

To investigate the role of miR-205, target prediction using softwares such as miRanda, TargetScan and PicTar was performed (data in **Supplementary Data Sheets 1,2**). Several of the predicted targets are known to be involved in cancer progression (E2F1, E2F5, ERBB, PTEN), invasion and metastasis (ZEB1/2, LRP-1). mirTarBase analysis was performed to further identify the experimentally validated targets of miR-205. These indicated that some but not all of these genes are aberrantly expressed in C13K cisplatin-resistant cells.

### miR-205-5p Attenuated Cisplatin-Induced Cytotoxicity

We confirmed that the viability of C13K cells was significantly increased following treatment with cisplatin for 48 h compared to OV2008 cells, and the IC50 values of cisplatin in C13K cells (107 µmol/L) were higher than the corresponding values in OV2008 (37 µmol/L), which indicated that C13K cells exhibited cisplatin resistance (**Figures 2A–C**).

To investigate whether miR-205**-**5p is associated with cisplatin-induced cytotoxicity in ovarian cancer cells, we examined the cell viability of C13K and OV2008 cells transfected with miR-205**-**5p inhibitor or mimics following cisplatin treatment for 48 h. We observed that both cell viability and IC50 in C13K decreased upon downregulation of miR-205**-**5p (**Figures 2D–F**), whereas both of them in OV2008 elevated upon upregulation of miR-205**-**5p (**Figures 2G–I**). These data suggested that miR-205**-**5p attenuated cisplatin-induced cytotoxicity and enhanced cisplatin resistance in ovarian cancer cells.

### miR-205-5p Inhibits Cisplatin-Induced Apoptosis

To explore whether miR-205**-**5p accounts for cisplatin-induced apoptosis in ovarian cancer cells, we measured cell apoptosis in both C13K and OV2008 cells transfected with miR-205**-** 5p inhibitor or mimics following cisplatin treatment for 48 h. We found that C13K cells apoptosis increased upon downregulation of miR-205**-**5p (**Figures 3A,B**), whereas OV2008

assay. miR-205-5p expression was measured by qPCR in C13K cells (D–F) or OV2008 cells (G–I). The data represent mean ± SD of three independent experiments (∗P < 0.01).

cells apoptosis reduced upon upregulation of miR-205**-**5p (**Figures 3C,D**).

### miR-205-5p Contributes to Cisplatin-Resistance in Ovarian Cancer Cells via Targeting PTEN/AKT Pathway

Given the involvement of PTEN in chemotherapy resistance including cisplatin in cancer (Lee et al., 2005; Juric et al., 2015), it is important to reveal whether PTEN, one of miR-205 target genes, is linked to miR-205 upregulation in cisplatin-resistant ovarian cancer cells. We examined PTEN mRNA and protein expression by using qPCR and Western blotting. Our results showed remarkable lower expression of PTEN mRNA and protein in C13K cells compared to that in OV2008 (**Figures 4A,B**). We further measured the expression of PTEN and AKT in C13K and OV2008 cells transfected with miR-205**-**5p inhibitor

or mimics following cisplatin treatment for 48 h. We found that PTEN increased and phospho-AKT (p-AKT) decreased in C13K upon downregulation of miR-205**-**5p (**Figures 4C,D**). On the contrary, PTEN decreased and p-AKT increased in OV2008 upon upregulation of miR-205**-**5p (**Figures 4E,F**).

#### DISCUSSION

In the last few years, drug resistance in various cancers has been linked with aberrant expression of miRNAs, suggesting that miRNAs might play important roles in chemoresistance and chemosensitivity (Koster et al., 2013). Although resistance to current chemotherapeutics represents a significant barrier to the improvement of the long-term overcome of patients with ovarian cancer, chemotherapy is still one of the most effective therapies for ovarian cancer patients. Although altered miRNAs are differentially expressed in ovarian cancer, the potential role of miRNAs in the induction of drug resistance, particularly in platinum resistance, has not been fully investigated. Growing evidences demonstrate that more and more miRNAs contribute to cisplatin-resistance in cancer (Yang et al., 2008; Wang et al., 2013; Ge et al., 2016; Tung et al., 2016). In this study, we identified 113 miRNAs which were significantly differentially expressed, including 32 upregulated miRNAs and 81 downregulated miRNAs. Most of them (such as miR-205, miR-200c, miR-100, miR-155, miR-125b, miR-214, miR-199a, miR-199b and miR-199a) are involved in cell differentiation, proliferation and apoptosis, and cancer development, progression and metastasis.

In present report, we suggests that upregulation of miR-205-5p contributes to cisplatin resistance in C13K ovarian cancer cells. However, evidences showed the paradoxical regulation of miR-205 on chemotherapy resistance in various cancers. Specifically, increased miR-205 level and paclitaxel resistance were noted in CD133+ ovarian cancer stem cells compared to adherent OVCAR3 cells (Nam et al., 2012). Additionally, increased miR-205 level was correlated with high proliferation, invasion and migration rates, enhanced resistance

to carboplatin both in lung cancer A549 cells and H1975 cells by inhibiting cellular apoptosis and augmenting cancer cell survival (Zarogoulidis et al., 2015). However, downregulation of miR-205 was noted in pancreatic cancer cell lines resistant to gemcitabine (Bera et al., 2014) and prostate cancer cell lines resistant to docetaxel (Puhr et al., 2012). Overexpression of miR-205 resensitized drug-resistant cancer cells and acted as a tumor suppressor miRNA (Puhr et al., 2012; Mittal et al., 2014; Ippolito et al., 2016; Chaudhary et al., 2017). As we know, miR-205 is involved in both physiological and pathological processes including cell proliferation, apoptosis, angiogenesis and mesenchymal transition by targeting either oncogenes or tumor suppressor genes such as PHLPP-2, CTGF, CYR61, ESRRGr, PTEN, ErbB3, E2F1, E2F5, Zeb1&2, HBx, PCNA, Her 2&3, VEGF-A, Bcl2, etc (Vosgha et al., 2014), which results in "dual roles" of cell regulation. Nevertheless, more extensive and detailed studies are required to explore the dual role of miR-205 in drug resistance associated with cancer chemotherapy.

The tumor suppressor PTEN, one of miR-205 target genes, plays an essential role in cancer development via modulating cell cycle progression (Qu et al., 2012; Cai et al., 2013). It was found that PTEN deletion in malignant gliomas, endometrial cancer, melanomas, breast cancer and other tumors, indicating that the PTEN gene has an important function in suppressing transformation to a malignant tumor (Habib et al., 2011). PTEN levels were significantly decreased in progesterone-resistant endometrial cancer Ishikawa cells and that the miR-205 inhibitor increased PTEN expression in these cells (Zhuo and Yu, 2017). Besides, PTEN acts as a phosphatase to dephosphorylate phosphatidylinositol (3,4,5) trisphosphate (PIP3), which results in inhibition of AKT signaling pathway (Cantley and Neel, 1999). It has been implicated that PTEN is involved in chemotherapy resistance including cisplatin in breast cancer, prostate cancer, melanoma, and EMT (Lee et al., 2005; Juric et al., 2015). Therefore, we speculated that miR-205 confers cisplatin resistance upon ovarian cancer cells through targeting PTEN. In this study, we confirmed that inhibition of miR-205-5p upregulated PTEN expression and subsequently attenuated its downstream target p-AKT, which inversed C13K cells from cisplatin resistance to sensitivity. Contrarily, overexpression of miR-205- 5p downregulated PTEN expression and subsequently increased p-AKT level, which turned OV2008 cells from cisplatin sensitivity into resistance.

Our findings suggest that miR-205 plays an important role in the loss of PTEN expression and the development of cisplatinresistant in ovarian cancer cells. Nevertheless, this research has limitation. All the data was based on cisplatin-resistant C13K ovarian cancer cells and its cisplatin-sensitive OV2008 parental cells. Therefore, we should note that differential expression patterns of miRNAs in C13K and OV2008 cells in this study are not applicable for ovarian cancer in general. More ovarian cancer cell lines need be checked by modulating miR-205 level and drug resistance. However, it was found that miR-205 regulated the proliferation and invasion of ovarian cancer OVCAR-3 cell line via suppressing PTEN/SMAD4 expression (Chu et al., 2018). miR-205 level was increased in ovarian cancer and it promoted the invasive behavior of ovarian cancer cell lines (OVCAR-5, OVCAR-8, and SKOV-3) (Wei et al., 2017).These studies support our conclusion, although the roles of miR-205 in the induction of drug resistance has not been investigated.

#### CONCLUSION

Our data suggest that miR-205-5p mediates downregulation of PTEN and upregulation of p-AKT, which contributes to cisplatin-resistance in C13K human ovarian cancer cells. Biological or pharmacological intervention based on miR-205-5p may be a new promising strategy to inverse the cisplatin resistance in human ovarian cancer cell.

#### AUTHOR CONTRIBUTIONS

fgene-09-00555 November 19, 2018 Time: 15:31 # 7

ZX and CP conceived and designed the experiments and explained the data. XS, LX, and XM performed the experiment. ZX, CP, XS, LX, and XM analyzed the content of the data with the help of JH, YD, and JH. XS, LX, and XM wrote the manuscript with the help of ZX and CP.

#### FUNDING

This work was supported by grants from the National Natural Science Foundation of China (81301084 and 31400916),

#### REFERENCES


the Doctoral Program of Higher Education of China (20130142120057) and the Bureauof Science and Technology of Wuhan Municipality (20150601010 10058).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene. 2018.00555/full#supplementary-material

DATAT SHEET S2 | miR-205 targets prediction using softwares such as miRanda, TargetScan, PicTar and mirTarBase analysis.



**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 reviewers SK, FW, and handling Editor declared their shared affiliation at the time of review.

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

digital media

of impactful research

article's readership