# NEW ADVANCES IN RNA TARGETING

EDITED BY : Olga N. Ilinskaya, Hector A. Cabrera-Fuentes, Derek John Hausenloy and Marina A. Zenkova PUBLISHED IN : Frontiers in Pharmacology

#### 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 978-2-88963-775-1 DOI 10.3389/978-2-88963-775-1

#### 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

# NEW ADVANCES IN RNA TARGETING

Topic Editors:

Olga N. Ilinskaya, Kazan Federal University, Russia Hector A. Cabrera-Fuentes, University of Giessen, Germany Derek John Hausenloy, University College London, United Kingdom Marina A. Zenkova, Institute of Chemical Biology and Fundamental Medicine (RAS), Russia

Citation: Ilinskaya, O. N., Cabrera-Fuentes, H. A., Hausenloy, D. J., Zenkova, M. A., eds. (2020). New Advances in RNA Targeting. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-775-1

# Table of Contents

## *05 Editorial: New Advances in RNA Targeting*

Olga Ilinskaya, Derek J. Hausenloy, Hector A. Cabrera-Fuentes and Marina Zenkova

*08 Current Development of siRNA Bioconjugates: From Research to the Clinic*

Ivan V. Chernikov, Valentin V. Vlassov and Elena L. Chernolovskaya

*33 MicroRNA Post-transcriptional Regulation of the NLRP3 Inflammasome in Immunopathologies*

Gulcin Tezcan, Ekaterina V. Martynova, Zarema E. Gilazieva, Alan McIntyre, Albert A. Rizvanov and Svetlana F. Khaiboullina


Svetlana Miroshnichenko and Olga Patutina

*79 Transcriptome Profiling Reveals Pro-Inflammatory Cytokines and Matrix Metalloproteinase Activation in Zika Virus Infected Human Umbilical Vein Endothelial Cells*

Svetlana Khaiboullina, Timsy Uppal, Konstatin Kletenkov, Stephen Charles St. Jeor, Ekaterina Garanina, Albert Rizvanov and Subhash C. Verma

*96 Bioinformatic Study of Transcriptome Changes in the Mice Lumbar Spinal Cord After the 30-Day Spaceflight and Subsequent 7-Day Readaptation on Earth: New Insights Into Molecular Mechanisms of the Hypogravity Motor Syndrome*

Maksim Sergeevich Kuznetsov, Artur Nicolaevich Lisukov, Albert Anatolevich Rizvanov, Oksana Victorovna Tyapkina, Oleg Aleksandrovich Gusev, Pavel Nicolaevich Rezvyakov, Inessa Benedictovna Kozlovskaya, Elena Sergeevna Tomilovskaya, Evgeny Evgenievich Nikolskiy and Rustem Robertovich Islamov

*106 Novel Peptide Conjugates of Modified Oligonucleotides for Inhibition of Bacterial RNase P*

Darya Novopashina, Mariya Vorobyeva, Anton Nazarov, Anna Davydova, Nikolay Danilin, Lyudmila Koroleva, Andrey Matveev, Alevtina Bardasheva, Nina Tikunova, Maxim Kupryushkin, Dmitrii Pyshnyi, Sidney Altman and Alya Venyaminova

*114 Catalytic Knockdown of miR-21 by Artificial Ribonuclease: Biological Performance in Tumor Model*

Olga A. Patutina, Svetlana K. Miroshnichenko, Nadezhda L. Mironova, Aleksandra V. Sen'kova, Elena V. Bichenkova, David J. Clarke, Valentin V. Vlassov and Marina A. Zenkova

*127 RNases Disrupt the Adaptive Potential of Malignant Cells: Perspectives for Therapy*

Vladimir Alexandrovich Mitkevich, Irina Yu Petrushanko and Alexander Alexander Makarov

*135 Characterization of the* Puumala orthohantavirus *Strains in the Northwestern Region of the Republic of Tatarstan in Relation to the Clinical Manifestations in Hemorrhagic Fever With Renal Syndrome Patients*

Yuriy N. Davidyuk, Emmanuel Kabwe, Venera G. Shakirova, Ekaterina V. Martynova, Ruzilya K. Ismagilova, Ilsiyar M. Khaertynova, Svetlana F. Khaiboullina, Albert A. Rizvanov and Sergey P. Morzunov


Olga Chinak, Ekaterina Golubitskaya, Inna Pyshnaya, Grigory Stepanov, Evgenii Zhuravlev, Vladimir Richter and Olga Koval

*178 A New Antisense Phosphoryl Guanidine Oligo-2*′*-O-Methylribonucleotide Penetrates Into Intracellular Mycobacteria and Suppresses Target Gene Expression*

Yulia V. Skvortsova, Elena G. Salina, Ekaterina A. Burakova, Oksana S. Bychenko, Dmitry A. Stetsenko and Tatyana L. Azhikina

*187 Evolutionary Trends in RNA Base Selectivity Within the RNase A Superfamily*

Guillem Prats-Ejarque, Lu Lu, Vivian A. Salazar, Mohammed Moussaoui and Ester Boix

*204 Immunotherapy Based on Dendritic Cell-Targeted/-Derived Extracellular Vesicles—A Novel Strategy for Enhancement of the Anti-tumor Immune Response*

Oleg Markov, Anastasiya Oshchepkova and Nadezhda Mironova

*228 Are Small Nucleolar RNAs "CRISPRable"? A Report on Box C/D Small Nucleolar RNA Editing in Human Cells*

Julia A. Filippova, Anastasiya M. Matveeva, Evgenii S. Zhuravlev, Evgenia A. Balakhonova, Daria V. Prokhorova, Sergey J. Malanin, Raihan Shah Mahmud, Tatiana V. Grigoryeva, Ksenia S. Anufrieva, Dmitry V. Semenov, Valentin V. Vlassov and Grigory A. Stepanov

*244 Differential Expression of HERV-W in Peripheral Blood in Multiple Sclerosis and Healthy Patients in Two Different Ethnic Groups* Rachael Tarlinton, Belinda Wang, Elena Morandi, Bruno Gran, Timur Khaiboullin, Ekatarina Martynova, Albert Rizvanov and Svetlana Khaiboullina

# Editorial: New Advances in RNA Targeting

Olga Ilinskaya1\*, Derek J. Hausenloy 2,3,4,5,6\*, Hector A. Cabrera-Fuentes 3,7 and Marina Zenkova<sup>8</sup>

<sup>1</sup> Institute of Fundamental Medicine and Biology, Kazan (Volga Region) Federal University, Kazan, Russia, <sup>2</sup> Cardiovascular & Metabolic Disorders Program, Duke-National University of Singapore Medical School, Singapore, Singapore, <sup>3</sup> National Heart Centre, National Heart Research Institute Singapore, Singapore, Singapore, <sup>4</sup> Yong Loo Lin School of Medicine, National University Singapore, Singapore, Singapore, <sup>5</sup> The Hatter Cardiovascular Institute, University College London, London, United Kingdom, <sup>6</sup> Cardiovascular Research Center, College of Medical and Health Sciences, Asia University, Taichung, Taiwan, <sup>7</sup> Institute of Biochemistry, Justus-Liebig-University Giessen, Giessen, Germany, <sup>8</sup> Laboratory of Biochemistry of Nucleic Acids, Institute of Chemical Biology and Fundamental Medicine of Russian Academy of Science, Novosibirsk, Russia

Keywords: oligonucleotides, noncoding RNAs, RNases, bioconjugates, delivery

Editorial on the Research Topic

New Advances in RNA Targeting

#### Edited by:

Lei Xi, Virginia Commonwealth University, United States

#### Reviewed by:

Ting C. Zhao, Boston University, United States

#### \*Correspondence:

Olga Ilinskaya ilinskaya\_kfu@mail.ru Derek J. Hausenloy derek.hausenloy@duke-nus.edu.sg

#### Specialty section:

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

Received: 03 March 2020 Accepted: 25 March 2020 Published: 17 April 2020

#### Citation:

Ilinskaya O, Hausenloy DJ, Cabrera-Fuentes HA and Zenkova M (2020) Editorial: New Advances in RNA Targeting. Front. Pharmacol. 11:468. doi: 10.3389/fphar.2020.00468 Recent discoveries have implicated specific genes in the pathogenesis of a variety of cancers, neuronal and infectious diseases, and generating a new trend in drug development. As such, scientists have been searching for new ways to turn off the target gene(s) causing the disease without affecting the DNA sequence itself. Figuratively speaking, this therapeutic approach needs "bow and arrows" to hit the RNA target. The main concept of RNA-targeting therapy is the use of molecular "arrows" selectively hitting the structure and functional activity of specific RNAs to alter the expression of disease-relevant genes. A large number of investigational RNA-targeting drugs are currently under clinical development, although clinical translation has yet to be realized. This means that manipulating the functions and structures of such complex and dynamic targets as RNA requires a detailed analysis of existing approaches and the development of novel ones.

In this Research Topic Frontiers in Pharmacology, are several studies investigating RNA-targeted therapeutics involving chemically modified oligonucleotides [antisense oligonucleotides, aptamers, small interfering RNAs (siRNAs), microRNAs (miRNAs), and synthetic messenger RNAs (mRNAs)], a wide range of small molecules inhibiting interactions between protein and RNA, and RNases of different origin. New alternatives to antibiotics are now in great demand for the effective treatment of microbial infections. Novopashina et al. have developed novel oligonucleotides to inhibit RNase P, an essential bacterial enzyme, to inhibit bacterial growth, thereby providing a novel therapeutic for combating microbial infections. Patutina et al. have designed a novel platform for knockdown of miRNA targets, which is based on synthetic, sequencespecific ribonucleases. MiRNases (peptides capable of RNA cleavage) can be conjugated to miRNAtargeted oligodeoxyribonucleotides, rendering them resistant to nuclease-resistant within the conjugate design, without resorting to chemically-modified nucleotides. This miRNase-design platform can be easily adapted to therapeutically target pathogenic microRNAs that are overexpressed in many diseases.

There are several examples in which the analysis of transcriptomic datasets have provided insights into underlying disease mechanisms, disease epidemiology, and have the potential to discover new RNA gene targets for treating a wide variety of medical conditions. Tezcan et al.

**5**

provide a comprehensive overview on the role of microRNA post-transcriptional regulation of the NLRP3 inflammasome in different immune related conditions. This has been shown to be effective in animal models but whether this can be translated into the clinical setting remains to be determined. Khaiboullina et al. have examined transcriptomic datasets of human umbilical vein endothelial cells (HUVECs) following Zika infection and identified several potential cytokine mediators of endothelial permeability. A bioinformatic study by Kuznetsov et al. investigated the changes in the transcriptome which occurred in the mouse lumbar spinal cord after a 30-day space flight and 7-day re-adaptation period on Earth, and reported new insights into the mechanisms underlying the hypogravity motor syndrome. In an interesting article by Tarlinton et al., analysis of peripheral monocyte gene expression found a correlation between regional changes in expression of human endogenous retrovirus W (HERV-W) and the prevalence of multiple sclerosis between different ethnic populations in Britain and the republic of Tatarstan. In a similar approach, Davidyuk et al. have investigated the relationship between the presence of Puumala orthohantavirus strains in small animals captured in the Republic of Tatarstan and Finland, and correlated the findings with the clinical features of hemorrhagic fever with renal syndrome.

Although synthetic oligonucleotides can be designed to bind to target RNA and modify the latter's function, the broader potential of these compounds as therapeutics has remained untapped because their delivery to cells has been limited. In this regard, Skvortsova et al. have demonstrated that antisense oligonucleotide derivatives can be used to target gene expression and inhibit the growth of intracellular mycobacteria. In this study they showed that the new RNA analogue, phosphoryl guanidine oligo-2'-o-methylribonucleotide, could be efficiently taken up by intracellular microorganisms with strong antisense activity, thereby providing a new treatment strategy for tuberculosis, and potentially preventing the emergence of drugresistant strains of mycobacteria.

Most of the human genome encodes RNA that do not code for protein. Noncoding RNAs may modulate gene expression and onset and progression of disease, positioning them as new therapeutic targets for drug discovery. Miroshnichenko and Patutina, provide an overview of review one of the different approaches for regulating the function of short noncoding RNAs, particularly miRNAs. The latter are viable targets for anticancer therapeutic, given that miRNAs play a key role in modulating a large number of signaling pathways involved with cell proliferation, apoptosis, migration, and invasion. Anticancer therapy using antisense oligonucleotide constructs have been shown to control miRNA activity, and these include a variety of strategies such as small RNA zippers, miRNases, miRNA sponges, miRNA masks, anti-miRNA oligonucleotides, and synthetic miRNA mimics. Furthermore, small RNA zipper technology may be utilized to ablate function of endogenous siRNAs and Piwi-interacting RNAs (piRNAs).

In the last few years, CRISPR–Cas systems have been introduced as a powerful mode of RNA-editing strategy, that provides an important alternative to DNA editing which can cause so called "off-target effects"—unwanted mutations in other parts of the genome. Filippova et al. have shown that small nucleolar RNAs (snoRNA) in human cells can be gene edited using CRISPR/Cas9 cleavage.

Over many years, RNases have been investigated as potential antitumor agents given their selectivity and toxicity against certain transformed cells. However, the mechanisms underlying their selective cytotoxic effects remain unclear, and may include controlling RNA hydrolysis products, and selective suppression of specific genes. Elucidating the underlying mechanisms requires understanding of the transcriptome of RNAase treated cells. In this regard, exogenous RNases can modify the redox potential of key proteins (e.g., NF-kB, p53) by suppressing reactive oxygen species (ROS) production in tumor cells, thereby increasing the susceptibility of cancer cells to apoptotic cell death and attenuating uncontrolled division of cancer cells. In most situations, the cytotoxic efficacy of RNases is dependent on their ability to be taken up by the cancer cells. Mitkevich et al. provide an overview of the potential role of exogenous RNases in mediating the adaptive response of tumor cells which allow the latter to remain active despite changes to the micro-environment including acidic and hypoxic factors. Mironova and Vlassov describe a large number of tumorassociated intracellular RNAs and extracellular RNAs, which can be targeted by exogenous RNAases, as therapeutic strategies for treating a variety of different tumors.

Prats-Ejarque et al. have analyzed the RNase A superfamily using kinetic assays and molecular dynamics simulations to identify the structural motifs for nucleotide recognition in RNases which make up the host defense, thereby providing a strategy for structure-based drug discovery.

Several articles have addressed the problem of delivering RNA-targeting therapeutics into diseased cells. In order to find an effective "bow" to direct the therapeutic agent to the desired cellular target, novel approaches are needed. Conjugating therapeutics with antibodies that have the ability to recognize cell-specific surface receptors can be employed to target drugs to particular cancer cells, but this technology has a number of limitations. Nanoparticle-delivery of therapeutics has emerged as an alternative approach to deliver RNA-targeting drugs. In this regard, Chernikov et al. used bioconjugation, which is the covalent binding of siRNAs with biogenic molecules (such as lipophilic proteins, aptamers, antibodies, ligands, peptides, or polymers). Bioconjugates make very good nanoparticles as they do not require a positive charge to form complexes, are less recognized by components of the immune system, and are less cytotoxic because of their small size. Markov et al. have reviewed the role of exosomes as an alternative to synthetic nanoparticles. Extracellular vesicles may be used as natural vectors for delivery of RNA and other therapeutics targeted to tumor cells, Tlymphocytes, and dendritic cells. Therefore, extracellular vesicles have the therapeutic potential to be used as novel cellfree anti-tumor vaccines providing an alternative to dendritic cell-based vaccines. Chinak et al. have shown that cellpenetrating peptides may be used to transport cargo into cells. They were able to show that non-covalently associated nucleic acids could be delivered into cancer cells in vitro using recombinant protein lactaptin.

Khojaewa et al. have explored the potential of natural and synthetic zeolites to deliver the RNase, binase, as a potential antitumor drug. They used a simple approach based on immobilizing the antitumor RNase on natural minerals of the zeolite group. Bacterial RNase were shown to complex with clinoptilolite and this increased cytotoxicity, a therapeutic approach which can applied using zeolitezeolite-based complexes with RNA-targeting therapeutics for treatment of colorectal cancer, and when combined in a cream it can be used to treat malignant skin neoplasms.

Well-tolerated humans vaccines based on viral mRNAs with optimized sequences have the therapeutic potential to treat infectious diseases, effective protein translation, and stimulation of immune response can persist for several days. Furthermore, the safety of vaccines can be provided by cellular RNases that have the ability to target viral mRNA. It is also necessary to pay attention to the identified antiviral potential of bacterial RNases: exogenous RNase from Bacillus pumilus has been shown to inhibit the replication of Middle East respiratory syndrome-related coronavirus (MERS-CoV) and human coronavirus 229E (HCoV-229E) (Müller et al., 2017). This raises the possibility of using mRNA-based vaccines as well as bacterial RNases to combat against the current COVID-19 pandemic.

#### REFERENCE

Müller, C., Ulyanova, V., Ilinskaya, O., Pleschka, S., Shah Mahmud, R., and S1, (2017). A Novel Antiviral Strategy against MERS-CoV and HCoV-229E Using Binase to Target Viral Genome Replication. BioNanoScience 7 (2), 244–299. doi: 10.1007/s12668-016-0341-7

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.

In the near future, tasks to make the RNA-targeting molecules more potent and less immunogenic as well as to increase their delivery and prolonged action should be pursued.

### AUTHOR CONTRIBUTIONS

DH and OI wrote this article. HC-F and MZ have made a direct and intellectual contribution to the work. All authors have approved the article for publication.

### FUNDING

OI was supported by Russian Government Program of Competitive Growth of Kazan University and RFBR project No. 17-00-00060, MZ was supported by RFBR project No. 17- 00-00062 (K). DH was supported by the British Heart Foundation (CS/14/3/31002), Singapore MHNMR Council (NMRC/CSA-SI/0011/2017) and Collaborative Centre Grant scheme (NMRC/CGAug16C006), and the Singapore MEAR Fund Tier 2 (MOE2016-T2-2-021). This article is based upon work from COST Action EU-CARDIOPROTECTION CA16225 supported by COST.

Copyright © 2020 Ilinskaya, Hausenloy, Cabrera-Fuentes and Zenkova. 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.

# Current Development of siRNA Bioconjugates: From Research to the Clinic

#### Ivan V. Chernikov, Valentin V. Vlassov and Elena L. Chernolovskaya\*

Laboratory of Nucleic Acids Biochemistry, Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia

Small interfering RNAs (siRNAs) acting via RNA interference mechanisms are able to recognize a homologous mRNA sequence in the cell and induce its degradation. The main problems in the development of siRNA-based drugs for therapeutic use are the low efficiency of siRNA delivery to target cells and the degradation of siRNAs by nucleases in biological fluids. Various approaches have been proposed to solve the problem of siRNA delivery in vivo (e.g., viruses, cationic lipids, polymers, nanoparticles), but all have limitations for therapeutic use. One of the most promising approaches to solve the problem of siRNA delivery to target cells is bioconjugation; i.e., the covalent connection of siRNAs with biogenic molecules (lipophilic molecules, antibodies, aptamers, ligands, peptides, or polymers). Bioconjugates are "ideal nanoparticles" since they do not need a positive charge to form complexes, are less toxic, and are less effectively recognized by components of the immune system because of their small size. This review is focused on strategies and principles for constructing siRNA bioconjugates for in vivo use.

#### Edited by:

Hector A. Cabrera-Fuentes, University of Giessen, Germany

#### Reviewed by:

Luiza Iuliana Hernandez, Linköping University, Sweden Arun Samidurai, Virginia Commonwealth University, United States

> \*Correspondence: Elena L. Chernolovskaya elena\_ch@niboch.nsc.ru

#### Specialty section:

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

Received: 04 February 2019 Accepted: 08 April 2019 Published: 26 April 2019

#### Citation:

Chernikov IV, Vlassov VV and Chernolovskaya EL (2019) Current Development of siRNA Bioconjugates: From Research to the Clinic. Front. Pharmacol. 10:444. doi: 10.3389/fphar.2019.00444 Keywords: RNAi, siRNA, bioconjugate, chemical modifications, patterns of chemical modifications

### INTRODUCTION

Small interfering RNAs (siRNAs) are the most promising type of RNA-based therapeutic oligonucleotide drug, since their mechanism of action is catalytic and each siRNA molecule can inactivate several target RNA molecules in a sequence-specific manner. Since the discovery of RNA interference (RNAi) and the development of the first oligomeric RNAs that trigger RNAi in mammalian cells, significant progress has been made in the development of therapeutic siRNAs (Fire et al., 1998; Crooke et al., 2018). Chemical modifications of RNA have been developed that modulate their activity and stabilize them in biological fluids (Hoerter and Walter, 2007); some progress has been made in the development of methods for the delivery of siRNAs to cells (Hassler et al., 2018). Several siRNA-based drugs are undergoing clinical trials, and one drug patisiran (Onpattro) is approved for use in the clinic (Garber, 2018). However, the outstanding potential of siRNAs as therapeutic drugs has not yet been fully implemented. A number of unsolved problems remain: it is essential to develop an effective means of delivering siRNAs to certain types of cells; it is also necessary to create modified versions of siRNAs that are stable, effectively silence target RNAs, and do not cause side effects. These problems are a consequence of the properties of siRNAs, which are large, polyanionic molecules that are unstable in biological media and are capable of causing unwanted immune responses when they enter cells.

### RNAi

Induction of RNAi occurs when double-stranded RNA (dsRNA) enters the cell; e.g., when transfected with dsRNA, infected with RNA-containing viruses (De Paula et al., 2007), or when endogenously formed in cells as a result of transposon or non-coding RNA expression (Khvorova et al., 2003). Mechanism of RNAi is divided in to two phases: in the first stage (initiation phase), long dsRNA is cleaved by the endoribonuclease Dicer into siRNAs, short dsRNAs (21–23 bp) with two nucleotides protruding at the 3′ -ends. In the second stage (effector phase), the multiprotein RNA-induced silencing complex (RISC) is formed, which, after activation, performs recognition and sequence-specific cleavage of the target mRNA (**Figure 1**). It has been shown that RNAi in mammalian cells can be induced by chemically or enzymatically synthesized siRNAs that mimic dsRNA Dicer cleavage products (Nakanishi, 2016); in this case, the RNAi mechanism involves only the effector phase.

In the first stage of RISC assembly, the R2D2 protein (in Drosophila) or its analog (in other species), which contains two dsRNA binding domains and the Dicer binding domain, binds the siRNA. R2D2 recognizes and binds the thermodynamically more stable 5′ end of the duplex, which allows further binding of Dicer (Tomari et al., 2004), whose dsRNA-binding domain has specificity for 3′ -overhangs (Ma et al., 2004). Thus, the intermediate RISC loading complex (RLC) is formed. Following formation of the RLC, Dicer interacts with argonaute-2

(Ago2), presumably with the participation of the PAZ domains (Bernstein et al., 2001).

At the last stage of RISC assembly, Ago2 cuts and causes dissociation of one of the siRNA strands ("passenger strand"), resulting in the formation of an activated RISC<sup>∗</sup> . Ago2 and the remaining siRNA strand ("guide strand") are the main components of activated RISC (Aronin, 2006; Addepalli et al., 2010); however, a number of other proteins may also be part of this complex (Rana, 2007; Ohrt et al., 2008). The selection of the strand that is included in RISC<sup>∗</sup> is determined by the orientation of the Dicer-R2D2 heterodimer relative to the siRNA; since R2D2 interacts with the thermodynamically more stable end of the duplex, the most active siRNAs are those with a 5′ end of the sense strand more thermostable than the 5′ end of the antisense strand. Ago2 cuts both the siRNA passenger strand and the target mRNA (Liu et al., 2004); however, siRNA strand dissociation can be carried out without cleavage. Moreover, it is assumed that human Ago2 causes strand dissociation, mainly by a mechanism that does not require its cleavage (Muhonen et al., 2007; Park and Shin, 2015), therefore, the total melting point of the duplex can contribute to the efficiency of siRNA interfering activity. Recognition of the mRNA target by RISC<sup>∗</sup> occurs in several stages, wherein the "seed" region (the siRNA region from 2 to 8 nucleotides from the 5′ end of antisense strand) plays an important role. First, an initial screening of the sequence for three nucleotides (2–4 nucleotides from the 5′ end of the siRNA strand) occurs (Chandradoss et al., 2015). After the triplet is recognized, the fifth nucleotide from the 5′ end of the antisense strand interacts with the target mRNA, which contributes to conformational changes, opening nucleotides 6–8 and 13–16 for interaction (Schirle et al., 2014). The complementary interaction of the siRNA strand with the mRNA provides an advantageous conformation to cleave the mRNA between nucleotides 10– 11 relative to the 5′ end of the siRNA, which occurs via the PIWI domain of Ago2 (Jinek and Doudna, 2009). After cutting and dissociating from the complex, the target RNA and passenger strand of the siRNA are degraded by ribonucleases. Released RISC<sup>∗</sup> can participate in subsequent cycles of cleavage in a catalytic manner (Haley and Zamore, 2004; Aronin, 2006; Leuschner et al., 2006).

Due to the high affinity of RISC<sup>∗</sup> to single-stranded RNA, binding efficiency of RISC<sup>∗</sup> with the target mRNA is almost an order of magnitude greater than that of antisense oligodeoxynucleotides (ASOs) with the same sequence in which binding to the target mRNA occurs only through complementary interaction (Ameres et al., 2007). Thus, the concentration of siRNA at which an effective decrease in the expression of the target gene is observed is two to three orders of magnitude lower compared to antisense oligonucleotides (Lemaitre et al., 1987; Subramanian et al., 2015).

### BARRIERS FOR siRNA TO THEIR TARGETS

There are a number of biological barriers that impede the effective action of RNA in mammalian cells. First, since siRNAs are polyanions, they are unable to penetrate directly through the hydrophobic cell membrane and can enter the cell only by endocytosis or pinocytosis. However, in order to implement the silencing effect, endocytosed siRNA must penetrate the endosome membrane and exit into the cytoplasm, otherwise it will undergo cleavage by ribonucleases (Varkouhi et al., 2011), or will leave the cell via exocytosis (Shukla et al., 2016). When siRNA enters the cytoplasmic space, it can also be cleaved by cytoplasmic ribonucleases (Whitehead et al., 2009) or its concentration can decrease due to the division of target cells.

Despite the high specificity of the action, some siRNA can cause a number of non-targeted effects that prevent its use in high concentrations due to the toxicity they cause. The most significant non-targeted effect of siRNA is unwanted activation of the system of innate immunity under the action of certain motifs in the siRNA sequence. When interacting with the membrane surface or in the endosome, immunostimulating motifs can be recognized by Toll-like receptors (TLR3/7/8) (Oosenbrug et al., 2017; Pirher et al., 2017), inducing the production of interferons (α or β) and inflammatory cytokines that activate immune response and induce global changes in gene expression pattern (Mansoori et al., 2016). Other non-target effect of siRNAs is the displacement of endogenous micro-RNA from RISC, which can disrupt the natural regulation pathways in the cell. Also, the sense strand can be introduced in RISC<sup>∗</sup> and suppress the expression of non-target genes, similar effects can be caused by the antisense strand of siRNA which bind to partially homologous non-target mRNA. In last case the block of translation does not include the cleavage of mRNA (Huntzinger and Izaurralde, 2011).

In the transition to the level of the organism, there are new factors that reduce the effectiveness of the siRNA, such as: siRNA filtration by the kidneys, siRNA capture by immune cells, cleavage by serum ribonucleases, the endothelial barrier (Kanasty et al., 2012, p. 511). Due to the presence of these barriers, siRNA have reduced bioavailability and unfavorable pharmacokinetics in vivo, which necessitates the use of high doses of siRNA and makes it not always possible to achieve the desired effect. This review examines approaches to solving the above problems based on chemical modifications of siRNAs; namely, introducing unnatural nucleotides into the siRNA structure and attaching molecules to siRNAs, ensuring the interaction of conjugates with biological structures that increase the efficiency and specificity of siRNAs as potential drugs.

### CHEMICAL MODIFICATIONS OF siRNA

Chemical modifications may affect the properties of siRNA: sensitivity to ribonucleases, recognition by the RNAi system, hydrophobicity, toxicity, duplex melting temperature, and conformation of the RNA helix (Manoharan, 2004; Behlke, 2008; Chen et al., 2018). Modifications can be divided into modifications of ribose, phosphates, and nucleobases (**Table 1**; **Figure 2**). Each type of modification is reviewed separately below.

#### Ribose Modifications

Among all ribose modifications, substitutions in the 2′ -position most effectively protect siRNAs against the action of serum nucleases, as the 2′OH group participates in the cleavage of TABLE 1 | The effect of chemical modifications on siRNA properties.





RNA by endoribonucleases (Findlay et al., 1962; Breslow and Chapman, 1996). In this case, the size of the substituent at the 2′ -position of ribose affects the properties of the modified base. When the hydrogen of the 2′OH group is replaced with a relatively small methyl residue (2′ -O-methyl modification [2′O-Me]) (Bobst et al., 1969), there is stabilization of the 3′ -endo ribose conformation, which provides the A-type RNA helix essential for RNAi. The introduction of 2′O-Me modifications into siRNA promotes its protection against nucleases both in vitro (Volkov et al., 2009) and in vivo (Liu et al., 2014; Chernikov et al., 2017). Moreover, the introduction of these modifications reduces the immune response (Judge et al., 2006). These properties make 2′O-Me modification of siRNA one of the most attractive strategies for introducing siRNA-based drugs into the clinic (Khvorova, 2017; Ray et al., 2017). However, replacement of each nucleotide in the siRNA with a 2′O-Me modified one leads to inhibition of RNAi (Czauderna et al., 2003); even the replacement of 50% of the nucleotides in siRNAs with 2 ′O-Me could inhibit this process. The presence of a hydrophobic residue at the 2′ position alters the overall structure of siRNA and increases the thermal stability of the duplex, which interferes with its effective incorporation into RISC and dissociation of the passenger strand.

Since the size of the modification may contribute to an increase in the nuclease resistance of siRNA (Cummins et al., 1995), attempts have been made to introduce more voluminous substituents into the 2′ position of ribose (2′ -O-methoxyethyl [2′O-MOE] Prakash et al., 2005; Zanardi et al., 2018, 2′Oallyl Amarzguioui et al., 2003, 2′O-benzyl Kenski et al., 2012, and other modifications); however, these substitutions more significantly inhibited RNAi than 2′O-Me.

Among the large number of bulky 2′ substituents, the 2′O-MOE ribose is one of the few modifications that stabilizes the 3′ endo ribose conformation and increases the melting temperature of the duplex more effectively than 2′O-Me (Dorn et al., 2004). The introduction of 2′O-MOE into siRNA without inhibiting RNAi is possible only along the flanks of the duplex and the central part of the antisense strand (Prakash et al., 2005). Availability of 2′O-MOE in the central part of the antisense strand (9 or 10 nucleotides) was recently shown to increase the biological activity of siRNA by reducing the probability of inclusion of the sense strand in RISC (Song et al., 2017), although direct confirmation of the mechanism of modified strands selection was not provided. Apparently, the presence of a bulk modification in this position may sterically affect the interaction with a Dicer-R2D2 heterodimer or the formation of RISC (Koller et al., 2006).

Unlike the substitution of hydrogen in the 2′OH group, the replacement of oxygen by 2′ -fluorine (2′F) is more consistent with the original structure of RNA, effectively stabilizing the 3′ endo ribose conformation (Manoharan et al., 2011). Introduction of 2′F into all nucleotides of the duplex only slightly reduces the efficiency of RNAi (Blidner et al., 2007; Deleavey et al., 2010). This modification protects siRNAs from the action of nucleases in vitro and in vivo (Manoharan et al., 2011; Fucini et al., 2012); however, the introduction of modifications in 50% of the siRNA nucleotides could lead to toxicity (Shen et al., 2015; Janas et al., 2017). In 2016, the third stage of clinical trials of the Alnylam conjugate of N-acetylgalactosamine and siRNA containing 50% 2′F revealed its cardiotoxicity (Garber, 2016). Since these effects were not detected in previous stages of clinical trials (Zimmermann et al., 2017) and the toxicity of 2 ′F modified siRNA for the heart was not previously shown, these results may be occasional and not related to the properties of the conjugate. On the other hand, it was shown that under the action of ASO containing 50% 2′F, the expression profiles of a number of genes in vitro (Shen et al., 2015) and in vivo (Shen et al., 2018) were altered. In confirmation of the toxicity hypothesis of siRNAs containing 2′F, another study showed that the introduction of 2′F at the ends of the duplex alters the localization of siRNA from cytoplasmic to nuclear (Ohrt and Schwille, 2008). These results suggested that only a limited amount (no more than 25%) of 2′F modifications (Ray et al., 2017) could be introduced for therapeutic applications. However, a decrease in the proportion of 2′F analogs from 50 to 25% did not lead to a decrease in hepatotoxicity in rats and mice following intravenous administration of high doses (100–200 mg/kg) of siRNA and N-acetylgalactosamine conjugate (Janas et al., 2018).

Although other positions in ribose, such as 4′ carbon, can be modified [4′ S Gore et al., 2012, 4′C-aminomethyl-2′ -O-methyl Takahashi et al., 2012, and 4′C-O-methyl-2′ -O-methyl Harp et al., 2018 (**Table 1**)] and such modifications protect siRNAs from nucleases in vitro efficiently, these modifications are not widely used in biomedical research because they significantly inhibit RNAi (Deleavey and Damha, 2012).

Ribose modifications are not limited to substitutions in structure; nucleic acid analogs with a modified structure of the furanose cycle, such as derivatives containing 6-membered (hexitol [HNA] Fisher et al., 2009, cyclohexenic [CeNA] Nauwelaerts et al., 2007, and altritol [ANA] Fisher et al., 2007 nucleic acids) and 7-membered rings (oxepanic nucleic acid [ONA] Sabatino and Damha, 2007), bicyclic (locked nucleic acids [LNA] Braasch et al., 2003, 2′ -deoxymethanocarbanucleosides [MCs] Terrazas et al., 2011), tricyclic (tricyclo-DNA [tc-DNA] Goyenvalle et al., 2015), and acyclic (unlocked nucleic acid [UNA] Jensen et al., 2008; Langkjaer et al., 2009) derivatives, can protect siRNAs from the action of nucleases and in some cases (CeNA, LNA, and UNA) do not inhibit RNAi (Herdewijn and Juliano, 2007; Deleavey and Damha, 2012). Among the 6-membered nucleic acid derivatives, CeNA is most suitable for modifying siRNA, since its complementary interaction with RNA stabilizes the duplex, increasing the melting point by 1.5◦C per modified base and increases the oligoribonucleotide resistance to degradation in serum (Wang et al., 2001). Bicyclic derivatives (LNA) can even more significantly increase the melting temperature of siRNA. In the case of LNA, affinity for the complementary strand is increased by 2–8◦C per nucleotide due to the extra cycle between 2′ and 4′ carbon, which fixes the 3′ endo ribose conformation (Julien et al., 2008). However, the introduction of this modification into siRNA strongly affects its interfering activity and the antisense strand is especially sensitive to this modification; even one LNA modification of its first nucleotide from the 5′ end completely inhibits RNAi (Elmen et al., 2005). Conversely, conformationally more flexible acyclic derivatives, such as UNA or glycolic nucleic acid (GNA), can destabilize the duplex, reducing the melting point by 5–8 and 5–18◦C per nucleotide, respectively (Laursen et al., 2010; Schlegel et al., 2017).

Since thermal asymmetry of the duplex makes a primary contribution to "guide" strand selection, modifications stabilizing the duplex formed by the 3′ end of the antisense strand and 5′ end of the sense strand and, conversely, modifications destabilizing the duplex formed by the 3′ end of the sense strand and 5′ end of the antisense strand can increase the efficiency of RNAi by providing favorable duplex thermal asymmetry. Thus, the introduction of LNA, UNA, or GNA at different ends of the duplex can lead to an increase in siRNA efficiency by increasing the probability of incorporation of the antisense strand into RISC (Vaish et al., 2011). Moreover, due the change in thermal asymmetry of siRNA, the probability of incorporation of the sense strand in RISC and the following suppression of expression of non-target genes those have regions complementary to the sense strand of the siRNA decreases.

The antisense strand of siRNA can also block the translation of non-target mRNA by complementary interaction with the "seed" region. It has been shown that a decrease in the melting temperature of this region leads to a decrease in the efficiency of suppression of non-target genes (Jackson et al., 2006). Therefore, the introduction of UNA or GNA modifications in the "seed" region of the antisense strand of siRNAs contributes to an increase in the specificity of action (Bramsen et al., 2009; Janas et al., 2017).

An interesting strategy to increase the specificity of the siRNA is introducing nicks in the middle of the sense strand of the siRNA (small internally segmented interfering RNA [sisiRNA]) (Bramsen et al., 2007). sisiRNAs have a greater specificity of biological action because RISC containing sense strand is not formed. sisiRNAs are less stable compared to siRNAs of the same sequence; thus, LNA modifications are introduced to stabilize sisiRNAs. However, it has been shown that such a duplex design did not contribute to a significant increase in the biological activity of siRNA in vitro (Hong and Nam, 2016) or in vivo (Mook et al., 2010).

### Phosphate Modifications

Ribose modifications do not interfere with changes in the phosphate structure, and since that modifications are directly involved in nuclease cleavage (Frazao et al., 2006), it is reasonable that such modifications could effectively protect siRNA from degradation. Replacement of the oxygen of phosphate with sulfur (phosphorothioate [PS] Eckstein, 1970, 2014) or boron (boranophosphate [BP] Hall et al., 2006) has been shown to protect siRNA from the action of ribonucleases in vitro and in vivo (Soutschek et al., 2004). The introduction of PS modifications to both strands of the duplex inhibits RNAi to some extent (Schwarz et al., 2004). However, the introduction of PS modifications into oligonucleotides facilitates their penetration into cells in the absence of transfection agents due to non-specific binding to cell receptors and penetration by clathrin-dependent endocytosis (Wang et al., 2018). On the other hand, due to the non-specific interaction of PS oligonucleotides with serum proteins and cell receptors (Lee et al., 1999), activation of the complement system and leukocyte infiltration of the corresponding organs (Iannitti et al., 2014) may occur. Therefore, for clinical use of siRNA, the number of PS modifications should be reduced to 5–50%, depending on the intended dose of siRNA.

Unlike PS, the introduction of BP into the central part of the antisense strand strongly inhibits the action of RNAi; however, a partially modified pattern (25–75%) may increase the efficiency of RNAi and resistance to ribonuclease (Hall et al., 2004, 2006). At the same time, according to the work of Hall et al. (2004), where unpublished comparison data of PS and BP modifications of siRNA is mentioned, BP was shown to exhibit approximately twice as effective protection of siRNA from nucleases. If so, this modification could address some of the issues regarding the biomedical use of siRNA. One of the main limitations of the use of BP for siRNA modification is the lack of an optimized method of synthesizing large quantities of BP-modified siRNA; therefore, novel methods must be developed to assess the therapeutic potential of BP-modified siRNAs in vivo.

The introduction of modifications replacing the phosphodiester bond with an amide bond (Selvam et al., 2011) contributes to the protection of siRNA from the action of nucleases (Iwase et al., 2007), but their effect on the efficiency of RNAi is uncertain. The introduction of an amide bond between 10, 11, and 12 nucleotides, despite the absence of a phosphodiester bond, has been shown to increase the inhibitory effect of the modified siRNA, presumably due to the formation of additional hydrogen bonds between the amide group and Ago2 (Mutisya et al., 2017).

The presence of phosphate at the 5′ end of the "guide" strand of siRNA is essential for RNAi (Frank et al., 2010), while siRNA with 5′ -hydroxyl possesses biological activity, since such siRNA is effectively phosphorylated inside cells (Weitzer and Martinez, 2007). When blocking phosphorylation of the 5 ′ -hydroxyl, siRNA does not exhibit interfering activity (Chen et al., 2008). Chemical modifications of the first nucleotide from the 5′ end of the antisense strand can interfere with intracellular phosphorylation (Allerson et al., 2005; Chen et al., 2008; Kenski et al., 2012); however, the introduction of chemically stable phosphate [5′ -(S)-C-methyl (Prakash et al., 2015), 5 ′ -methylphosphonate (Lima et al., 2012), and 5′ (E) vinylphosphonate (Elkayam et al., 2017)] at the 5′ end of the antisense strand can restore activity (Lima et al., 2012; Prakash et al., 2015). The introduction of 5′ (E)-vinylphosphonate solves this problem most efficiently. It has been shown that such modification of the antisense strand not only improves binding to Ago2 (Elkayam et al., 2017) 2-fold, but also leads to an increase in the accumulation and stability of siRNA conjugates containing cholesterol (Haraszti et al., 2017) or N-acetylgalactosamine (Elkayam et al., 2017) in the organs of mice following subcutaneous injection. Therefore, 5′ (E) vinylphosphonate modification of the antisense strand is a promising strategy for the development of therapeutic drugs based on siRNA.

### Nucleobase Modifications

Substitutions of nucleobases with various modified analogs [pseudouridine, 2′ thiouridine, dihydrouridine (Sipa et al., 2007), 2,4-difluorobenzene (Somoza et al., 2006), 4 methylbenzimidazole (Somoza et al., 2008), hypoxanthine (Addepalli et al., 2010), 7-deazaguanin (Eberle et al., 2008), N2 -alkyl-8-oxoguanine (Kannan and Burrows, 2011), N<sup>2</sup> -benzylguanine (Puthenveetil et al., 2006), and 2,6-diaminopurine (Chiu and Rana, 2003)] are designed to increase the thermal stability of the duplex by increasing the efficiency of the formation of hydrogen bonds with complementary nucleotides. However, such modifications reduce the efficiency of RNAi and do not contribute to an increase in siRNA resistance to nuclease action (Peacock et al., 2011). Nucleobase modifications in small amounts (up to 10%) could reduce immune reactions and improve the thermodynamic siRNA profile (Sipa et al., 2007; Anderson et al., 2010). The presence of such modifications in the patterns of chemical modification of siRNA can contribute to the optimization of the therapeutic properties of siRNA; however, this approach has not yet found wide application as similar effects can be obtained by introducing other modifications.

### PATTERNS OF CHEMICAL MODIFICATIONS OF siRNAs

siRNA is degraded in vivo as a result of its cleavage by endonucleases on pyrimidines (Turner et al., 2007) and exonucleases from both the 3′ and 5′ ends (Hsu and Stevens, 1993; Terrazas et al., 2013); thus, it is essential that siRNAs contain chemical modifications at cleavage sites to improve siRNA nuclease resistance to achieve biological activity of bioconjugates in vivo. However, the introduction of certain chemical modifications in siRNA is limited by inhibition of its interfering activity and toxicity. Thus, the introduction of modifications in siRNA is determined by the balance between the number of modifications sufficient for siRNA to be nontoxic, while retaining interfering activity and nuclease resistance. Introducing the 2′O-Me modification into siRNA can lead to inhibition of RNAi if the siRNA contains more than two consecutive 2′O-Me modifications in a row (Czauderna et al., 2003; Manoharan et al., 2011), while 2′O-Me modification of every second nucleotide does not block RNAi (Czauderna et al., 2003). Thus, an important parameter affecting RNAi is not only the number of introduced modifications, but also their location in the duplex.

One of the approaches to finding a balance between interfering activity and nuclease resistance of siRNA with 2′O-Me modifications is 2′O-Me selective modification of nucleasesensitive siRNA sites (Volkov et al., 2009). siRNA is subjected to serum cleavage at pyrimidines (Turner et al., 2007); however, replacing all pyrimidines with 2′O-Me analogs completely inhibits the interfering ability of siRNA (Manoharan et al., 2011). It has been shown that the main sites of siRNA cleavage in serum are generally CA, UA, and UG sites. Introduction of 2′O-Me at these sites will preserve the interfering activity of siRNA, increase serum nuclease resistance (Volkov et al., 2009), and provide longterm suppression of target gene expression (Petrova Kruglova et al., 2010; Chernikov et al., 2017).

The main limitation of the introduction of 2′F modifications in siRNA is their toxicity, although siRNA conjugates containing 2 ′F modifications on pyrimidines are protected from the action of ribonucleases and, unlike 2′O-Me (Manoharan et al., 2011), possess biological activity (Wolfrum et al., 2007). The introduction of PS into siRNAs is also limited by the toxicity of the resulting oligonucleotides, and since PS-modified siRNAs have been shown to be highly protected against exoribonucleases (Eckstein, 2014; Kel'in et al., 2016), PS modification is used only to replace two or three terminal nucleotides in siRNA (Soutschek et al., 2004). Most of the other chemical siRNA modifications are primarily introduced along the terminal regions of the duplex for various reasons; e.g., the effect of RNAi on proteins, thermal asymmetry, and protection against nucleases (Deleavey and Damha, 2012). It is important that modifications designed to change the properties of siRNAs that are significant for its therapeutic potential can be used together, complementing one another and ensuring biological activity and siRNA resistance to nucleases more efficiently than each modification separately (Deleavey et al., 2010). Recent studies have used combinations of chemical modifications to achieve maximum effect in vivo.

Since after replacing each second nucleotide with the 2′O-Me analog, half of the siRNA molecule is unprotected from the action of nucleases, the introduction of 2′F modifications into the unmodified portion of the duplex was proposed. It has been shown that siRNA molecules with alternating 2′O-Me and 2′F modifications are stable in mouse plasma and suppress expression of the target gene by several orders of magnitude more efficiently compared to unmodified siRNA (Allerson et al., 2005). Successful use of fully modified siRNA molecules based on alternating 2′O-Me and 2′F modifications was demonstrated when studying the properties of singlestranded siRNAs (ssRNAs). ssRNAs with this pattern, with the addition of several PS, 2′O-MOE, and 5′ (E)-vinylphosphonate modifications, exhibited biological activity in vitro and in vivo (Lima et al., 2012; Prakash et al., 2015), and modeling of the complex of this siRNA with Ago2 showed that the modifications did not sterically block the interaction of the ssRNA with Ago2 (Schirle et al., 2016).

Despite the fact that duplex cleavage in serum at internal nuclease sensitive sites makes the greatest contribution to siRNA degradation, siRNA cleavage can also occur at other sites. A fully modified siRNA, containing alternating 2′O-Me and 2′F modifications, was compared with a partially modified siRNA, containing a combination of 2′O-Me and 2′F modifications located along the pyrimidines and the ends of the duplex; both patterns contained PS modifications on the 3′ -overhangs (Hassler et al., 2018) (**Figure 3**). A cholesterol conjugate of the fully modified siRNA more effectively reduced expression of the Htt gene in HeLa cells compared to a cholesterol conjugate of the partially modified siRNA; the concentrations of siRNA at which expression of the target gene was 50% suppressed were 70 and 170 nmol/l, respectively (Hassler et al., 2018).

In contrast to serum siRNA cleavage, siRNA degradation in lysosomal hepatocyte extracts occurs mainly by 5′ -exonucleases; therefore, siRNA could be further stabilized by PS modifications of the 5′ ends in the N-acetylgalactosamine conjugate to increase the duration and efficiency of its inhibitory effect (Nair et al., 2017). Subcutaneous administration of 10 mg/kg of conjugate of fully modified at 2′ ribose positions siRNA and N-acetylgalactosamine was shown to cause a 30% decrease in target gene expression at the protein level for 10 days, while addition of the conjugate with modified 5′ ends reduced the protein level by 80% for 40 days. Conjugates of fully modified at the 2′ positions and stabilized by PS modifications at both the 3 ′ and 5′ ends siRNAs with cholesterol and docosahexaenoic acid were examined in vivo (Hassler et al., 2018). The accumulation of conjugates in the liver, kidney, and spleen 24 h after intravenous and subcutaneous injections was studied. Accumulation of all mentioned above conjugates was two orders of magnitude higher compared to similar conjugates of partially modified siRNA (2′O-Me and 2′F modifications located on pyrimidines and

duplex ends and PS modifications on 3′ -overhangs). However, no control experiment was performed to determine the contribution of the introduction of PS at the 5′ ends to the stability of siRNA conjugates in this study. Since the contribution of 2′O-Me modifications to the nuclease resistance of siRNA is greater than that of 2′F modifications (Cummins et al., 1995; Takahashi et al., 2009), attempts were made to increase the proportion of 2′O-Me modifications in siRNAs containing 50% 2′O-Me and 50% 2 ′F (Khvorova, 2017; Foster et al., 2018). For this purpose, 1,890 different siRNAs were synthesized, aimed at five different genes, varying in sequence (15 variants) and pattern of 2′O-Me and 2′F modifications (Foster et al., 2018). The optimal introduction of 2 ′O-Me or 2 ′F modifications for each position in the siRNA was determined via in vitro analysis of primary mouse hepatocytes. Two modification patterns were selected, containing 23% and 18% 2′F modifications, the biological activities of which were not less than that of the parent siRNA. Analysis of the biological activities of three different siRNA sequences in vivo showed that a lower content of 2′F modifications (18%) was the most effective. Subcutaneous administration of 1 mg/kg of the siRNA and Nacetylgalactosamine conjugate to primates showed that the newly selected siRNA (18% 2′F modifications) reduced the protein level by ∼70% for 70 days, while the parent siRNA suppressed the expression of the target gene AT by ∼40% for 40 days. It has been shown that a conjugate of fully modified siRNA (23% 2′F, 73% 2 ′O-Me, and one dNMP) and N-acetylgalactosamine suppressed PCSK9 gene expression in the liver of patients following a single subcutaneous injection of ∼6 mg/kg by ∼70% for 6 months (Ray et al., 2017).

When siRNAs are delivered as part of a bioconjugate, they are especially sensitive to the action of nucleases, and the duration of biological action in vivo, and the dose and frequency of drug administration depends on nuclease resistance. Therefore, it is important to pay particular attention to this parameter when creating therapeutic drugs based on siRNA bioconjugates.

### BIOCONJUGATES

The use of bioconjugation as a method of delivering siRNA to cells involves forming siRNA conjugates with (1) biomolecules capable of specifically binding receptors on the cell membrane [folate Thomas et al., 2009, antibodies Song et al., 2005; Dassie et al., 2009; Xia et al., 2009, aptamers Aronin, 2006; McNamara et al., 2006, some peptides Cesarone et al., 2007;

Lau et al., 2012, and carbohydrates Nair et al., 2014], (2) molecules able to penetrate the cell by natural transport mechanisms [cholesterol (Lorenz et al., 2004) and vitamins Nishina et al., 2008], or (3) molecules capable of interacting non-specifically with the cell membrane [positive electrostatic charge and hydrophobicity Kwiatkowska et al., 2013; Meade et al., 2014] (**Supplementary Table 1**; **Figure 4**). In addition to the nature of the biogenic molecule, the structure of the linker that binds the siRNA and the biomolecule affects the efficiency of accumulation and the biological activity of the siRNA. In particular, the ability of the linker to be cleaved when the conjugate enters the cells prevents a decrease in the efficiency of RNAi associated with the inhibition of RISC assembly. Disulfide (Turner et al., 2005) and thioether bonds (Meade et al., 2014), pH sensitive bonds [hydrazone (Dovydenko et al., 2016), carboxymethylmaleic anhydride Rozema et al., 2007], or photosensitive bonds (β-[bis (4-methoxyphenyl) phenylmethoxy]-2-nitrobenzeneethanol linker Yang et al., 2018) are used as cleavable bonds. Conjugates containing linkers that are stable under the experimental conditions (Lorenz et al., 2004) are widely used, and the structure of the conjugate plays a key role. The most commonly used types of bioconjugates are reviewed below.

### Lipophilic siRNA Derivatives

Lipids and cholesterol were suggested as the first ligands for conjugation with siRNAs, since they were supposed to ensure the interaction of siRNAs with the cell membrane due to their lipophilic properties and because of endogenous transport mechanisms (Letsinger et al., 1989). Cholesterol is not only part of the membrane, but is also transported into cells by low-density lipoproteins (LDL particles) and high-density lipoproteins (HDL particles) (Brunzell et al., 2008), which bind to corresponding receptors. Absorption of all lipoproteins by cells is carried out through the recognition of protein components of the particles by clathrin-dependent receptor-mediated endocytosis, using LDL and scavenger receptor class B member 1 (SR-BI) receptors that recognize LDL and HDL particles, respectively (Goldstein et al., 1985; Yvan-Charvet et al., 2008).

It has been shown that siRNA and cholesterol derivatives, as well as other lipophilic siRNA derivatives, are able to form complexes with HDL and LDL particles under certain conditions, which, in turn, can bind to the corresponding receptors (Wolfrum et al., 2007). It has also been shown that the hydrophobicity of a lipophilic molecule determines the efficiency of binding a lipophilic conjugate with lipoproteins (Wolfrum et al., 2007). Thus, a cholesterol residue introduced into siRNA provides effective binding to LDL and HDL particles and, as a result, higher activity compared to other lipophilic derivatives. However, the assumption of the penetration of lipophilic siRNA derivatives in the composition of such complexes into cells by receptor-mediated endocytosis has not been confirmed (Wolfrum et al., 2007). The transmembrane protein SIDT1 (Wolfrum et al., 2007) participates in the penetration of complex of lipophilic siRNA derivatives and lipoproteins. SIDT1 facilitates the penetration of dsRNA into cells, by forming channels for diffusion or by facilitating the penetration of dsRNA indirectly via interaction with other proteins (Feinberg and Hunter, 2003; Wolfrum et al., 2007). On the other hand, SIDT1 has a binding domain that interacts with steroid molecules and its localization in the cell depends on the presence or absence of cholesterol in the membrane (Mendez-Acevedo et al., 2017). It has also been shown that the SIDT1 homolog SIDT2 is involved in cellular transport of cholesterol (Mendez-Acevedo et al., 2017) and dsRNA without lipids (Nguyen et al., 2017; Takahashi et al., 2017). However, the specific role of SIDT1 in the penetration of complexes of lipophilic siRNA derivatives with lipoproteins has not been established.

Penetration of siRNA cholesterol conjugates without a carrier was studied in HeLa cells (Gilleron et al., 2015). For this purpose, the expression of genes important for endocytosis (DNM1L, CLTC, CAV1, CDC42, and RAC1) in cells was suppressed, and accumulation of the fluorescently labeled cholesterol conjugate was evaluated. Accumulation of the cholesterol conjugate was reduced by ∼40% compared to accumulation in untreated cells only when expression of DNM1L and CLTC, which participate in clathrin-dependent endocytosis, was inhibited.

In another study (Ly et al., 2017), clathrin-dependent intracellular transport of cholesterol conjugates was investigated by determining the colocalization of fluorescently-labeled endosome proteins and ligands of clathrin-dependent endocytosis with a fluorescently-labeled cholesterol conjugate. In this work, clathrin-dependent endocytosis was shown to account for 25% of the total intracellular transport of the cholesterol siRNA derivative in the cell. It should be noted that endocytosis is characterized by the ability to sort absorbed endosomal contents for recirculation or degradation (Lakadamyali et al., 2006). Depending on the type of receptor and the content of the endosome, sorting can occur at different stages of endocytosis; sorting and recycling of the LDL receptor takes an average of ∼6 min (Brown and Goldstein, 1976). Analysis of the kinetics of accumulation of cholesterol-modified siRNA revealed that in the first 60 min after the addition of cholesterol derivatives of siRNA to HeLa cells, only 5% of the conjugate was recycled, and 20% of the conjugate was sorted to the degradation pathway (Ly et al., 2017). Thus, it can be assumed that with systemic administration of the cholesterol conjugate, the primary route of transport is interaction with lipoproteins and penetration by receptor-mediated endocytosis into cells expressing the corresponding receptor (LDL or SR-BI receptor). Then, likely at some stage of intracellular transport, lipoproteins and the cholesterol-siRNA conjugate dissociate and the siRNA enters the cytoplasm, where it participates in RNAi. Indeed, many studies have shown that the addition of cholesterol to the 5′ and 3′ ends of the sense strand and the 3′ end of the antisense strand provides manifestation of the biological activity of siRNA upon delivery without a carrier in vitro (Lorenz et al., 2004; Cesarone et al., 2007; Moschos et al., 2007; Alterman et al., 2015; Chernikov et al., 2018) and in vivo (Soutschek et al., 2004; Wolfrum et al., 2007; Byrne et al., 2013; Khan et al., 2016; Haraszti et al., 2017). In most cell lines, the cholesterol-siRNA conjugate shows higher biological activity compared with other conjugates of siRNA and lipophilic derivatives; e.g., lithocholic acid derivatives, saturated fatty acids (C12–C22) (Wolfrum et al., 2007; Prakash et al., 2015), unsaturated fatty acids (Nikan et al., 2016, 2017), or tocopherol (Nishina et al., 2008).

The biodistribution of various lipophilic siRNA conjugates has been extensively studied in a recent paper (Biscans et al., 2018). It was shown that cholesterol conjugates were more effectively retained in the body (62%) compared with other lipophilic conjugates (27–62%). Following subcutaneous injection, the cholesterol conjugates accumulated in almost all organs: liver, kidney, adrenal glands, spleen, pancreas, heart, muscle, fat, thymus, lung, injection site, ovaries, and testes. At the same time, cholesterol conjugates accumulated most effectively in the liver, adrenal glands, spleen, and in the skin at the site of administration (Biscans et al., 2018). Cholesterol derivatives accumulated in other organs with the same or lesser efficiency than other lipophilic conjugates; e.g., it was shown that a conjugate of siRNA and saturated fatty acid (docosanoic, C21) accumulated more efficiently than a cholesterol conjugate and inhibited expression of the target gene (Htt or Ppib) in muscles (20 and 30% inhibition of Htt and Ppib, respectively) and fat (50% and 30% inhibition of Htt and Ppib, respectively) after subcutaneous injection (20 mg/kg) (Biscans et al., 2018). In this study, it was shown that the main factor determining the nature of the biodistribution of conjugates is their lipophilicity. Conjugates of siRNA with lower lipophilicity; i.e., derivatives

of retinoic acid, lithocholic acid, and docosahexanoic acid with greater efficiency than cholesterol conjugates accumulated in the kidneys, bladder, and lungs of the mouse after subcutaneous injection (Biscans et al., 2018). This fact is consistent with previous data that showed that more lipophilic conjugates bind more efficiently to serum components, and thus are not excreted by the kidneys (Wolfrum et al., 2007; Osborn et al., 2018).

Lipophilic derivatives after subcutaneous or intravenous injection do not penetrate the blood-brain barrier (BBB) (Biscans et al., 2018). Therefore, attempts were made to directly inject derivatives into the brain of the mouse to suppress gene expression in the brain (Alterman et al., 2015; Nikan et al., 2016, 2017). Since docosahexaenoic acid is the most common polyunsaturated fatty acid in the mammalian brain, conjugation of siRNA with docosahexaenoic acid more effectively suppressed the expression of the target gene (Nikan et al., 2016) than other lipophilic conjugates (Alterman et al., 2015). Injection of the siRNA-docosahexaenoic acid conjugate into the brain striatum of the mouse (∼1.25 mg/kg) caused a decrease in the mRNA level of the target gene (Htt) not only in the striatum (73%) but also in the cortex (52%) of the corresponding hemisphere (Nikan et al., 2016). However, no decrease in Htt mRNA was observed in the striatum and the cortex of the opposite hemisphere. An assessment of the toxicity of the conjugate in the brains of animals, at a dose 20 times higher than that in a study of biological activity, showed that this conjugate does not elicit an immune response or neuronal death. The therapeutic significance of suppressing the expression of the Htt gene was shown in another study in a mouse model of Huntington's disease using ASO (Kordasiewicz et al., 2012). It was shown that suppression of the expression of both alleles (mutant and wild-type) resulted in restoration of a healthy animal phenotype. Therefore, the use of docosahexaenoic acid for conjugation with siRNA may be a promising approach for the treatment of various hereditary neurodegenerative diseases, including Huntington's disease. Thus, such conjugates can be considered as a universal platform for siRNA delivery throughout the body since lipophilic siRNA derivatives can accumulate and exhibit biological activity in a variety of tissues and organs.

### siRNA and Peptide Conjugates

Some proteins and peptides are able to penetrate into the cell due to endogenous transport mechanisms, as well-transfer other molecules into the cell. The main mechanisms of peptide transport include binding to surface proteins, glycoconjugates [targeted peptides Pooga et al., 1998; Alberici et al., 2013], or anionic cell lipids, followed by absorption by endocytosis, membrane penetration (cell penetrating peptides [CPPs] Vives et al., 1997; Thoren et al., 2000; Console et al., 2003; Heitz et al., 2009; van den Berg and Dowdy, 2011; Lee et al., 2013; Gagat et al., 2017), membrane lysing, or pore formation in the membrane [lytic peptides (Meyer et al., 2009)]. There are two main ways of obtaining such peptides: using phage display, or using parts of proteins that perform similar functions in nature. In this section, the main approaches for siRNA delivery by peptide are reviews.

The ability of peptides to specifically interact with certain proteins on the cell surface due to specific elements in their tertiary structure was used for targeted delivery of siRNAs. Different targeted peptides were conjugated to siRNAs and such conjugates possessed biological activity in vitro (Cesarone et al., 2007; Alam et al., 2011; Alberici et al., 2013) and in vivo (Liu et al., 2014). For example, a conjugate of siRNA and the peptide "CSKC," which mimics insulin-like growth factor 1 (IGF-1), effectively penetrated MCF7 cells expressing the IGF-1-specific receptor and suppressed expression of the IRS1 target gene by 60% without the help of transfection agents (Cesarone et al., 2007).

One of the most successful examples of targeted peptides is the cyclic RGD (cRGD) peptide. cRGD is part of the iRGD peptide obtained by selecting a phage library of cyclic peptides for binding to a xenograft PC-3 tumor (Sugahara et al., 2009). cRGD binds to αVβ3/5 integrins that are expressed at a high level in tumor cells and vascular endothelium cells (Dubey et al., 2004; Weis and Cheresh, 2011). Conjugation of this peptide with siRNA contributed to the accumulation and manifestation of the biological activity of siRNA in tumor cells in culture (Alam et al., 2011) and in vivo following six intravenous injections in animals with xenograft A549 tumors (reduction of VEGFR2 gene mRNA by 55%), which was accompanied by a decrease in tumor growth (Liu et al., 2014) (**Supplementary Table 1**). Attempts to increase the number of RGD peptides in the siRNA conjugate were made; however, the introduction of additional molecules of this peptide did not have a direct dose-dependent effect on biological activity. Assessment of biological activity to suppress expression of the GFP gene in M21+GL<sup>3</sup> cells showed that a siRNA conjugate with two cRGD peptides has negligible activity (20%), a conjugate with four cRGD peptides showed moderate biological activity (37%), and a conjugate with three cRGD peptides had the highest biological activity (73%) (Alam et al., 2011).

Covalent attachment of peptides can not only increase the efficiency of siRNA accumulation in cells, but also ensure the specificity of their actions in target cells. A peptide with the "LEVDG" sequence attached to siRNA blocks RISC (Koehn et al., 2010) assembly; however, this sequence is specifically recognized and cleaved by caspase-4, which is expressed in Jeg-3 choriocarcinoma cells. Thus, following the introduction of anti-STAT3 siRNA conjugated with the "LEVDG" peptide into Jeg-3 cells, effective (up to 70%) suppression of target gene expression was observed (**Supplementary Table 1**), whereas in the control HEK293 cells not expressing caspase-4, suppression of the STAT3 gene was not observed (Koehn et al., 2010).

Due to the presence of positively charged amino acids in its composition and the secondary structure, CPP peptides, such as penetratin (Moschos et al., 2007), transportan (Muratovska and Eccles, 2004) and trans activator of transcription (Tat) (van den Berg and Dowdy, 2011) are able to penetrate the cell membrane, as well as deliver covalently attached nucleic acids into cells (Chiu et al., 2004; Muratovska and Eccles, 2004). It has been shown in a number of studies (Muratovska and Eccles, 2004; Cesarone et al., 2007) that CPP-siRNA conjugates exhibit biological activity when added to cells. However, the use of such conjugates in vivo is limited because they are toxic and can elicit an immune response (Boeckle et al., 2005; El-Andaloussi et al., 2007; Moschos et al., 2007).

Another limitation to the use of CPPs as siRNA delivery agents is the formation of complexes of positively charged CPP with siRNA, which prevents the siRNA from interacting with RISC components. To avoid this, neutralizing or shielding the negative charge of the siRNA by introducing chemical modifications [tertbutyl-S-acyl-2-thioethyl phosphotriester (tBu-SATE) (**Table 1**, 2)] has been proposed (Meade et al., 2014; Hamil and Dowdy, 2016). The introduction of tBu-SATE modifications into the siRNA conjugate made it possible to obtain siRNA conjugates with several cationic peptides without inhibiting RNAi. At the same time, the biological activity of such conjugates directly depended on the number of CPP molecules of the Tat peptide present (Meade et al., 2014). Conjugates containing four Tat peptides more effectively suppressed the expression of a target gene in cells compared with a conjugate with two or three Tat peptides. However, despite promising in vitro results (Meade et al., 2014; Kolosenko et al., 2017), this siRNA conjugate has not yet been studied in vivo.

Another successful example of the use of CPP as a ligand for conjugation with siRNA is the skin penetrating and cell entering (SPACE) peptide, obtained by the phage display method by selecting peptides that penetrate the epidermis (Hsu and Mitragotri, 2011). Attachment of the SPACE peptide to siRNA promoted penetration of the siRNA through the epidermis and dermis following application on the skin surface. A single application of ∼12 mg/kg of the siRNA conjugate with the SPACE peptide suppressed the expression of IL10 and GAPDH in the epidermis by 28 and 47%, respectively. The inclusion of such a conjugate in the composition of lipoplexes enriched with the SPACE peptide results in a more effective downregulation of the expression of the target gene (GAPDH).

To solve the problem of effective in vivo accumulation of siRNA, an interesting approach using siRNA containing a chemical modification at the 3′ end of the sense strand capable of forming a covalent bond between siRNA and albumin following intravenous injection of siRNA was proposed (Lau et al., 2012). For this purpose, succinimidyl 4-[N-maleimidomethyl] cyclohexane-1-carboxylic acid was attached to siRNA using the amino group at the 3′ end of the sense strand; the resulting molecule ("activated siRNA") could interact with albumin to form a disulfide bond. When BALB/c mice were injected with the "activated siRNA" capable of binding albumin, there was a more efficient accumulation of siRNA in the myocardium compared to unmodified siRNA, as well as a decrease in the mRNA level of the IGF-IR target gene by 35% (**Supplementary Table 1**). However, the toxic effect of the "activated siRNA" has not yet been investigated.

Another approach for the delivery of siRNA to target cells uses its conjugation with lytic peptides, which facilitate the release of siRNA from the endosome (Varkouhi et al., 2011). The main mechanisms that increase the efficiency of the release of siRNA from the endosome are (1) formation of transmembrane pores by peptides due to their amphiphilicity and ability to form complexes [melitin, cytolytic peptide from bee venom (Meyer et al., 2009), ricin, and ribosome-inactivating protein from the oil (Sun et al., 2004)]; (2) protonation of the main groups of peptides with a decrease in the pH of the endosome, followed by an increase in the osmotic pressure inside and rupture of its membrane [polyhistidine (Chen et al., 2017)]; or (3) local membrane destabilization due to the fusogenic properties of proteins and membrane penetration of the endosome [glycoprotein H from the herpes virus (Tu and Kim, 2008), hemagglutinin-2 domain of the influenza virus (Wadia et al., 2004; Lee et al., 2011), and diphtheria toxin domain (Barati et al., 2002)]. Despite the efficiency of the action of lytic peptides, the primary issue is toxicity, since the formation of pores and increased osmotic pressure inside the endosome implies its destruction. In this case, the approach in which the membrane is locally destabilized is less toxic, since it does not lead to significant damage to the endosome and therefore is the most promising. However, to date, an effective and non-toxic endosomolytic agent based on peptides has not been developed.

The use of siRNA and peptide conjugates for siRNA delivery is a promising approach; however, at present, the efficiency and specificity of delivery provided by peptides does not reach the level at which they do not exhibit toxic and immunogenic effects.

### siRNA and Receptor Ligand Conjugates

The main factors affecting the efficiency of specific delivery to target cells are the efficiency of ligand binding to the receptor and the degree of expression of the receptor on the membrane surface. Typically, the interaction of the ligand with the receptor is characterized by high specificity, so introduction of these molecules contributes to effective targeted delivery when covalently attached to molecules (Nikam and Gore, 2018). The most successful example of this strategy is the use of N-acetylgalactosamine as a ligand for siRNA delivery because its interaction with the asialoglycoprotein receptor (ASGPR) occurs with high efficiency (K<sup>d</sup> = 2.5 nM) and this receptor is expressed at a high level in hepatocytes (0.5–1 × 10<sup>6</sup> molecules per cell) (Spiess, 1990). Conjugation of siRNA with N-acetylgalactosamine contributes to efficient delivery of siRNA and the conjugate suppresses expression of the target gene (PCSK9) by 70% following a single subcutaneous injection of ∼6 mg/kg (Ray et al., 2017) (**Supplementary Table 1**). The level of ASGPR in hepatocytes is so high that reducing its expression by 50% does not reduce the biological activity of the siRNA and N-acetylgalactosamine conjugate; only suppression by 95% blocks the action of this conjugate (Willoughby et al., 2018). For these reasons, and due to inexpensive synthesis, siRNA and Nacetylgalactosamine conjugates are among the most promising prototypic drugs for introduction in the clinic for the treatment of liver diseases.

Conjugation with folic acid was proposed for specific accumulation of siRNA in tumor cells (Thomas et al., 2009). Folic acid is a precursor of tetrahydrofolate, which is essential for the synthesis of nucleotides de novo, and thus is essential for dividing cells. Folic acid penetration into cells occurs via receptor-mediated endocytosis, via the glycoprotein folic acid receptor, which binds strongly to folate (K<sup>d</sup> = 10−<sup>10</sup> M). It has been shown that expression of folic acid receptor in tumor cells of different origin is significantly higher compared with the expression level in normal cells (Parker et al., 2005; Xia and Low, 2010). Presumably, the penetration of folate-containing siRNAs also occurs via receptor-mediated endocytosis (Low et al., 2008); therefore, the penetration efficiency of folate-containing siRNAs into tumor cells is significantly higher compared with the efficiency of conjugate penetration into normal cells. In vivo, fluorescently-labeled siRNA and folate conjugates were shown to accumulate more efficiently in a mouse tumor compared to unmodified siRNA (Thomas et al., 2009). However, the use of siRNA and folate conjugates is limited to experimental purposes due to sophisticated synthesis.

### siRNA and Aptamer Conjugates

Aptamers are synthetic oligoribonucleotides (molecular weight, ∼6–30 kDa) with a complex tertiary structure, which allows specific binding to molecules (Zhou and Rossi, 2017). It has been shown that the attachment of aptamers to siRNAs contributes to specific accumulation in certain cell types (Catuogno et al., 2018). For instance, siRNA conjugation with the A10 aptamer specific to prostate specific membrane antigen (PSMA) promoted effective delivery of siRNA to prostate tumor cells, and the biological activity of the conjugate was comparable to that observed in the case of siRNA delivered by lipoplexes (McNamara et al., 2006). The conjugate silenced PKL1 and Bcl2 genes with 85 and 90% efficiency, respectively, in the xenographic prostate LNCaP tumor-bearing mice model after 10 intratumoral injections; a decrease in tumor growth and regression were also observed (McNamara et al., 2006) (**Supplementary Table 1**). In the same xenograft tumor model, another shRNA (short hairpin RNA) and aptamer A10-3 to PSMA conjugate also exhibited biological activity (65% inhibition of the PRKDC gene), but tumor regression occurred after two intratumoral injections only in combination with ionizing radiation (Ni et al., 2011) (**Supplementary Table 1**).

A conjugate of siRNA and the A-1 aptamer specific for the gp120 surface protein of human immunodeficiency virus 1 (HIV-1) capsid was synthesized as an anti-HIV-1 drug (Neff et al., 2011). In humanized Rag2−/<sup>−</sup> γc <sup>−</sup>/<sup>−</sup> (RAG-hu) mice 3 weeks after infection with HIV-1, weekly administration (0.38 mg/kg, intravenous) of this conjugate reduced the concentration of viral RNA in the plasma of animals by 105 times. However, a few weeks after treatment, the amount of viral RNA in the plasma was restored almost to the initial level (**Supplementary Table 1**).

Another interesting property of some aptamers is their ability to penetrate the BBB. For instance, it has been shown that the aptamers Gint4.T and GL21.T, specific to beta-type plateletderived growth factor receptor (PDGFRβ), can penetrate the in vitro model of the BBB (Esposito et al., 2016). This siRNA conjugate is able to accumulate in xenograft glioblastomas after several intravenous injections, suppress expression of the target gene (STAT3) by 60%, and reduce the rate of tumor growth (Esposito et al., 2018).

In addition to RNA aptamers, DNA aptamers have also been used for the delivery of siRNAs (Lai et al., 2014). The G-quadruplex-forming G-rich deoxyoligonucleotide AS1411 specifically binds nucleolin, an oncogene protein expressed at a high level in many cancer cell types (Ireson and Kelland, 2006). The AS1411-based DNA aptamer aptNCL conjugated with siRNA enables its delivery and biological activity in lung cancer cells in vitro and in vivo (Lai et al., 2014).

Although siRNA and aptamer conjugates have high biological efficiency at the experimental level, their use in the clinic has been limited by such factors as nuclease cleavage, filtration by the kidneys, polyanion effects, and the immune response. Selection of a new specific sequence of an aptamer to a specific object (systematic evolution of ligands by exponential enrichment [SELEX]) is fast, but the resulting aptamers do not always have high specificity for the target antigen (Yan and Levy, 2018). Nevertheless, in most cases, conjugation of siRNA with aptamers provides reproducible specific delivery of siRNA to target cells, and the possibility of obtaining aptamers directed to any protein on the surface of the cell membrane suggests this may be a promising approach. Therefore, the use of a modified SELEX protocol to search for chemically modified aptamers and conjugation with fully modified siRNAs can increase the effectiveness and duration of the therapeutic effect of aptamerbased conjugates of siRNA and their introduction into clinical practice (Hori et al., 2018).

### Antibody-siRNA Conjugates

Antibody-siRNA conjugates (ARCs) have been successfully used for targeted delivery of siRNA to specific types of cells expressing receptor-antigens; however, the effectiveness of ARCs varies significantly. For example, it has been shown that an ARC with an antibody against the insulin receptor suppresses expression of the target gene by 90% in HEK293 cells at a concentration of 115 nM (**Supplementary Table 1**) (Xia et al., 2009). Another ARC with an antibody to the Lewis-Y protein inhibited the expression of the target gene by 60% at a concentration of 300 nM only when the cells were treated with chloroquine, an agent that inhibits endosome maturation (**Supplementary Table 1**) (Ma et al., 2011). However, the non-covalent siRNA complex with the same antibody, formed by electrostatic interaction of oligoarginine and siRNA, showed 60% biological activity (300 nM) in the absence of chloroquine. It is likely that the efficiency of endosomal escape mediated by chloroquine or oligo-arginine is an important factor for the biological activity of ARCs. The biological activity of both covalent and non-covalent siRNA complexes and antibodies has been shown in vivo in a number of studies (Song et al., 2005; Xia et al., 2007; Baumer et al., 2015; Sugo et al., 2016; Ibtehaj and Huda, 2017). However, a systemic comparison of the effectiveness of the biological activity of ARCs differing in target antigens was carried out only in one study (Cuellar et al., 2014), which showed that along with the level of expression of the receptor-antigen, the type of intracellular transport of the receptor influences the interfering activity of the ARC. However, no direct correlation was found between the type of penetration of the antibody complex with the receptorantigen and the biological activity of the ARC. Such a factor as the presence of a cleavable bond between the siRNA and antibody did not affect the interfering activity of the ARC. Since in this study (Cuellar et al., 2014), the efficiency of the binding of antibodies to the corresponding receptors was not compared, it is not possible to evaluate the efficiency of their dissociation and the degree of endosomal escape of the conjugates. However, it is likely that this is a significant factor in determining the biological activity of ARCs.

The mechanism of penetration into cells of an ARC with the TENB2 antibody exhibiting high biological activity was studied (Cuellar et al., 2014). It was shown that the silencing of genes associated with clathrin-dependent receptor-mediated endocytosis led to a decrease in the efficiency of gene silencing by ARC. However, the suppression of the expression of RAB5C and HPS4, which are associated with intracellular transport, increased the silencing activity of the ARC. The product of the RAB5C gene likely sorts endosomal contents to the recycling pathway (Chen et al., 2014), while the product of the HPS4 gene is involved in regulation of RNAi (Lee et al., 2009). It has been shown that the product of the HPS4 gene reduces the amount of RLC and RISC proteins in the cell by increasing the frequency of lysosomes merging with multivesicular bodies, where, presumably, RNAi proteins are located (Lee et al., 2009). Thus, the suppression of HPS4 gene expression leads to an increase in the efficiency of RNAi in cells and a corresponding increase in the activity of the conjugate. The effectiveness of PPID gene silencing by the ARC with the TENB2 antibody after three intravenous injections of 24 mg/kg in xenograft PC3-TENB2-high tumorbearing nude mice was only 33% (**Supplementary Table 1**) (Cuellar et al., 2014).

In another study (Sugo et al., 2016), an Fab antibody fragment, an immunoglobulin molecule segment that binds an antigen that has lower affinity for the target receptor than antibody, was used for conjugation with siRNA. After 4 weekly intramuscular injections (∼3.6 mg/kg) of the conjugate of siRNA and Fab fragment to the transferrin receptor, the MSTN mRNA level was decreased by 72%, which increased the average running distance of mice by 24% in the peripheral arterial disease model (Sugo et al., 2016). Intramuscular injection of only ∼0.05 mg/kg of the conjugate of siRNA and Fab fragment to the transferrin receptor resulted in suppression of the HPRT target gene at the injection site by 55%. High biological activity of the siRNA and Fab fragment conjugate in muscle cells following intravenous injection was demonstrated by Avidity Bioscience. The mRNA level of the target MSTN gene was decreased by 90% and lasted for 20 days following a single intravenous injection of this conjugate, while the antigen for the Fab fragment was also a transferrin receptor. Fab fragments more efficiently than antibodies escape endosomes into the cytoplasmic space after being absorbed by cells. This is likely due to lower receptor binding efficiency and low molecular weight (55 kDa). Therefore, this approach is promising for the targeted delivery of siRNA to cells; however, a direct comparison of ARCs with conjugates of siRNA and Fab fragments has not yet been carried out.

Conjugation of siRNAs with antibodies to deliver siRNA to target cells has several advantages compared with the conjugation of siRNAs with other molecules, such as high ligand binding efficiency (K<sup>d</sup> < 10−9−10) and prolonged presence in blood due to high molecular weight (∼150 kDa). However, the immune response and low efficiency of the endosomal yield are the main disadvantage of this approach. Thus, further optimization, including the use of humanized antibodies or Fab fragments, endosomolytic agents, and fully modified siRNAs, is required for effective use of ARCs in the clinic.

### siRNA and CpG Oligonucleotide Conjugates

As an alternative method of siRNA delivery to target cells, systems that provide an efficient release of siRNA from endosomes to the cytoplasm are used. For example, conjugation of DsiRNA with CpG-containing oligodeoxyribonucleotides leads to recovery of the interfering activity of DsiRNA in cells expressing the TLR9 receptor due to endosomal release mediated by TLR9 (Nechaev et al., 2013). Thus, the conjugate is biologically active only in cells expressing the TLR9 receptor, such as cells of the immune system: B-lymphocytes, dendritic cells, and macrophages, as well as in some types of cancer (Zhang et al., 2013). The therapeutic effect of the siRNA and CpG oligonucleotide conjugate has been shown in various tumorbearing mouse models following systemic administration of the conjugate over several weeks (Kortylewski et al., 2009; Zhang et al., 2013; Hossain et al., 2014). However, since the injection of CpG oligonucleotides leads to the activation of cytokines and interleukins, their use is limited. Also, its application in vivo is limited to local injections due to rapid degradation in serum. Introduction of chemical modifications to such DsiRNA conjugates to increase nuclease resistance will likely change the interaction of the conjugate with Dicer (Nechaev et al., 2013).

Further, to suppress the expression of the target gene (STAT3), researchers conjugated the CpG oligonucleotide with the DNA duplex, which is part of the promoter of the STAT3 gene, so that when it enters the nucleus of the target cell, this duplex binds to the corresponding transcription factor and blocks its transcription (Sen et al., 2012). This conjugate showed a therapeutic effect in a mouse model of acute myeloid leukemia after several intravenous injections (5 mg/kg) (Zhang et al., 2016). The first stage of clinical trials of this conjugate for the treatment of B-cell non-Hodgkin's lymphoma is planned for 2019.

### Dynamic Polyconjugates

Dynamic polyconjugates, which contain two types of cleavable bonds, were developed by Arrowhead Pharmaceuticals to facilitate the endosomal escape of siRNA (Rozema et al., 2007). A polyconjugate is an amphiphilic polymer consisting of poly- (butyl-aminovinyl ether) (PBAVE), to which polyethylene glycol residues and a targeted ligand molecule (N-acetylgalactosamine) are attached using an acid-cleavable carboxy dimaleimide anhydride linker. siRNA molecules are connected to PBAVE through linkers containing disulfide bonds that can be cleaved in the cytoplasm of the cell. Thus, following penetration of the dynamic polyconjugate by receptor-mediated endocytosis into the cell and entry into the acidic environment of the endosome, the carboxy-diimide anhydrite bonds are cleaved and N-acetylgalactosamine and polyethylene glycol dissociate from the polyconjugate. Following this, the newly formed amino groups on PBAVE are protonated, which leads to a decrease in endosomal pH, an increase in osmotic pressure, and rupture of the endosomal membrane. Conjugates of PBAVE and siRNA are released into the cytoplasm, followed by cleavage of the disulfide bond and detachment of the siRNA from the polymer (Rozema et al., 2007). Due to

effective endosomal escape, dynamic polyconjugates exhibit high biological activity: suppression of the target gene F7 was observed in cynomolgus monkeys with 99% efficiency for 80 days following a single intravenous injection (5 mg/kg) (Rozema et al., 2015). Another drug based on a dynamic polyconjugate is the PBAVE polymer, conjugated with polyethylene glycol or N-acetylgalactosamine residues (NAG-PBAVE), but without covalent attachment of siRNA (Wong et al., 2012). In this case, NAG-PBAVE is injected with cholesterolmodified siRNA. Intravenous injection of the cholesterol-siRNA conjugate results in accumulation of siRNA mainly in the liver; the NAG-PBAVE component also accumulates in this organ. This co-administration increases the biological activity of the cholesterol-siRNA substantially: 75% suppression of the target gene (ApoB) in the livers of rhesus monkeys was observed over 30 days following one intravenous injection of the drug (2 mg/kg siRNA and 15 mg/kg NAG-PBAVE) (Wong et al., 2012).

Another drug developed by Arrowhead Pharmaceuticals represents a conjugate of siRNA and cholesterol administered together with a conjugate of N-acetylgalactosamine and melitinlike peptide (NAG-MLP) (Wooddell et al., 2013). Melitin-like protein is an endosomolytic pore-forming peptide capable of increasing the efficiency of the release of the cholesterol conjugate from endosomes and, therefore, its biological activity. It has been shown that after single intravenous injection of the cholesterolsiRNA conjugate (1 mg/kg) together with NAG-MLP (6 mg/kg), expression of the F7 target gene in the mouse liver was suppressed with 99% efficiency, while the cholesterol-siRNA conjugate (10 mg/kg) alone, without NAG-MLP, reduced expression of the F7 gene only by 20% (Wooddell et al., 2013). This co-administration system was used to treat chronic hepatitis B virus in patients in clinical trials. A 90% decrease of hepatitis B surface antigen (HBsAg) was observed for 50 days after a single injection of 4 mg/kg of a mixture of two cholesterol-siRNA conjugates (anti-HBx and anti-preC-C) and NAG-MLP.

However, clinical trials of several drugs based on PBAVE and melitin for the treatment of liver diseases were halted due to high toxicity demonstrated in a non-human primate study (Turner et al., 2018). The company switched to TRiM technology based on the conjugation of siRNA and N-acetylgalactosamine; however, the specific structure of the drug has not yet been disclosed (Wooddell et al., 2017) (see section "siRNA and receptor ligand conjugates").

### siRNA CONJUGATES IN THE CLINIC

Onpattro (Patisiran), the first commercially available siRNAbased drug, was released for the treatment of hereditary transthyretin polyneuropathy in August 2018 by Alnylam Pharmaceuticals (Adams et al., 2018; Garber, 2018; Solomon et al., 2019). Onpattro is an anti-TTR siRNA containing several 2′O-Me modifications in complex with a cationic lipid, phospholipid, cholesterol, and a conjugate of polyethylene glycol and lipid. Its administration every 3 weeks for 18 months contributes to a significant reduction in the symptoms of the disease compared with patients taking placebo. However, the fact that its use has to be combined with corticosteroids, acetaminophen, and antihistamines is evidence of the proinflammatory effect of the lipids used in Onpattro. Moreover, the side effects of this drug include redness, nausea, headache, pain in the back and abdomen, and breathing difficulties. Due to these reasons, subsequent drugs developed by Alnylam Pharmaceuticals do not use lipids for delivery and are presented as conjugates of siRNA and N-acetylgalactosamine. Currently, six drugs based on this platform are at the most advanced steps of development in Alnylam pipline; three are in the third stage, and three are in the second and first stages of clinical trials (Huang, 2017). Moreover, these conjugates all have the same structure and chemical modifications (2′O-Me, 2′F, and PS) and differ only in siRNA sequences and chemical modification patterns.

Advanced products under development by other companies (Dicerna Pharmaceuticals, Arrowhead Pharmaceuticals, and Silence Therapeutics) that suppress gene expression in hepatocytes are based on covalent conjugates of siRNA and N-acetylgalactosamine (Crooke et al., 2018; Nikam and Gore, 2018; Springer and Dowdy, 2018). Thus, significant success was achieved in the delivery of siRNA to liver cells; the search for new targets for siRNA and the determination of the dose required for a therapeutic effect will expand the range of drugs aimed at the liver (Zatsepin et al., 2016; Shen and Corey, 2017). The design of systems for delivery to organs is a fast developing area, however, currently such drugs are only at the preclinical stage (Benizri et al., 2019), successful delivery to such organs as the kidneys may be the next step (Khvorova and Watts, 2017).

### CHALLENGES AND LIMITATIONS OF USING siRNA BIOCONJUGATES IN CLINICS

The use of chemical modifications in siRNA conjugates significantly improved their bioperformance allowed to solve such problems as: some of non-target effects—the cellular immune response is reduced by the presence of 2′O-Me siRNA modifications (Judge et al., 2006); the probability of RISC<sup>∗</sup> binding to non-target mRNA molecules can be reduced by decreasing the melting temperature of the seed region of the siRNA by introducing UNA or GNA modifications (Janas et al., 2017); the use of a fully modified siRNA molecules increases the time of inhibition of the target gene up to half of the year (Ray et al., 2017). Bioconjugation strategies described above can improve the ability of conjugates to accumulate in certain organs and penetrate certain types of target cells without the help of transfection agents or other means of delivery have been developed. However, there are still a number of unsolved problems that limit the possibility of transfer of drugs from the laboratory bench to the clinic. The main problem of this kind is the low bioavailability of siRNA conjugates and unfavorable pharmacokinetics, which, together with the rather high cost of obtaining such drugs in quantities necessary to achieve a therapeutic effect, impedes their use in the clinic. The low bioavailability of siRNA conjugates is primarily due to the fact that in order for siRNA to enter the cytoplasm of a target cell after systemic administration, it has to overcome numerous barriers—the endothelial barrier when leaving the bloodstream to the tissue, as well as to escape siRNA from the endosome to the cytoplasm. The overcoming of these barriers is complicated by the unfavorable pharmacokinetics of such drugs, associated with their relatively low molecular weight, which lies below the filtration limit, due to which the drugs are rapidly removed from the bloodstream by renal filtration. In this regard, the development of alternative routes of administration, such as subcutaneous or local in which there is a deposit and the gradual release of the drug is promising, as well as approaches aimed at increasing the duration of the circulation of the drug in the bloodstream (Nair et al., 2017). Another option—the targeted delivery, which can be implemented only for some organs and tissues with sufficient efficiency, moreover, using the advantage of specific binding to target cells does not cancel the dependence of this process on the concentration of drug in the blood or interstitial fluid and the duration of maintenance of an effective concentration. The necessity to administer high doses of drugs to achieve a therapeutic effect raises the problem of specificity and possible side effects, the severity of which increases at high concentrations. It can be expected, that along with non-specific effects associated with the suppression of partially homologous targets, which can be eliminated by sequence selection and conjugate design, immunostimulation, the metabolic effects of unnatural analogs, including cumulative and long-term ones, as well as the intervention of exogenous siRNA into cellular regulatory systems of miRNA in competition for RISC, can become the main non-specific effects important for the safety of clinical use. In this regard, the main challengers of biomedical research are increasing the bioavailability, biological activity and targeting of delivery, which will reduce the therapeutic doses of drugs based on siRNA conjugates.

#### CONCLUSIONS

The introduction of molecules of natural origin into the composition of siRNA is a promising approach for nonviral delivery and has clear advantages over other approaches

#### REFERENCES


(physical methods, delivery using cationic lipids, and polymers): specificity of penetration into target cells and absence of toxic effects (Lee et al., 2016; Benizri et al., 2019). The primary difficulty in designing bioconjugates is the necessity of selecting specific ligands for individual applications due to the specificity of ligand-receptor interactions. From this point of view, the use of lipophilic siRNA conjugates is less specific, since LDL receptors are expressed at a high level by various cell types; however, it could be beneficial if high selectivity of delivery to certain cell types is not required and accumulation of the drug into non-target cells does not cause undesirable effects (Turanov et al., 2018). The latest patterns of chemical modifications can reduce the ID<sup>50</sup> and increase the duration of the biological effect of siRNA conjugates. As a result, the application of siRNAbased drugs in clinical practice in the next few years may significantly increase.

#### AUTHOR CONTRIBUTIONS

IC wrote the manuscript. VV and EC critically analyzed and corrected the text. EC came up with the concept.

#### FUNDING

This work was supported by Russian Scientific Foundation grant 14-14-00697 and the Russian State Funded Budget Project (# AAAA-A17-117020210024-8).

#### ACKNOWLEDGMENTS

The authors thank Daniil Gladkikh for help preparing the figures.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar. 2019.00444/full#supplementary-material

potency and stability compared to unmodified small interfering RNA. J. Med. Chem. 48, 901–904. doi: 10.1021/jm049167j


performance of GalNAc-siRNA conjugates. Mol. Ther. 26, 708–717. doi: 10.1016/j.ymthe.2017.12.021


treated with a phosphorothioate oligodeoxynucleotide. Int. Immunopharmacol. 2, 1657–1666. doi: 10.1016/S1567-5769(02)00142-X


using site-specific chemical modification. Nucleic Acids Res. 34, 4900–4911. doi: 10.1093/nar/gkl464


models with reduced asialoglycoprotein receptor expression. Mol. Ther. 26, 105–114. doi: 10.1016/j.ymthe.2017.08.019


**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 Chernikov, Vlassov and Chernolovskaya. 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.

# MicroRNA Post-transcriptional Regulation of the NLRP3 Inflammasome in Immunopathologies

Gulcin Tezcan<sup>1</sup> , Ekaterina V. Martynova<sup>1</sup> , Zarema E. Gilazieva<sup>1</sup> , Alan McIntyre<sup>2</sup> , Albert A. Rizvanov<sup>1</sup> and Svetlana F. Khaiboullina1,3 \*

1 Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, <sup>2</sup> Centre for Cancer Sciences, Faculty of Medicine and Health Sciences, University of Nottingham, Nottingham, United Kingdom, <sup>3</sup> Department of Microbiology and Immunology, University of Nevada, Reno, Reno, NV, United States

#### Edited by:

Hector A. Cabrera-Fuentes, University of Giessen, Germany

#### Reviewed by:

Saverio Bellusci, University of Giessen, Germany Federica Laudisi, University of Rome Tor Vergata, Italy Moritz Haneklaus, University of Cambridge, United Kingdom

> \*Correspondence: Svetlana F. Khaiboullina sv.khaiboullina@gmail.com

#### Specialty section:

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

Received: 12 January 2019 Accepted: 08 April 2019 Published: 01 May 2019

#### Citation:

Tezcan G, Martynova EV, Gilazieva ZE, McIntyre A, Rizvanov AA and Khaiboullina SF (2019) MicroRNA Post-transcriptional Regulation of the NLRP3 Inflammasome in Immunopathologies. Front. Pharmacol. 10:451. doi: 10.3389/fphar.2019.00451 Inflammation has a crucial role in protection against various pathogens. The inflammasome is an intracellular multiprotein signaling complex that is linked to pathogen sensing and initiation of the inflammatory response in physiological and pathological conditions. The most characterized inflammasome is the NLRP3 inflammasome, which is a known sensor of cell stress and is tightly regulated in resting cells. However, altered regulation of the NLRP3 inflammasome is found in several pathological conditions, including autoimmune disease and cancer. NLRP3 expression was shown to be post-transcriptionally regulated and multiple miRNA have been implicated in post-transcriptional regulation of the inflammasome. Therefore, in recent years, miRNA based post-transcriptional control of NLRP3 has become a focus of much research, especially as a potential therapeutic approach. In this review, we provide a summary of the recent investigations on the role of miRNA in the post-transcriptional control of the NLRP3 inflammasome, a key regulator of pro-inflammatory IL-1β and IL-18 cytokine production. Current approaches to targeting the inflammasome product were shown to be an effective treatment for diseases linked to NLRP3 overexpression. Although utilizing NLRP3 targeting miRNAs was shown to be a successful therapeutic approach in several animal models, their therapeutic application in patients remains to be determined.

#### Keywords: NLRP3, inflammasome, microRNA, inflammation, disease

### INFLAMMASOME

#### Structure

In 2002, the ground breaking work published by Martinon et al. (2002) has demonstrated the role of the inflammasome, a multi-protein complex, in the activation of pro-inflammatory caspases. The authors described the multistep process of the inflammasome assembly which is initiated by the detection of pathogen-associated molecular patterns (PAMPs) or danger signals released by damaged cells (Duncan et al., 2009; Ichinohe et al., 2010; Costa et al., 2012). Several inflammasome sensors were later identified including the nucleotide-binding oligomerization domain (NOD) like receptors (NLRs), the absent in melanoma-2 like receptors (ALRs) and pyrin (Ting et al., 2008).

In the past decade our understanding of NLR containing inflammasomes structure and assembly mechanisms has advanced considerably, largely due to their potential involvement in pathogenesis of several diseases (Hoffman et al., 2001; Alexander So and Borbála Pazár, 2010; Song et al., 2017). NLRs contain three domains, an N-terminal domain, a NOD, and a C-terminal leucine-rich repeat (LRR) (Inohara and Nunez, 2003). The N-terminal domain contains a caspase recruitment domain (CARD) or pyrin domain (PYD), which function to interact with downstream molecules, such as apoptosis-associated speck-like protein containing (ASC) (Inohara and Nunez, 2003; Schroder and Tschopp, 2010). The NOD domain is linked to LRR detecting PAMPs (Boekhout et al., 2011). Upon sensing PAMPs, the NLRs polymerize followed by the interaction between the PYD or CARD domains of LLR and ASC (Stutz et al., 2013). Once activated the inflammasome adopts a wheel-like structure (Hu et al., 2015), where CARD–CARD interactions are essential for recruiting pro-caspase 1 (PC1) into close proximity with the complex (Faustin et al., 2007). PC1 becomes proteolytically cleaved by the CARD domain releasing an active caspase 1 (AC1) p10/p20 tetramer (Martinon et al., 2002; Kanneganti et al., 2006; Boucher et al., 2018).

### NLR Inflammasomes

This family of inflammasomes includes two subgroups based on the presence of CARD or pyrin in the N terminus. Only nucleotide-binding domain leucine-rich repeats proteins (NLRP)1, NLRP3, and NLRC4 were shown to form inflammasomes that produce AC1 (Mao et al., 2014). In contrast, NLRP6, NLRP9b, and NLRP12 are believed to form inflammasomes, but their roles as inflammasome sensors are less recognized (Anand et al., 2012; Vladimer et al., 2012; Zhu et al., 2017).

#### NLRP1

NLRP1 was the first identified cytosolic receptor capable of forming active inflammasomes (Martinon et al., 2002). PYD, NBD, and LRR domains, a 'function-to find' domain (FIIND) and a C-terminal CARD are the structural components of NLRP1 (Jin et al., 2013b). Our knowledge of NLRP1 function comes largely from studying animal models. It appears that NLRP1 senses and protects against microbial pathogens, as was shown using a mouse model of Bacillus anthracis and Shigella flexneri infection (Boyden and Dietrich, 2006; Sandstrom et al., 2019). Additionally, NLRP1 inflammasomes facilitate parasite clearance and protection as demonstrated in Toxoplasma gondii infection in mouse and rat models (Cirelli et al., 2014; Gorfu et al., 2014). The clinical relevance of NLRP1 inflammasomes against Toxoplasma gondii is also evident in individuals with specific single-nucleotide polymorphisms in the NLRP1 gene, which are linked to congenital toxoplasmosis (Witola et al., 2011).

Aberrant activation of NLRP1 is linked to a pathogenesis of inflammatory diseases. Polymorphisms in the NLRP1 gene are linked to Crohn's disease, rheumatoid arthritis (RA) and systemic sclerosis (Finger et al., 2012). Although the mechanism of NLRP1 activation remains largely unknown, recently, the failure of inflammasome inhibition by dipeptidyl dipeptidase 9 (DDP9), linked to antigen processing (Zhong et al., 2018), was demonstrated to play role in pathogenesis of an autoimmune diseases (Zhong et al., 2018). The authors identified that a single mutation in the FIIND domain of NLRP1 abrogates binding to DPP9, triggering over activation of the inflammasome in autoinflammatory disease AIADK.

#### NLRC4

Similar to NLRP1, NLRC4 establishes protection against infectious pathogens (Mariathasan et al., 2004; Franchi et al., 2006; Zhao et al., 2011). In the absence of stimulus, NLRC4 remains inactive, where its NBD domain retains a closed conformation by binding to the winged helix domain (Tenthorey et al., 2014). NLRC4 activation is indirect, and it requires NLR family apoptosis inhibitory proteins (NAIPs) for the initial sensing of the microbial ligand (Rayamajhi et al., 2013; Yang et al., 2013; Kortmann et al., 2015). NAIPs trigger NLRC4 oligomerization, which is essential for inflammasome activation (Hu et al., 2015). Loss of the control over NLRC4 expression and subsequent production of AC1 and release of IL-1β by macrophages was suggested to play role in the pathogenesis of inflammasome linked autoinflammation (von Moltke et al., 2012; Canna et al., 2014). Also, a missense mutation in the NLRC4 gene was found in familial cold autoinflammatory syndrome (Kitamura et al., 2014). Multiple mutations in NLRC4 were identified in several autoinflammatory diseases including atopic dermatitis, periodic fever, and fatal or near-fatal episodes of autoinflammation (Nakamura et al., 2010; Canna et al., 2014; Bonora et al., 2015). These data suggest that NLRC4 plays role in protection against microbial pathogens and autoinflammation.

#### NLRP6

NLRP6 is an inflammasome which plays a role in gut health and maintaining mucosal response to pathogens (Elinav et al., 2011; Anand et al., 2012). A microbial metabolite, taurine, was identified as an NLRP6 activator (Levy et al., 2015). The NLRP6-taurite axis appears to be essential for the health of the gut mucosa and microbiome. Taurite produced by the normal microbiota activates NLRP6 which prevents dysbacteriosis by promoting production of antimicrobial peptides (Levy et al., 2015).

#### NLRP12

NLRP12 is intracellular protein expressed in cells of myeloid lineages (Arthur et al., 2010). NLRP12 inflammasome expression can be downregulated by microbial ligands (Williams et al., 2005; Lich et al., 2007) via canonical and non-canonical inhibition of NF-κB (Zaki et al., 2011; Allen et al., 2012). Several ligands were identified as NLRP12 activators including microbes (Allen et al., 2012; Vladimer et al., 2012).

### ALR Family Inflammasomes

ALR family inflammasomes contain an N-terminal PYD and a C-terminal hematopoietic interferon-inducible nuclear protein with 200-amino acid repeat (HIN200) domain (Cridland et al., 2012). ALR inflammasomes sense cytosolic double stranded DNA (dsDNA) (Burckstummer et al., 2009; Ferreri et al., 2010). Absent

in melanoma 2 (AIM2) is the best characterized member of ALR inflammasomes. Similar to other ALR family members, AIM2 senses dsDNA; however, it appears that dsDNA recognition is independent of nucleic acid sequence as it could bind to both, microbial and host genomic material (Jin et al., 2012). dsDNA binding to HIN200 causes its dissociation from the PYD domain (Jin et al., 2012), allowing the freed PYD domain to interact with ASC, and inflammasome assembly (Jin et al., 2013c). AIM2 was implicated in the recognition of microbial, host and tumor derived dsDNA (Davis B.K. et al., 2011; Choubey, 2012; Dihlmann et al., 2014).

### Pyrin

Pyrin is an inflammasome sensor complex, which contains a N-terminal PYD, central B-box and coiled-coil domain, and a C-terminal B30.2/SPRY domain (Heilig and Broz, 2018). Pyrin was proposed to sense the changes in actin cytoskeletal dynamics as it was found co-localized with stress actin filaments (Xu et al., 2014). Microtubules promote ASC recruitment and the oligomerization (Gao et al., 2016); however, the physiological relevance of this interaction remains largely unknown. Also, microbial toxins which cause impairment of Rho GTPase activity were identified as strong activators of the pyrin inflammasome (Dumas et al., 2014; Xu et al., 2014).

Several monogenic autoinflammatory syndromes were linked to pyrin inflammasome dysregulation including familial Mediterranean fever (FMF), pyrin-associated autoinflammation with neutrophilic dermatosis, pyogenic arthritis, pyoderma gangrenosum, acne, etc. (Jamilloux et al., 2018). FMF is the most investigated pyrin inflammasome disease, characterized by repeating, self-limited, episodes of fever and polyserositis (Bernot et al., 1998). FMF is linked to a mutation in the Mediterranean Fever (MEFV) gene in a region encoding the B30.2 domain of pyrin (Omenetti et al., 2014). Also, the high prevalence of FMF within certain populations could indicate a selective pressure to preserve this mutation (Schaner et al., 2001).

### Pyroptosis

Pyroptosis is an inflammatory form of programmed cell death linked exclusively to PC1 activation (Hilbi et al., 1998). AC1 is a product of several inflammasomes: NLRP1, NLRP3, NLRC4, and AIM2. Therefore, pyroptosis is often associated with inflammasome activation. Pyroptosis differs from apoptosis in many respects including lack of DNA fragmentation (Watson et al., 2000) and sustained structural integrity of the nucleus (Zychlinsky et al., 1992). Also, pyroptosis is characterized by cell membrane pore formation, which causes cell swelling in contrast to apoptosis, where cells shrink (Fink and Cookson, 2006). Additionally, an increased intracellular osmotic pressure generates large spherical protrusions of the membrane in pyroptotic cells, which coalescence and rupture (Ona et al., 1999). Multiple studies revealed the role of pyroptosis in clearance of microbial pathogens (Sansonetti et al., 2000; Tsuji et al., 2004; Lara-Tejero et al., 2006). However, over activation of AC1 could lead to pyroptosis associated tissue damage and autoimmunity (Ona et al., 1999; Siegmund et al., 2001; Frantz et al., 2003).

## NLRP3 INFLAMMASOMES

### Molecular Mechanism of Activation

NLRP3 is the most characterized inflammasome, and its expression is tightly regulated in resting cells (Bauernfeind et al., 2009). While NLRP3 levels in unstimulated cells are insufficient to trigger assembly of an active inflammasome complex, sensing of pathogen ligands or danger signals, triggers complex formation and pro-inflammatory cytokine production. There are multiple stimuli shown to activate NLRP3 including ATP, toxins, K<sup>+</sup> efflux, reactive oxygen species and mitochondrial dysfunction (Dostert et al., 2008; Piccini et al., 2008). Upon sensing the stimulus, the nucleotide binding domain (NBD) polymerizes initiating PYD–PYD oligomerization with ASC (Lu A. et al., 2014). The CARD of ASC recruits PC1, which becomes cleaved liberating AC1 (Boucher et al., 2018). It appears that within the large family of inflammasomes, NLRP3 is the main PC1 activator (Agostini et al., 2004; Davis E.E. et al., 2011). Inflammatory AC1 liberates functional IL-1β and IL-18 (Afonina et al., 2015), pleotropic cytokines regulating inflammation and innate immune response (Garlanda et al., 2013).

The classic pathway of NLRP3 activation requires two steps: priming and activation (**Figure 1**). Toll-like receptor (TLR), FAS-associated death domain protein and IL-1R ligands were identified as NLRP3 priming stimuli (Allam et al., 2014; Gurung et al., 2014; He Y. et al., 2016). The priming step includes transcriptional activation of NLRP3 via NF-κB signaling (Bauernfeind et al., 2009; Costa et al., 2012); however, it fails to initiate functional inflammasome formation, which requires a second stimulus (Jo et al., 2016). The second signal can be provided by multiple pathogen and danger associated ligands (Franchi et al., 2012; Koizumi et al., 2012), promoting the assembly of an adaptor (ASC) and PC1. The formed complex cleaves the PC1, which subsequently processes and releases functional IL-1β and IL-18 (Alnemri et al., 1996).

### EPIGENETIC FACTORS AND POST-TRANSCRIPTIONAL MECHANISMS REGULATING NLRP3 INFLAMMASOME ACTIVATION

The term "epigenetic" was originally presented by Waddington (1956) to describe regulation of gene expression during the embryogenesis. Since then, definition of "epigenetic" has changed, and now refers to a stably heritable modulation of gene expression without altering DNA sequence (Berger et al., 2009). Epigenetic factors include DNA methylation at cytosine followed by guanine (CpGs) nucleotide and histone posttranslational modifications (Peschansky and Wahlestedt, 2014). Initially, epigenetic control was demonstrated in normal development and differentiation; however, its role in pathogenesis of acute

cytokines, etc.) required for the upregulation of NLRP3 and pro-IL-1β transcription and protein synthesis. Signal 2 is an activation trigger (ATP, toxins, viral RNA, etc.) which is essential for formation of an active NLRP3 inflammasome. The second stimulus promotes NLRP3, PC1, pro-IL-1β, and pro-IL-18 protein synthesis. The N-terminal NBD of NLRP3 polymerizes initiating PYD–PYD oligomerization with ASC. The CARD of ASC recruits PC1, which become cleaved liberating AC1. Inflammatory AC1 liberates functional IL-1β and IL-18, pleotropic cytokines regulating inflammation and innate immune response.

and chronic inflammation has become increasingly recognized (Bayarsaihan, 2011).

## DNA Methylation

DNA methylation is dynamic and changes during the embryonic development and differentiation (Berger, 2007). It was shown that DNA methylation silences genes to ensure monoallelic expression, prevent endogenous retrovirus expression and transposon actions (Walsh et al., 1998; Bourc'his et al., 2001; Bourc'his and Bestor, 2004). DNA methylation is essential for normal cell function; however, its role in the pathogenesis of several diseases has also been confirmed (Wei et al., 2016; Vento-Tormo et al., 2017). DNA demethylation is often detected near promoters, suggesting that gene overexpression could play role in pathogenesis of many pathologies (Ryan et al., 2010; Bierne et al., 2012). NLRP3 inflammasome expression can also be regulated by changes in gene methylation status. For example, NLRP3 gene expression is silenced in health which appears to be essential for inhibiting inflammation (Ryan et al., 2010; Bierne et al., 2012; Wei et al., 2016). However, demethylation and, subsequent, overexpression of NLRP3 was linked to pathogenesis of cryopyrin-associated periodic syndromes (CAPS) (Vento-Tormo et al., 2017) and Mycobacterium tuberculosis infection (Wei et al., 2016).

## Histone Modifications

The effect of epigenetic modification of histones was studied using several inflammatory models (Bayarsaihan, 2011). Histone acetylation is essential for initiation of an activation phase of inflammation, which is characterized by the release of pro-inflammatory cytokines via CREB, mitogen-activated protein kinases (MAPKs), nuclear factor-κB (NF-κB) and signal transducer and activator of transcription (STAT) factors (Escobar et al., 2012). In contrast, histone deacetylations regulate the late, an attenuation phase of inflammation (Villagra et al., 2010). It appears that inflammasome activation can also be regulated by affecting the acetylation status of histones, as it was recently shown by Liu C.C. et al. (2018). The authors demonstrated upregulation of NLRP3 in patients diagnosed with painful neuropathy, which could be prevented by inhibition of histone acetylation.

### Non-coding RNAs

fphar-10-00451 April 29, 2019 Time: 15:11 # 5

In addition to epigenetic modulation non-coding RNAs are also involved in NLRP3 regulation (Bayarsaihan, 2011), as was demonstrated in the setting of inflammation caused by microbial and viral infection (Li et al., 2010; Ryan et al., 2010; Bierne et al., 2012; Jin et al., 2013a; Chen and Ichinohe, 2015). This inflammation is post-transcriptionally regulated via non-coding RNAs targeting inflammasome components, where mRNA stability and inhibition of translation were most commonly affected (Bayarsaihan, 2011).

### Post-transcriptional Regulation of NLRP3 Inflammasomes: MicroRNA (miRNA)

MicroRNAs are endogenous conservative, single-stranded non-coding RNAs which are 19–24 nucleotides long. Usually, miRNAs are derived from transcripts with a hairpin structure and are loaded into the Argonaute protein within a silencing complex (Hutvagner and Zamore, 2002; Mourelatos et al., 2002; Bartel, 2004). The inhibitory effect of miRNAs is explained by their binding to the untranslated regions (UTRs) of transcripts which modulates the stability and translation of the target mRNA (**Figure 2**) (Ruvkun, 2001; Filipowicz et al., 2008; Bartel, 2009; Coll and O'Neill, 2010). miRNAs can modulate the expression of histone modifies including histone deacetylases and DNA methyltransferases resulting in modulation of histone modifications and DNA methylation (Tuddenham et al., 2006; Fabbri et al., 2007).

NLRP3 activation is tightly regulated where two signals are required to initiate functional inflammasome formation. The first signal includes cell priming with TLR ligands (Bauernfeind et al., 2009; Franchi et al., 2009). Therefore, it could be suggested that targeting TLR expression will impact the inflammasome activity. Indirect regulation of TLR expression includes modulation of the downstream pathways molecules, which has been shown in injuries, inflammation and cancer

FIGURE 2 | miRNA regulation of NLRP3 inflammasome expression. (A) Priming signal triggers NLRP3, PC1, IL-1β, and IL-18 transcription and protein synthesis. Activation signal initiates inflammasome formation and release of AC1. AC1 proteolytically cleaves pro-IL-1β and pro-IL-18, liberating active cytokines. (B) Suppression of NLRP3 protein translation and inflammasome formation by miRNA. Priming stimulus triggers NLRP3 transcription; however, miR-223, miR-22, miR-30e, and miR-7 bind to the UTR region of NLRP3 mRNA and interrupt protein translation. Absence of NLRP3 protein leads failure of the inflammasome protein complex formation, when the second stimulus present.

(Coll and O'Neill, 2010; Sheedy et al., 2010; Nahid et al., 2011; Anzola et al., 2018; Tan et al., 2018; Zhi et al., 2018). TLR4 ligands are the most studied priming signals of NLRP3 activation (Groslambert and Py, 2018). It was shown that the TLR ligand binding increases the level of several miRNAs, including miR-155, miR-146a, miR-21, and miR-132, which were linked to inhibition of TLR4/MyD88/NF-κB signaling (Coll and O'Neill, 2010; Sheedy et al., 2010; Nahid et al., 2011; Anzola et al., 2018; Tan et al., 2018; Zhi et al., 2018). It appears that upregulation of miRNAs is a component of a negative feedback mechanism designed to down-modulate inflammatory cytokine production after response to microbial stimuli (Ceppi et al., 2009).

A direct inhibitory effect of let-7 family miRNAs on TLR4 mRNA has been demonstrated (Chen et al., 2007). Let-7 miRNA regulation of TLR4 was shown to occur via post-transcriptional suppression (Androulidaki et al., 2009). It was suggested that let-7 miRNA downregulation of TLR4 could have detrimental effect on host defense against microbes, promoting microbial survival and propagation (Chen et al., 2005; Muxel et al., 2018). Post-transcriptional regulation of TLR signaling and its impact on diseases are reviewed by Nahid et al. (2011).

Active inflammasome complex formation requires a second signal, initiating substantial NLRP3 transcription (Dostert et al., 2008; Piccini et al., 2008). During this transcriptionally active phase, NLRP3 mRNA could be regulated by miRNA, as was shown by miR-223 (Bauernfeind et al., 2012). According to an in silico analysis, miR-223 can bind to a highly conserved region of the 30UTR of NLRP3 mRNA and subsequently interfere with protein translation (Lewis et al., 2005). Interestingly, miR-223 appears to be an important NLRP3 regulator in leukocytes (Bauernfeind et al., 2012; Haneklaus et al., 2012), where the miRNA levels have been shown to vary in different leukocyte subsets. For example, this miRNA was found absent in T and B lymphocytes (Bauernfeind et al., 2012; Haneklaus et al., 2012). In contrast, the miR-223 was demonstrated in myeloid cells, where it was highest in neutrophils, followed by macrophages and dendritic cells (Bauernfeind et al., 2012). It has been suggested that this miRNA plays role in granulocyte production and regulation of inflammation (Johnnidis et al., 2008; Neudecker et al., 2017). Decreased production of pro-inflammatory cytokines such as IL-1β and IL-18 was demonstrated in cells treated with miR-223 or its mimics (Neudecker et al., 2017; Ding Q. et al., 2018). These data suggest that miR-223 could be a potential target for regulation of NLRP3 expression, where increased miRNA could reduce inflammasome activation and, subsequently, abrogate the inflammation (Bauernfeind et al., 2012; Haneklaus et al., 2012).

Since several miRNAs could regulate expression of a single transcript (Krek et al., 2005), it is likely that in addition to miR-223, other miRNAs can alter NLRP3 transcription (**Figure 3**).

Numerous studies have identified that pathogens, trauma and cancer can cause abnormal expression of miRNAs which impair NLPR3 inflammasome function disrupt the functional complex formation and its signaling (**Table 1**).

#### miRNA in Regulation of Inflammasome in Infections

Inflammasome activation is an important component of infectious pathogens surveillance and antimicrobial immune and inflammatory responses. This inflammasome was shown to be activated by several bacterial pathogens including


Staphylococcus aureus, Salmonella typhimurium, Listeria monocytogenes, Mycobacterium, Streptococcus pyogenes, Neisseria gonorrhoeae as well as fungi such as Candida albicans and Aspergillus fumigatus (Franchi et al., 2006; Mariathasan et al., 2006; Miao et al., 2006; Craven et al., 2009; Duncan et al., 2009; Harder et al., 2009; Hise et al., 2009; Joly et al., 2009; Munoz-Planillo et al., 2009; Broz et al., 2010; Carlsson et al., 2010; McElvania Tekippe et al., 2010; Said-Sadier et al., 2010). NAIP/NLRC4 inflammasome can protect against Salmonella Typhimurium and C. rodentium invasion by bacteria expulsion from intestinal epithelial cells together with IL-18 and eicosanoid lipid mediators release (Nordlander et al., 2014; Sellin et al., 2014; Rauch et al., 2017). It appears that NLRP3 activation is essential for establishing the inflammatory milieu in the target tissue and augmenting the phagocytic capacity of the local macrophages (Master et al., 2008; Melehani and Duncan, 2016; Cohen et al., 2018). Enhanced macrophage bactericidal activity is the most commonly identified mechanism of inflammasome antimicrobial effect (Master et al., 2008; Cohen et al., 2018). Additionally, NLRP3 activation induced death of macrophages was described as an effort to prevent microbial propagation and spread (Miao et al., 2010; Sagulenko et al., 2013). However, there is a growing body of evidence suggesting that there is a threshold of NLRP3 activity, which acts as a safeguard mechanism to prevent inflammasome over-activation. It appears that aberrant NLRP3 activation could have a detrimental effect on tissues homeostasis and compromise barrier integrity (Bortolotti et al., 2018; McKenzie et al., 2018). It is this detrimental effect of the inflammasome over-activation that is often employed by microbes to ensure

spread and propagation (Duncan et al., 2009; Harder et al., 2009; Carlsson et al., 2010).

Microbial virulence factors often act as NLRP3 activators. For example, it was shown that the detrimental (to the host) role of Esx1, a membrane lysis factor of Mycobacterium (Stanley et al., 2003), is linked to inflammasome activation (Carlsson et al., 2010). Two virulence factors of group A Streptococcus (GAS), M protein and streptolysin O, were also identified as contributing into NLRP3 activation and IL-1β production (Harder et al., 2009; Valderrama et al., 2017). Both virulence factors are commonly detected in association with invasive GAS infections, including necrotizing fasciitis and toxic shock syndrome. Therefore, NLRP3 activation by virulent factors could promote microbe propagation and aid their escape from immune clearance.

Restoring the NLRP3 activation threshold could be a novel therapeutic approach for treatment of invasive infections. In this respect, miRNA may be a tool to regain control over NLRP3. It has been shown that miR-223 expression is consistently high in NLRP3 responsive cells, suggesting the high efficacy of this miRNA in prevention of inflammasome over-activation (Bauernfeind et al., 2012). Dorhoi et al. (2013) demonstrated that miR-223 is upregulated in the blood and lung parenchyma of patients diagnosed with tuberculosis. Also, data collected using animal models confirmed the link between deletion of miR-223 and increased susceptibility to Mycobacterium tuberculosum infection (Dorhoi et al., 2013). Similarly, a protective role of miR-223 in Staphylococcus aureus infection was demonstrated by Fang et al. (2016). Additionally, the effect of targeting TLR4 for NLRP3 regulation in Listeria monocytogenes infection was demonstrated by Schnitger et al. (2011). The authors identified that, miR-146a can directly inhibit TLR4 receptor expression, which can downregulate inflammasome activity (Schnitger et al., 2011).

Many viruses can activate inflammasomes, including Influenza virus, Hepatitis C virus, Herpes simplex virus-1, etc. (Delaloye et al., 2009; Ichinohe et al., 2010; Ito et al., 2012; Kaushik et al., 2012; Negash et al., 2013; Triantafilou et al., 2013a,b; Wu et al., 2013; Ermler et al., 2014; Chen and Ichinohe, 2015). Inflammasome activation appears to be essential for anti-viral protection, serving as viral genome sensors and triggering innate immune response (Muruve et al., 2008; Lupfer et al., 2015). The protective role of inflammasomes was shown in influenza virus infection as an increased viral clearance was NLRP3 dependent (Allen et al., 2009). Also, inflammasome activation improved the survival rate in an animal model of influenza (Ichinohe et al., 2009). Thomas et al. (2009) demonstrated that, the innate immune response activation by NLRP3 inflammasomes is essential for animal protection. However, our understanding of the mechanisms of inflammasome antiviral defense remains limited (Anand et al., 2011).

Some viruses were shown to post-transcriptionally regulate inflammasome expression to benefit self-replication and propagation (Kieff and Rickinson, 2007; Rickinson and Kieff, 2007). For example, miRNA suppression of inflammasomes was shown in Epstein Barr Virus (EBV) infected cells (Kieff and Rickinson, 2007; Rickinson and Kieff, 2007). It appears that, EBV can avert NLRP3 inflammasome activation by expressing miRNAs encoded by three BHRF1-regions and 40 BART-regions of the viral genome (Albanese et al., 2016; Tagawa et al., 2016; Farrell, 2018). Additionally, two miRNAs encoded by EBV, miR-BART11-5p and miR-BART15, were identified by Haneklaus et al. (2012), which could bind to the 3<sup>0</sup> -UTR of NLRP3, the same site targeted by miR-223, and inhibit the inflammasome. It remains to be determined whether these viral miRNA could be used as therapeutic targets.

#### miRNA Regulation of Inflammasome in Autoimmune Diseases

Autoimmune diseases are often the result of a dysregulated immune response, characterized by inflammation and organ damage (Chang, 2013; Yang and Chiang, 2015). Chronic inflammation is frequently identified as a predisposing factor for an autoimmune reaction (Yang and Chiang, 2015). Multiple mechanisms were suggested to explain prolonged inflammation leading to autoimmunity; where failure to control inflammasome activation was recently identified in some autoimmune conditions (Yang and Chiang, 2015). It has been established that in addition to inflammation, an increased secretion of IL-1β and IL-18, can stimulate proliferation and organ distribution of the effector T cells, which can cause tissue damage (Oyanguren-Desez et al., 2011; Celhar et al., 2012). Therefore, targeting the inflammasome could be suggested to restore control over the inflammatory and immune response. Therapeutic potentials of several NLRP3 targeting miRNAs were investigated in autoimmune diseases such as inflammatory bowel diseases (IBDs) (Neudecker et al., 2017), RA (Xie Z. et al., 2018), type 1 diabetes (T1D) (Yang and Chiang, 2015), type 2 diabetes (T2D) (Yang and Chiang, 2015), and systemic lupus erythematosus (SLE) (Zhu et al., 2012).

#### **Inflammatory bowel diseases (IBDs)**

Inflammatory bowel diseases are characterized by chronic inflammation of the intestine and comprise two disorders Crohn's disease and ulcerative colitis. It is believed that the pathogenesis of IBDs is associated with dysregulation of innate and adaptive immune responses, triggered by microbial antigens. This could result in chronic inflammation of the digestive tract and damage to the intestinal mucosa (Fiocchi, 1998). The role of the inflammasome in intestinal inflammation is controversial. Zaki et al. (2010) reported that, NLRP3 induced production of IL-18 in intestinal epithelial cells can be protective, and contributes to epithelium integrity in experimental colitis. In contrast, Seo et al. (2015) have demonstrated the role of inflammasome in exacerbation of an intestinal pathology. The damaging effect of the inflammasome was also confirmed by Shouval et al. (2016), who identified that IL-1β inhibition improves the course of IBDs. It appears that increased IL-1β levels and tissue damage in IBDs are linked to NLRP3 activation in myeloid leukocytes infiltrating the gut tissue (Neudecker et al., 2017). The role of the inflammasome in IBDs pathogenesis was also confirmed by using a miR-223 deficient animal model of colitis (Neudecker et al., 2017). miR-223 deficient mice develop experimental colitis manifesting with colonic ulceration, inflammatory leukocyte infiltration and tissue injury which resembles closely IBDs (Neudecker et al., 2017). Tissue injury in these mice was linked to an enhanced NLRP3 expression and elevated IL-1β (Neudecker et al., 2017). Treatment of animals with miR-223 mimetics alleviated symptoms of the colitis which coincided with reduced NLRP3 RNA and IL-1β levels (Neudecker et al., 2017). This data presents miR-223 as a novel biomarker and therapeutic target in subsets of IBDs and colitis (Polytarchou et al., 2015).

#### **Rheumatoid arthritis (RA)**

fphar-10-00451 April 29, 2019 Time: 15:11 # 9

Rheumatoid arthritis is a chronic, systemic inflammatory disease affecting joints as well as skin, eyes, lungs, heart, and blood vessels (Scott et al., 2010). It was suggested that RA pathogenesis is related to activation of the NLRP3/IL-1β axis, where inflammasome activation was linked to worsening symptoms of the disease (Xie Q. et al., 2018). It was shown that activation of NLRP3 leads to an abundant expression of IL-1β (Guo et al., 2018), which can trigger T helper type 17 (Th17) cell differentiations and osteoclasts activation in RA (Dayer, 2003; McInnes and Schett, 2011; Zhang et al., 2015b). Th17 cells play a central role in RA pathogenesis, by maintaining chronic inflammation, recruiting neutrophils and promoting joint degradation (Cai et al., 2001; Shahrara et al., 2009; Leipe et al., 2010). Recently, an indirect effect of miR-33 on NLRP3 activation was demonstrated in RA (Xie Q. et al., 2018), which could be explained by miRNA controlled dysregulation of mitochondrial function (Schroder et al., 2010; Zhou et al., 2011; Miao et al., 2014; Ouimet et al., 2015). Xie Q. et al. (2018) suggested that miR-33 increases mitochondrial oxygen consumption and accumulation of reactive oxygen species which upregulates expression of NLRP3 and PCA1 in RA. Also, both miR-33 expression and NLRP3 inflammasome activity were found to be higher in RA monocytes as compared to controls (Xie Q. et al., 2018). These findings indicate that miR-33 could play an indirect role in pathogenesis of RA through NLRP3 inflammasome activation. Additional studies will provide more insight into the miRNA regulation of NLRP3 in RA and its therapeutic and prognostic implications.

#### **Type 1 diabetes (T1D)**

Type 1 diabetes is caused by autoimmune targeted elimination of pancreatic β cells islet (Kloppel et al., 1985). It was shown that TLRs play an essential role in the pathogenesis of T1D (Xie Z. et al., 2018). Upregulated expression of TLR4 as well as increased activity of the downstream targets was demonstrated in monocytes from T1D (Devaraj et al., 2008). Increased expression of activated TLRs was explained as a reaction to a high levels of circulating ligands in TID (Devaraj et al., 2009). Also, epigenetic regulation was associated with an aberrant TLR signaling and an increased IL-1β expression in T1D (Grishman et al., 2012). Several miRNAs were found altered in pre-TID patients, where levels of nine miRNAs (miR-146a, miR-561, and miR-548a-3p, miR-184, and miR-200a) were decreased, and two miRNAs (miR-30c and miR-487a) were increased (Grieco et al., 2018). Supporting these results was data published by Wang G. et al. (2018) demonstrating lower levels of miR-150, miR-146a, and miR-424 compared to controls. One of the most consistent findings was the decreased miR-146a levels in T1D. It appears that miR-146a deficiency could play role in T1D exacerbation and increased IL-1β and IL-18 expression (Bhatt et al., 2016). Increased IL-1β levels could indicate inflammasome activation in T1D, although the role of inflammasome in the disease pathogenesis remains largely unknown.

#### **Type 2 diabetes (T2D)**

Circulating autoantibodies to β cells, self-reactive T cells and the glucose-lowering efficacy of some immunomodulatory therapies are suggestive of the autoimmune nature of the T2D (Itariu and Stulnig, 2014). Interestingly, a role for miRNA regulation of gene expression was demonstrated in T2D, where Balasubramanyam et al. (2011) have shown reduced miR-146a which was associated with increased NF-κB, TNF-α and IL-6 mRNA levels. It is the same miRNA, which was found implicated to pathogenesis of T1D (Xie Z. et al., 2018), indicating potential similarities in the pathogenesis of both diseases. Recently in vivo studies demonstrated that miR-146a deficiency could increase expression of M1 and suppress expression of M2 markers in macrophages collected from patients with diabetes (Bhatt et al., 2016). Macrophage polarization occurs in the presence of IFNγ (M1) or IL-4 (M2) (Nathan et al., 1983; Stein et al., 1992) and is linked to pro-inflammatory and anti-inflammatory activities, respectively. M1 macrophages were shown to support inflammation by producing pro-inflammatory cytokines, including the inflammasome product IL-1β (Bhatt et al., 2016). Therefore, a link could be suggested between low miR-146a levels and inflammasome activation in M1 cells. More investigation is required to identify the connection between miR-146a and inflammasome activation and the role of this in T2D pathogenesis.

#### **Systemic lupus erythematosus (SLE)**

Systemic lupus erythematosus is an autoimmune disease caused by the loss of immune tolerance to ubiquitous autoantigens (Tsokos, 2011). Inflammation plays essential role in SLE pathogenesis (Yang et al., 2014; Rose and Dorner, 2017), where high levels of circulating proinflammatory cytokines are commonly detected (Yao et al., 2016; Mende et al., 2018). Inflammasome activation is proposed as one of the mechanisms underlying increased proinflammatory cytokine level in SLE (Kahlenberg and Kaplan, 2014). This assumption is supported by a report where IL-1β deficient mice were found to be resistant to experimental SLE (Voronov et al., 2006). Also, an increased expression of NLRP3 and AC1 have been reported in SLE nephritis biopsies (Kahlenberg et al., 2011). Kahlenberg and Kaplan (2014) have shown that SLE macrophages are highly reactive to innate immune stimuli, often leading to inflammasome activation. Therefore, targeting inflammasome activity could be a novel approach for SLE treatment. The expression of several miRNAs targeting the inflammasome and its products were found differentially expressed in SLE. For example, Wang et al. (2012) have demonstrated high levels of circulating miR-223, which was shown to inhibit NLRP3, in SLE.

Also, reduced levels of circulating miR-146a, which regulates priming of TLRs, was found in SLE plasma (Wang et al., 2012). Interestingly, expression of miR-23b, which indirectly inhibits IL-1β responses, was shown to be downregulated in inflammatory lesions of SLE patients and animal model (Zhu et al., 2012). More studies are required to determine the role of miRNAs in pathogenesis of SLE and their therapeutic potential.

#### miRNA Regulation of Inflammasome in Neurodegenerative Disorders

Inflammasome products, IL-1β and IL-18, were shown to be essential for the health and functional competence of the nervous system (McAfoose and Baune, 2009; Dinarello et al., 2012). NLRP3 expression was demonstrated in microglia and astrocytes, which could explain the constitutive level of these cytokines in the brain (McAfoose and Baune, 2009; Dinarello et al., 2012; Savage et al., 2012; Minkiewicz et al., 2013; Cho et al., 2014; Lu M. et al., 2014). Interestingly, higher than normal levels of IL-1β and IL-18 were found in several neurodegenerative disorders, suggesting that over-activation of inflammasomes may play a role in pathogenesis of these diseases (Cho et al., 2014; Lu M. et al., 2014; Denes et al., 2015; Mamik and Power, 2017; Song et al., 2017). The significance of miRNA in the regulation of inflammasome activation in the pathogenesis of neurodegenerative diseases remains largely unknown. However, the role of an aberrant miRNA in regulation of NLRP3 expression was previously demonstrated in Parkinson's disease (PD).

Parkinson's disease is a neurodegenerative disease which is characterized by progressive loss of dopaminergic neurons in substantia nigra compacta (Gasser, 2009). It is believed that accumulation of α-Syn fibrillary aggregates in the brain, most notably in the nigral dopaminergic neurons, induces the neuroinflammation (Eriksen et al., 2003). According to Zhou Y. et al. (2016), α-Syn can activate NLRP3 inflammasomes in microglia leading to an increased production of IL-1β. The authors also demonstrated that, miR-7 and miR-30e analogs can inhibit NLRP3 inflammasome mediated neuroinflammation in the brain and protect dopaminergic neurons (Zhou Y. et al., 2016). It appears that the anti-inflammatory effects of miR-7 and miR-30e are associated with their targeting of NLRP3 mRNA in microglial cells. Interestingly, decreased miR-7 and miR-30e expression was demonstrated in PD, which could lead to the loss of the regulatory control of α-Syn induced NLRP3 activation (Li D. et al., 2018).

#### miRNA Regulation of the Inflammasome in Cardiovascular Diseases (CVDs)

The physiological significance of inflammation is confirmed as it facilitates elimination of destructive stimuli and pathogens. However, aberrant inflammatory responses could cause tissue damage, tissue fibrosis and chronic diseases (Liu D. et al., 2018). Inflammation is recognized as a major risk factor for CVDs (Zhou et al., 2018), where chronic inflammasome activation was shown to contribute to the pathogenesis of atherosclerosis, ischemic and non-ischemic heart diseases (Zhou et al., 2018). Therefore, regulation of inflammasome activity using miRNA could be used for treatment and prevention of CVDs. Currently, strong evidence for the role of NLRP3 activation has been demonstrated in pathogenesis atherosclerosis.

Atherosclerosis is a form of CVD characterized by narrowing of the blood vessel lumen due to plaque formation, continuous dyslipidemia and inflammation (Ross, 1993). Chronic inflammation is commonly found in and around the atherosclerotic plaques which has an adverse effect on the arterial wall structure and function (Bernhagen et al., 2007). It is believed that atherogenic lipid mediators, involved in the formation of chronic inflammation in atherosclerotic plaque (Chen et al., 2006), can trigger peripheral blood monocytes migration and differentiation into macrophages within the intima of the arterial wall (Chen et al., 2006). T cells were also detected in atherosclerotic lesions (Kleemann et al., 2008), where, together with activated macrophages, they were shown to secrete proinflammatory mediators such as interferons, interleukins, and proteases (Østerud and Bjørklid, 2003; Shashkin et al., 2005; Tabas, 2005; Chen et al., 2006). IL-1β expression was identified in the early phase of atherosclerotic plaque formation and this stimulates secretion of additional cytokines and chemokines (Kleemann et al., 2008). Therefore, inflammasome activation in macrophages and T cell within the atherosclerotic lesion contributes to the pathogenesis of chronic inflammation.

miR-22, a miRNA inhibiting NLRP3, is decreased in peripheral blood mononuclear cells from coronary atherosclerosis (Chen B. et al., 2016), suggesting that upregulation of this miRNA could have therapeutic potential in CVD. Supporting this assumption, Huang W.Q. et al. (2017) investigated the effect of miR-22 on the NLRP3 inflammasome and endothelial cell damage in an in vivo model of coronary heart disease. The authors demonstrated that miR-22 mimics could decrease the release of inflammatory cytokines such as IL-1β and IL-18 by suppressing NLRP3 expression in monocytes and macrophages (Huang W.Q. et al., 2017). Two additional miRNAs, miR-9 and mir-30e-5p were found to indirectly affect inflammasome activation in atherosclerosis (Wang Y. et al., 2017; Li P. et al., 2018). It appears that miR-9 could indirectly suppress inflammasome activation by targeting an atherogenic lipid mediator, oxidized low-density lipoprotein (oxLDL), in atherosclerosis (Liu W. et al., 2014). In another report, Wang Y. et al. (2017) reported that miR-9 inhibits NLRP3 inflammasome activation induced by oxLDL in human THP-1 derived macrophages and peripheral blood monocytes in an in vitro atherosclerosis model. miR-9 targets Janus kinase 1 (JAK1) pathway (Wang Y. et al., 2017) inhibiting expression of NF-κB p65 which is required for the first step of NLRP3 inflammasome activation (Wang Y. et al., 2017). In addition, miR-30c-5p was linked to an indirect regulation of NLRP3 expression in atherosclerosis (Li P. et al., 2018). Li P. et al. (2018) reported that miR-30c-5p protects human aortic endothelial cells (HAECs) from the oxLDL insult by targeting FOXO3. The authors showed that miR-30c-5p can suppress FOXO3 expression and, consequently, decrease levels of NLRP3, AC1, IL-18 and IL-1β in HAECs (Li P. et al., 2018). As evidence emerges supporting the role of NLRP3 in the pathogenesis of atherosclerosis, targeting the inflammasome becomes an attractive therapeutic approach, where miRNAs could be suitable novel tools.

#### miRNA in Regulation of Inflammasome in Cancer

The role of the inflammasome in tumorigenesis remains controversial. Some reports indicate that NLRP3 inflammasome activation and IL-18 signaling protect against colorectal cancer (Karki et al., 2017), whereas progression of breast cancer, fibrosarcoma, gastric carcinoma, and lung metastasis were shown to be supported by the inflammasome (Okamoto et al., 2010; Kolb et al., 2014). Inflammasome regulation is complex, where multiple factors are implicated, making identification of the key regulatory elements challenging. As the inflammasome involvement in pathogenesis of some malignancies becomes more evident, understanding the regulatory mechanisms could lead to the discovery of novel therapeutic targets for cancer treatment.

#### **Hepatocellular carcinoma (HCC)**

Hepatocellular carcinoma (HCC) is a frequent sequelae of hepatitis B and hepatitis C viral infection (Perz et al., 2006). It is understood that these viruses activate NLRP3 inflammasomes causing hepatocyte pyroptosis, apoptosis and fibrosis (Kofahi et al., 2016). However, HCC tissue analysis failed to detect inflammasome activation; in fact, it was found to be significantly down-regulated when compared to the adjacent normal tissue (Zhu et al., 2011; Wei et al., 2014). To explain this inconsistency, Wei et al. (2014) suggested that NLRP3 expression is dynamic changing during the progression of HCC. It appears that NLRP3 expression was increased in liver cells at the early stages of transformation, while inflammasome levels were decreased in malignant cells when compared to adjacent normal tissue (Wei et al., 2014). Interestingly, levels of miR-223, a negative regulator of NLRP3, were found to be increased in Hep3B cells derived from HCC (Wan et al., 2018). Increased miR-223 was shown to coincide with tumor growth, suggesting a role in post-transcriptional mechanisms in malignant progression. In addition to NLRP3, miR-223 was shown to target erythrocyte membrane protein band 4.1 like 3 (EPB41L3) and FOXO1 (Li and Rana, 2014; Kim et al., 2017). FOXO1 transcription factor binds to the thioredoxin-interacting protein (TXNIP) and regulates genes involved in cell death as well as the oxidative stress responses (Kim et al., 2017). TXNIP interacts with the NLRP3 inflammasome and activates AC1 in murine β-cells (Zhou et al., 2010). In addition, miR-223 appears to be released systemically, where the level of this miRNA in the plasma was significantly lower in HCC cases (Giray et al., 2014). In addition to miR-223, decreased circulating miR-30e, which also targets NLRP3, was found in HCC cases (Bhattacharya et al., 2016). Therefore, it could be suggested that analysis of serum levels of miR-223 and miR-30e could be used for diagnosis of HCC as well as an indicator of the efficacy of anticancer therapeutics.

#### **Colorectal cancer (CRC)**

Data on the role of NLRP3 in colorectal cancer (CRC) pathogenesis is inconsistent, where some evidence suggests a pro-tumorigenic role for the inflammasome, while others identified that the inflammasomes protects against tumor (Allen et al., 2010; Huber et al., 2012; Guo et al., 2014; Wang et al., 2016). Inflammasome expression analysis also demonstrated contradicting results where Wang et al. (2016) reported high NLRP3 in mesenchymal-like colon cancer cells, while Allen et al. (2010) demonstrated decreased inflammasome expression in colitis-associated cancer. Inflammasome contribution to tumorigenesis varies depending on the target cell type in the intestinal tissue (Lissner and Siegmund, 2011). According to Lissner and Siegmund (2011), inflammasome activation is required to maintain integrity of the epithelium. However, aggravated activation of the inflammasome stimulates intestinal inflammation, which could have a detrimental effect on epithelium permeability and increase its leakage (Lissner and Siegmund, 2011). It was identified that damage to the intestinal epithelium could trigger NLRP3 activation and secretion of IL-18, a proinflammatory cytokine (Huber et al., 2012). Subsequently, it was shown that IL-18 could reduce the expression of IL-22 binding protein (IL-22BP) and increase levels of IL-22 (Huber et al., 2012). Although IL-22 is protective against malignancies, aberrant over expression of IL-22 could trigger gut epithelial cell transformation and CRC development (Huber et al., 2012). Therefore, it is believed that IL-18, a NLRP3 product, has a promoting role in CRC development (Huber et al., 2012).

Targeting the inflammasome was suggested as a potential approach for treatment of CRC (Guo et al., 2014). NLRP3 expression was shown to be regulated by multiple miRNAs in various diseases (Haneklaus et al., 2012; Feng et al., 2018; Wan et al., 2018; Xie Q. et al., 2018). However, the role of miRNAs in cancer pathogenesis is not straight forward. There are inconsistent results regarding the expression status of miR-223, a known regulator of NLRP3 expression, in CRC cell lines and primary tumors. In a clinical study, the expression of miR-223 was found to be significantly higher in stage III/IV patients (Ding J. et al., 2018). However, levels of miR-223 vary significantly in colon tumor derived cell lines (Ding J. et al., 2018). Wu et al. (2012) reported reduced expression of miR-223 in a HCT116, a CRC cell line. In contrast, several research groups demonstrated up-regulation of miR-223 in CRC cell lines and primary tissues (Wang F. et al., 2017; Ju et al., 2018; Wei et al., 2018). Similar to these results, Ju et al. (2018) demonstrated up-regulation of miR-223 in SW620, a CRC cell line. It was identified that high expression of miR-223 suppresses FoxO3a and enhances cancer cell proliferation (Ju et al., 2018). It appears that the protumorigenic effect of Foxo3a is via NF-κB activation, which is essential for upregulation of the inflammasome linked proinflammatory signaling pathways (Thompson et al., 2015).

Unlike miR-223, data on miR-22 expression status in CRC consistently demonstrates that miR-22 expression is significantly lower in CRC tissues and cell lines (Zhang et al., 2012, 2015a; Li B. et al., 2013; Xia et al., 2017; Liu Y. et al., 2018). Also, absence of miR-22 was shown to positively correlate with increased cancer cell proliferation, migration, invasion, and metastasis (Zhang et al., 2012, 2015a; Li B. et al., 2013; Xia et al., 2017; Liu Y. et al., 2018). Multiple genes were identified as targets for miR-22 including TIAM1 (Li B. et al., 2013), BTG1 (Zhang et al., 2015a), HuR (Liu Y. et al., 2018), and SP-1 (Xia et al., 2017). Among these genes, only SP-1 gene expression was linked to inflammasome regulation (Hofmann et al., 2015). According to Hofmann et al. (2015), Sp-1 protein could contribute to

NLRP3 inflammasome activation in monocytes in chronic recurrent multifocal osteomyelitis. However, the role of Sp-1 in activation of the NLRP3 inflammasome in CRC tumor tissues and monocytes remains largely unknown. Recent finding revealed that, in addition to miR-22, another negative regulator of NLRP3, miR-30e, is absent in CRC tumors as compared to normal colon tissues (Laudato et al., 2017). However, the role of miR-30e in CRC pathogenesis remains unknown.

#### **Gastric cancer (GC)**

It was shown that NLRP3 inflammasome activation promotes gastric cancer (GC) cells proliferation (Li S. et al., 2018). Over expression of miR-223 supports GC invasion and metastasis in primary GC tumors (Haneklaus et al., 2012). Additionally, Li S. et al. (2018) reported that increased NLRP3 expression in GC tumors and macrophages negatively correlates with miR-22 expression. The authors also demonstrated that constitutive expression of miR-22 dramatically decreases NLRP3 mRNA expression and IL-1β secretion in macrophages (Li S. et al., 2018). Therefore, the effect of targeting NLRP3 expression with miRNAs in tumors and immune cells may vary depending on tumor and/or cell type.

#### **Oral squamous cell carcinoma (OSCC)**

High NLRP3 expression was found in oral squamous cell carcinoma (OSCC) cells and tissues (Wang H. et al., 2018). A role for NLRP3 supporting OSCC proliferation and growth was demonstrated in several reports. Wang G. et al. (2018) demonstrated a positive correlation between NLRP3 expression and tumor size, lymph node status and IL-1β expression in OSCC tissue specimens and in vivo models of OSCC. Also, the authors showed that, silencing of NLRP3 in OSCC cell lines reduced cell proliferation, migration, and invasion in vitro (Wang H. et al., 2018). Additionally, high expression of the NLRP3 inflammasome mediates chemoresistance in OSCC (Feng et al., 2018). Therefore, downregulation of NLRP3 could have a therapeutic potential in OSCC.

Surprisingly, high expression of miR-223, which targets NLRP3, was found in primary OSCC tissue (Manikandan et al., 2016). In silico analysis identified a Ras Homolog Family Member B (RHOB) as a potential target for miR-223 in OSCC (Manikandan et al., 2016). It appears that miR-223 could indirectly suppress NLRP3 and TLR4/NF-κB signaling via RHOB (Yan et al., 2019). These data provide a novel potential target for OSCC treatment, where miR-223 inhibition of NLRP3 could be attained through RHOB.

Overexpression of miR-22 in OSCC was shown to reduce NLRP3 activation and decrease OSCC malignancy (Feng et al., 2018). miR-22 levels were shown to be inversely correlated with NLRP3 expression and miR-22 levels were significantly lower in OSCC compared to adjacent non-cancerous tissue (Feng et al., 2018). The inhibitory effect of miR-22 on OSCC migration was confirmed using a lentiviral expression system. As expected an inhibitor of miR-22 promoted OSCC spread (Feng et al., 2018). The 3<sup>0</sup> -UTR of the NLRP3 gene was identified as a miR-22 target site (Feng et al., 2018). It appears that NLRP3 promotes OSCC growth and tumor spread, which makes miR-22 a potential therapeutic target for cancer treatment. Two miRNAs, miR-223 and miR-22, were identified as inhibiting the inflammasome and, subsequently, suppressing tumor growth. Therefore, the anti-tumor effect of these molecules in OSCC warrants further investigation.

#### **Cervical cancer (CC)**

Human papillomavirus (HPV) infection and persistent chronic inflammation were identified as fundamental for the pathogenesis of cervical cancer (CC) (de Castro-Sobrinho et al., 2016; Kriek et al., 2016). HPV can cause chronic inflammation by inducing TLR4 expression and impairing the TLR4-NF-κB pathway (Wang et al., 2014; He A. et al., 2016).

Wu et al. (2012) reported reduced expression of miR-223, which targets NLRP3, in the CC cell line HeLa. The authors also demonstrated that over-expression of miR-223 inhibits tumor cell proliferation by targeting FOXO1 (Wu et al., 2012). In addition, another direct post-transcriptional regulator of NLRP3, miR-22, was found to be down-regulated in CC cell lines and tissues (Xin et al., 2016; Wongjampa et al., 2018). Furthermore, Wongjampa et al. (2018) reported an inverse correlation between histone deacetylase 6 (HDAC6) and miR-22. It was previously shown that HDAC6 directly binds to NLRP3 via its ubiquitin-binding domain to regulate NLRP3 inflammasome expression (Hwang et al., 2015). As NLRP3 plays a role in the pathogenesis of HPV induced chronic inflammation, miR-223 and miR-22, both of which regulate inflammasome activation, could be potential therapeutic tools for the treatment of CC.

#### **Glioblastoma (GBM)**

High NLRP3 inflammasome activation and high levels of inflammasome products are found in malignant glioblastoma (GBM) (Basu et al., 2004; Tarassishin et al., 2014). Increased IL-1β, a major NLRP3 inflammasome product, was linked to the release of VEGF and MMPs, angiogenic factors, in human astrocytes and GBM cells (Suh et al., 2013). Therefore, it could be suggested that inflammasome activation favors GBM growth and spread.

Several miRNAs were shown to regulate inflammasome expression, where decreased miRNA levels could promote GBM growth and invasion. Ding Q. et al. (2018) demonstrated that miR-223, which is effective at reducing NLRP3 inflammasome levels in several tumors (Wu et al., 2012), was decreased in GBM tissues (Ding Q. et al., 2018). However, a conflicting report from Cheng et al. (2017) indicated that miR-223 is overexpressed in GBM cell lines. Similar findings were also reported in GBM stem like cells and GBM tissues (Huang B.S. et al., 2017). Similarly there are conflicting data regarding miR-223 targets and phenotypic impacts. A miR-223-3p mimic inhibited tumor cell proliferation and migration, effects that were due to a reduction in proinflammatory cytokines IL-1β and IL-18 in GBM cell lines (Ding Q. et al., 2018). Also, nuclear factor I-A (NFIA) was a target of miR-223 in GBM cell lines and was found to decrease tumorigenesis in the CNS (Glasgow et al., 2013). The pro-tumorigenic effect of miR-223 was linked to suppression

of the tumor suppressor paired box 6 (PAX6) (Cheng et al., 2017). By targeting PAX6, miR-223 could promote GBM stem cell chemotherapy resistance (Huang B.S. et al., 2017). The mechanism underlying the diverse effects of miR-223 on GBM growth and metastasis remains largely unknown. However, it could be suggested that the stage of tumorigenesis plays a role in the effect of miR-223 in GBM.

Levels of miR-22 and miR-30e, two post-transcriptional regulators of NLRP3, are low in GBM tissues (Li W.B. et al., 2013; Chakrabarti et al., 2016; Chen H. et al., 2016). In addition to targeting NLRP3, miR-22 can also directly target the 3<sup>0</sup> -UTRs of SIRT1 (Li W.B. et al., 2013), and miR-22 mimics decrease the expression of SIRT1 protein in GBM cell lines (Li W.B. et al., 2013). Interestingly, several studies have demonstrated that SIRT1 can suppress NLRP3 (Ma et al., 2015; Jiang et al., 2016; Zhou C.C. et al., 2016). It could be proposed that the decreased levels of miR-22 could fail to control NLRP3 expression, which could enable GMB tumorigenesis.

### FUTURE ASPECTS FOR CLINICAL APPROACHES

The role of the NLRP3 inflammasome in the pathogenesis of several diseases was demonstrated, including CAPS, autoimmune disorders and cancers (Aganna et al., 2002; Martinon et al., 2006; Masters et al., 2009; Bauer et al., 2010; Wen et al., 2011). An increased IL-1β level, commonly found in these diseases, is a strong indicator of NLRP3 inflammasome activation. Also, the body of evidence suggests that IL-1β plays a central role in disease pathogenesis. Therefore, targeting IL-1β, a NLRP3 inflammasome product, appears to be a rational therapeutic approach. The efficacy of anti-IL-1β therapy was demonstrated in CAPS, where both the symptoms and severity of the disease were alleviated using either an IL-1β receptor antagonist or anti-IL-1β antibodies (Hoffman et al., 2008; Dinarello, 2009; Lachmann et al., 2009). A similar approach targeting IL-1β was successfully applied to treat NLRP3 inflammasome associated autoimmune diseases and cancer (Larsen et al., 2007; Lust et al., 2009). These data provide compelling evidence for the NLRP3 inflammasome as a potential therapeutic target for treatment of the diseases associated with an elevated level of IL-1β. In this respect, miRNAs have therapeutic potentials as they could target NLRP3 preventing its expression and, consequently, averting IL-1β production.

miRNA based replacement and silencing therapeutic approaches were tested in several preclinical and clinical studies (Li and Rana, 2014). miRNAs and miRNA-targeting oligonucleotides approaches (mimic and/or anti-miR technologies) appear to be more effective when compared to small-molecule drugs due to their ability to effect concurrently multiple gene targets (Li and Rana, 2014). Anti-miR-122 oligonucleotide, Miravirsen, was the first miRNA-based therapeutic used to treat hepatitis c infection (Lindow and Kauppinen, 2012; van der Ree et al., 2016). Currently Miravirsen is in a phase II clinical trial (van der Ree et al., 2016). Several phase I clinical trials and pre-clinical studies using miRNA-targeting oligonucleotide technologies targeted to Let-7, miR-10b, miR-21, miR-34, miR-155, miR-221, and others, have demonstrated positive results (Moles, 2017). miRNA-targeting oligonucleotides are designed to bind to their targeted miRNA (Li and Rana, 2014). miRNAs generally target more than one gene in the same signaling pathway (Li Z. et al., 2011; Li and Rana, 2014). This feature of miRNAs makes them valuable as therapeutic candidates (Li and Rana, 2014).

However, there are still multiple obstacles to overcome, including target specificity and the potential toxicity of miRNA-targeting oligonucleotides (Merhautova et al., 2016). First, the limited specificity, anti-miRs generally target nucleotide sequences on miRNAs which can be present on multiple miRNAs within the same family (Hogan et al., 2014). Chemical modifications of anti-miRs have been suggested to improve their specificity (Hogan et al., 2014). Second, when administered without a carrier molecule, their effect may be limited and they can be cleared by the liver and kidney (Bennett and Swayze, 2010). Third, anti-miRs can be sensed and eliminated by receptors of the innate and adaptive immune responses (Diebold et al., 2004; Heil et al., 2004). To overcome this limitation, tissue specific antibody coated chemically engineered polymer-based nanoparticles and carrier proteins have been developed to improve the specificity and efficacy of delivery. For example, the therapeutic efficiency of miR-223 was improved by using nanoparticle lipid emulsions as a delivery method, in animal model of colitis (Neudecker et al., 2017). These exciting results demonstrate great potential for miRNA-based treatments of diseases linked to NLRP3 dysfunction.

Our understanding of the role of the inflammasome in disease pathogenesis is still limited and is hampering development of the miRNA targeting therapeutics against the inflammasome. However, exciting discoveries in fundamental and preclinical research in recent years have demonstrated great potential for miRNA targeting in the treatment of diseases linked to NLRP3 dysfunction.

### AUTHOR CONTRIBUTIONS

GT and SK contributed to the conception and design of the study. ZG organized the database. SK wrote the first draft of the manuscript. GT, EM, ZG, AM, AR, and SK wrote sections of the manuscript. All authors contributed to manuscript revision, read and approved the submitted version.

### FUNDING

AR was supported by a personal state assignment 20.5175.2017/6.7 of the Ministry of Education and Science of Russian Federation. Kazan Federal University was supported by the Russian Government Program of Competitive Growth (5-100).

### REFERENCES

fphar-10-00451 April 29, 2019 Time: 15:11 # 14




and interleukin 1beta in salmonella-infected macrophages. Nat. Immunol. 7, 576–582. doi: 10.1038/ni1346


effect of V8 in LPS-induced human cervical cancer SiHa cells. Inflammation 39, 172–181. doi: 10.1007/s10753-015-0236-8


autoinflammatory syndromes. Pathog. Dis. 76:fty020. doi: 10.1093/femspd/ fty020



expression in patients with glioblastoma multiforme. Chin. Med. J. 126, 2881–2885.



response in leishmania amazonensis-infected macrophages. Front. Immunol. 9:2792. doi: 10.3389/fimmu.2018.02792



results in decreased microRNA-122 levels without affecting other microRNAs in plasma. Aliment. Pharmacol. Ther. 43, 102–113. doi: 10.1111/apt.13432



**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 Tezcan, Martynova, Gilazieva, McIntyre, Rizvanov and Khaiboullina. 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.

# Zeolites as Carriers of Antitumor Ribonuclease Binase

#### Vera Khojaewa<sup>1</sup> , Oleg Lopatin<sup>2</sup> , Pavel Zelenikhin<sup>1</sup> \* and Olga Ilinskaya<sup>1</sup>

<sup>1</sup> Department of Microbiology, Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, <sup>2</sup> Department of Mineralogy and Lithology, Institute of Geology and Petroleum Technologies, Kazan Federal University, Kazan, Russia

Natural and synthetic zeolites have many applications in biomedicine and nutrition. Due to its properties, zeolites can absorb therapeutically active proteins and release them under physiological conditions. In this study we tested the clinoptilolite, chabazite, and natrolite ability to be loaded by antitumor ribonuclease binase and the cytotoxicity of the obtained complexes. We found the optimal conditions for binase loading into zeolites and established the dynamic of its release. Cytotoxic effects of zeolitebinase complexes toward colorectal cancer Caco2 cells were characterized after 24 and 48 h of incubation with cells using MTT-test. Zeolites were toxic by itselfs and reduced cells viability by 30% (clinoptilolite), 40% (chabazite), and 70% (natrolite) after 48 h of incubation. Binase complexes with clinoptilolite as well as chabazite always demonstrated enhanced toxicity (up to 57 and 60% for clinoptilolite and chabazite, respectively) in comparison with binase and zeolites separately. Our results contribute to the perspective development of binase-based complexes for therapy of colorectal cancer for or the treatment of malignant skin neoplasms where the complexes can be used in pasty form.

Keywords: zeolites, chabazite, clinoptilolite, natrolite, cytotoxic RNAse, binase, immobilization

### INTRODUCTION

There are about 40 naturally occurring tectosilicate minerals in zeolite group, the most commonly mined isometric forms include chabazite and clinoptilolite, the fibrous form is mainly represented by natrolite. Chemical differentiation of zeolites is related to the ratio of SiO2/Al2O<sup>3</sup> and water content. Zeolites with an Al/Si ratio of 0.20–0.40 are leafy, others with an Al/Si ratio up to 0.50 are isometric or mostly isometric, and with an Al/Si ratio of 0.60–1.00 are predominantly fibrous. The crystalline structure of zeolites is formed by tetrahedral SiO2/<sup>4</sup> and AlO2/<sup>4</sup> groups, united into a three-dimensional framework pierced by cavities and channels which size is 0.2–1.5 nm. The internal cavities and the channels are filled with molecules of water. The open frame-cavity structure of zeolites has a negative charge, which is compensated by counterions (metal, ammonium, alkylammonium, and other cations).

Zeolites are capable to exchange cations and reversible dehydrate. Pores in zeolite let small molecules pass through but trap larger ones; that is why they are referred as molecular sieves. Alumina-rich zeolites are attracted to polar molecules such as water, while silica-rich zeolites work better with nonpolar molecules. Advances in material synthesis lead to engineering of hierarchically organized zeolites with multilevel pore architecture which combine unique chemical functionality with efficient molecular transport (Mitchell et al., 2015). Natural and synthetic zeolites are used as

#### Edited by:

Ali H. Eid, American University of Beirut, Lebanon

#### Reviewed by:

Marina A. Zenkova, Institute of Chemical Biology and Fundamental Medicine (RAS), Russia Irina Yurievna Petrushanko, Engelhardt Institute of Molecular Biology (RAS), Russia Maria Ignat, Institute of Macromolecular Chemistry "Petru Poni", Romania

> \*Correspondence: Pavel Zelenikhin pasha\_mic@mail.ru

#### Specialty section:

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

Received: 27 December 2018 Accepted: 08 April 2019 Published: 03 May 2019

#### Citation:

Khojaewa V, Lopatin O, Zelenikhin P and Ilinskaya O (2019) Zeolites as Carriers of Antitumor Ribonuclease Binase. Front. Pharmacol. 10:442. doi: 10.3389/fphar.2019.00442

**55**

drying agents, as detergents, and in water and air purifiers. Zeolites are also marketed as dietary supplements to treat cancer, diarrhea, autism, herpes, and hangover, and to balance pH and remove heavy metals in the body. In vivo studies, micronized zeolite has been shown to reduce the spread of cancer and increase the effect of the chemotherapy drug doxorubicin (Zarkovic et al., 2003). Up today, zeolites have not been studied as an anticancer drug in human clinical trials. A review by Memorial Sloan-Kettering Cancer Center concluded that none of the benefits seen in animals occurs in humans<sup>1</sup> . However, different zeolite forms must be distinguished: fibrous mordenite is not allowed for medical use, erionite inhalation toxicity is associated with high incidence of malignant mesothelioma (Elmore, 2003; de Assis et al., 2014). At the same time, TMAZ <sup>R</sup> , a natural isometric zeolite clinoptilolite with enhanced physicochemical properties, is the basis of the dietary supplements Megamin <sup>R</sup> and Lycopenomin <sup>R</sup> ("Tribo Ming," Croatia), which have demonstrated antioxidant activity in humans. Litovit <sup>R</sup> ("Nov," Novosibirsk, Russia) that removes heavy metals and has radioprotective properties is also manufactured on the basis of clinoptilolite. The composition synthesized from naturally occurring non-toxic zeolites was patented in United States against buccal mucosa and lung squamous epithelial cell cancers (Kaufman, 2001).

Taken together, this data indicates that adsorptive and ionexchange properties of some zeolites could be applied in medical practice. In our study, several zeolites allocated as possible candidates for loading of anticancer therapeutics. We tested isometric clinoptilolite and chabazite ability to absorb therapeutic protein and realize it, in comparison to this ability of fibrous natrolite. We chose binase (ribonuclease (RNase) from Bacillus pumilus) as a therapeutic protein. RNases are potential antitumor drugs due to their cytotoxicity and due to their influence at some tumor cells functions. RNases have demonstrated the ability to overcome multidrug resistance and to enhance the cytotoxicity of a variety of anticancer agents (Suri et al., 2007; Zelenikhin P.V. et al., 2016). Binase triggers apoptotic response in cancer cells expressing RAS oncogene which is mutated in a large percentage of prevalent and deadly malignancies (Ilinskaya et al., 2001; Cabrera-Fuentes et al., 2012, 2013). Other microbial RNases, cationic mutants of RNAse Sa, for example, possess similar selective activity to oncotransformed cells (Ilinskaya et al., 2002). The specific antitumor effect of binase toward RAStransformed cells is due to its direct binding of RAS protein and inhibition of downstream signaling (Ilinskaya et al., 2016). The expression of oncogenes, in particular, AML1-ETO and kit, was shown to determine the selective sensitivity of cells to the binase action. Moreover, the anti-metastatic effect of binase was demonstrated in animal models. Binase at doses of 0.1– 1 mg/kg, which produced effective suppression of tumor growth and metastasis, showed positive effect on the liver of tumorbearing mice expressed in a significant reduction of the liver parenchyma destructive changes, and return to the normal level the liver regenerative potential (Sen'kova et al., 2014). Thus, this bacterial RNase can be considered a perspective antitumor agent because of its targeted activity toward certain oncogenes expressing cancer cells.

The present study is aimed at the search for a biocompatible mineral carrier that allows the safe delivery and long-term action of binase needed for treatment of ras-expressing malignances, especially colorectal cancer. The delivery of proteins to the intestine is known to be complicated by their degradation in digestive tract with subsequent loss of therapeutic activity. Therefore, the prolonged release of antitumor agents from composite pills or rectal suppositories can provide certain advantages. Similarly, these advantages are inherent in therapeutic application of pasty form for the treatment of malignant skin neoplasms. Here, we found the optimal conditions for binase loading into zeolites and established the dynamic of its release. Cytotoxic effects of zeolite-binase complexes toward colorectal cancer cells were compared with cytotoxicity of enzyme or zeolite. Our results contribute to the perspective development of binase-based complexes for therapy of colorectal cancer.

### MATERIALS AND METHODS

#### Binase

The guanyl-preferring RNase from B. pumilus, binase (monomer of 12.2 kDa, 109 amino acid residues, pI 9.5), was isolated from culture fluid of native binase producer as homogenous protein using the three-step procedure described earlier (Dudkina et al., 2016). The binase catalytic activity was determined by measurement of high-polymeric yeast RNA hydrolysis products according to modified method of Anfinsen (Kolpakov and Il'inskaia, 1999).

#### Zeolites

Chabazite [(Ca,Na2,K2,Mg)Al2Si4O<sup>12</sup> × 6H2O], the mineral of trigonal syngony, crystallizes in the triclinic crystal system with typically rhombohedral shaped crystals. The crystals are typically twinned, and both contact twinning and penetration twinning may be observed. Crystals of local chabazite up to 5 cm in size have pseudocubic forms, are pale orange with pearly tint and are characterized by a high degree of stoichiometry. In our study we used the samples from Sokolovo-Sarbaisky ore complex, Kazakhstan.

Clinoptilolite [(Na,K,Ca)2−3Al3(Al,Si)2Si13O<sup>36</sup> × 12H2O] forms as white to reddish tabular monoclinic tectosilicate crystals. We used samples from Tatar-Shatrashan deposit of zeolite-bearing rocks, Russia. This mineral of the monoclinic syngony exists on the specified deposit in a fine-dispersed state, which is part of a polymineral aggregate consisting of a clayey and siliceous phase (the so-called zeolite-bearing rock). The maximum amount of zeolite in this unit can reach 50%.

Natrolite [Na2Al2Si3O<sup>10</sup> × 2H2O] often occurs in compact fibrous aggregates, the fibers having a divergent or radial arrangement. Natrolite is a mineral of rhombic syngony, in the zones of metasomatic processing of alkaline igneous rocks of the Kola Peninsula forms large (up to 1 m) mono- and polycrystalline aggregates of snow-white color with a characteristic silky shine.

<sup>1</sup>https://www.mskcc.org/cancer-care/integrative-medicine/herbs/zeolite

We used natrolite from the Khibiny Mountains of the Kola Peninsula, Russia.

### Loading/Unloading Procedure

Zeolite powder obtained after grinding in an electric mill was treated with concentrated filtered hydrochloric acid HCl to remove various impurities, washed with MQ-water and dried using dry heat oven at 160◦C. Each sample (5 mg) was mixed with binase solution in 96% ethanol (1 mg/ml), thoroughly vortexed (V-1 plus, Biosan, Latvia) until a homogeneous suspension, and then sonicated in ice for 5 min, 35 kHz, 130 W (Sapphire, Russia) to disintegrate the aggregates. Afterward samples were incubated for 2 h with gentle shaking (Mini Rocker-Shaker MR-1, Biosan, Latvia) at 25◦C. To estimate the part of non-immobilized binase, protein concentration by optical density at 280 nm and RNase catalytic activity of the supernatant obtained after centrifugation (5 min, 4300 g, Eppendorf 5415R, Germany) were measured. The sediments were dried at 50◦C and stored at room temperature. To analyze the release of immobilized enzyme, the samples were suspended in MQ-water and incubated at room temperature for 2, 4, or 6 h. After centrifugation, the binase catalytic activity and protein concentrations were measured in the supernatant.

### Cell Cultures

Colon adenocarcinoma cells (Caco2) were obtained from Russian cell culture collection (Saint-Petersburg, Russia). Cells were grown in RPMI 1640 medium supplemented with penicillin (100 U/mL), streptomycin (100 U/mL), 2 mM glutamine (Sigma-Aldridge, United States), and 10% fetal bovine serum (HyClone, United States) at 37◦C in a humidified atmosphere with 5% CO2. Cells were seeded into 96-well plates and grown 12 h; then tested samples dissolved in fresh medium were added into plates. After 24 and 48 h of incubation the MTT assay was performed.

### MTT-Assay

Cell viability was measured according to mitochondrial dehydrogenase activity tested by standard procedure based on the reduction of MTT tetrazolium dye. Cells (10<sup>4</sup> per well in 96-well plate, CELLTREAT Scientific Products, United States) have grown overnight, then cultural fluid was discarded and fresh medium with test samples (or with an equivalent volume of water for negative control) was added. After 24 and 48 h culture medium was replaced with dimethyl sulfoxide (Sigma-Aldrich, United States) to dissolve formazan crystals, when absorption was measured at 570 nm (xMark, Bio-Rad, United States). As a positive control inducing cell death 1% Triton was used.

### Transmission Electron Microscopy

Zeolite samples with 96% ethanol solution were sonicated during 10 min, 35 kHz, 130 W (Sapphire, Russia) to disintegrate the aggregates. A droplet of diluted zeolite samples was placed onto carbon-coated grids and left to evaporate. Specimens were inspected using a Hitachi HT7700 Exalens transmission electron microscope (Hitachi High-Tech Science Corporation, Japan) at resolution 1.4 Å. TEM bright field images were recorded at 100 kV accelerating voltage using a AMT XR-81 CCD camera (3296 × 2742, 8 megapixel, 5.5 mm pixel size).

### Statistics

Statistical data analysis and plotting were performed by means of GraphPad Prism6 software (United States). The statistically significant level was taken as p ≤ 0.05.

## RESULTS

Chabazite is morphologically represented by small oval or pseudocubic particles with a diameter about 100∼200 nm aggregated into regular round-shaped particles (Ø ∼2 µm). The same small particles are typical for clinoptilolite, but they form amorphous structures of different size. Small particles of natrolite are partially leafed or polygonal forms (**Figure 1**).

FIGURE 1 | TEM images of three different zeolites (Chabazite – A, Clinoptilolite – B, and Natrolite – C) grinded in an electric mill up to particles of micrometer size. bar = 200 µm.

Finely crushed samples of three different zeolites in the form of micrometer particles were used for binase immobilization. Initially, during the selection of binase loading conditions we used aqueous solution of enzyme but the measutment of unloaded protein concentration and RNase activity of the supernatant showed the lack of enzyme immobilization. Therefore we used 96% ethanol to solve the enzyme before loading. RNase activity in ethanol solution was almost the same as in water, 1.116 ± 0.013 × 10<sup>6</sup> units/mg and 1.588 ± 0.020 × 10<sup>6</sup> units/mg, correspondingly. More than 80% of the protein was found to adsorbe on all zeolites, whereby residual catalytic activity measured in supernatant was very low. The best results were obtained with chabazite (**Table 1**). Full release of binase from chabazite takes 6 h, for clinoptilolite this time period is 4 h. Natrolite kept residual amount of protein more than 6 h. The main part of protein (more that 80% of immobilized one) released from all three zeolites was found in solution already after 2h of incubation (**Table 2**). RNase activity of released binase was comparable to the activity of pure binase in water. Staying in natrolite reduced the catalytic activity of the enzyme released after 2 h up to 57%. This effect disappeared after 4 h of incubation. Opposite, staying inside chabazite slightly activated the binase catalytic activity (**Table 2**).

The cytotoxicity of pure zeolites and zeolites loaded with binase was studied on human colon adenocarcinoma cell line Caco2. Each type of zeolites (300 µg/ml) was examined on possible toxic effects after 24 and 48h of incubation with growing cells (**Figure 2**). After addition of the same amount of water used for zeolite suspension the cells viability reduced on 18% compared to growth without any supplements. Pure chabazite inhibited cell viability by less than 40% during all time of cultivation, clinoptilolite showed inhibitory effect of approximately 50% after 24 h decreased after 48 h up to 30%. Natrolite was more toxic, its inhibitory effect increased from 30% at 24 h to 70% at 48 h of cell growth.

Pure binase at concentration 100 µg/ml reduced cell viability by 60% only after 48 h of incubation. Higher concentration (300 µg/ml) affected cell viability already after 24 h (inhibition reached 40%), after 48 h inhibitory effect was 58% (**Figure 2**). Binase immobilized in natrolite or clinoptilolite increased their toxicity during 24 h, then this increase for natrolite, but not for clinoptilolite, was abolished. Complexes of clinoptilolite as well as chabazite with binase always demonstrated enhanced toxicity in comparison with binase and zeolites separately (**Figure 2**).

### DISCUSSION

Binase possesses selective toxicity toward certain tumor cells in vitro (Mitkevich et al., 2011, 2013, 2015; Zelenikhin P. et al., 2016; Zelenikhin P.V. et al., 2016; Makeeva et al., 2017) and in vivo (Mironova et al., 2013; Sen'kova et al., 2014).

The expression of certain oncogenes (ras, kit, AML1-ETO) is a marker of tumor cells susceptibility to binase apoptogenic action. In some cases RNases catalytic activity may be an important factor for their cytotoxicity manifestation (Ilinskaya and Vamvakas, 1997; Makarov and Ilinskaya, 2003). However, the sensitivity of malignant breast cancer cells to binase apoptosis inducing effect was not shown to correlate with the level of cellular RNA catalytic degradation (Zelenikhin P. et al., 2016). This effect was also demonstrated for oncogene kit transformed cells (Mitkevich et al., 2010). Using quantitative RT-PCR with RNA samples isolated from the binase-treated transgenic myeloid progenitor cells FDC-P1-N822K expressing the activated kitoncogene (mutation Asn822Lys), we have found that the amount

TABLE 1 | The amount of binase loaded onto zeolite and silica samples from ethanol solution<sup>a</sup> .


<sup>a</sup>The protein concentration was measured using absorption spectrophotometry at 280 nm (SmartSpec Plus, Bio-Rad, United States); the catalytic activity was estimated as described in section "Materials and Methods." The initial concentration of protein (1 mg/ml) and RNase catalytic activity in ethanol solution (1.116 ± 0.013 × 10<sup>6</sup> units/mg) used for enzyme loading was taken for 100%. Values are means ± SD. Experiments were performed in triplicate with five independent replications in each series.

TABLE 2 | The amount of binase protein released from zeolite samples into MQ-water and catalytic activity of the released enzyme.


<sup>a</sup>The amount of loaded binase was taken for 100%. <sup>b</sup>Catalytic activity RNase dissolved in MQ-water (1.588 ± 0.02 × 10<sup>6</sup> units/mg) was taken for 100%. <sup>c</sup>n – Not measured. Values are means ± SD. Experiments were performed in triplicate with five independent replications in each series.

of mRNA of the kit oncogene gene was reduced by half. This means that binase effect to tumor cells is specific and is determined by presence of cells specific molecular targets, which can be certain RNA as well as proteins, in particular, RAS (Ilinskaya et al., 2016).

Therefore, its delivery and prolonged action could have benefits during application against cancer.

Zeolites are a group of calcium and sodium aqueous aluminosilicates similar in composition and properties. In the gut, these silicates could act as adsorbents, catalysts, detergents or anti-diarrheic agents to their absorption potential and ionexchanger properties. Zeolites themselves are widely used in agriculture as adsorbents. In animals, zeolite supplementation of feed resulted in a reduction in number of poultry pathogens without damaging the beneficial bacteria (Prasai et al., 2017). Dietary administration of small particle size clinoptilolite can effectively reduce concentration of aflatoxins in dairy cattle milk (Katsoulos et al., 2016). So, the detoxificant role of zeolites is already evident in agro and in zoothecnical fields [for review see Laurino and Palmieri, 2015). We started our study from two simple approvals. First, clinoptilolite application in medicine is allowed, preparations "Tribo Ming" (Croatia), and "Nov" (Russia) based on this zeolite are available for purchase in pharmacies. Natural clinoptilolite with enhanced physicochemical properties is the basis of the dietary supplements Megamin and Lycopenomin, which have demonstrated antioxidant activity in humans (Ivkovic et al., 2004).

We have also studied the possibility of other zeolites, chabazite and natrolite to serve as carriers for binase. Chabazite was studied previously as an agent for wastewater purification (Lee et al., 2016; Montégut et al., 2016), natrolite was described as an environmentally benign catalyst (Nasrollahzadeh et al., 2017). Our results have shown that all three zeolites used at this study have the possibility to absorb binase, an antitumor bacterial protein. The zeolites crystalline structure is formed by tetrahedral SiO2/<sup>4</sup> and AlO2/4, groups, joined by common vertices into three-dimensional framework, penetrated by cavities and channels 2–15 Å in size. The surface of zeolites has a negative charge, compensated by counterions (metal cations, ammonium, and other ions) and water molecules. Washing the zeolites with acid allowed us to get rid of carbonate impurities, and the subsequent washing with water and alcohol removed counterions and released the negative charge necessary for sorption of cationic binase (PI 9.5) due to electrostatic interactions.

We found the conditions suitable for loading more than 80% of protein from ethanol solution during 2 h with gentle shaking by room temperature. Natrolite demonstrated slowly decreasing absorption ability compared to clinoptilolite and chabazite. It probably could be connected to its fibrous nature (**Table 1**). On the other hand, binase was released from natrolite more slowly than from clinoptilolite and chabazite and did not reached 100% output during 6 h (**Table 2**). It could be a positive fact for prolonged binase action, but natrolite itself possessed cytotoxicity increasing along the time of incubation with the cells up to high value about 40% with the same cytotoxicity value as pure binase. Therefore, clinoptilolite and chabazite have some preferences for use as binase carriers. Complex of clinoptilolite with binase induced the cell death comparable to pure binse after 48 h, but during the first 24 h of incubation the release of binase from clinoptilolite induced higher cytotoxicity as pure binase (**Figure 2**). This data are in accordance with previously obtained results about the capacity of clinoptilolite to be useful in medicine. Zeolite-containing mixture (Hydryeast) maintaining mucosal immune homeostasis and epithelial integrity, is known to have a suppressive effect on colitis (Lyu et al., 2017). In humans, zeolite supplementation exerted beneficial effects on intestinal wall integrity and accompanied by mild antiinflammatory effects in aerobically trained subjects (Lamprecht et al., 2015). Treatment of cancer-bearing mice and dogs with micronized zeolite clinoptilolite led to improvement of the overall health status, prolongation of life span and decrease of tumor size in some cases. Combined treatment with doxorubicin and clinoptilolite resulted in strong reduction of the pulmonary

metastasis count increasing anticancer effects of doxorubicin (Zarkovic et al., 2003). Clinoptiolite is also used in water filters, to soil improvement, wastewater treatment and remidiation, in veterinary medicine (in gastrointestinal tract treatment). Chabazit does not have such widespread use.

So, our results could rise especial interest concerning a binase with chabazite complex. First of all, this complex was always more cytotoxic toward Caco2 cells then chabazite or binase themselves. Then, chabazite has low cytotoxicity. Finally, binase release from chabazite is time-dependent (**Figure 2**). Moreover, the catalytic activity of binase was slightly stimulated during staying inside of chabazite (**Table 2**) possibly due to interaction with cations released from this carrier. It means that chabazite-binase complex could be a perspective anticancer agent.

Binase cytotoxicity has grown with concentration increasing during 24 h of incubation. At 48 h of incubation, the difference in cytotoxicity of 100 and 300 µg/ml binase was not significant (**Figure 2**). This could be probably caused by the fact that absorption of binase by cells occured rather quickly, especially in first hours, and reached a practical maximum at 6 h. At this time (6 h) we previously described a permeability peak for trypan blue-labeled albumin macromolecule across cell membrane of cancer lung epithelial cell monolayers treated with RNase (Cabrera-Fuentes et al., 2013). Probably, during prolonged incubation, binase adsorption slows down, which leads to cytotoxicity of the two used concentrations differences leveling at 48 h incubation. Over time, we observed increasing toxicity of natrolite, which formed the needleshaped fibrous aggregates, and damaging cells. Therefore, natrolit cannot be recommended as a carrier of potential therapeutic proteins.

Now, application of zeolites as materials for various therapeutic substances delivery include antitumor ones is intensively studied. The composition synthesized from naturally occurring non-toxic zeolites had a 100% kill rate within 72 h against buccal mucosa and lung squamous epithelial cell cancers and was non-toxic to healthy human cells (Kaufman, 2001). Zeolite-based nanoparticles used in generating timecontrolled release of 5-fluorouracil from zeolite preparations showed anti-cancer effect toward Caco-2 monolayers (Spanakis et al., 2014). Earlier, we have demonstrated that binase-halloysite

#### REFERENCES


complex doubled anticancer efficiency of binase due to its perfect absorption by cells and longer release reducing the viability of human colon adenocarcinoma cells Colo320 by 60% (Khodzhaeva et al., 2017). The same level of toxicity toward human adenocarcinoma Caco2 cells was obtained for chabazite-binase complex. At the first time, we have shown that not only clinoptiolite but also chabazite could be used as carriers for new antitumor agents inducing prolonged cytotoxicity toward cancer cells. Moreover, chabazite could help to counteract oxidative stress in apparently healthy subjects exposed to different oxidative stress risk factors affecting the levels of different antioxidant enzymes (gluthatione peroxidase, superoxide dismutase, and gluthatione reductase (Dogliotti et al., 2012). Our results indicates that (a) the toxicity of chabazite is insignificant in magnitude and does not increase with time; (b) its complex with binase exhibits cytotoxicity increasing with time due to release of binase from the complex; (c) the level of complex toxicity is slightly higher in comparison with pure binase. These facts could open the prospect of using chabazite as a carrier for potential therapeutics proteins.

### AUTHOR CONTRIBUTIONS

VK, OL, and OI planned the experiments. VK and PZ performed the experiments. OL and OI analyzed the data. VK and OI wrote the manuscript.

### FUNDING

The study was performed within the Russian Government Program of Competitive Growth of Kazan Federal University and supported by the Russian Foundation for Basic Research (Project No. 17-00-00060).

### ACKNOWLEDGMENTS

The authors thank Mr. Y. Osin for technical help with TEM images.


(binase). Biochim. Biophys. Acta 1863, 1559–1567. doi: 10.1016/j.bbamcr.2016. 04.005


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

Copyright © 2019 Khojaewa, Lopatin, Zelenikhin and Ilinskaya. 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.

# Enhanced Inhibition of Tumorigenesis Using Combinations of miRNA-Targeted Therapeutics

#### Svetlana Miroshnichenko and Olga Patutina\*

Laboratory of Nucleic Acids Biochemistry, Institute of Chemical Biology and Fundamental Medicine SB RAS, Novosibirsk, Russia

The search for effective strategies to inhibit tumorigenesis remains one of the most relevant scientific challenges. Among the most promising approaches is the direct modulation of the function of short non-coding RNAs, particularly miRNAs. These molecules are propitious targets for anticancer therapy, since they perform key regulatory roles in a variety of signaling cascades related to cell proliferation, apoptosis, migration, and invasion. The development of pathological states is often associated with deregulation of miRNA expression. The present review describes in detail the strategies aimed at modulating miRNA activity that invoke antisense oligonucleotide construction, such as small RNA zippers, miRNases (miRNA-targeted artificial ribonucleases), miRNA sponges, miRNA masks, anti-miRNA oligonucleotides, and synthetic miRNA mimics. The broad impact of developed miRNA-based therapeutics on the various events of tumorigenesis is also discussed. Above all, the focus of this review is to evaluate the results of the combined application of different miRNA-based agents and chemotherapeutic drugs for the inhibition of tumor development. Many studies indicate a considerable increase in the efficacy of anticancer therapy as a result of additive or synergistic effects of simultaneously applied therapies. Different drug combinations, such as a cocktail of antisense oligonucleotides or multipotent miRNA sponges directed at several oncogenic microRNAs belonging to the same/different miRNA families, a mixture of anti-miRNA oligonucleotides and cytostatic drugs, and a combination of synthetic miRNA mimics, have a more complex and profound effect on the various events of tumorigenesis as compared with treatment with a single miRNAbased agent or chemotherapeutic drug. These data provide strong evidence that the simultaneous application of several distinct strategies aimed at suppressing different cellular processes linked to tumorigenesis is a promising approach for cancer therapy.

Keywords: oncogenic miRNA, cancer, antisense oligonucleotide, miRNA mimic, chemotherapy, temozolomide, gemcitabine

### INTRODUCTION

Tumorigenesis represents a complex process characterized by several hallmarks including fast, uncontrolled cell proliferation, reprogramming of energy metabolism, resistance to cell death and replicative senescence, immortalization, evasion of immune surveillance, abundant vascularization, infiltrated growth, and metastasis (Markopoulos et al., 2017). Each of these events is regulated by numerous non-coding molecules, such as long non-coding RNAs and small non-coding RNAs,

#### Edited by:

Hector A. Cabrera-Fuentes, University of Giessen, Germany

#### Reviewed by:

Fan Jiang, Shandong University, China Qing Lyu, University of Rochester, United States Maria Caffo, University of Messina, Italy

> \*Correspondence: Olga Patutina patutina@niboch.nsc.ru

#### Specialty section:

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

Received: 11 February 2019 Accepted: 17 April 2019 Published: 16 May 2019

#### Citation:

Miroshnichenko S and Patutina O (2019) Enhanced Inhibition of Tumorigenesis Using Combinations of miRNA-Targeted Therapeutics. Front. Pharmacol. 10:488. doi: 10.3389/fphar.2019.00488

in particular miRNAs (Zheng et al., 2017). MiRNAs appear to be the key markers of pathological conditions and significant therapeutic targets, since tumorigenesis is commonly associated with aberrant miRNA expression and an imbalance between tumor suppressor and oncogenic miRNAs.

The present review mainly encompasses publications from 2015 to 2018, and contains very recent data on miRNA turnover and function in normal tissues and during pathological states, as well as the latest achievements in the development of sequencespecific approaches for regulating miRNA function and activity. The main informative load of the review focuses on a comparative analysis of the antitumor efficiency of in vitro and in vivo application of various combinations of current miRNA-based therapeutics, in addition to their efficacy in conjunction with chemotherapy. The most promising combination schemes that exhibit additive or synergistic influence on tumor growth in vitro and in vivo are illuminated in the present paper.

### DIFFERENT SCENARIOS OF miRNA BIOGENESIS

miRNA biogenesis is a complicated process accomplished by several enzyme complexes. In contrast to exogenously created siRNAs, miRNAs are generated endogenously by RNA polymerase II from protein-coding sequences or individual miRNA promoters (Lee et al., 2004; **Figure 1**). The produced transcript is called pri-miRNA and comprises a set of miRNA copies encoded in hairpin structures. Each hairpin, consisting of a 33-35 bp stem and terminal loop, is recognized and specifically processed by the Drosha/DGCR8 enzyme complex (**Figure 1B**). Drosha/DGCR8 forms short hairpin RNA with a 2-nt overhang at the 3<sup>0</sup> -end called pre-miRNA, which is further transported from the nucleus by the protein complex, EXP5/Ran, and disposed of in the cytoplasm using GTP hydrolysis as an energy source. Further transformations of premiRNA mediated by the Dicer enzyme include recognition of the 3 <sup>0</sup> overhang, excision of the loop from the hairpin structure, and formation of a linear miRNA duplex (**Figure 1**). Such a duplex consists of two strands differentiated by their stability; a guide strand representing mature miRNA that performs the regulatory function and a passenger strand that is usually degraded by intracellular exonucleases. During the final step of miRNA biogenesis, proteins belonging the Ago family conduct distinction of the strands, duplex melting, and initiation of passenger strand degradation. Moreover, Ago proteins not only provide conclusive maturation of miRNA but also control miRNA activity as regulators of gene expression (Daugaard and Hansen, 2017). Such a canonical scheme of biogenesis is typical for the majority of intracellular miRNAs; however, some molecules are known to be formed by alternative pathways, which are characterized by the absence of one or several processing steps.

In Dicer-independent biogenesis, canonically formed premiRNA escapes Dicer processing and interacts directly with the Ago2 protein following export to the cytoplasm (Daugaard and Hansen, 2017). Since Ago2 is not able to catalyze the further maturation of pre-miRNA due to its hairpin structure, it induces a single-stranded break in the hairpin region and initiates the exonuclease degradation of the passenger strand of pre-miRNA, which finishes with the formation of conventional mature miRNA. Another alternative avenue for miRNA maturation is Drosha/DGCR8-independent biogenesis (**Figure 1C**). In such a scheme, the pri-miRNA molecule represents copies of sequences called "miRtron" that comprise the acceptor and donor sites for splicing. The splicing of pri-miRNA replaces the processing by Drosha/DGCR8 and promotes the formation of pre-miRNA, further maturation of which occurs according to the canonical scheme of biogenesis (Daugaard and Hansen, 2017; **Figure 1C**).

There exists another biogenesis pathway that does not depend on the action of either Drosha/DGCR8 or Dicer complexes, and can be realized in the presence of "agotron" sequences in the genome (**Figure 1A**). Being subjected to splicing, "agotron" represents an analog of "miRtron" that binds to Ago proteins following transcription and translocation to the cytoplasm. "Agotron" regulates gene expression in the same manner as miRNA, through binding to mRNA targets; however, it is a molecule consisting of multiple copies of mature miRNAs (Daugaard and Hansen, 2017).

All biogenesis pathways result in the formation of mature miRNA molecules and consequent assembly of the miRISC complex. This ribonucleoprotein complex serves as a catalytic engine for the posttranscriptional regulation of gene expression, during which miRNA binds to particular mRNA via its seed region at the 5<sup>0</sup> -end of the molecule. Proteins of the miRISC complex, such as Ago, Snd1, and MTDH, are responsible for translation inhibition, mRNA target degradation, and promotion of the protection of miRNA from the action of intracellular nucleases, determining the duration of life and functional period of mature miRNA (Kim et al., 2009). The key interaction for assembling and function of miRISC complex is appeared to be co-operation of scaffold protein TRNC6B and protein with ribonuclease activity Ago2. Interactions between tryptophan residues in Ago-binding domain (ABD) of TRNC6B and tryptophan-binding pockets of Ago2 were recently discovered to promote the phase transition (Sheu-Gruttadauria and MacRae, 2018). This process represents formation of viscoelastic TRNC6B - Ago2 droplets heterogeneous on the atomic level since Ago2 interacts not only with rigorously defined tryptophan residues but also with random indole groups of tryptophans in ABD of TRNC6B. Apart from aforementioned proteins different miRNAs and their direct mRNA targets are included in the composition of droplets during the phase transition. It was evaluated that only one couple of miRNA–mRNA is introduced in the droplet suggesting the occurrence of selection process for particular mRNA molecules (Sheu-Gruttadauria and MacRae, 2018). In vitro experiments demonstrated that TRNC6B - Ago2 droplets are formed to degrade mRNAs (Sheu-Gruttadauria and MacRae, 2018). It becomes possible since Ago2 protein retains its catalytic activity and is able to promote target mRNA slicing. In addition, the active subunits of de-adenylation CCR4-NOT complex were also found in the droplets aimed at mRNA de-adenylation after its slicing (Sheu-Gruttadauria and MacRae, 2018). Thus, novel data give evidence that mRNA degradation following binding with miRNA

occurs in separate heterogeneous viscoelastic structures that concentrate particular mRNA targets and protein components for its degradation. These results turn around the present perceptions about miRISC complex opening the new horizons for miRNA-based investigations. In particular, it raises a question about existence of special structural organization of miRNAs and their mRNA targets in cell cytoplasm necessary for the regulation of its function and decay.

### INTRACELLULAR miRNA TURNOVER

While the biogenesis and function of miRNAs are described in detail, the mechanism of cellular miRNA degradation remained unclear until relatively recently. Since 2010, increasing amounts of data have become available regarding miRNA degradation as a result of miRNA binding to non-coding RNAs (Ameres et al., 2010). In particular, it has been shown that the level of miRNA-27 falls significantly following binding to viral H. samiris U-rich non-coding RNAs or the m169 transcript that was defined as natural inhibitor of miRNA-27, which enter the cell as a result of infection with Herpesvirus Samiri or Cytomegalovirus, respectively (Cazalla et al., 2010; Marcinowski et al., 2012). It has been stressed that such a phenomenon is accompanied by the elongation of miRNA with several U or A nucleotides, followed by 3<sup>0</sup> -end trimming of the miRNA molecule. This process is called target RNA-directed miRNA degradation (TDMD), and in confirmation of such a mechanism, several enzymes, namely terminal uridine-transferases (TUT) and 3<sup>0</sup> -exonucleases (for example, DIS3L2), have been found to take part in TDMD and carry out the transfer of uridines to the 3<sup>0</sup> -end of miRNA and the consequent miRNA 3<sup>0</sup> -trimming, respectively (Haas et al., 2016; **Figure 1**). The most important parameter determining the efficiency of TDMD is the degree of complementarity between the miRNA and its RNA target. It has been identified that the tight binding between the 3<sup>0</sup> -end of miRNA and the 5<sup>0</sup> -end of its target must occur for the initiation of TDMD. A 2-nt mismatch may cause greater than a three-fold decrease in the efficiency of TDMD, and a 4-nt mismatch may completely abolish miRNA degradation. Moreover, if there is a central bulge in the miRNA/RNA target complex, it should comprise no more than 5 nt, otherwise TDMD will not take place (De la Mata et al., 2015). It is likely that the different levels of complementarity between mammalian miRNAs and their RNA targets promoted separation of the pathways of TDMD and miRNA-mediated posttranscriptional regulation of gene expression. It should be stressed that conventional binding of target mRNA to the miRNA seed region does not initiate miRNA degradation due to the low degree of complementarity (Ameres et al., 2010).

Another factor that may determine the balance between TDMD and target RNA decay is miRNA abundance. According to a study in neuronal cells, low-abundance miRNAs, for example miRNA-132, bind to RNA targets with high complementarity and initiate TDMD, whereas highly abundant miRNAs, such as miRNA-138, miRA-128, and miRNA-124, mediate target RNA decay. Confirmation of such a phenomenon showed that the high intensity of TDMD observed for miR-132 fell dramatically after the level of miRNA-132 was elevated following transfection with synthetic mimics (De la Mata et al., 2015).

Finally, its non-cumulative nature is the last important feature of TDMD. An increase in the number of binding sites for miRNA within target mRNA significantly improves its inhibitory effect but does not stimulate TDMD. To illustrate, it has been shown that the addition of miRNA binding sites to the m169 transcript structure has no impact on TDMD efficiency (Haas et al., 2016); however, a two-fold decrease in the miRNA-132 degradation rate was observed following an increase in the number of binding sites in the RNA target from one to four (De la Mata et al., 2015).

Thus, TDMD represents a non-cumulative degradation of miRNA accomplished by two enzymes and resulting from miRNA binding to full-length complementary targets. It should be noted that terminal uridine transferases and 3<sup>0</sup> -exonucleases that carry out TDMD interact with Ago family proteins (Haas et al., 2016), suggesting that there is co-regulation of the miRNA biogenesis, functioning, and degradation processes.

### THE ROLE OF miRNAs IN ONCOPATHOLOGIES. OLIGONUCLEOTIDE-BASED APPROACHES TO THE MODULATION OF miRNA FUNCTION

Recently, a vast amount of studies have been published stating the direct involvement of miRNAs in malignant growth. Aberration of miRNA expression is not uncommon during the initiation and progression of various diseases, including cancer (Rupaimoole et al., 2016; Singh and Sen, 2017). Investigators have discriminated between oncogenic miRNAs that promote tumor development and tumor suppressor miRNAs that impede tumorigenesis (Nucera et al., 2016; O'Bryan et al., 2017). Oncotransformation and metastasis during each step can be accompanied by the downregulation of tumor suppressor miRNAs and the hyperexpression of oncogenic miRNAs. During the initiation stage, associated mainly with uncontrolled cell proliferation, clusters of oncogenic miRNAs, in particular miRNA-17-92 and miRNA-106b, appear to be the main regulators (Tan et al., 2014; Li H. et al., 2017). Cell immortalization, which involves intense cell division and evasion of apoptosis, may be partly connected to the action of such oncogenic miRNAs: miRNA-125b that plays a role in the inhibition of apoptosis; miRNA-221/222 that enhance the proliferative potential of cells; miRNA-130b that provides resistance to chemotherapy; and miRNA-21 that exhibits a complex pathological influence on various cellular functions (Masoudi et al., 2018; Zhang et al., 2018). Substantial vascularization, which furnishes the forming neoplasia with ceaseless nutrition, is the consequence of the angiogenic action of multiple miRNAs, including miRNA-9, miRNA-27b, miRNA-130a, miRNA-210, miRNA-191, and miRNA-378 (Markopoulos et al., 2017). However, the process of tumor cell dissemination (metastasis), is prompted by the activity of miRNA-155, miRNA-9, miRNA-10b, and miRNA-21 (Devulapally et al., 2015; Mignacca et al., 2016), during which, miRNA-181b, miRNA-193a, and miRNA-29a are also engaged in modulating the adhesive properties of cells (Mohamad et al., 2016;Liu et al., 2017). Currently,manipulation of miRNAs associatedwith tumorigenesisis of considerable scientific and practical interest.

Among the ways by which to modulate the level of a particular miRNA, restoration of tumor suppressor miRNA by straightforward transfection of synthetic miRNA mimics (Zhang Y. et al., 2016; Tian et al., 2017) or transformation of cells with vectors expressing deficient miRNAs has been investigated in most depth (Michelfelder and Trepel, 2009), and a number of positive results have already been obtained in this area (**Figures 2A,B**). Transfection of acute promyelocytic leukemia cells with synthetic miRNA-218 reduces viability, inhibits intracellular DNA synthesis, promotes cell cycle arrest in the G0/G<sup>1</sup> phase, and triggers apoptosis, leading to a two-fold increase in the number of apoptotic cells (Wang Y. et al., 2017). Delivery of miRNA-1193 and miRNA-455 decreases the invasive potential of breast cancer and non-small cell lung cancer up to three times as compared with the control (Li et al., 2016; Li X. et al., 2017). Treatment of nasopharyngeal carcinoma, colorectal cancer, and non-small lung cancer with synthetic miRNA-497, miRNA-495, and miRNA-142, respectively, leads to a 2.5-fold suppression of migration and proliferation of tumor cells and a two-fold tumor growth retardation in vivo (Wang et al., 2015; Wang Z. et al., 2017; Yan et al., 2017). Expression of miRNA-26 in an adenoviral vector inhibits tumorigenesis, induces apoptosis, and suppresses cancer cell growth in a model of Myc-dependent liver cancer (Kota et al., 2009). Expression of miRNA-15a in a lentiviral vector considerably reduces colony formation and proliferation of endometrial cancer cells (Wang Z.M. et al., 2017).

Restoration of tumor suppressor miRNA levels for the treatment of oncopathologies represents a relatively promising strategy. À number of therapeutics based on miRNA mimics are currently undergoing clinical trials. TargomiRs, which represent minicells containing an miRNA-16 mimic for malignant pleural mesothelioma treatment, successfully passed Phase I clinical trials by EnGeneIC Ltd. (Van Zandwijk et al., 2017). In 2013, clinical trials for MRX34 – a synthetic miRNA-34a mimic – were initiated; however, regrettably, in 2017, Mirna Therapeutics Inc., ascertained that this agent promotes adverse events (Beg et al., 2017). The recruitment of patients for Phase II clinical trials of the drug, MRG-201, which represents the mimic of miRNA-29 initiated by miRagen Therapeutics Inc., is currently being executed (ClinicalTrials.gov Identifier: NCT03601052). Despite being aimed at the treatment of keloids, such a therapeutic agent may be tested as an antitumor vehicle, since miRNA-29 plays a relevant tumor suppressive role in various cancers, including myeloid leukemia, esophageal squamous cell carcinoma, and gastric cancer (Qi et al., 2017).

Based on the application of synthetic constructions, a number of technologies have been developed to inhibit the hyperfunctions of oncogenic miRNAs. Such compounds may be aimed at both direct miRNA sequestration and interruption of miRNA regulatory activity through the interaction with mRNA targets, but not at the miRNAs themselves. For instance, miRNA sponges induce a significant decrease in the level of functionally active oncogenic miRNAs operating as additive sites for miRNA binding (Tay et al., 2014; **Figure 2C**). Many sponge constructs

have been developed to inhibit various miRNAs, including miRNA-10b, miRNA-21, miRNA-19, miRNA-155, miRNA-23b, miRNA-221/222, miRNA-9, and miRNA-140 (Haraguchi et al., 2009; Ma et al., 2010; Chen et al., 2013; Liu et al., 2014; Mignacca et al., 2016). The in vitro effects of their application are: a 40% decrease in the invasive and migratory potentials of tumor cells (Liang et al., 2016); up to a 50% inhibition of cell proliferation (Mignacca et al., 2016); and a two-fold increase in the sensitivity to chemotherapeutics, in particular doxorubicin (Gao et al., 2015). In vivo administration of miRNA sponges results in a reduction in angiogenesis and a two-fold reduction in the number of metastases formed in the case of glioma and colorectal cancer (Ma et al., 2010; Liu et al., 2014).

Other antisense oligonucleotide-based agents for miRNAmediated therapy are miRNA masking oligonucleotides or target protectors, which impede the binding of oncogenic miRNAs to their targets via an interaction with the 3<sup>0</sup> -untranslated region of mRNAs followed by the reactivation of normal activity of genes that were previously repressed (Choi et al., 2007; **Figure 2D**). In 2012, miRNA masking oligonucleotides were used for the first time to investigate miRNA function in tumorigenesis. These experiments resulted in the evaluation of the stimulatory impact of TP63 protein expression on breast cancer cell proliferation, which proved to be directly regulated by miRNA-196a2<sup>∗</sup> (Kim et al., 2013). Since then, the functional relationships between dozens of miRNAs and their mRNA targets have been ascertained by means of miRNA masking oligonucleotides, including miRNA-203 and the LASP-1 (LIM and SH3 Protein 1) gene, miRNA-17 and miRNA-20à and the NOR-1 (Neuron-derived Orphan Receptor-1) gene, miRNA-29-b-1 and the SPIN1 (Spindlin 1) gene, and miRNA-27à and the CALR (calreticulin) gene. These interactions appear to be crucial for tumorigenesis events such as an increase in the invasive, proliferative, angiogenic, and migratory potentials, and evasion of immunogenic apoptosis (Colangelo et al., 2016; Drago-Ferrante et al., 2017). Evidence highlights the promising application of miRNA masking oligonucleotides as inhibitors of oncogenic miRNA function in tumor cells. For instance, effective proliferation suppression and apoptosis induction have been described for miRNA masks that prevent the interaction of miRNA-522 with DENND2D mRNA; the negotiation of glioblastoma cell resistance to temozolomide has been evaluated using target protectors that impede the interplay between miRNA-9 and PTCH1 mRNA; and the inhibition of angiogenesis has been achieved by blocking the binding of the miRNA-30 family to DLL4 (Delta-like 4) mRNA (Bridge et al., 2012; Munoz et al., 2015; Zhang T. et al., 2016). However, despite this progress and promising in vitro results, there still exists no data regarding the application of miRNA masking oligonucleotides in vivo.

In 2017, a novel type of oligonucleotide construction, named small RNA zippers, was designed, which can block miRNA function by forming a duplex with multiple copies of miRNAs, linking them together in an end-to-end manner (Meng et al., 2017; **Figure 2E**). In vitro application of zippers targeted to oncogenic miRNA-17 and miRNA-221 in a breast cancer cell line has been shown to decrease the level of targeted miRNAs by up to 90%, followed by the inhibition of cancer cell migration and a 1.5-fold increase in the sensitivity to doxorubicin (Meng et al., 2017). This is the only article that states the application of small RNA zippers; however, these first data show

a high therapeutic potential of such constructions, the biological effect of which is comparable with the efficiency of current antisense oligonucleotides.

Another technique becoming increasingly popular nowadays is application of clustered regularly interspaced short palindromic repeats (CRISPR)-associated nuclease 9 (Cas9) – CRISPR/Cas9 systems. They consist of Cas9 endonucleases cloned from Streptococcus pyogene and single guide RNA (sg RNA). The latter, in turn, represents two sequences: (1) CRISPR RNA (crRNA) that is complementary to target DNA sites and responsible for its recognition and binding; and (2) trans-activating CRISPR RNA (trancrRNA) that is partially complementary to crRNA and is essential for maintenance of Cas9 nuclease activity. The CRISPR/Cas9 systems distinguish target DNA sequences that are located in direct proximity to protospacer adjacent motif (PAM) and induce the doublestranded breaks. Hereafter, the repair system patches the breaks in non-homologous end joining manner with variable sizes of insertions or deletions followed by inhibition of target molecule expression and activity. To suppress functions of oncogenic miRNA, investigators usually apply CRISPR/Cas9 systems that introduce mutations to the Drosha/Dicer processing site of miRNA precursors (pri- or pre-miRNAs) that leads to the cancelation of further biogenesis and decrease in the level of mature miRNA in cells, in average, by 55–96% (Yang et al., 2018). The consequences of inhibition of oncogenic miRNAs such as miRNA-17, miRNA-21, miRNA-141 and miRNA-3188 using specific CRISPR/Cas9 systems are 1.5-2-fold inhibition of proliferation and invasion, up to 5-fold decrease in migration as well as two-fold effective induction of cancer cells apoptosis as compared to control (Aquino-Jarquin, 2017; Huo et al., 2017). In addition, the CRISPR/Cas9 was found to inhibit epithelial-mesenchymal transition and significantly increase the sensitivity of tumor cells to chemotherapeutics, in particular, cisplatin and paclitaxel (Aquino-Jarquin, 2017; Huo et al., 2017). CRISPR/Cas9 systems is out of question the perspective approach to downregulate the oncogenic miRNAs and though its effectiveness may vary for different miRNA targets, this feature is balanced out by high stability and duration of its effect that may last 30 days for both in vitro and in vivo models (Chang et al., 2016). Moreover, it should be noted that increased specificity of CRISPR/Cas9 turns it to the highly precise instrument for inhibition of particular oncogenic miRNAs from one miRNA cluster.

One of the promising approaches for decreasing the hyperexpression of miRNAs is the application of synthetic antisense oligonucleotides (or anti-miRNA ONs; **Figure 2F**). Antisense oligonucleotides are single-stranded DNAs, 15-20 nt in length, which have historically been used to inhibit mRNA targets and are currently applied to modulate the activity of miRNAs. Antisense oligonucleotides function as competitive inhibitors of miRNAs, which complementarily interact with mature molecules and cause steric hinderance of function or complete degradation mediated by RNase H (Boutla et al., 2003). A number of modifications of antisense oligonucleotides, including structural changes to the sugar backbone such as 20O-Methyl (20OMe), 2<sup>0</sup> -Fluoro (20F), 20O-Metoxyethyl (20MOE), Locked Nucleic Acids (LNA), and Peptide Nucleic Acids (PNA), as well as chemical modifications to the phosphodiester bonds, in particular phosphorothioate (PS) and N-mesyl- (µ-) and methoxyethyl-phosphoramidate have been developed, allowing a significant increase in the therapeutic potential of antisense oligonucleotides via the enhancement of nuclease resistance and binding affinity, and facilitation of penetration across the cell membrane (Lennox and Behlke, 2010; Watts, 2013; Miroshnichenko et al., 2019; **Supplementary Figure S1**). Under the action of different chemically modified antisense oligonucleotides, it is possible to attain a considerable reduction in oncogenic miRNA activity, and consequently to inhibit various events of tumorigenesis. For instance, the inhibition of oncogenic miRNAs, such as miRNA-23a and miRNA-106b∼25 cluster, has been found to promote a decrease in the proliferative potential of cancer cells by up to 75% (Zhang R. et al., 2016; Quan et al., 2017). In addition, application of antisense technology may cause an approximate two-fold decrease in cellular motility and a two-fold increase in the percentage of apoptotic cells following transfection with anti-miRNA ONs targeted, in particular, to miRNA-21, miRNA-221, miRNA-17, miRNA-18a, and miRNA-191 (Haghpanah et al., 2016; Liang et al., 2017; Quan et al., 2017). Experiments in breast cancer cells, glioblastoma cells, and medulloblastoma cells have shown that the anti-metastatic and anti-apoptotic effects of oligonucleotides targeted to miRNA-191, miRNA-10b, miRNA-221, or miRNA-222 are two-fold higher (Quintavalle et al., 2011; Pal and Greene, 2015; Sharma et al., 2017). In vivo administration of antisense oligonucleotides has been found to reduce tumor growth (Mercatelli et al., 2008; Li et al., 2009). For instance, ONs targeted to miRNA-10b cause greater than a two-fold retardation of tumor growth and a considerable increase in the survival of mice with intracranial glioblastoma (Tepluyk et al., 2016). Administration of anti-miRNA-155 ONs to mice suffering from lymphoma led to a five-fold reduction in tumor weight as compared with the control (Babar et al., 2012). Moreover, antisense oligonucleotides may also have an inhibitory effect on metastasis in vivo. Administration of anti-miRNA-182 ONs was shown to promote a two-fold decrease in the number of liver metastases at the latest stage of melanoma (Huynh et al., 2011), while inhibition of miRNA-10b led to complete elimination of metastases in the lymph nodes of xenograft mice with breast cancer (Yoo et al., 2017). Furthermore, antisense miRNA-based oligonucleotides may also suppress angiogenesis and stimulate tumor infiltration by macrophages in vivo (Kong et al., 2014). In addition, antisense oligonucleotides may serve as effective therapeutic agents to treat pathologies which require overcoming the blood-brain barrier including glioblastoma and neuro-degenerative diseases. Initially, antisense oligonucleotides were not supposed to be delivered to brain systematically, but to date the significant progress in the area of delivery of such agents beyond the blood-brain barrier has been achieved. Two main types of delivery provided passing through the blood-brain barrier are well investigated today, in particular, direct and peripheral delivery. Direct methods include intratumoural or intraventicular administration of antimiRNA ON using osmotic pomp that is effective in terms of

glioblastoma treatment (Kim et al., 2016). Peripheral delivery of antisense oligonucleotides implies (1) subsequent administration of ONs with agents that contribute in re-arrangement of blood-brain barrier such as angubudin-1 leading to transient dissociation of proteins from tight junctions of blood brain barrier followed by size-selective leakage of therapeutic ON; or (2) conjugation of ONs with different groups, such as ligands to the receptors on the surface of blood-brain barrier or argininerich cell penetrating peptides to cross the barrier by receptormediated endocytosis, inverted micelle or pore formation as well as direct membrane translocation depending on the type of conjugated group (Evers et al., 2015; Zeniya et al., 2018). All these approaches showed high specificity and effectiveness in vivo without any adverse effects. Successful application of antisense oligonucleotides in vivo gave an impetus to initiate clinical trials. MRG-106 (miRagen Therapeutics, Inc.), an antimiRNA-155 drug for the treatment of a wide spectrum of leukemias and lymphomas, and RG012 (Regulus Therapeutics Inc.), an oligonucleotide targeted to miRNA-21 for curing Alport syndrome, have already entered clinical trials. Thus, antisense technology represents a highly effective and rapidly progressing area of research related to miRNA.

Finally, one of the recent directions in miRNA-based therapy is the application of artificial ribonucleases (aRNases) that are conjugates of miRNA-targeted oligonucleotides and catalytic constructs (Gaglione et al., 2011; Danneberg et al., 2015; Patutina et al., 2017, 2018; **Figure 2G**). One of the variants of effective constructions is miRNA-targeted metaldependent synthetic RNases that represent conjugates contained (1) peptide nucleic acids (PNA) for highly effective target RNA binding and (2) diethylenetriamine (DETA) or a three-amino acids peptide HGG with Cu2<sup>+</sup> ion, as a co-factor, aimed at the oxidative or acid-base cleavage of hsa-miR-1323, which is involved in the development of neuroblastoma (Gaglione et al., 2011). Another aRNase targeted to oncogenic miRNA, in particular miR-20a from oncogenic cluster miRNA-17∼92, comprises metal-independent artificial RNase consisting of a PNA oligonucleotide with tris(2-aminebenzimidazole) which degrades miRNA substrate, presumably, in RNase A-like manner in the bulge forming upon the hybridization (Danneberg et al., 2015). To date, the most successful miRNA-based conjugates, miRNAses, represent conjugates of oligonucleotides and the catalytic peptide [(LRLR)G]<sup>2</sup> that cleave clinically relevant miRNA targets, miRNA-21 and miRNA-17 by trans-esterification reaction (Patutina et al., 2017, 2018). These miRNases have shown a significant reduction in target miRNA levels in tumor cells, followed by the restoration of key tumor-suppressor proteins and considerable inhibition of cell proliferation without any off-target effects.

Currently established miRNA-based therapeutics represent a spectrum of oligonucleotide constructions with different mechanisms of oncogenic miRNA activity modulation (**Figure 2**). In particular, miRNA sponges, small RNA zippers, and some types of antisense oligonucleotides that do not exhibit RNase H-activating ability (20OMe, LNA, PNA etc.) effectively suppress the activity of oncogenic miRNAs by forming a steric duplex. MiRNA masks act as competitive inhibitors of miRNAs by allowing restriction of the influence of one separated miRNA on a particular target via complementary binding to corresponding mRNAs. CRISPR/Cas9 systems downregulate miRNAs by inclusion of mutations to the miRNA precursors sequences and prevention of miRNA biogenesis. Antisense oligonucleotides that are recognized by RNase H, in addition to aRNases, irreversibly degrade target miRNAs and effectively inhibit their function in cells. These approaches have demonstrated the relatively high in vitro and in vivo efficiency and have laid the foundation for further investigation of these miRNA-based strategies in combined applications.

### EFFECTS OF THE SIMULTANEOUS APPLICATION OF SEVERAL miRNA-BASED THERAPEUTICS ON TUMORIGENESIS IN VITRO AND IN VIVO

In the present review, three strategies for the combined application of miRNA-based therapeutics are considered: (1) simultaneous application of more than one antisense oligonucleotide targeted to oncogenic miRNAs; (2) transfection of tumor suppressor miRNA mimic combinations; (3) treatment with antisense oligonucleotides or mimics in conjunction with chemotherapeutics. **Supplementary Table S1** summarizes the results of reported preclinical studies on the combined application of miRNA-based therapeutics in vitro and in vivo. This table contains a detailed description of all the effects of mono- and combination therapy for each individual study described below, with a view to keeping track of the efficiency of combination therapy. The most impressive results showing a synergistic effect of combination therapy are listed in **Table 1**.

In in vitro experiments, the final concentration of the applied ON mix was mainly 50 or 100 nM (Lee et al., 2015; Zhang R. et al., 2016; Li Y. et al., 2017); however, in some cases, the concentration of the mix was significantly lower (10 or 30 nM; Matsubara et al., 2007; Lu et al., 2009) or, vice versa, reached 1.2-2 µM (Brognara et al., 2015; Zhang et al., 2015a), since such combinations were applied in the absence of any delivery vehicle. It should be stressed that the concentration of each oligonucleotide in the mix was less than the final concentration in proportion to the number of compounds used.

Cancer cells were transfected with pre-miRNAs or miRNA mimics at a concentration of 10-50 nM each (Cheng et al., 2017; Zeng et al., 2018; Jiang et al., 2019), and only in one case of concurrent application of an miRNA mimic and chemotherapy was the concentration of the miRNA mimic 200 nM (Huang et al., 2014).

In combination with miRNA-based agents, chemotherapeutics were applied at different concentrations depending on the cancer cell type: temozolomide was used to treat glioma and glioblastoma at 25-400 µM (Tan et al., 2018; Zeng et al., 2018); gemcitabine was applied to various types of pancreatic cancer cell lines at concentrations of 500 nM to 20 µM (Chaudhary et al., 2017; Tu et al., 2019); sunitinib was used to treat glioblastoma cells at 15 or 20 µM (Costa et al., 2013; Liu et al., 2015), as


Frontiers in Pharmacology | www.frontiersin.org

fphar-10-00488 May 15, 2019 Time: 16:33 # 8

well as at 1 µM in the case of kidney cancer cell treatment (Khella et al., 2015); and cetuximab was added to colorectal cancer and hepatocellular carcinoma cell lines at 10-13 µM (Huang et al., 2014; Zhou et al., 2015).

In vivo experiments were conducted using miRNA-based therapeutics at doses from 1 to 4 mg/kg (Mittal et al., 2014; Dai et al., 2016; Li Y. et al., 2017) and chemotherapies were administrated at the following doses: gemcitabine, 40 mg/kg (Chaudhary et al., 2017; Mondal et al., 2017); temozolomide, 10- 20 mg/kg (Fan et al., 2015; Zeng et al., 2018); paclitaxel, up to 5 mg/kg (Dai et al., 2016; Tu et al., 2019); and sunitinib, 30 mg/kg (Costa et al., 2015).

### COMBINATION OF ANTI-miRNA OLIGONUCLEOTIDES FOR OVERALL INHIBITION OF ONCOPATHOLOGICAL STATES

The relatively high biological activity of a particular anti-miRNA oligonucleotide in vitro and in vivo suggested that combined administration of several miRNA-targeted agents may reinforce the antitumor effect. Two strategies have proven to be effective. The first is related to the application of several oligonucleotides targeted to multifunctional oncomiRs, such as miRNA-155, miRNA-21, and miRNA-10b, which modulate multiple events of tumorigenesis (Pfeffer et al., 2015; Wang, 2017). The second strategy includes co-inhibition of miRNAs that belong to one oncogenic miRNA cluster, which usually combines mono- and polyfunctional regulators, as in the case of the miRNA-183-96- 182 cluster, where miRNA-96 mostly stimulates cell proliferation, whereas polyfunctional miRNA-182 and miRNA-183 intensify the invasive and migratory potentials of cancer cells and promote chemotherapy resistance (Ren et al., 2014; Yan et al., 2014; Zhang et al., 2015c; Ma et al., 2016).

Treatment of cancer cells with various oligonucleotide mixes has resulted in considerable enhancement of therapy efficiency in comparison with the effects of a single anti-miRNA ON. Simultaneous silencing of miRNA-17 and miRNA-20à in the miRNA-17∼92 cluster or three miRNAs in the miRNA-183/182/96 cluster leads to a 3- or 4-fold greater inhibition of human colon and lung cancer cell viability (Matsubara et al., 2007; Zhang et al., 2015a; **Table 1** and **Supplementary Table S1**). A combination of miRNA-130a and miRNA-495 antisense inhibitors promotes a two-fold greater inhibition of angiogenesis in gastric cancer (Lee et al., 2015; **Supplementary Table S1**). Application of cocktails containing oligonucleotides targeted to miRNAs from the miRNA-106b∼25 cluster, including miRNA-106, miRNA-93, and miR-25, and miRNAs from the miRNA-221/222 cluster, leads to a 1.5-fold to 4-fold greater rate of apoptosis induction than with each oligonucleotide separately (Brognara et al., 2015; Zhang R. et al., 2016; **Table 1** and **Supplementary Table S1**). Moreover, various mixes of antimiRNA oligonucleotides, in particular anti-miRNA-221 and antimiRNA-222, and anti-miRNA-10b and anti-miRNA-21, as well as anti-miRNA ONs to all miRNAs from the miRNA-106b∼25 cluster, are two-fold stronger in terms of decreasing the proliferative potential of glioma and glioblastoma cells (Zhang et al., 2009, Dong et al., 2012; Zhang R. et al., 2016; **Table 1** and **Supplementary Table S1**). A reduction in the invasive and migratory potentials as great as 3.5-fold may be reached using mixes of anti-miRNA-99b, anti-miRNA-let-7-e, and antimiRNA-125a oligonucleotides or anti-miRNA-21 and antimiRNA-10b oligonucleotides in comparison with monotherapy (Dong et al., 2012; Ma et al., 2017; **Table 1** and **Supplementary Table S1**). It should be noted that in most cases, oligonucleotide cocktails have a simultaneous influence on several different events of tumorigenesis (Li L. et al., 2014; Zhang R. et al., 2016; Ma et al., 2017; **Figure 3**).

Successful application of antisense inhibitor mixes promoted the development of oligonucleotide constructs able to provide combined delivery of anti-miRNA agents to cells. Two very similar structures were independently established, namely multitarget antisense oligonucleotides (MTg-AMO) and multipotent miRNA sponges, which represent long oligonucleotides containing several sites for binding to the miRNA of interest. The substantial difference between these inhibitors is the method of intracellular delivery, with MTg-AMOs being co-transfected with delivery agents including lipofectamine, and miRNA sponges being expressed using viral vectors. Application of an miRNA-21/miRNA-155/miRNA-17 MTg-AMO and an miRNA sponge complementary to miRNA-21, miRNA-155, miRNA-221, and miRNA-222 were found to considerably decrease cancer cell viability and proliferation, with the effect of multitarget inhibitors being up to 60% more effective than that of separate inhibitors (Lu et al., 2009; Jung et al., 2015; **Supplementary Table S1**). Moreover, using a miRNA sponge targeted to miRNA-17, miRNA-18a, miRNA-19, and miRNA-92, it is possible to attain up to a 6-fold greater inhibition of lymphoma cell growth as compared with the effect of miRNA sponges targeted to individual miRNAs (Kluiver et al., 2012; **Supplementary Table S1**). In addition, a significant increase in the percentage of apoptotic cells may be reached by the application of multipotent sponges targeted to the miRNA-183-96-182 cluster or containing simultaneous binding sites for miRNA-17 and miRNA-20a (Li P. et al., 2014; Niu et al., 2014).

In vitro results of the application of oligonucleotide cocktails targeted to oncogenic miRNAs are relatively promising. The most outstanding of these proves that the application of anti-miRNA oligonucleotide mixes results in a 4- to 7-fold decrease in the invasive potential of tumor cells (Dong et al., 2012; Zhang R. et al., 2016; **Supplementary Table S1**), up to a 5-fold induction of apoptosis of cancer cells (Brognara et al., 2015; **Supplementary Table S1**), and up to a 9-fold suppression of cancer cell growth (Kluiver et al., 2012; **Supplementary Table S1**); whereas the average pro-apoptotic and anti-proliferative effects of single anti-miRNA oligonucleotide therapy do not exceed 20-25%. Nevertheless, there is only one publication that demonstrates the in vivo application of an antisense oligonucleotide mix that consisted of anti-miRNA-221 and anti-miRNA-222 ONs, and provided only a 1.5-fold greater inhibition of tumor growth as compared with each ON used alone (Zhang et al., 2009; **Supplementary Table S1**).

### APPLICATION OF SYNTHETIC miRNA MIMIC MIXES FOR AN ENHANCED EFFECT ON A PARTICULAR CELLULAR FUNCTION

Another approach for the creation of combinations of miRNAbased therapeutics is the concurrent application of synthetic miRNA mimics. The most often used are mimics of miRNA-34a, miRNA-99a, and members of the miRNA-145 family. The employment of such synthetic miRNA mimics arose from the key regulatory roles of these tumor suppressor miRNAs in cell cycle surveillance, focal adhesion, and cytokine-cytokine receptor interactions in cancer cells of various histogenesis via adjustment of fundamental signaling pathways including p53, Notch, TGF-β, IGF-1R, and mTOR (Li and Zheng, 2017; Huang et al., 2018; Ren et al., 2018). Therefore, restoration of normal expression levels of the proteins involved in these pathways and regulated by these miRNAs may account for normalization of cellular processes and inhibition of the development of pathological conditions.

In contrast to the application of antisense oligonucleotide mixes that exert a complex influence on several events of tumorigenesis simultaneously, miRNA mimic combinations more frequently exhibit a considerable effect on one particular cellular function (**Figure 3**). For instance, simultaneous treatment of non-small lung cancer cells with pre-miRNA-15a/16 and pre-miRNA-34a promotes an increase in the number of cells undergoing arrest in the G1-G0 phase of the cell cycle in a synergistic manner; approximately 55% of cells are under arrest, which is almost three-fold greater than the results seen with monotherapy (Bandi and Vassella, 2011; **Supplementary Table S1**). However, a synergistic pro-apoptotic effect of such a mix is not observed. Simultaneous transfection of nonsmall cell lung cancer cells with miRNA-34a and miRNA let-7b mimics promotes a significant decrease in invasive potential: the number of invading cells is 5- to 8-fold less, as compared with the application of let-7b or miRNA-34a mimics alone, respectively (Kasinski et al., 2014; **Table 1** and **Supplementary Table S1**). Application of a cocktail consisting of pre-miRNA-145 and pre-miRNA-141 causes a more profound inhibitory effect on cell migration than that of each premiRNA separately (Liep et al., 2016); concurrent treatment with miRNA-145 and miRNA-143 mimics leads to a two-fold greater reduction in colorectal cancer cell viability (Su et al., 2014; **Supplementary Table S1**). In some esophageal squamous cell carcinoma cells, transfection of pre-miRNA-99a in combination with pre-miRNA-100 causes a 1.5-fold greater decrease in cell proliferation in contrast with monotherapy (Sun et al., 2013; **Supplementary Table S1**).

According to certain results, the application of miRNA mimic mixes exerts a therapeutic effect on several cellular functions. For instance, a combination of miRNA-99a and miRNA-497 mimics more effectively triggers apoptosis, as well as reduces hepatocellular carcinoma cell viability in vitro, as compared with each mimic alone (Cheng et al., 2017; **Supplementary Table S1**).

The introduction of vectors driving the expression of both miRNA-34a and miRNA-497 provokes a 1.5-fold greater decrease in colony formation and cell viability than those driving expression of each mimic alone; however, neither an additive nor a synergistic influence on cell proliferation took place (Han et al., 2015; **Supplementary Table S1**). The viability and migratory ability of multiple myeloma cells were two-fold decreased following simultaneous treatment with pre-miRNA-137 and pre-miRNA-197 as compared with monotherapy (Yang et al., 2015; **Supplementary Table S1**). Joint restoration of miRNA-193a and miRNA-600 expression promotes a 1.5-fold greater reduction in colony formation and a up to 2.5-fold greater induction of apoptosis as compared with separate mimic treatment (Li et al., 2018; **Supplementary Table S1**). There exists only one example showing that the application of miRNA mimic cocktails can attain a synergistic effect on multiple events of tumorigenesis. Notably, restoration of tumor suppressor miRNA-34a and miRNA-126 levels in pancreatic adenocarcinoma cells has been shown to provoke the inhibition of cancer cell viability by up to 75%, which exceeds the effect of monotherapy by threefold. At the same time, a considerable decrease in the migratory and invasive potentials (up to two-fold), as well as significant apoptosis induction, were established; approximately 30-50% of apoptotic cells are observed, which is four- and two-fold higher than the effects of miRNA-34a and miRNA-126 mimics alone, respectively (Feng et al., 2017; **Figure 3**, **Table 1**, and **Supplementary Table S1**).

The observed effectiveness of miRNA mimic combination therapy is sufficient for successful in vivo effects. Application of miRNA mimics in vivo has been demonstrated to be relatively potent in terms of tumor growth inhibition, and contributes to up to a two- and three-fold greater reduction in tumor node volume and weight, respectively, in comparison with monotherapy (Kasinski et al., 2014; Cheng et al., 2017; **Supplementary Table S1**). Moreover, in the case of simultaneous administration of miR-497 and miR-34a mimics, the in vivo effects of this mix are proven to be up to 4-fold more potent than each mimic alone (Han et al., 2015; **Table 1** and **Supplementary Table S1**).

Results indicate the great potential of miRNA mimic application, and undoubtedly, the next step is the development of preclinical protocols for cooperative therapy of oncopathologies using combinations of synthetic miRNA mimics.

### COMBINATION OF miRNA-BASED THERAPEUTICS AND CHEMOTHERAPY TO REVERSE DRUG RESISTANCE

Chemotherapy remains one of the main methods for the treatment of oncopathologies. FDA-approved drugs, including temozolomide, gemcitabine, sunitinib, paclitaxel, and cetuximab, possess different mechanisms of antitumor action. Temozolomide, as an alkylating agent, promotes the initiation of apoptosis as a result of DNA damage during replication (Ohba and Hirose, 2016); sunitinib exhibits an anti-angiogenic effect by inhibiting tyrosine kinase receptors, in particular, the vascular endothelial growth factor receptor (Patel et al., 2016); gemcitabine, being a nucleotide analog bearing 2<sup>0</sup> -fluorine, blocks DNA synthesis (Song et al., 2016); cetuximab represents a monoclonal antibody that represses cell proliferation through the selective inhibition of the epidermal growth factor receptor (Li et al., 2015); and paclitaxel, belonging to the class of taxanes, exerts its cytostatic activity via suppression of the normal reorganization of microtubules during mitosis (Yardley, 2013). All these agents exhibit relatively high antitumor activity; however, the development of cancer cell resistance considerably reduces the efficacy of such therapy (Schlack et al., 2016). A plethora of evidence highlights the essential role of miRNAs in the development of resistant tumor cell phenotypes. A lack of sensitivity to several chemotherapeutics may be a consequence of the hyperfunction of oncogenic miRNAs such as miRNA-26a, miRNA-18a, miRNA-29b-1, miRNA-431, miRNA-4521, and miRNA-155, or may be the result of a significant decrease in the activity of tumor suppressor miRNAs, including miRNA-575, miRNA-642b, miRNA-4430, miRNA-203a, and miRNA-203b (Mikamori et al., 2017; Yamaguchi et al., 2017; Ge et al., 2018; Aako et al., 2019; Wang H. et al., 2019; Wang M. et al., 2019). Aberrant functions of miRNAs may impede the interactions between miRNAs and their mRNA targets, a few examples of which are miRNA-210-3p and the multidrug efflux transporter ABCC5 (also known as MRP5; Amponsah et al., 2017); miRNA-101 and DNA-dependent protein kinases (Hu et al., 2017); miRNA-125a and the pro-apoptotic protein A20 (Yao et al., 2016); miR-145 and the ribosomal protein S6 kinase p70S6K1 (Lin et al., 2016); miRNA-181b and the cylindromatosis protein CYDL (Takiuchi et al., 2013); miRNA-144-3p and the AT-rich interactive domain 1A protein ARD1A (Xiao et al., 2017); and miRNA-199a-5p and miRNA-375 with the PH domain and leucine-rich repeat protein phosphatase 1 PHLPP1 (Mussnich et al., 2015). An impaired interaction between miRNA and its mRNA target unescapably results in tumor cell resistance, since these targets are elements of key cellular signaling cascades such as NF-κβ, AKT, and JAK2/STAT3, which control apoptosis, cell cycle progression, DNA repair, RNA editing, and nucleotide synthesis (Wu et al., 2018; Chen et al., 2019). Based on these data, investigators have attempted to use combinations of miRNA-based therapeutics in conjunction with chemotherapeutics to overcome the development of resistance and to increase the efficacy of treatment. Two strategies for the combined application of miRNA-targeted agents and chemotherapeutic drugs have been developed. The first includes preliminary transfection of cancer cells with miRNA mimics followed by chemotherapy, and the second uses simultaneous treatment of miRNA-based therapeutics and chemotherapeutics (**Figure 3**).

Application of the first approach involves preliminary treatment with miRNA mimics that leads to the restoration of sensitivity of cells to chemotherapeutic agents and considerable enhancement of the effectiveness of subsequent chemotherapy. For instance, transfection with mimics of miRNA-429, miRNA-383, miRNA-101-3p, miRNA-195, miRNA-634, or miRNA-1294 results in a two- to 5-fold decrease in the values of EC<sup>50</sup> and IC<sup>50</sup> for gemcitabine, temozolomide, and paclitaxel

(Fan et al., 2016; Yu et al., 2017; Chen et al., 2018; Tan et al., 2018; Jiang et al., 2019; Tu et al., 2019; Wang H. et al., 2019; **Supplementary Table S1**). A preparatory increase in cell sensitivity to gemcitabine by transfection with the synthetic tumor suppressor miRNA-205 has been shown to promote a two-fold decrease in the migratory potential of pancreatic cancer cells (Mittal et al., 2014; **Supplementary Table S1**). Moreover, 79% of cells are arrested in the G0/G1 phase of the cell cycle following combination treatment, which is 1.5-fold greater as compared with the use of the miRNA-205 mimic alone (Chaudhary et al., 2017; **Supplementary Table S1**). Another example is the precursory addition of miRNA-151a mimics to glioblastoma cells followed by temozolomide treatment, which contributes to the inhibition of colony formation promoting a three-fold decrease in the number of colonies in vitro (Zeng et al., 2018; **Supplementary Table S1**). In another case, transfection of metastatic renal carcinoma cells with miRNA-222 mimics followed by sunitinib treatment is 6-fold and two-fold more effective than monotherapy with miRNA mimic and sunitinib, respectively, in terms of angiogenesis inhibition (Khella et al., 2015; **Supplementary Table S1**).

As in the case of preliminary miRNA-based treatment, the strategy of simultaneous application of miRNA-based therapeutics and chemotherapies exerts high efficiency in vitro, providing an influence on the various events of tumorigenesis. Co-incubation of glioblastoma and colorectal cancer cells with sunitinib and an anti-miRNA-21 oligonucleotide or sunitinib and an miRNA-145 mimic attains up to a 75% decrease in cell viability, which is on average 1.5-fold more effective than each agent alone (Costa et al., 2013; Liu et al., 2015; **Supplementary Table S1**). Combination therapy with an miRNA-383 mimic and paclitaxel, an anti-miRNA-21 oligonucleotide and gemcitabine, and an miRNA-146 mimic and cetuximab promotes an average of two- to 2.5-fold more profound apoptosis induction in comparison with single-agent therapy (Huang et al., 2014; Li Y. et al., 2017; Jiang et al., 2019; **Supplementary Table S1**). In some exceptional cases, the percentage of apoptotic cells may be up to 6-fold higher than with mimic monotherapy (Dai et al., 2016; **Table 1** and **Supplementary Table S1**). An approximate 3.5-fold effective reduction in the invasive or migratory potentials of osteosarcoma, lung or pancreatic cancer cells was observed for combinations of an miRNA-34a mimic and celecoxib, an miRNA-let-7b mimic and paclitaxel, and an miRNA-205 mimic and gemcitabine, respectively, as compared with single therapeutics (Mittal et al., 2014; Dai et al., 2016; Chen et al., 2017; **Table 1** and **Supplementary Table S1**).

The efficiency of combination therapy with chemotherapeutics and miRNA mimics or anti-miRNA oligonucleotides has also been confirmed in vivo. Different combinations of chemotherapy, including gemcitabine, temozolomide, sunitinib, or paclitaxel, and miRNA-based therapeutics such as synthetic miRNA-205, miRNA-151à, miRNA-1291, and miRNA-let-7b mimics, or an anti-miRNA-21 oligonucleotide, have been shown to promote considerable inhibition of tumor growth, leading to a two- to four-fold retardation of tumor growth in comparison with monotherapy with each agent (Costa et al., 2015; Dai et al., 2016; Chaudhary et al., 2017; Li Y. et al., 2017; Mondal et al., 2017; Yu et al., 2017; Zeng et al., 2018; Tu et al., 2019; **Table 1** and **Supplementary Table S1**). Moreover, in the case of concurrent miRNA-218 mimic and temozolomide administration, a 40-fold inhibition of tumor progression is achieved (Fan et al., 2015; **Table 1**). In addition, using combinations of antisense oligonucleotide targeted to miR-21 with gemcitabine achieves complete elimination of liver metastases as well as application of an miR-151a mimic and temozolomide significantly increases mouse survival rate (Li Y. et al., 2017; Zeng et al., 2018; **Figure 3**, **Table 1**, and **Supplementary Table S1**).

### CONCLUSION

All the described strategies based on the combined application of miRNA-based therapeutics, such as mixes of antisense oligonucleotides or miRNA mimics, as well as their combination with chemotherapeutics, exhibit high therapeutic efficacy. Although many publications evaluated neither the additive nor synergistic effects of the therapeutic mixes, in some, a significant enhancement of the therapeutic efficacy was demonstrated upon their application. In particular, an additive influence on the cell viability or invasion was observed for an Mtg AMO-21/155/17 (Lu et al., 2009) and an miRNA-101 mimic and gemcitabine (Fan et al., 2016) or for an miRNA-34a mimic with an miRNA-126 mimic or celecoxib, respectively (Chen et al., 2017; Feng et al., 2017; **Supplementary Table S1**). Moreover, miRNA-34a and miRNA-let-7b mimics and a miRNA-1291 prodrug with nabpaclitaxel-conjugated gemcitabine inhibit tumor growth in vivo in an additive manner (Kasinski et al., 2014; Tu et al., 2019; **Supplementary Table S1**). A very limited number of publications noted a synergistic influence of combinations on tumorigenesis (**Table 1**). For instance, mixes of anti-miRNA-221 and antimiRNA-222 R8 conjugates, anti-miRNA-21 and anti-miRNA-10b ONs, miRNA-34a and miRNA-126 mimics, or an miRNA-634 mimic and temozolomide promote a synergistic manifold enhancement of apoptosis induction (Dong et al., 2012; Brognara et al., 2015; Feng et al., 2017; Tan et al., 2018), inhibition of colony formation (Zhang et al., 2015b; Tan et al., 2018), and a decrease in the invasive and migratory potential of cancer cells as compared with monotherapy (Dong et al., 2012; Kasinski et al., 2014; Chen et al., 2017; **Table 1** and **Supplementary Table S1**). In addition, a number of combinations provide synergistic inhibition of tumor growth ex vivo and in vivo, such as miRNA-34a and miRNA-497 mimics, miRNA-205 and gemcitabine, and miRNA-let-7b and paclitaxel (Han et al., 2015; Dai et al., 2016; Chaudhary et al., 2017; **Table 1** and **Supplementary Table S1**). In the most outstanding cases, a 40-fold retardation of tumor growth was attained using an miRNA-218 mimic and temozolomide (Fan et al., 2015), and complete elimination of metastases employing an anti-miRNA-21 ON and gemcitabine was achieved (Li Y. et al., 2017; **Table 1** and **Supplementary Table S1**). These combinations may act as prototypes for therapeutic regimens against malignancies such as hepatocellular carcinoma, pancreatic cancer, and glioblastoma.

All the analyzed data suggest that the basis of successful application of miRNA-based therapeutics lies in the choice of optimal miRNA targets that regulate key cellular pathways and form complicated networks of interactions with protein and mRNA targets in particular tumor cells.

The analyzed data demonstrate that miRNA-based therapeutics generate an entirely novel therapeutic paradigm in cancer treatment. It should be noted that miRNA-based therapeutics possess significantly less toxicity in contrast to existing drugs, in particular chemotherapeutics. For this reason, miRNA-based modalities may be applied as adjuvant agents to chemotherapy, allowing significant reductions in the dosage of antitumor drugs currently used in the clinic. Application of miRNA-based therapeutic mixes is, without question, an evolving area of antisense technology, and combinations of anti-miRNA ONs, mimics, and chemotherapeutics may represent highly efficient approaches to treating oncopathologies and other miRNAassociated diseases in the near future.

### REFERENCES


### AUTHOR CONTRIBUTIONS

SM analyzed and published the data, and prepared the manuscript. OP revised and corrected the manuscript.

### FUNDING

This work was funded by Russian Science Foundation (Grants 14-44-00068 and 19-74-30011) and the Russian State funded budget project of ICBFM SB RAS # AAAA-A17-117020210024-8.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar. 2019.00488/full#supplementary-material



proliferation and migration. Breast Cancer Res. 16:473. doi: 10.1186/s13058- 014-0473-z


center study, efficacy and safety. Indian J. Cancer 53, 118–122. doi: 10.4103/ 0019-509X.180844



**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 Miroshnichenko and Patutina. 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.

# Transcriptome Profiling Reveals Pro-Inflammatory Cytokines and Matrix Metalloproteinase Activation in Zika Virus Infected Human Umbilical Vein Endothelial Cells

*Svetlana Khaiboullina1, Timsy Uppal 1, Konstatin Kletenkov 2, Stephen Charles St. Jeor 1,3, Ekaterina Garanina2, Albert Rizvanov 2 and Subhash C. Verma1\**

*1 Department of Microbiology and Immunology, University of Nevada, Reno, Reno, NV, United States, 2 Department of Exploratory Research, Scientific and Educational Center of Pharmaceutics, Kazan Federal University, Kazan, Russia, 3 Genequest LLC, Reno, NV, United States*

#### *Edited by:*

*Hector A. Cabrera-Fuentes, University of Giessen, Germany*

#### *Reviewed by:*

*Sonia Maria Oliani, São Paulo State University, Brazil Andre Laval Samson, Walter and Eliza Hall Institute of Medical Research, Australia*

*\*Correspondence:*

*Subhash C. Verma scverma@med.unr.edu* 

#### *Specialty section:*

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

*Received: 07 November 2018 Accepted: 17 May 2019 Published: 12 June 2019*

#### *Citation:*

*Khaiboullina S, Uppal T, Kletenkov K, St. Jeor SC, Garanina E, Rizvanov A and Verma SC (2019) Transcriptome Profiling Reveals Pro-Inflammatory Cytokines and Matrix Metalloproteinase Activation in Zika Virus Infected Human Umbilical Vein Endothelial Cells. Front. Pharmacol. 10.642. doi: 10.3389/fphar.2019.00642*

The deformities in the newborns infected with Zika virus (ZIKV) present a new potential public health threat to the worldwide community. Although ZIKV infection is mainly asymptomatic in healthy adults, infection during pregnancy can cause microcephaly and other severe brain defects and potentially death of the fetus. The detailed mechanism of ZIKV-associated damage is still largely unknown; however, it is apparent that the virus crosses the placental barrier to reach the fetus. Endothelial cells are the key structural component of the placental barrier. Endothelium integrity as semi-permeable barrier is essential to control the molecules and leukocytes trafficking across the placenta. Damaged endothelium or disruption of adherens junctions could compromise endothelial barrier integrity causing leakage and inflammation. Endothelial cells are often targeted by viruses, including the members of the *Flaviviridae* family such as dengue virus (DENV) and West Nile virus (WNV); however, little is known about the effects of ZIKV infection of endothelial cell functions. Our transcriptomic data have demonstrated that the large number of cytokines is affected in ZIKV-infected endothelial cells, where significant changes in 13 and 11 cytokines were identified in cells infected with PRVABC59 and IBH30656 ZIKV strains, respectively. Importantly, these cytokines include chemokines attracting mononuclear leukocytes (monocytes and lymphocytes) as well as neutrophils. Additionally, changes in matrix metalloproteinase (MMPs) were detected in ZIKV-infected cells. Furthermore, we for the first time showed that ZIKV infection of human umbilical vein endothelial cells (HUVECs) increases endothelial permeability. We reason that increased endothelial permeability was due to apoptosis of endothelial cells caused by caspase-8 activation in ZIKV-infected cells.

Keywords: Zika virus, inflammation, cytokines, cell permeability, matrix metalloproteinase

### INTRODUCTION

Zika virus (ZIKV) is an approximately 11 kb positive-sense single-stranded enveloped RNA virus, classified into the *Flavivirus* genus within the *Flaviviridae* family, along with yellow fever virus (YFV), dengue virus (DENV), and West Nile virus (WNV). It is a rapidly emerging arbovirus (arthropod-borne virus), which was initially isolated from the serum of a febrile rhesus macaque in the Zika Forest of Uganda in 1947 and later on from the humans in Nigeria (Musso and Gubler, 2016). The virus, earlier thought to be restricted to the African and Asian continents, has now increased global attention due to its rapid spread throughout the Americas (Terzian et al., 2018) since its first detection in Brazil in May 2015.

Approximately 80% of ZIKV infections are asymptomatic, and the most common symptoms include fever, arthralgia, rash, myalgia, edema, vomiting, and non-purulent conjunctivitis (Moghadas et al., 2017). However, ZIKV infection in pregnant women has been linked to congenital microcephaly and other birth defects seen in neonates, such as placental insufficiency, fetal growth retardation, and fetal death (Chibueze et al., 2017; Ribeiro et al., 2018). ZIKV has also been linked to Guillain–Barré syndrome, a rare but serious auto-immune disorder (Parra et al., 2016). In humans, ZIKV is transmitted primarily *via* female mosquitoes, *Aedes aegypti* bite, through the skin of the infected host, followed by infection of permissive cells through specific receptors. Skin is thus believed to be the initial site of ZIKV replication, and from there, virus disseminates crossing blood– tissue barriers and could be detected in the brain, muscles, and placenta (Hamel et al., 2015; Aliota et al., 2016; Miner et al., 2016). Endothelial cells are the key elements of blood–brain barrier and a part of the placental blood barrier, which have been recognized as the site of flavivirus replication (Gowen et al., 2010; Beatty et al., 2015; Vervaeke et al., 2015). Earlier research studies on related flaviviruses, primarily DENV and YFV, showed that infected endothelial cells can produce inflammatory cytokines and chemokines, which further attract leukocytes to the site of virus propagation (Huang et al., 2000; Khaiboullina et al., 2005).

A recent study by Liu et al*.* (2016) reported that primary human endothelial cells are susceptible to productive infection by ZIKV. Specifically, both the African and South American ZIKV strains used in the study were found to efficiently infect and replicate in human endothelial cells, leading to the release of infectious virus. Conceptually, after its initial replication at the site of entry, ZIKV can spread *via* blood vessels while propagating in the infected endothelial cells (Liu et al., 2016). Recently, Szaba et al. (2018) have shown ZIKV antigens in embryonic endothelial cells and in the necrotic debris within the embryonic vasculature. These data strongly suggest the role of endothelial cells in propagation and dissemination of ZIKV to the fetus site.

Despite of intensive studies, our knowledge on alteration in transcriptional levels of cellular genes in ZIKV-infected endothelial cells remains limited. Also, the effect of ZIKV infection on the activation of cytokines in the endothelial cells is still poorly understood. The importance of cytokines for antiviral protection and cellular immune defense activation has been well established (Dinarello, 2007). Mediators induced in the infected endothelial cells may contribute to the loss of vascular integrity and recruitment of leukocytes. Therefore, knowledge on cytokine activation in infected endothelial cell will help to better understand the role of host immune responses in pathogenesis of ZIKV.

Here, we show the transcriptional profiles of cellular genes following *de novo* infection of ZIKV in endothelial cells. Interestingly, changes in the transcriptional levels of genes differed between cells infected with South American and African strains of ZIKV. Several pathways were identified to be activated in HUVECs infected with ZIKV. Our data, for the first time, demonstrate ZIKV strain-specific activation of cytokines and matrix metalloproteinase (MMPs). Also, we report an upregulated production of an active form of caspase-8 in ZIKVinfected endothelial cells. Taken together, these findings explain the inflammation and destruction commonly seen in ZIKVinfected tissues.

### MATERIALS AND METHODS

*Cell lines and Reagents.* Vero E6 cells were purchased from ATCC (American Type Culture Collection) and maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS, Atlanta Biologicals), 2 mM L-glutamine, 25 U/ml penicillin, and 25 μg/ml streptomycin. Human umbilical vein endothelial cells (HUVECs) were purchased from and cultured in M200 medium enriched with 50× large vessel endothelial supplement (Salvesen and Dixit; Thermo Fisher Scientific). All cell lines were grown at 37°C in a humidified chamber supplemented with 5% CO2.

MMP inhibitor GM6001 (10 µM), GM6001 negative control (10 µM), and caspase-8 inhibitor, Ac-IETD-CHO (50 µM/ml), were purchased from Santa Cruz Biotechnology (Santa Cruz, CA).

*Infection of HUVECs.* ZIKV strains, PRVABC59 (Human/2015, Puerto Rico, South America) and IBH30656 (Human/1968, Nigeria), were obtained from ATCC (Manassas, VA) and subsequently propagated in Vero cells. Monolayer Vero cells were inoculated with each ZIKV strain for 2 h. Unattached virus was removed by washing with medium before incubating with fresh medium. Virions were collected after 7 days following the removal of cell debris and quantified by qPCR. HUVEC monolayers were incubated with each ZIKV (multiplicity of infection (MOI) ~0.1) for 2 h. Virus was allowed to bind to the cells for 2 h at 37°C in a 5% CO2 atmosphere. Non-attached virus was removed by washing the cells with medium before incubating with the fresh medium. HUVECs were collected at indicated time points (3, 12, and 24 h) and stored at −80°C until used. Mock infection was carried out by incubation of HUVEC monolayers with supernatant from uninfected Vero cells for 2 h. In control experiments, ultraviolet (UV) light-inactivated ZIKV was used to infect HUVECs. ZIKV stocks were subjected to UV inactivation (1,200 µJ) in a UV Stratalinker 2400 (Stratagene).

*Plaque-forming assay for titrating ZIKV*. ZIKV titer was determined by plaque-forming assay as described previously (Khaiboullina et al., 2017). Briefly, infected Vero cells were overlaid with agarose (1%) containing DMEM medium supplemented with 10% FBS (HyClone, Logan, UT), 2 mM l-glutamine, 25 U/ml penicillin, and 25 μg/ml streptomycin. Seven days later, monolayers were fixed with 1% paraformaldehyde and stained with crystal violet (0.1%). All these assays were conducted under the biosafety level 2+ (BSL-2+) containment.

*Monocyte separation*. Cord blood buffy coats were obtained from the University of Colorado Cord Blood Bank, Aurora. Monocytes, CD14+ lymphocytes, were separated using Miltenyi magnetic bead separation kit (CD14 Microbeads; Miltenyi, Auburn, CA). Cells were rested overnight and cultured in DMEM medium supplemented with 10% FBS (HyClone, Logan, UT), 2 mM l-glutamine, 25 U/ml penicillin, and 25 μg/ml streptomycin.

Deidentified human cells were used in these assays, and all the experiments were done in accordance with guidelines of the University of Nevada, Reno. The Environmental and Biological Safety committee of the University of Nevada, Reno, approved the methods and techniques used in this study.

*Transwell migration assay*. Transwell system (5-µm pore size, Corning, Tewksbury, MA) was used to determine monocyte migration across HUVEC monolayer. HUVECs were seeded in the upper compartment and infected with PRVABC59 or IBH30656 ZIKV. Three days post-infection, monocytes (105 cells/insert) were added into the upper compartment. Monocyte migration across the HUVEC monolayer was analyzed 24 h later by detecting monocytes in the lower compartment.

*RNA extraction and Next Generation Sequencing (NGS).* Total RNA was extracted using Illustra RNAspin Mini kit (GE Healthcare). The RNA was quantified using NanoDrop UV spectrophotometer (ThermoFisher Scientific), and the quality of RNA was determined using Bioanalyzer (Agilent Technologies). The poly-A-containing mRNA was subjected to cDNA synthesis according to the TruSeq Stranded mRNA preparation guide (Illumina, San Diego, CA) to prepare the sequencing library. Validated and normalized libraries were diluted to a concentration of 4 nM, denatured with 0.2N NaOH, and then diluted to 20 pM with Hybridization buffer (HT1). The denatured and pooled libraries were pipetted onto a MiSeq Reagent Kit v3 and loaded on a HiSeq according to the manufacturer's recommended procedure. The FastQ data generated by the HiSeq (Illumina, San Diego CA) was annotated, and the sequence reads were analyzed by CLC workbench 10.0.0 (Qiagen, Germantown, MD) for the detection of viral and cellular genes. Differential expressions of cellular gene expressions based on the RPKM (reads per kilobase of transcript per million mapped reads) were determined by comparing with mock-infected cells using RNA-seq analysis tool of CLC Workbench.

*Transwell permeability assay*. HUVEC monolayers were placed on the transwell polycarbonate filters (0.45-μm pore size; Costar, Brumath, France). Monolayers were infected with ZIKV for 72 h. Transmembrane diffusion of FITC-dextran (70 kDa; Sigma) was used to detect changes in permeability of HUVEC monolayers, as described by Lander et al*.* (2014). FITC-dextran (1 mg/ml) was added to the upper compartment of the transwell system. Aliquots (100 μl) of culture medium were collected from the lower chamber at 3-, 6-, 12-, 24-, and 48-h intervals. The fluorescence was measured using a spectrophotometer (Fluoroskan Ascent; ThermoFisher Scientific). OD values of culture medium in the lower chamber of ZIKV-infected cells were presented as percent change to that in mock-infected HUVECs, which were set as 100%.

*Quantitative PCR (qPCR).* An aliquot of total RNA (40 ng) was used for synthesizing the cDNA (Superscript kit; ThermoFisher Scientific). Synthesized cDNA (1 μl for each target) was used for relative quantification of transcripts in a qPCR assay. ∆Ct values were calculated by normalizing with respective GAPDH Ct values, and the fold changes were calculated using ∆∆Ct method relative to the mock-infected control cells. The error bars represent standard deviation of three experiment replicates. Primer sequences are summarized in **Table 1**.

*Western blotting.* Total protein was collected in 0.1% solution of sodium dodecyl sulfate (SDS) at 24 h post-infection and normalized using Bradford Protein Assay. Proteins were separated on a 9% polyacrylamide gel (BioRad, Hercules, CA), transferred onto nitrocellulose membrane, and blocked [5% nonfat milk in Tris-buffered saline (TBS) and 0.5% Tween 20] for 1 h at room temperature. Membranes were then incubated (18 h, 4°C) with the rabbit anti-ZIKV envelope protein polyclonal antibodies (1:5000; GTX133314; GeneTex, Irvine, CA), rat anticaspase-8 (1:2,000, 645501; BioLegend, San Diego, CA), p-p38 mitogen-activated protein kinase (MAPK) (1:2,000, sc7973; Santa Cruz Biotechnology, Santa Cruz, CA), p38 MAPK (1:2,000, sc7149; Santa Cruz Biotechnology, Santa Cruz, CA), NF-κB (1:2,000, sc-8008; Santa Cruz Biotechnology, Santa Cruz, CA), Iκ-B (1:2,000; sc-1643, Santa Cruz Biotechnology, Santa Cruz, CA), and mouse anti-GAPDH (1:2,000; US Biological, Tustin, CA). The blots were washed (3×) with TBST and incubated with appropriate secondary antibodies conjugated with Alexa Fluor 680 (1:10,000, Molecular Probes, Carlsbad, CA). The membranes were scanned using the Odyssey scanner (Li-COR, Lincoln, NE).

*Cytokine detection*. The Bio-Plex human 21-Plex, 27-Plex, and 40-Plex and matrix metalloproteinase (MMP) 9-Plex (BioRad) were used for analyzing samples according to the manufacturer's recommendations. Fifty microliters of the sample was used for determining cytokine concentration, and the collected data were analyzed using Luminex 200 analyzer with MasterPlex CT control software and MasterPlex QT analysis software (MiraiBio division of Hitachi Software San Francisco, CA, USA). Each Bioplex analysis was conducted in triplicates, and each experiment was repeated three times.

*Protein interaction network analysis:* The Search Tool for the Retrieval of Interacting Genes/Proteins (STRING version 9.0) was used for analyzing interactions between cytokines with differential expression between ZIKV-infected and control


TABLE 2 | Cytokine and metalloproteinase activation in HUVECs infected with PRVABC59 ZIKV and IBH30656 ZIKV.


HUVECs (Franceschini et al., 2013). STRING analysis was conducted using high confidence (score 0.7). Cluster analysis was conducted using k-means with a value of k = 3.

*Immunofluorescence analysis.* Cells were fixed (3:1 methanol/ acetone) and stored at −80°C until use. Slides were permeabilized with 0.1% TritonX-100 for 30 min, washed (3×), and blocked (3% normal donkey serum, 0.5% BSA) for 60 min at room temperature. Monolayers were washed again (3×) and incubated with rabbit anti-ZIKV envelope polyclonal primary antibody (1:1,000; GeneTex, Irvine, CA), rat-anti VE cadherin antibody (1:100, 138101; Biolegend, San Diego, CA), or mouse antiannexinV (1:100, sc-74438; Santa Cruz Biotechnology, Santa Cruz, CA) for 1 h at room temperature, followed by incubation with goat anti-rabbit Alexa Fluor 488, donkey anti-rat Alexa Fluor 488 (1:5000; Molecular Probe, Carlsbad, CA), or goal anti-mouse Alexa Fluor 488 secondary antibody for 1 h at room temperature in the dark, respectively. The nuclei were stained with TO-PRO-3 (ThermoFisher Scientific, Waltham, MA). Cells were examined by Carl Zeiss LSM 780 confocal laser-scanning microscope.

*Statistical analysis.* Statistical analyses were performed using Prism 6.0 software (Graphpad Inc.), and the p-values were calculated using two-tailed t tests. An asterisk represents statistical significance on the graphs. Pathway analysis was done using the Pathway Studio MammalPlus (Elsevier) package. Analytes which differ statistically between groups were used for enrichment analysis. For the enrichment analysis, only analytes with the adjusted p < 0.05 were selected. The Benjamini– Hochberg method was used to control the false recovery rate.

**Genebank accession number:** The RNA-seq data are submitted to Gene Expression Omnibus (accession number GSE103114) and can be accessed by the following link: https:// www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE119232.

#### RESULTS

**ZIKV efficiently infects HUVECs**. HUVEC monolayers were infected with both the strains of ZIKV, PRVABC59 (South America) or IBH30656 (Nigeria), at a MOI of 0.1. HUVEC susceptibility to ZIKV infection and efficacy of virus replication was analyzed using next-generation sequencing of mRNA, real-time qPCR, Western blotting, and immunofluorescence assays. Mapping of the sequence reads with reference ZIKV genome revealed an accumulation of ZIKV transcripts in HUVECs, and a representative image of the mapped reads from 12 h post-infected samples is shown in **Figure 1**. The mockinfected HUVECs showed only one or two reads, as expected, while the ZIKV-infected cells showed a large number of reads (green and red lines representing reads) mapping to the viral genome, confirming that HUVECs were efficiently infected with both the strains of ZIKV (**Figure 1-I**; PRVABC59-panel B and IBH30656-panel C). To determine whether the incoming viral genome of ZIKV undergoes active replication, viral genome copies were quantified using qPCR at 24 and 72 h post-infection (hpi). Relative copies of ZIKV genome at these time points post-infection revealed an accumulation of ZIKV, confirming an active ZIKV replication in these cells (**Figure 1-II**). Detection of the viral envelope protein by Western blotting and immunofluorescence assays confirmed the transcriptional activation and expression of viral proteins following active replication of ZIKV (**Figure 1-III** and **IV**). The mock-infected cells lacking the band or signal for envelope protein confirmed the specificity of these assays (**Figure 1-III** and **IV**).

We also confirmed ZIKV replication in infected HUVECs by determining virions in the culture supernatant from 72 hpi cells through plaque-forming assay. Interestingly, both ZIKV strains, PRVABC59 and IBH30656, produced nearly similar number of plaques, 2 × 104 and 1.5 × 104 , respectively, per ml of the culture supernatant (**Figure 1-V**).

**PRVABC59 ZIKV affects significantly higher number of cellular pathways as compared to the IBH30656 ZIKV***.* Differential expression of cellular genes in ZIKV-infected HUVECs was analyzed by comparing with mock-infected control cells, and the genes with more than three-fold difference were selected for further analysis. Genes at early time postinfection (3 hpi) showed 110 differentially regulated genes in

HUVECs infected with ZIKV strain PRVABC59; lane 3: HUVECs infected with ZIKV strain IBH30656. (IV) Immunofluorescence analysis of ZIKV envelope protein in the infected HUVECs. Images were captured using Carl Zeiss LSM780 confocal laser-scanning microscope. HUVECs infected with PRVABC59 ZIKV: A: envelope localization, B: nuclei; C: merge of A and B, D-DIC. HUVECs infected with IBH30656 ZIKV strain: E: envelope localization, F: nuclei; G: merge of E and F, H-DIC. Mock-infected HUVECs: I: envelope localization; J: nuclei; K: merge of I and J, L-DIC. Bar represents 10 μm size. (V) Plaque assay analysis of ZIKV replication in HUVECs. Supernatants were collected 72 h post-infection and used to determine infectious virus.

PRVABC59-infected HUVECs, whereas only 36 genes were differentially expressed in HUVECs infected with IBH30656 (**Figure 2**). Importantly, 23 genes were commonly affected by both the strains of ZIKV, as shown in a Venn diagram. Similarly, genes analyzed from 12 hpi cells showed 65 genes affected by both the strains of ZIKV. However, the total number of genes affected by PRVABC59 was significantly higher (610) as compared to those affected by IBH30656 (112) in infected HUVECs upon progression of infection.

#### Identification of the Cellular Pathways Activated in ZIKV-Infected HUVECs Using Pathway Studio, MammalPlus

Differentially expressed cellular genes in these infected HUVECs were used for identifying pathways modulated by ZIKV. ZIKV infection affected multiple cellular genes and pathways at all the time points following infection including at 3 hpi. These pathways include leukocyte migration, leukocyte differentiation, platelet activation, focal cell junction assembly, and apoptosis. Interestingly, pathways involved in regulation of endothelial permeability, extracellular matrix turnover, platelets and neutrophil activation, as well as cytoskeleton reorganization were affected in ZIKV-infected HUVECs. Among these, PRVABC59 ZIKV infection led to an increase in the expression of cytochrome c, while IBH30656 infection resulted in a decreased level of cytochrome c (**Figure 3**).

**Cytokines and MMPs activation in ZIKV-infected HUVECs.**  Cytokine activation has been previously demonstrated in flavivirus-infected HUVECs (Lin et al., 2002; Khaiboullina et al., 2005). Therefore, we sought to determine whether ZIKV infection activates cytokine production in HUVECs. Culture supernatants from the mock-infected HUVECs as well as cells infected with either ZIKV strains were collected at 24 hpi and used for determining the levels of cytokines (**Table 2**). Out of 58 cytokines tested, the levels of 13 cytokines (IL-1β, IL-10, IL-15, IL-16, CCL5, G-CSF, HGF, LIF, MCSF, PDGFbb, CXCL1, CXCL9, and CXCL12) were affected in HUVECs infected with PRVABC59 ZIKV, while 11 cytokines (IL-15, CCL2, CCL5, bFGF, G-CSF, HGF, LIF, MCSF, CXCL1, CXCL11, and CXCL12) were altered in IBH30656 ZIKV-infected HUVECs as compared to the mock-infected controls (**Table 2**). Among these, eight cytokines (IL-1β, IL10, CCL5, G-CSF, CSF, CXCL1, and CXCL12) were upregulated and five (IL-15, IL-16, HGH, PDGFbb, and CXCL9) were downregulated in PRVABC59-infected HUVECs compared to the mock-infected HUVECs (**Figure 4**). Out of the 11 cytokines affected in IBH30656-infected HUVECs, 8 cytokines were higher (CCL2, CCL5, bFGF, G-CSF, LIF, M-CSF, CXCL1, and CXCL12), while 3 (IL-15, HGH, and CCXL11) were lower than the mockinfected control cells. Cytokines IL-15, CCL5, HGF, LIF, M-CSF, CXCL1, and CXCL12 were commonly upregulated by PRVABC59 and IBH30656. Interestingly, some cytokines were expressed in HUVECs infected with each ZIKV strain. IL-1β, IL-10, IL-16, PDGFbb, and CXCL9 were altered only in PRVABC59-infected HUVECs as compared to mock-infected control, while only three cytokines, namely, CCL2, bFGF, and CXCL11, were altered in IBH30656-infected HUVECs.

Cytokine activation and endothelial monolayer permeability is regulated by matrix metalloproteinases (MMPs), which act by degrading the extracellular matrix to facilitate the movement of molecules and cells across the blood–tissue barrier (Lakhan et al., 2013). Since changes in MMPs have been demonstrated in HUVECs infected with *Flaviviridae*, we sought to determine whether ZIKV infection will affect secreted MMP levels (**Table 3**). Indeed, the levels of secreted MMP8, MMP10, and MMP13 were altered in HUVECs infected with both the strains of ZIKV. Interestingly, the levels of MMP8 were reduced, while MMP10 and MMP13 levels were increased in ZIKV-infected as compared to the mock-infected cells.

**String analysis**. STRING analysis generates a protein interaction profile using different sources including interaction databases, text mining, and genetic interactions (Szklarczyk et al., 2015). Cytokines having significantly altered expression in HUVECs due to ZIKV infection were used as an input to the STRING tool. STRING provides different viewing options

FIGURE 3 | Pathway analysis of transcriptome data. Pathway analysis was done using the Pathway Studio MammalPlus (Elsevier) tool. Analytes that differ significantly between groups were used for enrichment analysis. Only pathways with p < 0.05 after correction for the multiple comparison were selected. Blue color highlights analytes, which were significantly lower in IBH30656 ZIKV-infected HUVECs. (A) PR vs.CON-12hpi – Extracellular matrix Turnover; (B) NIG vs.CON-12hpi – Extracellular matrix Turnover; (C) - PR vs. CON-12hpi -Neutrophil Activation via FCGR3B.

TABLE 3 | MMPs activation in HUVECs infected with PRVABC59 ZIKV and IBH30656 ZIKV.


including action, evidence, and confidence. STRING analysis identified two clusters in HUVECs infected with both the strains of ZIKV (**Figure 5**). The upregulated cytokines in IBH30656 infected HUVECs were in two clusters: G-CSF, HGF, LIF, and CXCL12 in the first cluster, and IL-15, CCL2, CCL5, CXCL1, CXCL11, and FGFb in the second cluster (**Figure 5A**). In PRVABC59-infected HUVECs, one cluster contained IL-1β, G-CSF, HGF, LIF, and CXCL12, while IL-15, IL-16, CCL5, CXCL1, and CXCL9 were in another cluster (**Figure 5B**). In STRING analysis connection between these two clusters, it appears that IL-1β regulates the expression of GM-CSF and CXCL1, while bFGF activates CCL2 and CCL5.

**ZIKV-induced trans-endothelial migration of monocytes.** MMPs and cytokines/chemokines play an important role in regulation of leukocyte migration across endothelium (Gerhardt and Ley, 2015). We sought to determine whether infection of HUVECs with ZIKV will enhance monocyte cross endothelial migration. We infected HUVECs with PRVABC59 and IBH30656 ZIKV at MOI of 0.1 and determined migrating monocytes, which showed a significant number of monocytes that migrated across the ZIKV-infected HUVEC monolayer as compared to the mock-infected control HUVEC monolayer (**Figure 6A**). Importantly, these migrated monocytes were ZIKV antigen positive, suggesting that they could contribute to the dissemination of ZIKV across the endothelial barrier (**Figure 6B**).

**ZIKV increased HUVEC permeability.** Several members of the *Flaviviridae* family are known to target endothelial cells and affect vascular endothelium permeability (Gowen et al., 2010; Beatty et al., 2015; Vervaeke et al., 2015). Change in permeability could lead to local edema and inflammation, which are hemorrhages and leukocyte extravasation (Nourshargh and Alon, 2014; Chanthick et al., 2018). Therefore, we sought to determine whether ZIKV infection disturbs the permeability of HUVEC monolayer *in vitro* using modified Boyden chamber system. Following infection of HUVECs with PRVABC59 or IBH30656 ZIKV (MOI 0.1), FITC-dextran was added into the upper compartment of the Transwell system. Aliquots of culture medium from the lower chamber were collected at selected time points to determine the presence of dextran molecule. Significantly higher amount of the 70-kDa FITC-dextran was detected in the lower compartment of ZIKV-infected endothelial monolayers as compared to mock-infected control cells (**Figure 7A**). Importantly, both ZIKV strains caused increased dextran permeability.

Since MMPs are implicated in pathogenesis due to an increased endothelial monolayer permeability (Alexander and Elrod, 2002), we sought to determine whether inhibition of MMPs will restore the permeability of ZIKV-infected HUVEC monolayer. HUVECs were infected with PRVABC59 or IBH30656 ZIKV (MOI 0.1) followed by changing the medium and replacing with fresh medium supplemented with MMP inhibitor GM6001 (10 µM) or GM6001 negative control (10 µM). HUVEC monolayer permeability was determined by FITC-dextran detection in the lower compartment of the Transwell system. Interestingly, inhibition of MMPs led to a restoration of ZIKV-mediated cellular permeability almost to a level in uninfected control cells (**Figure 7B**). MMP negative control did not affect the endothelial monolayer permeability and failed to restore permeability in ZIKV-infected cells (**Figure 7**C).

**ZIKV infection changed VE-cadherin expression.** VE-cadherin plays an important role in regulating endothelial permeability (Gavard, 2013). Therefore, we sought to determine whether changes in permeability could be explained by ZIKV affecting VE-cadherin expression (**Figure 8**). Strong VE cadherin expression was demonstrated in mock-infected HUVECs. VE cadherin accumulation was detected as continuous lines around each HUVEC cell (**Figure 8A**, white arrow). However, VE cadherin expression was changed in ZIKV-infected HUVECs, where its expression was lower in some areas of endothelial cells (**Figure 8B** and **C**, yellow arrow). Also, open spaces free of VE cadherin were detected in ZIKV-infected HUVECs (**Figure 8B** and **C**, red arrow).

**ZIKV-induced caspase-8 in HUVECs**. Increased endothelial permeability could be due to the ZIKV-induced endothelial cell death, which has been documented in cells infected with DENV, WNV, and Japanese Encephalitis virus, other members of *Flaviviridae* (Ghosh Roy et al., 2014). Our pathway data analysis indicates cytochrome c mobilization in ZIKV-infected HUVECs. The mitochondria activation may lead to an upregulation of caspase-8, which plays important role in the initiation of apoptosis (Elmore, 2007). Caspase-8 is also known to activate caspase-3, a main executional caspase (Elmore, 2007). Our data showed an increased level of full-length caspase-8 (Procaspase-8) and the release of an active form of caspase-8 in both ZIKVinfected cells (**Figure 9A**). Viral replication was confirmed by expression of ZIKV envelope protein (**Figure 9A**). An increase in the transcriptional levels of caspase-8 was determined in ZIKV-infected HUVECs (**Figure 9B**). Also, it appears that ZIKV replication is required for caspase-8 activation, as increased level of caspase-8 transcripts was found only in HUVECs infected with infectious virus, while these transcripts were absent in cells infected with UV-inactivated ZIKV (**Figure 9D**).

Since caspase-8 is an executive caspase in the apoptotic pathway, we sought to determine whether ZIKV triggers cell death. Apoptosis was evaluated by the detection of annexinV expression in ZIKVinfected cells. Increased annexinV expression was demonstrated in ZIKV-infected HUVECs using Western blot (**Figure 9A**) and immunofluorescence assays (**Figure 9C**). To determine the effects of ZIKV on HUVEC cell growth and viability, 4 × 105 /well cells were placed into the six-well plate before infection with PRVABC59 or IBH30656 at MOI of 0.1. Cells were trypsinized after 72 hpi and

control was calculated. The binary logarithm of these values was used for sorting of analytes and generation of "lollypop" diagram with "ggplot2" package.

counted using trypan blue (**Figure 9D**). The number of HUVECs in uninfected controls was 1.18 × 106 ± 7.7 × 104 , while the number of live cells was significantly reduced in sets infected with IBH30656 ZIKV (9.8 × 105 ± 8.5 × 104 ; p < 0.1) or PRVABC59 (3.3 × 105 ± 4.4 × 104 ; p < 0.0001) (**Figure 9E**).

**Caspase-8 activation linked cell death could contribute to HUVEC monolayer permeability.** To address this assumption, HUVEC monolayer permeability was analyzed *in vitro* using modified Boyden chamber system. HUVEC monolayers were infected with PRVABC59 or IBH30656 ZIKV (MOI 0.1) for 1 h, virus inoculum was removed, and fresh medium supplemented with Ac-IETD-CHO (50 µM), a caspase-8 inhibitor, was added. FITC-dextran was applied into the upper compartment of the Transwell system, and aliquots of culture medium from the lower

chamber were collected at selected time points. Significantly higher amount of the 70 kDa FITC-dextran was detected in the lower compartment of ZIKV-infected HUVEC monolayers as compared to mock-infected controls (**Figure 10**). Inhibition of caspase-8 decreased permeability of HUVECs infected with PRVABC59 or IBH30656 ZIKV. These data suggest that activation of caspase-8 alters the endothelial monolayer integrity of ZIKVinfected cells.

**ZIKV activates mitogen-activated protein kinase p38 and NF-κB kinases**. MAPK p38 and NF-κB are shown to be activated in cells infected with *Flaviviruses* (Marianneau et al., 1997; Kesson and King, 2001). Activation of these kinases is essential for upregulation of cytokines and matrix metalloproteinases (Prickett and Brautigan, 2007;Lawrence, 2009). Therefore, we sought to determine whether ZIKV infection also upregulates these kinases. HUVECs were infected with PRVABC59 (South America) or IBH30656 (Nigeria) at a MOI 0.1. Proteins were collected at 72 hpi and used for Western blot analysis of p38 and NF-κB expression (**Figure 11**). Both ZIKV increased expression of p38 MAPK and NF-κB kinases. Increased level of IκB was demonstrated in ZIKV-infected cells, suggesting its dissociation from NF-κB. Also, increased expression of phosphorylated p38 kinase was found in ZIKV-infected cells.

### DISCUSSION

ZIKV is a member of the Flavivirus family, which was recently shown to be associated with severe defects in newborns, microcephaly (Campos et al., 2015). Although infection is often asymptomatic and rarely presents with clinical symptoms, the newborn brain is targeted by the virus, causing tissue damage and inflammation (Chibueze et al., 2017; Moghadas et al., 2017). It appears that after initial replication at the site of entry, virus disseminates, reaching tissues and organs including placenta. Placenta is a complex barrier, where endothelial cells play central role in regulating the transfer of materials between the mother and the fetus (Griffiths and Campbell, 2015). Endothelium plays an important role in regulating barrier permeability and protecting fetus from pathogens (Pang et al., 2017). However, viruses targeting endothelial cells could compromise endothelial monolayer integrity and consequently can cross the organ barrier. Our data suggest that endothelial cells are susceptible to ZIKV infection and cause cytokine activation and increased endothelial monolayer permeability, which could contribute to pathogenesis of ZIKV infection.

Endothelial cells are susceptible to ZIKV infection (Papa et al., 2017; Roach and Alcendor, 2017; Peng et al., 2018), where virus replication was shown in human brain microvascular endothelial cells (HBMECs), retinal endothelial cells (RECs), and HUVECs. ZIKV infection of HBMECs and RECs led to activation of several proinflammatory cytokines including IL-6, CCL1, and CCL5 (Papa et al., 2017; Roach and Alcendor, 2017). To expand our understanding on inflammatory response in ZIKV-infected endothelial cells, we sought to determine changes in cytokines and chemokines in ZIKV-infected endothelial cells. Additionally, ZIKV effect on MMPs in HUVECs was investigated.

We show that ZIKV infection activates proinflammatory (IL-1β) cytokines as well as chemokines (CCL2, CCL5, GM-CSF, G-CSF, CXCL1, and CXCL12) in the endothelial cells. The cytokine upregulation could be explained by increased expression of p38 MAPK and NF-κB kinases in ZIKV-infected cells. These kinases are major regulators of cell responses to various stimuli including cytokine activation (Prickett and Brautigan, 2007; Lawrence, 2009). Interestingly, the cytokines activated in HUVECs suggests that multiple subsets of leukocytes could be attracted to the site of infection including mononuclear leukocytes (monocytes and lymphocytes) as well as granulocytes (neutrophils). Additionally, we found that the activation of cytokines in HUVECs was ZIKV strain dependent as 13 cytokines were affected in PRVABC59 infected cells as compared to 11 cytokines in IBH30656-infected cells. It could suggest that ZIKV infection promotes mononuclear leukocytes and neutrophil chemotaxis across the endothelial barrier. Recently, ZIKV-positive monocytes were demonstrated

FIGURE 7 | Transwell permeability assay. HUVECs were seeded onto Transwell inserts and infected with PRVABC59 or IBH30656 ZIKV (MOI 0.1) or mock. FITC-dextran was added into the upper compartment of the Transwell system. Aliquots of culture medium from the lower chamber were collected at selected time points to determine the presence of dextran molecule. Data are presented as a percent change in permeability as compared to mock-infected control. MMPs were inhibited using GM6001 (10 µM). Also, GM6001 negative control (10 µM) was used. (A) FITC-dextran permeability of HUVECs infected with PRVABC59 or IBH30656 ZIKV; (B) FITC-dextran permeability of HUVECs infected with PRVABC59 or IBH30656 ZIKV and treated with GM6001; (C) FITC-dextran permeability of HUVECs infected with PRVABC59 or IBH30656 ZIKV and treated with GM6001, negative control. \*\*\*p < 0.0001 by paired t test. Data shown are the mean ± Sterr of three independent experiments.

#### FIGURE 9 | Continued

At 72 hpi, total proteins were collected and used for the detection of envelop protein, caspase-8, and annexinV. GAPDH was detected as the loading control. Lane 1: mock-infected HUVECs, lane 2: HUVECs infected with ZIKV strain PRVABC59; lane 3: HUVECs infected with ZIKV strain IBH30656. (B) qPCR analysis of virus transcript accumulation and caspase-8 transcription activation in ZIKV-infected HUVECs. Relative copies of each transcript were calculated using ∆∆Ct method. (C) IFA analysis of annexinV expression in ZIKV-infected HUVECs. A: mock-infected control; B: ZIKV strain PRVABC59; C: ZIKV strain IBH30656. Bar represents 10 μm size. (D) Effect of ZIKV replication on caspase-8 expression. UV-inactivated and replication-competent ZIKV PRVABC59 and IBH30656 were used to infect HUVECs. Total RNA was collected at 72 hpi and analyzed using qPCR. Relative copies of the viral genomes (ZIKV) were calculated using ∆∆Ct method. (E) Effect of ZIKV infection on cell vitality: mock-infected and ZIKV-infected HUVECs (IBH30656 and PRVABC59 strains) were counted using trypan blue. Experiments were performed in triplicate for three independent times.

control.

in the blood of infected individuals (Lum et al., 2017). We have also shown that monocytes are susceptible to ZIKV infection and virus replication (Khaiboullina et al., 2017). Therefore, we suggest that ZIKV-infected endothelial cells attract monocytes carrying viral antigens and promote their migration into the tissue.

It is striking that the level of IL-1β was upregulated in HUVECs infected with PRVABC59 ZIKV, and it remained unaffected in cells infected with IBH30656 ZIKV. IL-1β is a product of activated inflammasomes (Netea et al., 2015). Inflammasomes are complex polymers assembled when pathogen recognition receptors (PRRs) are activated (Ferreri et al., 2010). Inflammasomes activate pro-caspase1, which proteolytically cleaves IL-1β (Netea et al., 2015) IL-1β is powerful pro-inflammatory cytokine shown to exacerbate damage during chronic disease and acute tissue injuries (Dinarello, 2010). Additionally, IL-1β release indicates caspase-1 activation. Caspase-1 can trigger pyroptosis, which is similar to apoptosis, but does not require the activation of death caspases (Brennan and Cookson, 2000; Jesenberger et al., 2000). Therefore, it could be suggested that IL-1β activation by PRVABC59 ZIKV could lead to a serious clinical outcome due to severe inflammation and pyroptotic cell death. The importance of this discovery is that PRVABC59 ZIKV, South American strain, belong to a group of ZIKV associated with newborn microcephaly, while IBH30656 ZIKV strain (African origin) has not been linked to microcephaly. We believe that severe damage to fetal brain tissue and consequent microcephaly could be the result of ZIKV-caused inflammasome activation by the South American strains of virus. We have shown that IL-1β is activated in HUVECs infected with both ZIKV strains, suggesting that the pathogenesis of ZIKV-caused microcephaly is complex, where crossing the placenta and targeting neuronal progenitors are essential. It could be suggested that these features are characteristic only for the South American ZIKV, while they are absent in African strains.

IL-15, which promotes proliferation of natural killer cells (NKs) and plays important role in innate antiviral defense (Biron et al., 1999), was downregulated in ZIKV-infected endothelial cells. IL-15 induces anti-viral defense by an activation of IFN-α production (Foong et al., 2009) and stimulation of NK activity (Azimi et al., 2000; Ashkar and Rosenthal, 2003). The mechanisms of IL-15 downregulation remain to be determined; however, it could be suggested that decreased interleukin production may contribute to virus replication.

We, for the first time, show that ZIKV infection affects MMP expression in endothelial cells. Interestingly, the levels of MMP10 and MMP13 were upregulated in infected HUVECs and the upregulation of MMPs production is commonly seen in dysregulated extracellular matrix leading to inflammation, angiogenesis, and leukocyte trafficking (Page-McCaw et al., 2007). Our most intriguing observation was that ZIKV infection led to a decreased level of MMP8 from the endothelial cells. Recent study by Gutierrez-Fernandez et al. has demonstrated that MMP8 has more complex role in inflammation as its deficiency prolongs inflammation and delays the wound healing (Gutierrez-Fernandez et al., 2007). Another study by Fang et al. (2013) demonstrated the MMP8 knockout mice have less endothelial cell sprouting, migration, and capacity to proliferate, which explains why the lack of this MMP delays wound healing. Increased MMP10 and MMP13 are further supporting the notion that ZIKV infection promotes inflammation and extracellular remodeling. MMP13 is absent in normally healing wounds but is largely expressed in chronic lesions (Toriseva et al., 2012). Also, increased expression of MMP13 was shown to promote inflammation and cartilage destruction (Singh et al., 2013). Our hypothesis for MMP's role in ZIKV-caused HUVEC monolayer leakage is supported by the data collected using MMP inhibitors. We showed that MMP inhibition restores the permeability of endothelial monolayer. Together, these data support the role of MMPs in ZIKV-induced inflammation and tissue destruction.

Our data demonstrate that ZIKV compromises endothelial monolayer permeability. Interestingly, increased permeability was found in HUVECs infected with both PRVABC59 and IBH30656 strains of ZIKV. These results contrast the previous findings showing the lack of changes in ZIKV-infected HBMECs monolayer permeability (Papa et al., 2017). The variation in endothelial permeability could be related to the physiological differences between HBMECs and HUVECs. It has been shown that, under stress, HBMECs can retain cobbled stone form and minimize the span of tight junctions per unit of the capillary length, which is essential to reduce permeability (Ye et al., 2014). It appears that HUVECs obtain spindle shape, which can increase the monolayer's permeability. Our findings corroborate the data recently published by Szaba et al. (2018), where embryonic endothelial cell necrosis was demonstrated in ZIKV-infected fetus. The compromised permeability of HUVECs could also be associated with ZIKV-caused cell damage.

We found that increased ZIKV permeability was associated with decreased VE cadherin expression and increased MMP expression. VE cadherin is found exclusively in endothelial cells, where it regulates expression of multiple junction molecules and important adhesion molecules (Taddei et al., 2008; Giampietro et al., 2012; Ye et al., 2014). Interestingly, it appears that low VE cadherin expression is associated with upregulation of MMPs (Llorens et al., 1998). This reciprocal expression of VE cadherin and MMPs is essential for regulation of endothelial cell growth, angiogenesis, and permeability (Kiran et al., 2011). We suggest that downregulation of VE cadherin in ZIKV-infected cells could be related to upregulation of MMPs.

This assumption is supported by our observation of an increased expression of an active form of caspase-8. Caspase-8 is an initiating caspase, which is activated by proteolytic processing in the cytoplasm (Salvesen and Dixit, 1999). Caspase-8 can be activated by the death receptors of the Fas/tumor necrosis factor receptor family (Kischkel et al., 1995). Once activated, caspase-8 can propagate the apoptotic signal by proteolytically releasing downstream executing caspases or by cleaving the BH3-interacting domain death agonist (BID), which migrates to mitochondria and triggers the release of cytochrome c (Luo et al., 1998). Once released, cytochrome c activates caspase-9, which plays a role in the execution phase of apoptosis (Juin et al., 1999). Interestingly, the pathway analysis revealed cytochrome c release in ZIKV-infected HUVECs; therefore, we suggest that ZIKV-caused endothelial cell apoptosis is due to caspase-8 induced cytochrome c release.

In conclusion, our data for the first time demonstrate that ZIKV triggers changes in transcription of genes involved in multiple cellular pathways. Also, the transcriptional activation was detected early following the infection with significantly more genes due to South American strain as compared to African strain. Changes in the cytokine and MMP levels by ZIKV could explain inflammation and tissue destruction caused by infection. Additionally, activated cytokines could contribute to monocyte differentiation and migration across the endothelium tissue barrier.

#### AUTHOR CONTRIBUTIONS

SK and TU performed the experiments. KK and EG performed the pathway analysis. SJ performed the sequencing. AR

#### REFERENCES


analyzed the data. SK and SV analyzed the data and wrote the manuscript.

#### FUNDING

RAA was supported by Program of Competitive Growth of Kazan Federal University and state assignment 20.5175.2017/6.7 of the Ministry of Science and Higher Education. This publication was made possible by a grant from the National Institute of General Medical Sciences (GM103440) from the National Institutes of Health. This work was supported by the institutional and departmental funds.

#### ACKNOWLEDGMENTS

We thank Genequest for assistance with next-generation RNA-sequencing.


interplay between VEGF-A165a, VEGF-A165b, PIGF and VE-cadherin. *Clin. Sci. (Lond)* 131, 2763–2775. doi: 10.1042/CS20171252


**Conflict of Interest Statement:** SJ was employed by Genequest LLC, Reno, NV.

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

*Copyright © 2019 Khaiboullina, Uppal, Kletenkov, St. Jeor, Garanina, Rizvanov and Verma. 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.*

## Bioinformatic Study of Transcriptome Changes in the Mice Lumbar Spinal Cord After the 30-Day Spaceflight and Subsequent 7-Day Readaptation on Earth: New Insights Into Molecular Mechanisms of the Hypogravity Motor Syndrome

*Maksim Sergeevich Kuznetsov1\*, Artur Nicolaevich Lisukov1, Albert Anatolevich Rizvanov2, Oksana Victorovna Tyapkina1,3, Oleg Aleksandrovich Gusev2,4, Pavel Nicolaevich Rezvyakov1, Inessa Benedictovna Kozlovskaya5, Elena Sergeevna Tomilovskaya5, Evgeny Evgenievich Nikolskiy1,2,3 and Rustem Robertovich Islamov1,3\**

*1 Department of Medical Biology and Genetics, Kazan State Medical University, Kazan, Russia, 2 Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, 3 Kazan Institute of Biochemistry and Biophysics, Federal Research Center "Kazan Scientific Center" of RAS, Kazan, Russia, 4 RIKEN Institute, Yokohama, Japan, 5 Institute of Biomedical Problems of Russian Academy of Sciences, Moscow, Russia*

The hypogravity motor syndrome (HMS) is one of the deleterious impacts of weightlessness on the human body in orbital space missions. There is a hypothesis that disorders of musculoskeletal system as part of HMS arise in consequence of changes in spinal motor neurons. The study was aimed at bioinformatic analysis of transcriptome changes in lumbar spinal cords of mice after a 30-day spaceflight aboard biosatellite Bion-M1 (space group, S) and subsequent 7-day readaptation to the Earth's gravity (recovery group, R) when compared with control mice (C group) housed in simulated biosatellite conditions on the Earth. Gene ontology and human phenotype ontology databases were used to detect biological processes, molecular functions, cellular components, and human phenotypes associated with HMS. Our results suggest resemblance of molecular changes developing in space orbit and during the postflight recovery to terrestrial neuromuscular disorders. Remarkably, more prominent transcriptome changes were revealed in R vs. S and R vs. C comparisons that are possibly related to the 7-day recovery period in the Earth's gravity condition. These data may assist with establishment of HMS pathogenesis and proposing effective preventive and therapeutic options.

Keywords: Bion-M1 biosatellite, 30-day spaceflight, 7-day postflight readaptation, mice lumbar spinal cord, hypogravity motor syndrome, transcriptome, bioinformatic study

#### *Edited by:*

*Derek John Hausenloy, University College London, United Kingdom*

#### *Reviewed by:*

*Leniz Nurullin, Kazan Institute of Biochemistry and Biophysics (RAS), Russia Petr Masliukov, Yaroslavl State Medical University, Russia*

#### *\*Correspondence:*

*Maksim Sergeevich Kuznetsov qmaxksmu@yandex.ru Rustem Robertovich Islamov islamru@yahoo.com*

#### *Specialty section:*

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

*Received: 10 January 2019 Accepted: 11 June 2019 Published: 11 July 2019*

#### *Citation:*

*Kuznetsov MS, Lisukov AN, Rizvanov AA, Tyapkina OV, Gusev OA, Rezvyakov PN, Kozlovskaya IB, Tomilovskaya ES, Nikolskiy EE and Islamov RR (2019) Bioinformatic Study of Transcriptome Changes in the Mice Lumbar Spinal Cord After the 30-Day Spaceflight and Subsequent 7-Day Readaptation on Earth: New Insights Into Molecular Mechanisms of the Hypogravity Motor Syndrome. Front. Pharmacol. 10:747. doi: 10.3389/fphar.2019.00747*

## INTRODUCTION

Absence of gravity causes changes in virtually all organs and systems of a living organism at the molecular, cellular, and tissue levels (Edgerton and Roy, 2000). Fundamental knowledge of these changes extends our understanding of the human body functioning in the extreme microgravity environment of outer space and offers a clearer view of preventive options needed for astronauts on long missions.

Investigations of weightlessness effects on living organisms began with the launch of dog Laika to space in 1957 and the flight of Yuri Gagarin in 1961. Since then, studies in the domain of space medicine have revealed a variety of spaceflight effects on human cardiovascular, neurovestibular, and musculoskeletal systems (Grigoriev and Potapov, 2013). The hypogravity motor syndrome (HMS) is considered to be a severe microgravity effect on astronauts (Grigoriev and Kozlovskaya, 1991), which is why success of future remote space missions will be highly dependent on how soon we get to the roots of HMS pathogenesis and be ready to offer methods of prevention on the molecular level. An important input to this effort comes from experiments with animals that have been exposed to spaceflight weightlessness (Moyer et al., 2016).

Mechanisms of HMS development are still poorly explored. Interestingly, pathognomonic signs are observed in skeletal muscles following spinal or peripheral nerve disorders (Hides et al., 2017). Earlier, it has been shown that the important trigger of HMS is violation of sensory impulses from the skin (Kozlovskaya et al., 1981). Possibly, absence of sensory stimuli from mechanoreceptors of the sole skin results in the functional activity of motor neurons innervating leg muscles. However, it is well known that in a motor unit, the neuron and skeletal muscle fibers interrelate through electrical impulses and informative molecular signaling (Chevrel et al., 2006; Sakuma and Yamaguchi, 2011; Baguma-Nibasheka et al., 2016). Thus, skeletal muscle plasticity depends on the condition of the different inputs to motor units. In light of this assumption and based on our earlier studies of the neuromuscular synapse, peripheral nerve, and spinal cord in a rat model of hypogravity [hind limb unloading model (HUM)] (Chelyshev et al., 2014; Islamov et al., 2015), we made a supposition that HMS pathogenesis is invoked by spinal motor neurons.

Under the auspices of the Federal Space Program, the Russian Federal Space Agency and the Institute of Biomedical Problems of the Russian Academy of Sciences undertook a vast program of space bioresearches aboard biosatellite Bion-M1 launched on April 19 and landed on May 19, 2013 (Andreev-Andrievskiy et al., 2014). To test the hypothesis that changes in spinal motor neurons provoke HMS, a full genome study of the mice lumbar spinal cord was performed after the 30-day Bion-M1 mission. The list of genes from the lumbar spine of mice with significant increases and decreases has been presented in our previous study (Islamov et al., 2016). Some changes in gene expression supported our hypothesis that molecular changes in spinal motor neurons are among the key factors in HMS pathogenesis. Meanwhile, the further comparative bioinformatic analysis based on the contemporary genetic and medical databases is needed to discover the possible molecular mechanism of HMS development. At the same time, it is known that back on Earth, astronauts undergo lengthy recovery or rehabilitation (Loehr et al., 2015). To date, the problem of rapid rehabilitation of astronaut's

activity is under the intensive study. Obviously, the changes in the transcriptome profile arising during the readaptation to the Earth's gravity may explain some clinical signs in astronauts after landing. The Bion-M1 research program included studies of mice immediately following the exposure to the 30-day spaceflight and after the 7-day readaptation on Earth. The chosen periods were designed for international collaborative investigation to obtain data on long-term exposure to weightlessness on varied physiological systems and provide data on behavioral readaptation of mice to Earth's gravity after the flight (Andreev-Andrievskiy et al., 2014).

The investigation was designed as a full-genome study and bioinformatic analysis of transcriptome changes in the lumbar spinal cord of mice that spent 30 days onboard Bion-M1 in comparison with their counterparts that were given a week of rehabilitation to the Earth's gravity. We examined transcriptome changes further using gene ontology (GO) and human phenotype ontology (HPO) databases for disclosure of the HMS pathogenesis and the relationship of HMS with the terrestrial neuromuscular diseases.

### MATERIALS AND METHODS

#### Animals and Treatment

Adult male C57BL/6J mice (4‒5 months of age, 25.1 ± 3.2 g) obtained from the "Puschino" animals breeding laboratory and nursery (Puschino, Moscow region, Russia) were divided in the flight and control groups. In its turn, the flight group was subdivided into the group of mice sacrificed 14 h following the biosatellite landing (space group, S) and the other one where the mice were given 7 days for readaptation (recovery group, R). Mice from the control (C) group were housed three per cage under simulated biosatellite conditions on Earth. Male mice were selected since they are physically stronger than female and have more stable hormonal status. The overview of the mice selection, housing, and training before they were launched, during the 30-day spaceflight and in postflight 7-day period of readaptation, was published by Andreev-Andrievskiy et al. (2014). The animal protocols including euthanasia were reviewed and approved by the Commission on Bioethics at the Institute of Mitoengineering of the Lomonosov Moscow State University (protocol no. 35 of November 1, 2012) and the Commission on Biomedical Ethics at the Institute of Biomedical Problems of the Russian Academy of Sciences (protocol no. 319 of April 4, 2013).

#### Full-Genome Study of the Mice Lumbar Spinal Cord

In our investigation, biospecimens were obtained from six experimental animals. Lumbar spinal cords were harvested from groups S (n = 2), R (n = 2), and C (n = 2). The spinal cords of group C were extracted simultaneously with S group. Spinal cords have been removed from the vertebral column by hydraulic ejection method. Ventral and dorsal roots were sectioned, and L1 to L5 region of the lumbar enlargement were isolated and frozen in liquid nitrogen immediately. The prepared tissue samples were processed for RNA isolation from both the gray and white matter.

The microarray analysis procedures for group R were performed as described previously for groups S and C (Islamov et al., 2016). In short, total RNA was extracted from the entire sample of lumbar enlargements using the RNeasy Mini Kit (Qiagen, Valencia, CA). RNA quality was confirmed with 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA) where 18S and 28S ribosomal RNA served as controls. For microarray analysis, 500 ng of RNA in the Mouse GE 4x44K v2 Microarray Kit (Agilent Technologies, Santa Clara, CA) was used. Microarray analysis procedures were conducted as per manufacturer's instructions. Quality control of the feature was performed using the settings recommended by Agilent Technologies. Raw microarray data were uploaded in ArrayExpress (accession: E-MTAB-7426).

#### Bioinformatic Study

We used R version 3.4.4 (R Foundation for Statistical Computing, Vienna, Austria) for all steps of data processing and analysis (code is available on github.com/bqmax/bion\_m1) (R Core Team, 2018). Background correction and quantile-based normalization were performed before removal of control probes and within-array replicates replacement with averages. We used the hierarchical clustering with complete agglomeration and Euclidean distance as a similarity measure and principal component analysis to assess general transcriptome profiles of the experimental spinal cords (Aziz et al., 2017). Linear models implemented in limma package were applied to assess differential genes expression. The Benjamini–Hochberg procedure was used to control the false discovery rate (Ritchie et al., 2015).

To determine differentially expressed genes, the followed cutoff rules were applied: pFDR-value <0.05 and absolute log2(fold change) > 2 (or absolute fold change >4).

For functional analysis of the obtained microarray data, we used two databases, namely GO and HPO as follows:


#### Real-Time PCR Analysis

Transcriptome data were validated by real-time PCR (RT-PCR) using SYBR Green method. Differentially regulated genes (Echs1, Ndufa9, Rplp2, Fbxo21, and Parl) were selected by both minimization of observed p-values and effect sizes (absolute fold changes). Ppib and Rpl7 genes were chosen as housekeeping references due to stable expression in different mouse tissues (Thomas et al., 2014; Bruckert et al., 2016). Primers with appropriate thermodynamic properties for these genes were designed using Vector NTI software (Invitrogen); specificity of these primers was confirmed using BLASTn. Random Hexamers (Trermo Fisher) and RevertAid transcriptase (Thermo Scientific) for reverse transcription and qPCRmix-HS SYBR (Eurogene) for quantitative PCR were employed. For RT-PCR data analysis, REST software (QIAGEN, GmbH) was used. The list of the target genes and corresponding sequences of designed primers are presented in **Table S1**.

### RESULTS

#### Hierarchical Clustering and Principal Component Analysis

Experimental observations form two distinct clusters, one of which consists of two samples from group R and the other is a compound of observations from groups S and C. Dendrogram on **Figure 1A** illustrates the arrangement of clades produced by hierarchical clustering. Similar results obtained with principal component analysis are visualized on the biplot (**Figure 1B**).

#### Microarray Expression Profiling

In our previous research, we employed the Mouse GE 4x44K v2 Microarray Kit to perform transcriptome analysis of the lumbar spinal cord of mice flown on Bion-M1. In comparison with group C, out of genes declared on the microarray platform increased and decreased expressions were documented in 134 genes and 41 genes, respectively. The list of up- and downregulated genes is published by Islamov R. and coauthors (Islamov et al., 2016).

In the present study, the same Mouse GE 4x44K v2 Microarray Kit was used for transcriptome profiling of the lumbar spinal

FIGURE 1 | Hierarchical clustering and principal component analysis. (A) Dendrogram representing hierarchical clustering of gene expression data. Observations from the R (recovery) group form a single cluster. Observations from the S (space group) and C (control) groups are in a compound cluster. (B) Principal component analysis of gene expression data. Observations from the S (space) and C (control) groups are mixed in PC1 (Dim 1) and PC2 (Dim 2) spaces.

cord from mice after 7 days of readaptation after the Bion-M1 mission (group R).

Comparative analysis between groups R and C using the declared cutoff rules revealed 1,994 downregulated and 1,726 upregulated genes. In comparison between R and S groups, there were 2,025 downregulated and 1,758 upregulated genes. For clarity, differential performance of gene expression in mice lumbar spinal cord in three comparisons is presented in the form of volcano plots (**Figure 2**). It is important to note that multiple groups comparison results in decrease of statistical power which did not reveal differentially expressed genes between S and C groups, as was shown earlier (Islamov et al., 2016). Full lists of genes considered as differentially expressed with appropriate statistics are presented in **Table S2**. **Figure 3A** represents top 50 differentially expressed genes after hierarchical biclustering.

#### Functional Profiling of Gene List Using g:Profiler

Based on the full list of differentially expressed genes, GO-based and HPO-based overrepresentation analysis was performed. **Table S3** contains all overrepresented GO:biological processes (n = 1,775), GO:molecular functions (n = 321), GO:cellular components (n = 236), and HPO terms (n = 43). **Figure 3B** represents top 15 overrepresented HPO and GO terms based on pFDR-value. Notably, top-rated biological processes terms are semantically associated with the nervous system and are consistent with revealed molecular functions, cellular components, and human phenotypes.

### Gene Ontology-Based Enrichment Analysis

This analysis allows discrimination of biological processes, molecular functions, and cellular components that are possibly involved with pathological processes in the spinal cord in the condition of weightlessness without loss of information about expression levels of genes in each experimental observation. The results of functional enrichment based on random forest are shown in **Table S4.** Ranking of enriched GO terms depends on random forest results (aggregation of Gini scores for each genes linked with term) and count of mapped genes. For downstream analysis, we selected the top 1,000 genes based on Gini score and only terms from biological processes namespace with p-value <0.01 and total genes count ≥10. Using binary biclustering, we identified "hot spots" of enrichment by bivariate maximization (implemented algorithm and R code are described in **Appendix S1**). The results of analysis are presented in **Figure 4**. This approach allowed to reveal 28 genes linked with six GO terms: GO:0006886 (intracellular protein transport); GO:0006913 (nucleocytoplasmic transport); GO:0007165 (signal transduction); GO:0015031 (protein transport), GO:0006184 (obsolete GTP catabolic process), and GO:0007264 (small GTPase mediated signal transduction).

#### Real-Time PCR Validation of Microarray Data

The RT-PCR results have confirmed the directions of the obtained transcriptome changes in five of five cases in R vs. S comparison and in four of five cases in S vs. C comparison (gene Parl was

terms (HP). Letter size of words is proportional to 1/-log10(pFDR-value).

upregulated in RT-PCR assay and downregulated in microarray approach). **Table S1** contains fold changes for five target genes estimated in both microarray and RT-PCR analysis. It should be noted that our results are consistent with the study on the reliability of the use of RT-PCR for verifying the microarray data (Morey et al., 2006).

### DISCUSSION

Spaceflight and exposure to microgravity cause specific changes in human skeleton (Smith et al., 2012) and skeletal muscles (Belavý et al., 2011; Narici and de Boer, 2011). It is known that the disorders seen in astronauts after a space mission are similar to terrestrial neuromuscular diseases in patients (Hides et al., 2017). Studies of the neuromuscular system plasticity in astronauts are beneficial to patients with similar disorders, and, vice versa, researches with patients may provide new options for the reconditioning of astronauts (Stokes et al., 2017). HMS is among the most untoward consequences of long-term orbital spaceflight. HMS is characterized by specific changes in skeletal muscles, particularly in the so-called postural muscles responsible for maintaining posture in the gravity field of the Earth (Grigoriev

and Kozlovskaya, 1991). Along with weakness and atrophy of skeletal muscles, the ratio of muscle fiber types, expression of muscle-specific proteins, contractile characteristics, and electrophysiological properties of skeletal muscle fibers are affected gravely (Shenkman, 2016). The physiotherapy methods applied to patients on Earth and experience of spaceflights has shown that the most effective way to maintain an astronaut's capacity for work and to be prepared to return to Earth is regular performance of a complex set of physical exercises (Grigoriev and Kozlovskaya, 1991) that astronauts have to accomplish several hours a day in order to prevent HMS (Loehr et al., 2015).

The proposed role of spinal motoneurons in HMS pathogenesis requires a deep insight into the molecular and cellular disorders in the spinal cord. However, up to date, little is known about microgravity effects on the central nervous system. Few reports on spinal cord of rodents after different periods of orbital flight are available. Thus, it was shown that succinate dehydrogenase activity was selectively decreased in the medium-sized motoneurons of the mice lumbar spinal cord after the 9-day spaceflight (Ishihara et al., 2006). In rat spinal motoneurons, the content of cytoplasmic proteins was significantly lowered 22 days after spaceflight (Gorbunova and Portugalov, 1976). Signs of myelin destruction and decreased number of myelin-forming cells in white matter (Povysheva et al., 2016) and changes in immonoexpression of choline acetyltransferase (ChAT) and neurofilament proteins in gray matter (Porseva et al., 2016) were shown in mice spinal cord after the 30-day exposure to microgravity. During the longest mission in space, which lasted 91 days, unfortunately, only three mice survived, and only data on spinal cord necropsy was presented (Cancedda et al., 2012). In general, these results demonstrate the evidence of negative effects of weightlessness on spinal cord, which may lead to HMS, although the particular mechanisms of HMS pathogenesis still need to be fully elucidated. In our investigation, we studied lumbar spinal cord housing motoneurons that innervate hind limb skeletal muscles having an antigravity function.

The contemporary "omics" technologies are widely employed in unveiling the mechanisms of human diseases. Microarray data analysis makes it possible to examine the transcriptome changes at the cell, tissue, or organic levels (Ewis et al., 2005). Numerous bioinformatics studies on gene expression profiling in the spinal cord of mice with neurotrauma or neurodegenerative diseases have been performed in recent years (Munro et al., 2012; Elliott et al., 2013; Zhu et al., 2017; Barham et al., 2018). There have been only few studies of spaceflight effects on the transcriptome profile in the bone tissue, immune system, and skeletal muscles of mammals (Nichols et al., 2006). A substantial progress in microgravity genomics has been made owing mostly to the investigations of rodent skeletal muscles. For instance, Gambara and coauthors obtained a global gene expression profile of the paraspinal skeletal muscle (longissimus dorsi) (Gambara et al., 2017a) as well as the slow-twitch (soleus) and fast-twitch (extensor digitorum longus) hind limb muscles (Gambara et al., 2017b) following the exposure onboard spacecraft Bion-M1. It was shown that microgravity strongly affected the transcriptome profile in the postural soleus muscle and slightly changed the gene expression pattern in the extensor digitorum longus and longissimus dorsi. These data pointed to the microgravity-sensitive muscle genes involved in pathogenesis of HMS.

In this study, for the first time, we investigated lumbar spinal transcriptomes of mice after their 30-day spaceflight on biosatellite Bion-M1 and subsequent 7-day readaptation on Earth. Transcriptome analysis of the obtained data was completed in three comparisons, i.e., spaceflight (S) vs. ground control (C), 7-day postflight recovery (R) vs. C, and R vs. S. The results that were received with the involvement of the GO and HPO databases suggest that molecular changes developed in the mice lumbar spinal cord during the flight are similar to those in consequence of terrestrial neuromuscular disorders (Hides et al., 2017; Stokes et al., 2017). Thus, discovered biological processes terms (nervous system development, synaptic signaling, and anterograde trans-synaptic signaling), molecular functions (signaling receptor binding, gated channel activity, and ion gated channel activity), and cellular components (plasma membrane part, synapse, and neuron projection) are highly linked with human phenotypes (electromyography: myopathic abnormalities, abnormality of muscle fibers, muscle stiffness), which are in line with signs of HMS.

Earlier, using the mice HUM, we hypothesized that HMS pathogenesis partly may be due to spinal motoneurons disorders (Islamov et al., 2011; Chelyshev et al., 2014). Taking into consideration our HUM findings demonstrating decreases in the gray and white matter areas, decrease in ChAT immunoexpression, changes in myelin gene expression, and phenotypic modifications of glial cells in lumbar spinal cords, the results of this investigation suggest that motoneurons contribute to the HMS development. These findings are in a very good agreement with the data obtained in the present bioinformatic analysis. The discovered GO-based biological processes, molecular functions, and cellular components in spinal cord support conclusions in our previous report (Povysheva et al., 2016) that demyelination in the central nervous system is a factor in the HMS development. Moreover, it is notable that discovered GO:0006886 (intracellular protein transport), GO:0006913 (nucleocytoplasmic transport), and GO:0015031 (protein transport) may be associated with the axonal transport. It is known that axonal proteins are synthesized in the motor neuron perikaryon and then are distributed over the axon by the mechanism of anterograde axonal transport. The distance to which the molecules are transported varies significantly. In fact, the length of human neural outgrowths may by more than 1 m. Thus, to our knowledge, this is the first report that molecular disorders in intracellular transport system may affect the axonal transport that may be one of the important mechanism of HMS pathogenesis. Furthermore, overrepresented GO terms based on differentially expressed genes have revealed biological processes and molecular functions that are involved in synaptic plasticity (chemical synaptic transmission and synaptic signaling), cell membrane permeability (ion channels, potassium channel activity, and voltage-gated ion channel activity), and cytoskeleton (cytoskeletal protein binding and actin binding) and may be the key factors in the HMS pathogenesis as well (**Table S3**). These findings suggest resemblance of molecular changes developing in space and during the postflight recovery to the HPO terms for terrestrial neuromuscular disorders.

Under the auspices of the Bion-M1 program, we also used the immunohistochemical assay to investigate reactions of lumbar motor neurons from the S, R, and C groups (Tyapkina et al., 2016). It should be noted that the decreased immunoexpression of synaptic proteins (synaptophysin and postsynaptic density protein 95) in motor neurons of mice after the spaceflight (groups S and R) is consistent with more than 15-fold upregulation of the corresponding genes (Syp and Dlg4) in group R. Moreover, Porseva et al. (2016) showed that the number of neurons containing ChAT and neurofilament proteins decreased in the thoracic section of the spinal cord in mice after the 30-day spaceflight (group S in our research). According to our results, in a week after landing (group R), the level of ChAT gene revealed a 17-fold increase; levels of Nefl, Nefm, and Nefh genes increased 176, 284, and 176 times, respectively. Changes in the level of proteins in motoneurons resulted, possibly, in increased expression of the gravity-sensitive genes during the readaptation period. Meanwhile upregulation or downregulation of certain genes in spinal cord tissue after 7-day readaptation period may be due to not only compensatory reaction to 30-day period of disuse of musculoskeletal system in space but also activity of the microgravity-sensitive genes and their hierarchical status in specific biological processes. Our results are also in line with the findings of Gambara et al. (2017a) demonstrating expression of microgravity-sensitive non-muscle-specific genes that match with genes and trends in expression identified in the corresponding groups in our study.

Thus, the bioinformatic analysis of transcriptome changes presented in the study provides a molecular evidence of HMS resemblance to the pathogenesis of the terrestrial neurological disorders. However, because of a significant loss of mice during the mission, we received very few animals (n = 2 in each group) for our study (Andreev-Andrievskiy et al., 2014) and, therefore, were unable to provide an acceptable power to identify the entire pool of neuron-specific target genes. At the same time, a vast majority of transcriptomic studies (microarray and RNA-seq) are conducted with a small sample size that makes them underpowered and exploratory. In our study, linear models realized in limma package to determine differentially expressed genes were used. The implemented analysis can be considered as de facto standard approach, especially in case of small sample sizes and multiple group comparison (Gentleman et al., 2005). This being so, we reason that our results should be considered as exploratory due to the insufficient strength of evidence.

For today, the longest continuous presence of human in space amounts to 438 days. This period is comparable with the estimated time of a Martian mission that includes the transits to Mars and back to Earth and a short stay on the planet. As the tasks facing astronautics become more challenging, this period is likely to extend. Still, even diligent pursuance of expressly developed preventive complexes may fail to preclude HMS development completely. According to the Bion-M1 program, the mice were launched for a 30-day stay in orbit. However, major disorders in the musculoskeletal system develop in the first 2 weeks of spaceflight (Akima et al., 2000; LeBlanc et al., 1995). Therefore, the 30-day period of exposure to microgravity may be appropriate for discovery of tissue-specific and gravity-sensitive genes and intracellular pathways involved in the HMS development. Severe functional impairment of postural and locomotor musculature was obtained in mice after the 30-day spaceflight on the board Bion-M1 biosatellite (Andreev-Andrievskiy et al., 2014). These independent findings are in line with GO biological processes [locomotion (GO:0040011), locomotory behavior (GO:0007626), regulation of locomotion (GO:0040012), and musculoskeletal movement (GO:0050881)] terms and HPO terms [myopathy (HP:0003198) and proximal muscle weakness (HP:0003701)] revealed in our work. We believe that our bioinformatics study will help future experiments aimed at disclosure of the HMS pathogenesis and suggest advanced methods of preventing and treatment of HMS and the similar terrestrial neuromuscular disorders.

### CONCLUSION

Comprehensive bioinformatic analysis of genes expression profiling in the mice lumbar spinal cord after the 30-day spaceflight with subsequent 7-day recovery revealed molecular cascades that may be involved in pathogenesis of HMS. These data may assist in unveiling HMS pathogenesis and development of novel effective preventive and therapeutic options. Moreover, it was shown that postflight readaptation is complicated with further molecular changes in the condition of normal gravity. However, it is necessary to take into consideration that the identified genes and pathways probably associated with HMS development may be triggered not only by weightlessness but also accelerations during spacecraft insertion and descent, exposure to space radiation, or attenuation of the Earth's magnetic field.

### ETHICS STATEMENT

The animal protocols including euthanasia were reviewed and approved by the Commission on Bioethics at the Institute of Mitoengineering of the Lomonosov Moscow State University (Protocol No. 35 of November 1, 2012) and the Commission on Biomedical Ethics at the Institute of Biomedical Problems of the Russian Academy of Sciences (Protocol No. 319 of April 4, 2013).

### AUTHOR CONTRIBUTIONS

EN and RI contributed to the study conception and design. AL, PR, OT, and OG contributed to the acquisition of data. MK, AL, AR, and RI contributed to the analysis and interpretation of data. MK, AL, PR, and RI contributed to the drafting of the manuscript. AR, IK, ET, and RI provided critical revision.

## FUNDING

The study was funded by grant RFBR 17-04-00385, grant of Presidium of the Russian Academy of Sciences "Fundamental research for biomedicine technology development." Albert Rizvanov was personally supported as a "leading scientist" by state assignment 20.5175.2017/6.7 of the Ministry of Science and Higher Education of Russian Federation.

### ACKNOWLEDGMENTS

The authors would like to thank Gogoleva N.E. and Gogolev Yu.V. (Kazan Institute of Biochemistry and Biophysics, Federal Research Center "Kazan Scientific Center" of RAS, Kazan, Russia) and Volkov K.D. (Kazan State Medical University) for assistance in some of the experiments. Kazan Federal University was supported by the Russian Government Program of Competitive Growth.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00747/ full#supplementary-material

TABLE S1 | Validation of selected differentially regulated genes in mice lumbar spinal cord in S (space) vs. C (control) and R (recovery) vs. S (space) comparisons.

TABLE S2 | Differentially expressed genes identified by comparison between the R (recovery) and C (control) groups and between the R (recovery) and S (space) groups.

TABLE S3 | Overrepresented GO and HPO terms based on differentially expressed genes with appropriate statistics.

TABLE S4 | GO terms enriched using random forest with Gini feature scoring.

DATA SHEET S1 | R code for binary biclustering and genes/GO terms selection.

### REFERENCES


motoneurons between the cervical and lumbar segments in the rat spinal cord after spaceflight and recovery. *Neurochem. Res*. 31, 411–415. doi: 10.1007/ s11064-005-9027-1


knockout mice with spinal cord injury. *Acta Histochem*. 119, 663–670. doi: 10.1016/j.acthis.2017.07.007

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

The reviewer, LN, declared a shared affiliation, though no other collaboration, with three of the authors, OT, EN, and RI, to the handling Editor at the time of review.

*Copyright © 2019 Kuznetsov, Lisukov, Rizvanov, Tyapkina, Gusev, Rezvyakov, Kozlovskaya, Tomilovskaya, Nikolskiy and Islamov. 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.*

# Novel Peptide Conjugates of Modified Oligonucleotides for Inhibition of Bacterial RNase P

*Darya Novopashina1,2\*, Mariya Vorobyeva1, Anton Nazarov3, Anna Davydova1, Nikolay Danilin2, Lyudmila Koroleva1,2, Andrey Matveev1, Alevtina Bardasheva1, Nina Tikunova1,2, Maxim Kupryushkin1, Dmitrii Pyshnyi1,2, Sidney Altman4,5 and Alya Venyaminova1*

*1 Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia, 2 Department of Natural Sciences, Novosibirsk State University, Novosibirsk, Russia, 3 Shemyakin-Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Moscow, Russia, 4 Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, CT, United States, 5 Division of Life Sciences, Arizona State University, Tempe, AZ, United States*

#### *Edited by:*

*Olga N. Ilinskaya, Kazan Federal University, Russia*

#### *Reviewed by:*

*Arun Samidurai, Virginia Commonwealth University, United States Santiago Grijalvo, Center for Biomedical Research in the Network in Bioengineering, Biomaterials and Nanomedicine (CIBER-BBN), Spain*

#### *\*Correspondence:*

*Darya Novopashina danov@niboch.nsc.ru*

#### *Specialty section:*

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

*Received: 15 March 2019 Accepted: 24 June 2019 Published: 19 July 2019*

#### *Citation:*

*Novopashina D, Vorobyeva M, Nazarov A, Davydova A, Danilin N, Koroleva L, Matveev A, Bardasheva A, Tikunova N, Kupryushkin M, Pyshnyi D, Altman S and Venyaminova A (2019) Novel Peptide Conjugates of Modified Oligonucleotides for Inhibition of Bacterial RNase P. Front. Pharmacol. 10:813. doi: 10.3389/fphar.2019.00813*

Novel alternatives to traditional antibiotics are now of great demand for the successful treatment of microbial infections. Here, we present the engineering and properties of new oligonucleotide inhibitors of RNase P, an essential bacterial enzyme. The series of 2'-*O*-methyl RNA (2'-OMe-RNA) and phosphoryl guanidine oligonucleotides were targeted to the substrate-binding region of M1 RNA subunit of the RNase P. Uniformly modified 2'-OMe RNA and selectively modified phosphoryl guanidine oligonucleotides possessed good stability in biological media and effectively inhibited RNase P. Their conjugates with transporting peptides were shown to penetrate bacterial cells (*Escherichia coli* and *Acinetobacter baumannii*) and inhibit bacterial growth.

Keywords: bacterial RNase P, inhibition of RNase P, modified oligonucleotides, oligo(2'-*O*-methylribonucleotides), phosphoryl guanidine oligonucleotides, peptide conjugates of oligonucleotides, penetration into bacterial cells, antibacterial activity

### INTRODUCTION

The design of novel compounds with antibacterial activity is one of the most acute issues of modern chemical biology, biotechnology, and medicine. Despite a broad spectrum of antimicrobial agents, the problems of the drug resistance and side effects remain unsolved until now (Guidry et al., 2014; Llor and Bjerrum, 2014). A promising strategy to overcome these problems could be a displacement of low-molecular-weight antimicrobial compounds targeting essential bacterial biomolecules and pathways for nucleic acid-based therapeutics targeting bacterial nucleic acids. Within this context, the development of oligonucleotides that specifically interact with bacterial RNAs, block their functions, and thereby inhibit bacterial growth is of particular interest (Bai et al., 2010). Bacterial RNase P, a tRNA-processing enzyme, is an attractive target for the design of antibacterial oligonucleotides (Guerrier-Takada et al., 1983; Altman, 2011). On the one side, RNase P is a key player of the well-established EGS (external guide sequence) technology (see, e.g., the reviews Davies-Sala et al., 2015; Derksen et al., 2015). Specially designed EGS oligonucleotides mimicking the 3'-fragment of the pre-tRNA substrate can address the enzyme to cleave the target sequence within specific bacterial mRNA. On the other side, as one of the essential bacterial enzymes, RNase P itself represents an attractive target for antibacterial agents. The enzyme contains an RNA subunit (catalytic M1 RNA), which gives a possibility to inhibit RNase P by complementary oligonucleotides and thus suppress the bacterial growth (Gruegelsiepe et al., 2006). One of the main advantages of this approach is the specific targeting of bacterial cells provided by huge differences between eukaryotic and bacterial enzymes, primarily their M1 RNA sequences (Klemm et al., 2016). Oligonucleotide inhibitors targeted to the bacterial RNase P should not cause any off-target effects on eukaryotic cells. The possibility of bacterial growth suppression by RNA, DNA, locked nucleic acid (LNA), and peptide nucleic acid (PNA) oligonucleotide inhibitors targeting certain M1 RNA regions was demonstrated earlier (Gruegelsiepe et al., 2003; Willkomm et al., 2003; Gruegelsiepe et al., 2006).

With all significant achievements in the design of oligonucleotide inhibitors of RNase P, there is plenty room for improvement of the resistance of these oligonucleotides to nuclease digestion and effectivity of their interaction with the enzyme, as well as penetration into bacterial cells.

Here, we present the design of novel modified oligonucleotides as RNase P inhibitors. A set of 2'-OMe-RNA and selectively modified phosphoryl guanidine oligonucleotides were generated and evaluated for their inhibiting properties. Conjugates of most prominent modified oligonucleotides with cell-penetrating peptides were shown to be capable of penetrating bacterial cells and suppress their growth.

#### MATERIALS AND METHODS

Tris(hydroxymethyl)aminomethane (Sigma-Aldrich, USA), acetonitrile (PanReac, Spain), acrylamide, N,N' methylenebisacrylamide (Acros Organics, Belgium), sodium perchlorate, ammonium persulfate, "Stains-Аll" dye, magnesium chloride, urea, xylene cyanol FF, bromophenol blue, potassium chloride (Fluka, Switzerland), Na2EDTA (AMRESCO, USA), fetal bovine serum (FBS, heat-inactivated, Invitrogen, USA), culture medium DMEM (Life Technologies, USA), γ-[32P]- ATP (120 TBq/mol, «Biosan», Russia), and other reagents and solvents supplied by Sigma-Aldrich, PanReac, and Acros Organics. 3-Maleimidopropanoic acid pentafluorophenyl ester (MPPf) was synthesized by analogy with Kida et al. (2007). Peptides bearing cysteine at N-terminus were obtained from Almabion (Russia): *Pept1*—CKWKLFKKIGAVLKVLTTG, *Pept2*—CRGW EVLKYWWNLLQY, *Pept3*—CHHHHHHHHHHHHHHHH, and *Pept4*—CINVLGILGLLGEALSEL.

C5 protein unit of *Escherichia coli* RNase P was prepared as described in Guerrier-Takada et al. (1983) and kindly provided by Prof. Khodyreva S.N. (ICBFM SB RAS, Novosibirsk, Russia); M1 RNA was synthesized by protocol (Guerrier-Takada et al., 1989) and kindly provided by Prof. Moor N.A. (ICBFM SB RAS, Novosibirsk, Russia). The DH5α strain of *E. coli* and the type ATCC (#19606) strain of *Acinetobacter baumannii* from the collection of thermophilic organisms and type cultures of ICBFM SB RAS were used for investigation of cell penetration and suppression of bacterial growth.

The radioactive 5'-[32P]-labeling of oligonucleotides was performed using four MBq [γ-32P]-ATP and T4 Polynucleotide Kinase (Thermo Scientific, USA) by standard protocol. The

isolation of 5'-[32P]-labeled oligonucleotides was performed with Micro Bio-Spin P30 columns (Bio-Rad, USA).

The gels were dried using Gel Dryer B35 instrument (Bio-Rad, USA) and radioautographed using Bio-Rad Exposure Cassette-K and photosensitive Kodak Storage Phosphor Screen SO230 (Bio-Rad, USA). The screen was scanned using Pharos FX (Bio-Rad Laboratories Inc., CA, USA) Phosphorimager; the images acquired were processed using Quantity One Analysis Software (Bio-Rad Laboratories Inc., CA, USA).

Fluorescence was measured in microplates Costar 96-Well Half-Area Black (Thermo Fisher Scientific, USA) using CLARIOstar instrument (BMG LABTECH, USA).

Water filtration system simplicity (Millipore, USA), spectrophotometer NanoDrop 1000 (Thermo Fisher Scientific, USA), thermomixers and centrifuges (Eppendorf, Germany), Speed-Vac Concentrator SVC-100H (Savant, USA), the gelelectrophoresis system (Helicon, Russia), and gel-documentation system Molecular Imager FX (Bio-Rad, USA) were also used.

#### Synthesis of Modified Oligonucleotides and Model RNA Target

Synthesis of modified oligonucleotides and model RNA target was carried out by the solid-phase phosphoramidite method on the ASM-800 synthesizer (Biosset, Russia) using protocols optimized for this instrument. 2'-*O*-*tert*-Butyldimethylsilyl (2'-*O*-TBDMS) protected RNA phosphoramidites, 2'-*O*-methyl RNA, and DNA phosphoramidites; solid supports with first nucleosides, modified solid supports for the synthesis of 3'-fluorescein; 3'-BHQ (Black Hole Quencher), and 3'-amino linker (aminohexanol) containing oligonucleotides were purchased from ChemGene (USA). Fluorescein phosphoramidite (Glen Research, USA) was used for the introduction of fluorescein residue on 5'-end of oligomers. DMS(O)MT-protected amino linker C6 (Lumiprobe, Russia) was used to prepare oligonucleotides bearing 5'-amino linker. Phosphoryl guanidine oligonucleotides were prepared in LLC «NooGene» (Russia) using protocols published earlier (Kupryushkin et al., 2014; Stetsenko et al., 2014).

All oligonucleotides and their derivatives were deblocked by standard protocols for the corresponding type of modification. Isolation of oligoribonucleotides, their modified analogs, and derivatives was performed using preparative electrophoresis in denaturating 15% PAAG. Oligodeoxyribonucleotides and phosphoryl guanidine oligonucleotides were isolated by highperformance liquid chromatography (HPLC) on Agilent 1 200 HPLC system (Agilent Technologies, USA) using Zorbax SB-C18 (4.6 × 150 mm) column in acetonitrile concentration gradient 0–50% in 20 mM triethylammonium acetate (pH 7.0) during 30 min and rate 2 ml/min.

#### Investigation of the Cleavage of Modified Oligonucleotides by Serum Nucleases

The treatment of 5'-[32P]-labeled oligonucleotides (**r-inh**, **m-inh**, **d-inh**) by 10% FBS in DMEM was carried out at 37°C. The 5-µl aliquots were taken after 15, 30, 60, 120, 240, and 360 min and 1 day, mixed with Stop Mix solution and analyzed by denaturating 15% polyacrylamide gel electrophoresis (PAGE).

#### Mass Spectrometry of Oligonucleotides and Their Peptide Conjugates

The mass spectra of the oligonucleotide conjugates were recorded on a Matrix-Assisted Laser Desorption Ionisation-Time-of-Flight (MALDI-TOF) Autoflex Speed mass-spectrometer (Bruker Daltonics, Germany). The mass spectra of phosphoryl guanidine oligonucleotides were obtained using Electrospray Ionisation Mass Spectrometry (ESI-MS) on the Agilent G6410A LC-MS/MS instrument (Agilent Technologies, USA).

#### Hydrolysis of RNA Target by RNase P

The hydrolysis of fluorescent RNA target (5'-**flu**pGUUUUCUUCGGUGGGGUUUCUUCCCCACCACCA-**BHQ**-3') at a concentration from 50 to 300 nM was carried out in 50 µl of a solution containing 10 mM Tris-HCl (pH 7.5), 10 mM MgCl2, 100 mM NH4Cl, 5 nM M1 RNA, and 50 nM of C5 protein at 37°С. RNA target was annealed for 2 min at 90°С and cooled up to 37°С. The RNA target solution was placed in the wells of Costar 96 Half-Area Microplate. The reaction was initiated by addition of the mixture of enzyme and inhibitor oligonucleotide. Fluorescence intensity was registered each minute using CLARIOstar instrument. The excitation wavelength was 483 nm, and the emission wavelength was 530 nm. The data were processed using Mars Data Analysis Software (BMG Labtech, USA).

#### Calculation of Kinetic Parameters of Hydrolysis of Fluorescent RNA Target by RNase P in the Presence of Inhibiting Oligonucleotides

The dependence of fluorescence intensity data from time was analyzed using the equation (1) in GraphPad Prism 5.0.4.533 software:

$$F = F\_{p^\ell} \left( 1 - e^{k\_{\text{obs}} \cdot t} \right) \tag{1}$$

where *F* is the intensity of fluorescence on 530 nm at the moment *t*, *Fp* is the fluorescence intensity upon stage (stable) phase, *kobs* is the pseudo-first order rate constant, and *t* is the reaction time. The values of *kobs* were used to calculate parameters *K*m and *V*max using equation (2) in the same software package.

$$k\_{\rm obs} = \frac{V\_{\rm max}}{K\_{\rm su} + S} \tag{2}$$

where *V*max is the maximum reaction rate, *K*m is the Michaelis constant, and *S* is the substrate concentration. The *k*cat value was calculated from the relationship *V*max *= k*cat *∙ E*0, where *E*0 is the enzyme concentration.

The *IC50* values were calculated using observed rate constants *kobs* and equation (3) for the concurrent inhibition in the same software package

$$k\_{\rm obs} = \frac{\left(\frac{K\dot{\imath}}{K\_m}\right)V\_{\rm max}}{IC\_{50} + I} \tag{3}$$

*kobs* is the pseudo-first order rate constant, *Vmax* is the maximal rate of reaction, *IC*50 is the half-maximal inhibitory concentration, *Ki* is the constant of the inhibition, *Km* is the Michaelis constant, and *I* is the concentration of inhibiting oligonucleotide.

#### Synthesis of Peptide Conjugates of Inhibiting Oligonucleotides

The solution of MPPf (1 mg, 3 μmol) in 20 μl of DMSO was added to the solution of 5'- or 3'-amino-modified oligonucleotides (**m-inh** or **pgd-inh3**) (120 nmol) in 5 μl of 0.02 M (4-(2-hydroxyethyl)-1 piperazineathanesulfonic acid)(HEPES) (pH 7.2) by portions of 10, 5, and 5 μl each 30 min. The reaction mixture was incubated at 37°C upon mixing at 1,200 rpm. After 30 min from the last addition of MPPf, the reaction mixture was precipitated by 2% NaClO4 in acetone, and the pellet was washed by acetone and dried in air. The precipitate was dissolved in 10 μl of 0.01 M HEPES (pH 7.2). The solution of the peptide (*Pept1*, *Pept2*, *Pept3*, or *Pept4*) in 20 μl of dimethylsulfoxide (DMSO) was added to the maleimide-modified oligonucleotide solution, and the reaction was carried out for 1–3 h at 37°C upon mixing at 1,200 rpm. The conjugates were isolated by electrophoresis in denaturating 12% PAAG (acrylamide:bisacrylamide, 30:0,5), eluted by 0.3 М NaClO4, and desalted using Amicon 3K (Millipore, USA). Conjugates were precipitated as Na+ salts.

#### Investigation of Conjugate Penetration to the Bacterial Cells

Cell penetration studies were carried out at the cultures of *E. coli* and *A. baumannii*. The night culture of bacterial cells was diluted at 100 times by growth medium LB (lysogeny broth, Luria-Bertani medium); then, the cells (3–5×106 cells per ml) were incubated for 2 h at 37°C upon swinging up to the optical density OD600 =0.35. The cell culture was prepared at the exponential phase of growth (5–6×106 cells/ml). Then cells were precipitated by centrifugation at 4,000×g for 4 min, and resuspended at LB medium containing peptide conjugate. The final concentrations of conjugates in the medium were 1 or 0.2 µМ. The cells were incubated for 1 h at 37°C upon swinging in the dark. The cells were precipitated by centrifugation at 4,000×g for 4 min; then, 100 μl of 0.9% NaCl solution was added to precipitate, and the procedure was repeated twice. The cell precipitate was resuspended in 100 μl of 4% formaldehyde solution in phosphate buffer and incubated for 30 min at room temperature upon swinging. Then, the cells were washed three times by sterile phosphate buffer and incubated with 4′,6-diamidino-2-phenylindole (DAPI) for additional 15 min.

The slides were prepared by placing 10 μl of cell suspension and 10 μl of antifade diamond solution (Life Technologies, USA) and covering with the 25×25-mm cover glass. Visualization was performed using the confocal laser-scanning microscope LSM 710 Carl Zeiss upon magnification at 630 times and excitation at 405 nm for DAPI and 488 nm for fluorescein isothiocyanate (FITC) (fluorescein). The pictures were analyzed with ZEN 2011 Black Edition software.

Flow cytometry was performed using the NovoCyte Instrument (ACEA Biosciences, USA).

#### Bacterial Growth Inhibition Experiments

Oligonucleotides and their conjugates were tested for inhibition of bacterial cell growth using *E. coli* (strain DH5α). Cell cultures were incubated with the oligonucleotide or oligonucleotidepeptide conjugate for 22 h in 96-well plate at 37°C and rotation at 530 rpm. In control samples, water was used instead of the conjugate solution. The growth of cultures was estimated by optical density at 595 nm using plate reader (Uniplan, Russia).

#### RESULTS AND DISCUSSION

To study the RNase P activity in the presence of oligonucleotide inhibitors, we employed the model synthetic hairpin RNA imitating the natural pre-tRNA substrate of RNase P. This RNA contained 5'-fluorescein and 3'-BHQ quencher residues (**Figure 1**).

Upon the RNA hydrolysis, fluorophore and quencher are moving away from each other, and the fluorescence arises. The values of the kinetic parameters for the hydrolysis of fluorescent RNA target by *E. coli* RNase P (*K*m = 83 ± 49 nM, *k*cat = 24 ± 5 min–1, see **Figure S1**, **Table S1**) are close to those for the hydrolysis of the native pre-tRNATyr (*K*m = 33 nM, *k*cat = 29 min–1) (Jiang et al., 2014). Therefore, we validated the appropriate catalytic activity of the enzyme and the feasibility of fluorescent RNA as a model substrate. The presence of fluorescent and quencher groups had no impact on the affinity of the substrate to the enzyme and the cleavage rate.

One of the most suitable sites for RNase P inhibition is the region of P15 loop taking part in recognition of CCA sequence on the 3'-end of the pre-tRNA substrate (Childs et al., 2003; Gruegelsiepe et al., 2003; Willkomm et al., 2003). Modified oligomers complementary to the 291–304 fragment of *E. coli* M1 RNA and containing 2'-*O*-methylated or LNA monomers specifically inhibited RNase P with practically the same IC50 values (Childs et al., 2003). PNA conjugates with peptides suppressed *E. coli* growth in cell culture experiments (Gruegelsiepe et al., 2006).

Based on these data, we designed 14-nt oligonucleotide inhibitors complementary to nucleotides 291–304 in the P15-loop (**Figure 2**, **Table 1**). The set of inhibitors included native oligoribonucleotide, oligo(2'-*O*-methylribonucleotide), oligodeoxyribonucleotide, and oligodeoxyribonucleotides with phosphoryl guanidine modifications in different positions.

TABLE 1 | The inhibiting oligonucleotides and IC50 values for hydrolysis of RNA target.


*The cleavage conditions: 5 nM M1 RNA, 50 nM C5 protein, 100 nM fluorescent RNA target, 0-125 nM inhibiting oligonucleotide, 10 mM Tris-HCl, pH 7.5, 10 mM MgCl2, 100 mM NH4Cl, 37°C. Nm – 2'-O-methylribonucleotide; Nx – phosphoryl guanidine deoxyribonucleotide; x - phosphate groups modified with 1,3-dimethylimidazolidine-2 imine residues, RNH2 = –(CH2)6NH2. The results represent mean values (± SD) from two independent experiments.*

**Figure 3** shows the typical curves for hydrolysis of fluorescent RNA target upon inhibition of RNase P by **r-inh**. We observed the progressive decrease of enzyme activity upon the increase of the inhibitor concentration. The values of the half-maximal inhibitory concentration (IC50) were obtained using the equation (3) for competitive inhibition, by analogy with (Sabatino and Damha, 2007) (**Table 1**).

The non-modified oligoribonucleotide **r-inh** provided the maximal inhibitory effect. The change of ribose for

FIGURE 3 | The influence of inhibiting oligoribonucleotide r-inh on hydrolysis of fluorescent RNA target. The cleavage conditions: 5 nM M1 RNA, 50 nM C5 protein, 100 nM fluorescent RNA target, 0–125 nM inhibiting oligonucleotide, 10 mM Tris-HCl, pH 7.5, 10 mM MgCl2, 100 mM NH4Cl, 37°C.

2'-*O*-methylated ribose (**m-inh**) or deoxyribose (**d-inh**) led to more modest inhibiting activity: the IC50 values increased approximately 10 times. There was no inhibiting effect in the case of uniformly modified phosphoryl guanidine oligonucleotide **pgdinh1**. Nevertheless, selectively modified phosphoryl guanidine oligonucleotides **pgd-inh2** and **pgd-inh3** containing modified phosphodiester linkagesin certain positions showed pronounced inhibiting activity. At that, **pgd-inh3** with phosphoryl guanidinemodified 3'-terminal fragment was more effective than **pgd-inh2**  with the same modifications introduced close to the 5'-end**.** We chose two most effective modified oligonucleotide inhibitors **m-inh** (IC50 = 90 nM) and **pgd-inh3** (IC50 = 100 nM) for further studies. These oligomers combine RNase P inhibiting activity with good stability in biological media (Novopashina et al., 2018) (see **Figure S2**), so we consider them as a prospective basis for the development of antibacterial agents. Of note, the high level of structural and functional conservation of bacterial RNase P (Altman, 2011) permits to extrapolate the regularities obtained for *E.coli* enzyme to the other bacterial species, particularly *A. baumannii* (Davies-Sala et al., 2018).

Currently, several oligonucleotide constructions have been proposed as antibacterial agents (Woodford and Wareham, 2008; Cansizoglu and Toprak, 2017; Narenji et al., 2017; Równicki et al., 2018; Xue et al., 2018; González-Paredes et al., 2019). In this context, the pivotal challenges are the resistance of oligonucleotides to the nuclease digestion, the capability to penetrate bacterial cells, and the efficiency of interaction with a target bacterial molecule. To solve the problem of cell penetration, we coupled antibacterial oligonucleotide constructs with cell delivery agents (Good, 2003; Good and Stach, 2011; Giedyk et al., 2019). Four peptides were chosen as transporters: 19-mer fragment CM18 of cecropin-A/melittin hybrid peptide capable to disturb membrane (Salomone et al., 2013; Fasoli et al., 2014); 15-mer fragment of HGP peptide of gp41 HIV protein, which enhances endosomolytic activity (Kwon et al., 2008; Kwon et al., 2010); 16-mer oligohistidine peptide (Н16) (Iwasaki et al., 2015); and 17-mer analog of bombolytine V, membrane-destroying antimicrobial peptide (AMP), with all basic residues replaced by the glutamine acid (Ahmad et al., 2015).

For synthesizing oligonucleotide-peptide conjugates, we used the strategy based on thiol-maleimide conjugation chemistry (**Figure 4**). The maleimide group was attached to the 5'- or 3'-amino-modified oligonucleotide using MPPf synthesized by analogy with (Gruegelsiepe et al., 2003). The peptides containing N-terminal cysteine reacted with the maleimide-modified oligonucleotide, giving the covalent conjugates (**Table 2**). The degrees of the oligonucleotides' conversion to the conjugates were about 70–90% by the HPLC data, depending on the type of the peptide. The conjugates were isolated by denaturating PAGE and analyzed by MALDI-TOF and ESI mass spectrometry (**Figure S7** and **S8**).

For cell penetration studies, we employed 5'-peptide conjugates of oligonucleotides bearing the 3'-fluorescein label. The levels of penetration in *E. coli* and *A. baumannii* were estimated using flow cytometry (**Figure 5**, **Table S2**), and the intracellular distribution of conjugates was visualized using confocal microscopy (**Figure S3**, **S4** and **Table S3**).

Fluorescent oligo(2'-O-methylribonucleotide) **m-inh-Flu** penetrated *E. coli* cells at the level of 2%. However, the levels of cell penetration were significantly higher for peptide conjugates of **m-inh-Flu**. Approx. 10–12% of oligonucleotides, depending on the peptide, were found in *E. coli*. Meanwhile, the levels of penetration in *A. baumannii* were approx. 3–6%. Peptide conjugates of phosphoryl guanidine oligonucleotide **pgd-inh3- Flu** demonstrated fewer levels of penetration into the bacterial cells. The best result was observed for *5'-Pept2***-pgd-inh3-***Flu*, which penetrated *A. baumannii* at the level of 6.5%.

The obtained results indicated that conjugation with peptides facilitates the penetration of modified oligonucleotides into bacterial cells of both types. Nearly in all cases, we observed better penetration for *E. coli* cells compared to *A. baumannii.*  This phenomenon might be explained by distinct differences either in the cell wall structure or in efflux systems of these two gram-negative bacteria. Therefore, the approach to intracellular delivery requires optimization when passing from one bacterium to another, even within the same class.

The oligonucleotides and their conjugates were tested on their antibacterial properties using *E. coli* as a target. We used approximately 60 mg/ml concentrations of oligonucleotide inhibitors, which are comparable with the minimum inhibiting concentration (MIC) for standard antibiotics (50–300 mg/ml) (Wannigama et al., 2019). Namely, we studied conjugates of **m-inh** and **pgd-inh3** with *Pept2* peptide at the 5'-end, either with or without 3'-fluorescein residue (**Figure 6**), and the same oligonucleotides with Pept2 at the 3'-end. Despite the relatively low level of the cell penetration for phosphoryl guanidine oligonucleotide, we observed the suppression of *E. coli* growth

#### TABLE 2 | The peptide conjugates of RNase P inhibiting oligonucleotides m-inh and pgd-inh3.


*Nm – Am,Um,Cm,Gm – 2'-O-methylribonucleotides, Nx – phosphoryl guanidine deoxyribonucleotide, Flu – fluorescein residue, Pept1– CKWKLFKKIGAVLKVLTTG, Pept2 – CRGWEVLKYWWNLLQY, Pept3 – CHHHHHHHHHHHHHHHH, Pept4 – CINVLGILGLLGEALSEL. 1\* – Determined by MALDI-TOF mass-spectrometry. 2\* – Determined by ESI mass-spectrometry.*

FIGURE 5 | The level (%) of oligonucleotide conjugates penetration in *Escherichia coli* and *Acinetobacter baumannii* obtained by flow cytometry. The concentration of the conjugates was 1 μM. The cells were incubated with conjugates for 1 h at 37°C. The results are mean values (± SD) from four independent experiments. The fluorescence of the sample of bacterial cells incubated with fluorescein-labeled peptide conjugate of the oligonucleotide is taken as 100%.

for 5'-peptide conjugates, and the presence of fluorescein residue enhanced the effect to some extent (**Figure 6A**). Inhibiting 2'-OMe RNA oligonucleotides suppressed the bacterial growth irrespective of the absence or presence of 5'-peptide. Upon that, although their level of cell penetration was higher than that for phosphoryl guanidines, the inhibiting effect was somewhat lower as compared to peptide conjugates of **pgd-inh3** (**Figure 6B**). For both types of modified oligonucleotides, peptide attached to the 3'-end had no impact on their inhibiting activities: unconjugated oligomers and their 3'-peptide conjugates suppressed the bacterial growth to the same extent. We suppose that the presence of bulk peptide fragment at the 3'-end causes steric hindrance for binding of 3'-CCA fragment to M1RNA in P15 loop site. We also observed all abovementioned regularities for conjugates with the peptides *Pept1*, *Pept3*, and *Pept4* (see **Figure S5** and **S6**).

Enhancement of antibacterial effect upon attachment of peptides to the 5'-end of modified oligonucleotide inhibitors of RNase P proves the feasibility of the proposed approach. We had not observed any correlation between the level of cell penetration of the oligonucleotide and their ability to suppress the bacterial growth. Peptide conjugates of 2'-OMe RNA demonstrated relatively good cell penetration, but rather low antibacterial activity. Relevant conjugates of the phosphoryl guanidine oligonucleotide were less effective in cell penetration but showed better results in suppression of bacterial growth. We hypothesize that the optimization of cell-penetrating properties of phosphoryl guanidine oligonucleotides would improve their antibacterial properties. Further studies are required to prove this suggestion, directed to revealing the roles of oligonucleotide and peptide counterparts in the cell penetration and growth suppression. With this knowledge, we would be able to optimize the structure of oligonucleotide-peptide conjugates inhibiting RNase P to improve their antibacterial activity.

### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the supplementary files.

### AUTHOR CONTRIBUTIONS

DN and MV conceived and designed the experiments. AN, AD, ND, LK, AM, AB, and MK performed the experiments. NT, DP, SA, and AV analyzed the data and co-wrote the paper.

### FUNDING

The research was carried out with financial support by the RFBR grant N 17-04-01892. In the part of the synthesis of modified oligonucleotides, the work was supported by the Russian State-funded budget project of ICBFM SB RAS # АААА-А17- 117020210021-7.

### ACKNOWLEDGMENTS

The authors are grateful to Donna Wesolowski (Yale University, USA), who kindly provided the plasmids for the preparation of the RNase P subunits, Prof. S.N. Khodyreva for preparation and isolation of C5 protein, and Prof. N.A. Moor for preparation and isolation of M1 RNA.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00813/ full#supplementary-material

### REFERENCES


acid active esters. *Chem. Pharm. Bull. (Tokyo)* 55, 685–687. doi: 10.1248/ cpb.55.685


**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 Novopashina, Vorobyeva, Nazarov, Davydova, Danilin, Koroleva, Matveev, Bardasheva, Tikunova, Kupryushkin, Pyshnyi, Altman and Venyaminova. 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.*

# Catalytic Knockdown of miR-21 by Artificial Ribonuclease: Biological Performance in Tumor Model

*Olga A. Patutina1, Svetlana K. Miroshnichenko1, Nadezhda L. Mironova1, Aleksandra V. Sen'kova1, Elena V. Bichenkova2, David J. Clarke2, Valentin V. Vlassov1 and Marina A. Zenkova1\**

#### *Edited by:*

*Ali H. Eid, American University of Beirut, Lebanon*

#### *Reviewed by:*

*Catherine Greene, Royal College of Surgeons in Ireland, Ireland Rute G. Matos, Universidade Nova de Lisboa, Portugal Antoni Benito, University of Girona, Spain*

> *\*Correspondence: Marina Zenkova marzen@niboch.nsc.ru*

#### *Specialty section:*

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

*Received: 13 May 2019 Accepted: 10 July 2019 Published: 08 August 2019*

#### *Citation:*

*Patutina OA, Miroshnichenko SK, Mironova NL, Sen'kova AV, Bichenkova EV, Clarke DJ, Vlassov VV and Zenkova MA (2019) Catalytic Knockdown of miR-21 by Artificial Ribonuclease: Biological Performance in Tumor Model. Front. Pharmacol. 10:879. doi: 10.3389/fphar.2019.00879*

*1 Laboratory of Nucleic Acids Biochemistry, Institute of Chemical Biology and Fundamental Medicine SB RAS, Novosibirsk, Russia, 2 School of Health Sciences, Faculty of Biology, Medicine and Health, University of Manchester, Manchester, United Kingdom*

Control of the expression of oncogenic small non-coding RNAs, notably microRNAs (miRNAs), is an attractive therapeutic approach. We report a design platform for catalytic knockdown of miRNA targets with artificial, sequence-specific ribonucleases. miRNases comprise a peptide [(LeuArg)2Gly]2 capable of RNA cleavage conjugated to the miRNAtargeted oligodeoxyribonucleotide, which becomes nuclease-resistant within the conjugate design, without resort to chemically modified nucleotides. Our data presented here showed for the first time a truly catalytic character of our miR-21-miRNase and its ability to cleave miR-21 in a multiple catalytic turnover mode. We demonstrate that miRNase targeted to miR-21 (miR-21-miRNase) knocked down malignant behavior of tumor cells, including induction of apoptosis, inhibition of cell invasiveness, and retardation of tumor growth, which persisted on transplantation into mice of tumor cells treated once with miR-21-miRNase. Crucially, we discover that the high biological activity of miR-21 miRNase can be directly related not only to its truly catalytic sequence-specific cleavage of miRNA but also to its ability to recruit the non-sequence specific RNase H found in most cells to elevate catalytic turnover further. miR-21-miRNase worked synergistically even with low levels of RNase H. Estimated degradation in the presence of RNase H exceeded 103 miRNA target molecules per hour for each miR-21-miRNase molecule, which provides the potency to minimize delivery requirements to a few molecules per cell. In contrast to the comparatively high doses required for the simple steric block of antisense oligonucleotides, truly catalytic inactivation of miRNA offers more effective, irreversible, and persistent suppression of many copy target sequences. miRNase design can be readily adapted to target other pathogenic microRNAs overexpressed in many disease states.

Keywords: artificial ribonuclease, oligonucleotide-peptide conjugate, RNA cleavage, RNase H, anti-miRNA therapy, miRNA-21

## INTRODUCTION

The well-recognized function of miRNAs as effective biological regulators and their role in the development of a broad spectrum of human pathophysiology offer oncology both novel biomarkers (Markou et al., 2013; Tahiri et al., 2014) and a promising paradigm shift in therapeutic targets (Frixa et al., 2015; O'Bryan et al., 2017; Zhou et al., 2017) from the protein-coding genome or already-expressed pathogenic proteins (Lennox and Behlke, 2011; Gambari et al., 2016).

To date, several technologies have been proposed to suppress excessive miRNA expression and function. Deep analysis of the structure of the key proteins involved in synthesis and maturation of miRNA, along with a large-scale ligand screening, has allowed selection of small molecule inhibitors suppressing enzymes involved in miRNA biogenesis broadly (Gumireddy et al., 2008; Lorenz et al., 2015; Di Giorgio et al., 2016). Much more selectively, miRNA-masking oligonucleotides can prevent the interaction of a miRNA with its target mRNA (Choi et al., 2007; Xiao et al., 2007; Wang, 2011), and anti-miRNA oligonucleotides can knock down target miRNA through formation of complementary complexes (Lennox and Behlke, 2011; Lennox et al., 2017; Miroshnichenko et al., 2019). Given the high doses often necessary, interest has grown in mopping up miRNA with larger structures with multiple miRNA binding sites, such as miRNA sponges (Ebert et al., 2007; Jung et al., 2015) and miRNA zippers, which tandemly bind the 3′- and 5′-ends of two miRNAs, thus forming an extended duplex with multiple copies of bound miRNA molecules (Meng et al., 2017).

Genome editing with the CRISPR/Cas9 system to decrease the level of mature miRNA (Chang et al., 2016; Huo et al., 2017) might be a potential alternative to binding many miRNA copies present and replenished in target cells. However, only sequenceselective artificial ribonucleases (aRNases) offer the potential to degrade many copies of miRNA in the cell. RNA-cleavage agents conjugated to oligonucleotide recognition motifs can selectively recognize specific miRNA sequences and hydrolyze the phosphodiester bonds within RNA molecules. However, in order to act as true enzymes, such artificial ribonucleases must exhibit catalytic turnover, which requires rapid release of RNA fragments after each cleavage event, followed by many subsequent attacks of other target RNA molecules. Metalion dependent (Dy3+, Eu3+, Cu2+, Zn2+) ribonucleases showed some promising results in terms of sequence-specific cleavage of RNA in a catalytic regime (Magda et al., 1997; Häner et al., 1998; Trawick et al., 2001; Niittymäki et al., 2004; Niittymäki and Lönnberg, 2006; Murtola et al., 2010). However, these aRNases were not amenable to application *in vivo* because of competing protein ligands and other bioavailable metals in cells and tissue, which render them virtually uncontrollable. Moreover, metal loss from their coordinating ligands led to concerns about their toxicity in humans. Metal-independent artificial ribonucleases offer advantages including reduced toxicities and the ability to work in intracellular conditions. However, most of the developed metal-independent aRNases are less active and show low to negligible catalytic turnover (Mironova et al., 2006; Mironova et al., 2007). Hybrid enzymes engineered by the fusion of oligodeoxyribonucleotide to staphylococcal nuclease efficiently hydrolyzed complementary RNA targets but did not leave the substrate after cleavage (Zuckermann and Schultz, 1989). This was overcome by creating a hybrid of *Escherichia coli* RNase H and a 9-mer oligodeoxynucleotide, when the catalytic turnover of the reaction of sequence-specific cleavage was achieved (Kanaya et al., 1992), but never investigated *in vivo*.

Recently we have reported the discovery of novel miRNAspecific metal-independent aRNases, here called miRNases, capable of selective cleavage of biologically significant miRNAs (Patutina et al., 2017a), where we presented a design concept, synthesis, and full characterization of a panel of the bioconjugates at their discovery stage with the significant structural and functional variations. Each of those miRNases comprised an oligodeoxyribonucleotide complementary to the miRNA of interest as a recognition motif conjugated to peptide acetyl-[(LeuArg)2Gly]2 capable of a site-specific scission of phosphodiester bonds within the target miRNA molecule. That pilot study allowed us to compare *back-to-back* the ability of such conjugates to cleave the target miR-21 in a sequencespecific manner to distinguish successful structural candidates from inactive counterparts. One of the most efficient miRNases from that series (i.e. 5′-h-9/14) was used for preliminary evaluation of a biological activity in order to estimate the future antitumor potential of these catalytic bioconjugates. As the selected miRNase targeted to the oncogenic miR-21 (miR-21-miRNase) showed efficient sequence-selective cleavage, leading to inhibition of miR-21 in lymphosarcoma cells and suppression of tumor cell proliferation (Patutina et al., 2017a), we undertake here a comprehensive study of its serum stability and biological performance. We demonstrate that specific deactivation of miR-21 by miR-21-miRNase induces apoptosis in tumor cells, suppresses their invasive properties, and decreases the proliferative rate of tumor cells *in vitro*. We show here for the first time that a single treatment of tumor cells with miR-21 miRNase persisted to retard subsequent tumor growth *in vivo*. We report here the discovery that the high efficiency of miRNA inactivation in cells by this miRNase is related to its ability to work in a true catalytic mode with multiple turnover, which is increased considerably by recruitment of the non-sequence specific intracellular RNase H.

### MATERIALS AND METHODS

### Peptide-Oligonucleotide Conjugates (POCs)

Oligodeoxyribonucleotides (with and without an aminohexyl linker attached to the 5′-terminal phosphate of the oligonucleotide, and 2′-O-methyl-modification (2′-OMe) (see **Supplementary Table 1**) were synthesized in ICBFM SB RAS (Russia) by Dr. V. Ryabinin and Dr. Maria Meschaninova. The catalytic peptide, acetyl-[(LeuArg)2Gly]2, was attached *via* its C-terminus to the amine group of the aminohexyl linker located at the 5′-terminus of respective oligonucleotides to prepare the miR-21 miRNase and luc-POC conjugates (Patutina et al., 2017a).

### 5**′**-RNA Labeling

miR-21, 5′-UAGCUUAUCAGACUGAUGUUGA-3′, was 5′-end labeled using γ-[32P]-ATP and T4 polynucleotide kinase (Thermo Scientific, USA) as previously described (Silberklang et al., 1979; Mironova et al., 2007).

#### Assay of Ribonuclease Activity and miR-21-miRNase in Multiple Turnover Conditions

Unlabeled miR-21 (10 µM, 25 µM, or 50 µM) and 105 cpm (Cherenkov's counting) of [32P]-miR-21 was incubated at 37°C with either miR-21-miRNase (5 µM) or RNase H (100 U/ml, plus 5 µM of either h-ODN or miR-21-miRNase) in buffer (20 mM Tris-HCl, pH 7.8, 40 mM KCl, 8 mM MgCl2, 1 mM DTT). Reactions were quenched by precipitation of RNA with 2% LiClO4 in acetone, and RNA cleavage products quantified in dried gels as described previously (Patutina et al., 2017a) using Molecular Imager FX and Quantity One software.

#### Nuclease Resistance Assay

Oligonucleotides or conjugates (0.1 µg/µl) were incubated at 37°C in Dulbecco's Modified Eagle Medium (DMEM) (Sigma, USA) with 10% fetal bovine serum (FBS; GE Healthcare, USA). Reactions were quenched in urea (8M) and immediately frozen in liquid nitrogen, prior to analysis of thawed samples in 12% PAAG/8M urea gels, using Tris-Borate-EDTA (TBE) as running buffer, Stains-All (MP Biomedicals, USA) and photographic gel documentation (VilberLourmat, France).

### Cell Lines

Mouse B16 melanoma cells, grown in DMEM/10% FBS/1% antibiotic antimycotic solution (ICN, Germany) and RLS40 lymphosarcoma cells, grown in IMDM/10% FBS/1% antibiotic antimycotic solution, were cultivated at 37°C in a humidified incubator with 5% CO2.

### Transfection of Cells With Oligonucleotides and Conjugates

B16 cells were pre-seeded (2 × 105 per well of 24-well plates) in DMEM/10% FBS a day before transfection. Medium was replaced with serum-free DMEM, prior to incubation for 4 h with h-ODN, 2′-OMe, Inh (hsa-miR-21 5 p inhibitor, Ambion, USA), miR-21-miRNase, or luc-POC, each (0.1 or 1 µM) pre-complexed with Lipofectamine (Thermo Fisher Scientific, USA) in Opti-MEM medium (Thermo Fisher Scientific, USA) according to manufacturer's instructions, then cells were cultivated in fresh culture standard medium for 24–72 h.

### Annexin V-FITC/PI Apoptosis Assay

At 24, 48, and 72 h after transfection, cells were harvested, washed twice with PBS, resuspended in Binding Buffer to a density of 2–5 × 105 /ml, and incubated with annexin V-FITC and propidium iodide (PI) (ab14085, Abcam, UK) in the dark at room temperature for 15 min, prior to flow cytometry analysis (Novocyte CEA Biosciences, USA).

### *In Vitro* Invasiveness Assay

Real-time cell analysis (xCELLigence, ACEABiosciences, USA) used CIM-Plates held at 37°C under an atmosphere of 5% CO2. Top chamber wells were coated with Matrigel (20 µl per well, diluted 1:40 with cold serum-free DMEM) and allowed to polymerize (4 h at 37°C under 5% CO2) prior to cell seeding. Bottom chamber wells were filled with DMEM/10% FBS (160µl each), prior to assembly of the chambers, when serum-free DMEM was added to the top wells (30 µl each) and equilibrated (1 h at 37°C under 5% CO2). At 4 h after transfection, B16 cells (4 × 104 ) were seeded into each top chamber well in serum-free DMEM and pre-incubated for 30 min at room temperature before monitoring (xCELLigence set to collect impedance data, reported as cell index at least once every 30 min).

#### Mice

Male 10–12 week-old CBA/LacSto (hereinafter, CBA) mice were kept in the vivarium of the Institute of Chemical Biology and Fundamental Medicine, SB RAS, with a natural light regime on a standard diet for laboratory animals [GOST (State Standard) R 5025892] in compliance with the international recommendations of the European Convention for the Protection of vertebrate animals used for experimental studies (1997), as well as the rules of laboratory practice in the performance of pre-clinical studies in the Russian State Standards (R 51000.3–96 and 51000.4–96). The experimental protocols were approved by the Committee on the Ethics of Animal Experiments with the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences.

### Tumor Transplantation Assay

RLS40 ascites were taken from CBA mice intraperitoneally injected with tumor cells (2 × 106 in 0.2 ml buffered saline) into their abdominal cavities. RLS40 cells, isolated from ascites fluid by filtration through LSM were divided into seven portions: 1) untreated cells; 2) the cells incubated with Lipofectamine ; the cells transfected with 1 µM of 3) h-ODN, 4) 2′-OMe, 5) Inh, 6) miR-21-miRNase, or 7) luc-POC each. After 4 h of transfection, cell suspensions (0.1 ml, 2 × 106 cells) were intramuscularly inoculated into the right lower limb of CBA mice for solid tumor development. As soon as tumors began to be palpable, the tumor volumes were measured every 2–3 days using calipers. Tumor volumes were calculated as V = (π/6 × length × width × height). Hepatic index was estimated as (liver weight/mouse weight) × 100% and as 5.0% for healthy CBA mice. Tumor doubling time (DT) was estimated as DT = (*t* − *t*0) × ln 2/(ln *V* − ln *V*0), where (*t* − *t*0) indicates the length of time between two measurements of tumor size and *V*0 and *V* denote the tumor volume at two points of the measurement (Ozono et al., 2004).

### STATISTICS

Data were statistically processed using two-way ANOVA taking into account two factors: antisense oligonucleotide and peptide. *Post hoc* testing was completed using Tukey test; *p* < 0.05 was considered to be statistically significant. The statistics package STATISTICA version 12.0 was used for this analysis.

### RESULTS

#### Structure of Nuclease-Resistant miR-21-miRNase

Stability of antisense oligonucleotides (AOSs) against nucleases in serum is essential for *in vivo* application, which typically requires nuclease-resistant chemically modified nucleotides. We avoid this here by design of a nuclease-resistant structure with a hairpin at the 3′-end and the catalytic peptide at the 5′-end. The hairpin oligonucleotide (h-ODN) comprised a 14-mer recognition motif at the 5′-terminus, which is complementary to the miRNA target, and a hairpin structure with a four-member apical loop and 9 nts stem at its 3′-terminus. The catalytic peptide, acetyl- [(LeuArg)2Gly]2, was covalently attached *via* its C-terminus to the 5′-end of the oligonucleotide through an aminohexyl linker (**Figure 1**). The biological and therapeutic properties of miR-21-miRNase was compared with the properties of miR-21-targeted hairpin deoxyribooligonucleotide (h-ODN) and 2′O-methyl-modified oligonucleotide (2′-OMe) unconjugated with the peptide, with a linear 14-mer deoxyribooligonucleotide (ODN) and with the control conjugate (luc-POC), in which the sequence of the hairpin was the same but the complementary part was replaced by the fragment of the luciferase gene with no homology in the mammalian genome (**Supplementary Table 1**). Although the free oligonucleotide (ODN) was chemically unmodified and rapidly degraded in serum, the hairpin ODN resisted dilute serum for a few hours but survived for over 48 h (see **Supplementary Figure 1**), when conjugated to the catalytic peptide within the miRNase (**Supplementary Figure 1**).

#### Induction of Apoptosis and Suppression of Invasiveness of Tumor Cells

By modulating tumorigenesis through the regulation of multiplicity of critical genes suppressing malignant growth, evasion from apoptosis, invasion, and metastasis, miR-21 functions as an oncogene (Melnik, 2015). Knockdown of miR-21 by miR-21 miRNase resulted in specific elevation of the level of the direct mRNA target of miR-21, tumor-suppressor programmed cell death 4 (PDCD4) (Patutina et al., 2017a), and so may also turn down the malignant behavior of tumor cells through induction of apoptosis and inhibition of invasion and metastasis. We studied this in B16 melanoma cells, which are known for maintaining aggressive growth, active migration, and metastasis. As indicated by the Annexin V-FITC/PI assay, miR-21-miRNase increased necrosis of melanoma B16 cells to a limited extent (up to 7%) similar (up to 6.5%) to the sham peptide-oligonucleotide conjugate, luc-POC (for luc-POC sequence see **Supplementary Table 1**) with a sequence of a firefly luciferase gene with no complementarity to the mammalian genome. However, miR-21-miRNase caused a significant increase in the proportion of cells in the apoptotic state with a maximum at 48 h after transfection: the signs of apoptosis were observed in 28% of the cells (cf 15% control), which was six times higher than that in the intact B16 cells (**Figures 2A**, **B**, **Supplementary Figure 2**).

Cells transfected with h-ODN (blue curve), or its methylated derivative 2′-OMe (violet curve) or the sham luc-POC (green curve), migrated at almost the same rate as cells incubated with Lipofectamine (grey curve) (**Figures 2C**, **D**). However, miR-21 miRNase (red curve) strongly inhibited the invasion potential of B16 cells by about 70% relative to control and sham luc-POC (*p*=0.0008). (**Figures 2C**, **D**, time point 40 h). Knockdown of miR-21 by commercial Inh (orange curve) showed only a delayed inhibition of cell invasion of approximately 40% relative to control and only 20% relative to luc-POC.

#### Persistence to Suppress Tumor Growth

To evaluate the ability of miRNase to suppress growth of tumor cells and dissemination in the body, an experiment in mice was conducted. RLS40 lymphosarcoma cells were treated with the miRNase *in vitro* and then intramuscularly transplanted to the right lower limb of CBA mice to form solid tumors with dissemination in the body. miR-21-miRNase efficiency was assessed in a specially collected portion of the treated cells: a 2-fold decrease in miR-21 level by miR-21-miRNase was confirmed by PCR analysis (24 h after transfection), and 1.7- and 2.4-fold increase in protein levels of miR-21 target proteins— PTEN and PDCD4—was shown by Western blot hybridization, respectively, (72 h after transfection) (**Supplementary Figure 3**) which is consistent with previously obtained data (Patutina et al., 2017a). Despite no further treatment of the cells or animals transplanted, the effects of miR-21-miRNase persisted to retard subsequent tumor growth, with a mean tumor volume 16.7 times less than in the control group, showing a 94% tumor volume reduction upon treatment (**Figures 3A**, **B**), which was a statistically reliable difference (*p* value was 10−6 against control). Similar inhibition of tumor growth could only be seen when dosing of the positive controls with Inh was much elevated (1 µM, **Figures 3A, B**) to a 10- to 20-fold higher level than the manufacturer's recommendations (0.05–0.1 μM). Whether the positive control was then specific or caused by a non-specific cytotoxic effect of Inh is unclear. The higher rate of tumor growth in the groups transplanted with RLS40 cells with no treatment (w/t), cells treated with Lipofectamine (LF) alone, Inh (0.1 μM), the methylated h-ODN 2′-OMe (1 μM), and luc-POC did not differ, and h-ODN retarded primary tumor growth at the limit of statistical significance (*p* ~ 0.05) by only about 2-fold (**Figures 3A**, **B**).

Tumor doubling time (DT) was increased to 12.77 ± 0.34 days by miR-21-miRNase, >5-fold greater than of untreated control (w/t) of 2.45 ± 0.49 days (**Figure 3C**). The groups LF, 2′-OMe, luc-POC and h-ODN did not show a statistically significant difference against the w/t, and neither did the positive control of Inh, until a much (10×) elevated level was used to increase

DT to a similar extent (13.90 ± 0.46 days). Relative liver weights remained similar to healthy animals for the miR-21-miRNase and excessive Inh treatments (see **Supplementary Table 2**), even though absence of metastases in the liver may not be expected to persist in tumors seeded from cells treated only before transplant.

### Synergistic Catalytic Recruitment of RNase H

It is desirable that a true artificial nuclease exhibits a reaction catalytic turnover as a result of eventual release of the cleaved RNA fragments followed by attack of the next target miRNA molecules. Using the synthetic 5′-[P32]-labeled miR-21, we evaluated the ability of miRNase to cleave multiple copies of miR-21, when significant excess (up to 10-fold) of miRNA target was used over miR-21-miRNase. Since the recognition motif of miR-21-miRNase was represented by the "naked" (e.g., unmodified) oligodeoxyribonucleotide, it was anticipated that in cellular environment the duplex formed between the miRNase and miR-21 can potentially be recognized by RNase H, which is present in virtually all cells. Therefore, we aimed to evaluate the mutual effect of both RNases on each other (RNase H and miRNase) by comparing their individual activity against miR-21 with their joint action at various excess of miRNA target over miRNase. To achieve this, we ran nine series of the parallel cleavage assays (see **Figures 4A**–**G**). Series A(I), C(IV), and E(VII) (**Figure 4**) measured the efficiency of miRNA cleavage by miRNase alone at the 2-fold, 5-fold, and 10-fold excess of the target RNA with respect to miRNase, respectively. Series A(III), C(VI), and E(IX) evaluated the cleavage efficiency of the RNase H alone towards the miR-21 target present at the same concentrations, while it was hybridized into a heteroduplex with the corresponding oligonucleotide h-ODN lacking the catalytic peptide. Series A(II), C(V), and E(VIII) measured the combined action of miRNase and RNase H present together in the reaction.

We compared the cleavage patterns and kinetics of the miR-21-miRNase with the non-sequence selective RNase H, which is present in virtually all cells. miR-21-miRNase alone cleaved miR-21 at the G3C4, G15A16, G18U19, and G21A22 sites (**Figures 4A**-I, **C-**IV, and **E**-VII), whereas RNase H cuts the target predominantly at the U6A7, C9A10, and G11A12 sites of miR-21 (**Figures 4A**-III, **C**-VI, and **E**-IX). Neither of these cleavage patterns was affected by the concentration of miR-21. The respective activities of miR-21-miRNase and RNase H in combination (**Figures 4A**-II, **C**-V, and **E**-VIII) provided clear evidence that the hetero-complex formed by miRNA

experiments.

and miRNase is also a substrate for RNase H. Target miRNA was effectively cleaved by RNase H when heteroduplexed to the hairpin targeting structure, whether alone (h-ODN) or when conjugated to the peptide (miR-21-miRNase). Moreover, miRNA cleavage occurred at both G-X sites, which are specific for miRNase, and at the sites sensitive to RNase H.

At 2-fold molar excess of miR-21 with respect to h-ODN (**Figure 4A**), the kinetics of miRNA cleavage in a heteroduplex with h-ODN by RNase H alone reached a plateau by 30 min of incubation, but with a total extent of cleavage of only 45% (**Figures 4A**-III, **B**—blue curve). Moreover, with an increase in miRNA concentration to 5- and 10-fold excesses (**Figures 4C**, **E**), the extent of miRNA cleavage by RNase H fell to about 10% and 5%, respectively (**Figures 4C**-VI, **E**-IX; **D** and **F**—blue curves), which presumably reflected the lesser proportions of the miRNA/h-ODN heteroduplex within the overall amount of miRNA present. However, the extent of miRNA cleavage by the miRNase alone was higher and clearly indicative of catalytic turnover. Indeed, the level of cleavage of miR-21 by miRNase alone was similar irrespective of the 2-, 5-, or 10-fold excess of the target over miRNase and progressed towards 65–68% over 24 h (**Figures 4A**-I, **C**-IV, **E**-VII; **B**, **D** and **F**—black curves), and 90% over 72 h (**Supplementary Figure 4A**). Clearly, this miR-21-miRNase structural design was capable of cleaving many copies of the target miRNA, with a greater efficiency than that achieved by a classical antisense pathway *via* RNase H recruitment.

However, as RNase H is ubiquitous in target cells, we also considered whether the simultaneous presence of RNase H affected degradation of miR-21 by this miR-21-miRNase. We discovered that the presence of RNase H led to a considerable increase in the rate and extent of miRNA cleavage, as compared to the individual action of each enzyme alone under identical conditions (**Figure 4**, compare II with I and III, V with IV and VI, and VIII with VII and

FIGURE 3 | Antitumor effect of miR-21-specific miRNase. (A) Retardation of RLS40 growth *in vivo* after treatment of RLS40 cells with antisense oligonucleotides, control conjugate (luc-POC), or miR-21-miRNase. (B) Tumor volume on day 19 after RLS40 transplantation. Mice were injected with RLS40 cells either without any treatment (w/t) or treated with Lipofectamine (LF), commercial miR-21 inhibitor at concentration 0.1 μM (Inh 0.1, recommended) and 1 μM (Inh 1, in accordance to concentration of miRNase used in the study), control conjugate (luc-POC, 1 μM), hairpin 2′OMe oligonucleotide (2′-OMe, 1 μM), hairpin antisense oligonucleotide (h-ODN, 1 μM) or miR-21-specific miRNase (miR-21-miRNase, 1 μM). Data were statistically analyzed using two-way ANOVA with *post hoc* Tukey test. Data are presented as mean ± s.e. *p*-value indicates a statistically reliable difference. (C) Doubling time of RLS40 tumors after treatment with oligonucleotides and conjugates.

FIGURE 4 | Cleavage of 5'-[32P]-labeled miR-21 by miR-21-miRNase and/or RNase H. Autoradiographs of 18% polyacrylamide/8 M urea gel, showing the patterns of cleavage of miR-21 at its 2- (A), 5- (C), and 10-fold (E) molar excess over miR-21-miRNase or h-ODN. Autoradiographs I, IV, and VII represent cleavage of miR-21 by miR-21-miRNase alone, whereas autoradiographs II, V, and VIII show cleavage of miR-21 by a mixture of miR-21-miRNase and RNase H. Autoradiographs III, VI, and IX demonstrate cleavage of the complex formed between miR-21 and h-ODN by RNase H alone (100 U/ml). Duplexes formed by 5′-[32P]-miR-21 (10, 25, and 50 μM) and h-ODN or miRNase (5 μM) were incubated at 37°C for 24–72 h. Lanes Im and T1: imidazole ladder and partial RNA digestion with RNase T1, respectively; "control": RNA was incubated in the absence of oligonucleotide or conjugate and in the presence of RNase H (100 U/ml). Diagrams (B), (D), and (F) show time dependency of cleavage of miR-21 by miR-21-miRNase alone, by a synergetic action of miR-21-miRNase and RNase H, or by RNase H (100 U/ml) alone when miR-21 was in a complex with h-ODN at 2-, 5- and 10-fold molar excess of miR-21 over miR-21-miRNase or h-ODN, respectively. (G) Positions of miR-21 cleavage induced by miR-21-miRNase (red arrows), RNase H in the duplex with h-ODN (black arrows), and by combination of miR-21-miRNase and RNase H. (H) The hypothetical representation of miR-21 cleavage by a combination of miRNase and RNase H.

IX). At a 2-fold excess of miR-21 over miR-21-miRNase, complete (~100%) RNA degradation was observed after 8 h of incubation (**Figures 4A**-II, **B**—red curve, **Supplementary Figure 4B**). With a 5-fold excess, such degradation took 24 h (**Figures 4C**-V, **D** red curve, **Supplementary Figure 4B**), and with a 10-fold excess, RNA degradation was complete within 48 h (**Figures 4E**-VIII, **F**—red curve, **Supplementary Figure 4B**).

By interpolating the half-life (τ/2) of miR-21 degradation, we estimated the catalytic turnover in the absence and presence of RNase H. RNase H elevated cleavage of miR-21 by the miR-21-miRNase by a factor of 14.9 times in the case of 2-fold excess of miR-21, whereas this was less pronounced at 3.0 and 2.4 times for greater excesses of miR-21 at 5-fold and 10-fold, respectively (**Table 1**). In the absence of RNase H, the catalytic turnover of miR-21-miRNase alone was estimated to be around 100 cleaved RNA molecules per molecule of miRNase in 1 h in the cases of 2- and 5-fold excess of miR-21, but reached 382 cleaved RNA molecules per hour at 10-fold miR-21 excess (**Table 1**). The presence of RNase H enhanced the catalytic turnover by factors of 15.6, 7.0, and 4.6 times for 2-fold, 5-fold, and 10-fold excess of miR-21, respectively, with the potential to reach up to 1,800 cleaved miR-21 molecules by one miR-21-miRNase molecule per hour. A possible explanation of such a synergistic effect is that the combined action of the miRNase and RNase H led to formation of very short cleavage fragments of three nucleotides and shorter (**Figures 4A**-II, **4C**-V, **4E**-III) that may rapidly dissociate from the hybridized complex and release the sequencespecific miR-21-miRNase, when it would become available for another cleavage event (see **Figure 4H**). Indeed, the almost complete cleavage of miRNA at 5-fold excess formed short 2 and 3-nucleotide fragments (**Supplementary Figure 5**).

## DISCUSSION

Short functional non-coding microRNAs are implicated in many types of cancer (Frixa et al., 2015; O'Bryan et al., 2017; Zhou et al., 2017). Amongst many miRNAs identified as important inducers of oncogenesis, miR-21 represents an extreme oncomiR, which is strongly involved in tumor onset and progression for many types of malignancies (Huang et al., 2013; Lianidou et al., 2016; Yu et al., 2018). Since miR-21 is abnormally overexpressed in major types of tumors, this oncomiR is the most considered target for anti-miRNA inhibitory studies. However, depending on the origin of tissue and degree of pathological state, the concentration of miRNAs in the cell can vary from several copies to 2–4 μM concentrations, which can be too great to knock down effectively by antisense approaches alone. The most widely used anti-miRNA oligonucleotides, including commercial inhibitors fabricated on the basis of nucleic acids, are usually the perfect complements to miRNA targets, which contain various chemical modifications at the sugar-phosphate backbone to enhance their nuclease resistance and improve binding affinity. However, the chemical substitutions used for generation of such inhibitors are often not compatible with the activity of intracellular RNase H, and their inhibitory effect relies on 1:1 stoichiometry of binding between the oligonucleotide-based inhibitor and miRNA target. Although the use of such oligonucleotide-based inhibitors shows detectable level of miRNA silencing, the absence of the catalytic amplification represents their disadvantage as compared with our biocatalytic agents, capable to degrade irreversibly multiple copies of pathogenic miRNA molecules, which will be particularly crucial to effectively suppress malignant growth. On the basis of our recent discovery of peptide-oligonucleotide conjugates (Patutina et al., 2017a; Staroseletz et al., 2017;


*1)τ/2—a half-life time of miR-21 in the presence of catalysts, the time during which 50% of miRNA molecules is cleaved.*

*2)Rt, reaction turnover—the number of miRNA molecules cleaved by one molecule of miRNase per 1 h. Rt was calculated according to equation Rt = [miR-21] × (number of bonds) × (cleavage, %)/[miRNase].*

Patutina et al., 2018) with desired biocatalytic properties against disease-relevant RNAs, we have designed miRNA-specific aRNase targeted to miR-21. Efficient knockdown of miR-21 by the miR-21-miRNase triggered a broad spectrum of biological responses, leading to the inhibition of pro-survival behavior of tumor cells, including inhibition of tumor cell proliferation by 50% (Patutina et al., 2017a), the initiation of apoptosis in 28% of the tumor cell population, and suppression of cell invasiveness by 60–70%. Furthermore, upon transplantation of the treated tumor cells into mice, the oncosuppressive effect of miR-21-miRNase persisted. Even a single treatment of tumor cells with miR-21 miRNase prior to transplantation led to a 50% reduction in miR-21 level and a drastic (94%) reduction in primary tumor growth in mice. Tumor doubling time increased by more than 5-fold (from 2.45 to 12.77 days) for tumor cells treated with miR-21 miRNase prior to transplantation, as compared to control groups without treatment or treated with a sham (luc-POC) or the hairpin oligonucleotide (h-ODN). The biological effects induced by this miRNase are comparable or superior to those of available miR-21 oligonucleotide inhibitors (Dong et al., 2012; Wagenaar et al., 2015; Wu et al., 2016; Wang et al., 2018; Zhou et al., 2018).

The design developed for this miRNase: i) enhanced the hybridization efficiency towards the miRNA target, probably due to the additional stacking interactions provided by the hairpin structure (Sunami et al., 2004; Patutina et al., 2017b); ii) bound only mature forms of miRNA; and iii) protected the oligonucleotide from nuclease degradation. It is well known that high nuclease susceptibility of natural oligodeoxyribonucleotides leads to short half-life and reduced antisense functionality *in vivo*, typically requiring the use of chemically modified oligonucleotides to avoid this rapid degradation. However, the rational design of nuclease resistance into these therapeutic miRNases was most likely provided by the hairpin targeting structure and the catalytic peptide location. The h-ODN with a hairpin at the 3′-end remained intact for a few hours but, with the catalytic peptide conjugated at the 5′-end of the hairpin oligonucleotide, persisted for more than 24 h. This nuclease resistance of the miR-21-miRNase undoubtedly contributed to the longer and stronger biological effect observed *in vitro* and *in vivo*. The nuclease resistance of this miRNase is not inferior to the stability of potent phosphorothioate DNA/LNA mixmer, which begins to degrade by 24 h of similar incubation (Lennox and Behlke, 2010). The obtained results correlate with the data reported earlier, where it was shown that 3′-end structures can improve resistance to snake venom phosphodiesterase, DNA polymerase I, and FBS and significantly increase functional potency for target RNA knockdown (Tang et al., 1993; Hosono et al., 1995). The presence of flanking duplexes also improves the ability of anti-miRNA oligonucleotides to invade RISC and serve functional enhancement (Vermeulen et al., 2007; Lennox and Behlke, 2011).

Clear evidence of the catalytic nature of miRNA cleavage by this miRNase was found when the miR-21 was present at 2-, 5-. or 10-fold excess over the miRNase. In contrast, RNase H alone did not exhibit any catalytic turnover, as the extent of miR-21 cleavage by this ribonuclease simply correlated with the proportion of the miR-21/h-POC hetero-complex within the overall amount of miRNA present, which is the expected result for a conventional antisense-mediated pathway involving RNase H recruitment. Interestingly, neither introduction of the hairpin nor the presence of the peptide within the structure of the miR-21-miRNase affected the ability of RNase H to recognize and cleave miR-21 hybridized with the oligonucleotide moiety of the miR-21-miRNase.

Particularly exciting though was the discovery of the synergistic effect from the joint action of this miRNase and RNase H towards miR-21. In isolation, both RNases cleaved totally different regions of miR-21, which can be easily identified from the PAAG analysis (**Figure 4**). However, when miR-21 formed a complementary duplex with h-ODN oligonucleotide, RNase H recognized this hetero-complex and cleaved miRNA at the "seed region," located close to the 5′-end (bases 2–8), which is known to be a "canonical" determinant in miRNA function (Lai, 2002). In contrast, miR-21 miRNase cleaved the 3′-region of the miR-21 molecule (bases 13–18), which represents a "3′-compensatory" or "beneficial 3′-paring" site in the miRNA-mRNA target recognition and plays a crucial role for miRNA specificity and functioning (Brennecke et al., 2005; Grimson et al., 2007; Wang et al., 2009; Robertson et al., 2010). However, when both ribonucleases were present (which appears likely *in vivo* in the intracellular environment), we observed a significant increase of the rate of miRNA cleavage. Earlier, using miRNA-targeted conjugates of a different structure, we have shown that the presence of RNase H can significantly enhance its activity and increase the rate of target RNA cleavage, without any positive effect on RNase H activity (Patutina et al., 2018). In this study, we discovered a different mechanism of synergy when both RNases showed mutual enhancement of their cleavage activity in the presence of each other, which was considerably higher than any simple additive effect expected from independent enzymatic action. The observed synergistic effect from this combined attack on miRNA molecules could be attributed to multiple cuts within miR-21 induced by these ribonucleases, thus leading to a formation of short cleavage products. This may trigger a rapid collapse of the hybridized complex and a release of both miRNase and RNase H, when each could then initiate a new attack on another miR-21 molecule (**Figure 4H**). This mechanism may well explain the high level of inactivation of oncogenic miRNA and the persistence of diverse suppression of malignant behavior of tumor cells upon treatment with this miR-21-miRNase.

We conclude that high therapeutic efficacy of peptideoligonucleotide conjugates is likely to arise from the following: 1) high resistance to nuclease degradation from structural design, 2) efficient cleavage of miRNA-target molecules in a multiple turnover or true catalytic manner, and 3) by synergistically harnessing intracellular RNases such as RNase H. Over the past few decades, many different artificial ribonucleases have been reported, with some cleaving RNA *in vitro* with high efficiency and selectivity. However, to the best of our knowledge, we report here the first example of metal-independent, RNAspecific peptide-oligonucleotide conjugates operating under physiologically relevant conditions within tumor cells, which maintained miRNA knockdown in tumors subsequently formed from cells transplanted into mice, without further treatment. These findings elevate these sequence-specific aRNases to worthy rivals to antisense and siRNA technologies for the development of new RNA targeting therapeutics.

The key discoveries of this research provide a solid experimental ground for facilitating a paradigm shift in the development of new therapeutic interventions to ultimately offer effective, safe and cost-efficient treatment of unresectable or metastatic tumors, which currently relies on the cytotoxic effect of chemotherapies or radiotherapy and inevitably leads to a severe toxicity in humans. A fundamental change in treatment from smallmolecule anti-cancer drugs, which often suffer from poor target selectivity, to highly selective RNA-targeting therapies provides a possibility to overcome such undesirable side effects arising from cytotoxic drug cocktails of combination regimens currently used in oncology. Our miRNA-selective chemical ribonucleases offer unprecedented opportunity to selectively knock down many copies of oncogenic miRNA in cells by irreversible cleavage in a truly catalytic manner with multiple turnovers, which is reinforced even further by recruitment of intracellular RNase H. This will ultimately boost potency, reduce dosage, and decrease cost of treatment. In the future, miRNases can be used both as a monotherapy and as an adjuvant therapy in combination with surgical treatment, radiotherapy, and chemotherapy. In the latter case, miRNases may allow to significantly decrease the dosage of cytostatics thereby reducing toxicity. Application of miRNases is a promising and rapidly evolving area of antisense technology, and its suitable combination with chemotherapeutics may represent highly efficient approaches to treating oncopathologies and other miRNA-associated diseases in humans.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript/supplementary files.

### ETHICS STATEMENT

This study was carried out in accordance with the international recommendations of the European Convention

#### REFERENCES


for the Protection of vertebrate animals used for experimental studies (1997), as well as the rules of laboratory practice in the performance of pre-clinical studies in the Russian State Standards (R 51000.3–96 and 51000.4–96), the Committee on the Ethics of Animal Experiments with the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences. The protocol was approved by the Committee on the Ethics of Animal Experiments with the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences.

#### AUTHOR CONTRIBUTIONS

MZ, EB, and OP conceptualized and designed the study. VV supervised the research. OP, SM, NM, AS, and EB performed the experiments and analyzed the data. OP, EB, DC, and MZ interpreted the data, wrote, edited, and revised the manuscript. All authors approved the final version of the manuscript.

#### FUNDING

This work was funded by Russian Science Foundation (Grants No. 14-44-00068 and No. 19-14-00250) and by Russian State funded budget project of ICBFM SB RAS # АААА-А17-117020210024-8, BBSRC (Grant No. BB/K012622/1), and EPSRC (Grant No. EP/ E003400/1).

#### ACKNOWLEDGMENTS

The authors are grateful to Mrs. Albina V. Vladimirova (ICBFM SB RAS) for cell maintenance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00879/ full#supplementary-material


distant metastasis and survival in kidney renal clear cell carcinoma. *J. Cancer* 9, 3651–3659. doi: 10.7150/jca.27117


**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 Patutina, Miroshnichenko, Mironova, Sen'kova, Bichenkova, Clarke, Vlassov and Zenkova. 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.*

# RNases Disrupt the Adaptive Potential of Malignant Cells: Perspectives for Therapy

*Vladimir Alexandrovich Mitkevich\*, Irina Yu Petrushanko and Alexander Alexander Makarov*

*Engelhardt Institute of Molecular Biology, Russian Academy of Sciences, Moscow, Russia*

Exogenous RNases are selectively toxic to tumor cells. The reasons for this selectivity are not quite clear and should be searched for in the properties that distinguish malignant from normal cells. During onco-transformation, cells acquire properties allowing them to adapt to the altered microenvironment, such as resistance to hypoxia, changes in intracellular pH, disruption of ion transport, reduced adhesion and increased mobility, and production of specific exosomes. These adaptation mechanisms distinguish malignant cells from normal ones and give them a competitive advantage, ensuring survival and spread in the organism. Here, we analyze if the directed cytotoxic effect of exogenous RNases is linked to the disruption of the adaptive potential of tumor cells and how it can be used in anticancer therapy.

#### *Edited by:*

*Jean-Michel Fustin, Kyoto University, Japan*

#### *Reviewed by:*

*Pavel Ivanov, Brigham and Women's Hospital, United States David Pulido-Gomez, University of Oxford, United Kingdom*

### *\*Correspondence:*

 *Vladimir Alexandrovich Mitkevich mitkevich@gmail.com*

#### *Specialty section:*

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

*Received: 10 April 2019 Accepted: 22 July 2019 Published: 16 August 2019*

#### *Citation:*

*Mitkevich VA, Petrushanko IY and Makarov AA (2019) RNases Disrupt the Adaptive Potential of Malignant Cells: Perspectives for Therapy. Front. Pharmacol. 10:922. doi: 10.3389/fphar.2019.00922*

Keywords: RNase, malignant cell, adaptation potential, external environment, redox state

### INTRODUCTION

Enzymes that hydrolyze RNA, ribonucleases (RNases), have several biological effects on cells. Stimulation of the growth of blood vessels, toxicity toward tumor cells, and antiviral activity are important therapeutic properties of RNases. To date, a significant number of RNases with a selective cytotoxic effect on tumor cells have been studied (Libonati et al., 2008; Makarov et al., 2008; Ardelt et al., 2009; Fang and Ng, 2011; Castro et al., 2013) making them potential therapeutic agents for cancer. However, the mechanism of their antitumor activity remains largely unclear. RNases are thought to exhibit cytotoxic activity by cleaving cellular RNA. The most accessible for them are RNAs that are not associated with proteins, i.e., transport and non-coding miRNAs. Thus, angiogenin cuts tRNA molecules inside anticodon loops, producing "stress-induced" RNA fragments, which leads to inhibition of translation initiation (Ivanov et al., 2011). The cytotoxicity of onconase, RNAse from *Rana pipiens*, is also associated with the degradation of tRNA in the cytosol of tumor cells with subsequent inhibition of protein synthesis and apoptosis (Saxena et al., 2002). Сytotoxic RNAse α-sarcin from *Aspergillus oryzae* specifically cleaves one bond in 28S rRNA, causing inhibition of translation and subsequent cell death (Olmo et al., 2001). However, numerous evidences indicate that inhibition of protein synthesis is not the only cause of apoptosis induced by RNases (Qiao et al., 2012). In particular, it has been shown that binase, RNase from *Bacillus pumilus*, directly interacts with K-RAS oncoprotein, disrupting its functioning (Ilinskaya et al., 2016). No correlation was found between the decrease in the level of RNA and the toxic effect of RNases. Thus, in the precursors of myeloid cells, binase reduced the level of RNA by 20% with 67% cell viability, and in the human kidney epithelium cells, RNA levels were halved while 85% of the cells remained viable (Ilinskaya et al., 2008). In the cells of acute myeloid leukemia Kasumi-1, which are extremely sensitive to binase, the total RNA level did not change even when the viability decreased by 95% (Mitkevich et al., 2011). Onconase caused apoptosis of mitogen-stimulated lymphocytes, without reducing the level of cellular RNA (Ardelt et al., 2003).

Another argument in favor of the fact that the degradation of available RNA is not the only cause of the cytotoxicity of RNases is that they selectively eliminate certain malignant cells. It is known that the malignancy of cells occurs as a result of a change in the processes of differentiation and control of division, which may be due to the expression of oncogenes. At the same time, several similar cancerous cell lines may respond differently to RNases (Altomare et al., 2010). It was found that cells expressing certain oncogenes, like *RAS, KIT, FLT3, AML1-ETO, E6,* and *E7*, acquire sensitivity to some bacterial RNases (Ilinskaya et al., 2002; Mitkevich et al., 2010; Mitkevich et al., 2011; Mitkevich et al., 2013; Mitkevich et al., 2017). Apparently, sensitivity to RNases occurs due to the activation and/or inhibition of certain signaling pathways and changes in the properties of onco-transformed cells. It is also obvious that the rate of cell division is not a decisive criterion for the toxic action of RNases. In accordance with our hypothesis, based on the study of the molecular mechanisms of the cytotoxic effect of exogenous RNases in recent years (Mitkevich et al., 2011; Mironova et al., 2013; Mitkevich et al., 2014; Mitkevich et al., 2015; Mitkevich et al., 2017), the reason for the selective toxicity of RNase to cancer cells should be sought in the properties of tumor cells that enable them to constantly adapt to the extracellular environment. Unlike normal cells that exist in conditions of permanent tissue homeostasis, actively proliferating malignant cells affect the composition and properties of their microenvironment and must continuously adapt to changing environmental conditions. The properties of malignant cells that ensure their survival, proliferation, and distribution include altered redox status, resistance to hypoxia, changes in the functioning of ion transport systems, increased intracellular pH, reduced adhesion, and production of specific exosomes. Despite an intensive study of the toxic properties of RNases from different organisms, information about their influence on the above properties of tumor cells is rather scarce. Here, we summarize the knowledge accumulated to date about influence of RNases on adaptive properties of malignant cells and consider the directions of future investigations for RNases application in the treatment of tumors.

#### RNASES AND INTRACELLULAR REDOX STATUS

Cancer cells possess a redox potential different from normal cells due to their accelerated metabolism. They demand high levels of reactive oxygen species (ROS) to maintain a high proliferation rate (Sosa et al., 2013). The thiol redox status, which depends on the ratio of reduced (GSH) and oxidized (GSSG) glutathione (Gonzalez-Dosal et al., 2011; Wang et al., 2013), is also altered in cancer cells. Under normal conditions, the GSH level in cells (1–5 mM) is 100-fold higher than GSSG. Under oxidative stress, this ratio can be reduced to 1 (Allen and Mieyal, 2012). However, many tumors show elevated levels of GSH emphasizing the link between the deregulations of GSH homeostasis and cancer (Tew and Townsend, 2011). Changes in the redox status lead to changes in the functioning of redox-sensitive proteins (kinases, transcription factors, ionic transporters, etc.) due to their redox modification (Gao and Schottker, 2017). In particular, the shift of intracellular redox conditions to the oxidized state induces protein glutathionylation that protects the thiol groups of proteins from irreversible oxidation and changes their function (Mieyal et al., 2008; Miller and Mieyal, 2015). In the case of viruses that provoke oncogenesis, a change in the redox status is necessary to "tweak" the cell to fit their needs. Thus, human papillomavirus (HPV) suppresses antioxidant systems in the cell for the functioning of redox-sensitive viral oncoproteins, E6 and E7, which bind to the tumor suppressors p53 and retinoblastoma protein, respectively, suppressing their activity, which stimulates uncontrolled proliferation and prevents apoptosis. The return of cells to normal redox status will reduce the activity of such oncoproteins. Redox active drugs, which are based on changing the cell's redox status, have already proven their effectiveness in the treatment of certain types of cancer including promyelocytic leukemia, esophageal, ovarian, non-small cell lung, colon, and breast cancer (Wondrak, 2009; Tew and Townsend, 2011). It has been established that RNases lead to a decrease in the level of ROS in tumor cells. So, onconase caused a decrease in ROS in cells of acute lymphoblastic leukemia, Jurkat, and several fibroblast cell lines (Ardelt et al., 2007). Binase caused a decrease in ROS level in Kasumi-1 cells, B16 mouse melanoma, and Jurkat cells (Mironova et al., 2013; Mitkevich et al., 2013; Burnysheva et al., 2016). RNases, by reducing ROS in tumor cells, alter the redox modifications of key proteins (such as NF-kB, p53, etc.), which, in turn, suppress the resistance of cancer cells to apoptosis. Thus, the "return" of the redox status of cancer cells by RNases to the values characteristic of normal cells lead to a decrease in their resistance to death and uncontrolled division by normalization of the redox-sensitive cellular systems.

It should be mentioned, that one of the key player in redox regulation of cells is cytosolic RNase inhibitor (RI) (Dickson et al., 2005). It binds to and inhibits extracellular RNases. It was postulated that only RNases that evade RI can kill cells. For example, RI regulates angiogenesis through very strong binding to angiogenin, which is a cytoprotective RNase. However, in the absence of RI, the angiogenin converts into a cytotoxic molecule (Thomas et al., 2018). Unlike mammalian RNases, RI does not inhibit the ribonuclease activity of cytotoxic bacterial RNases (Sevcik et al., 2002), but the absence of direct interaction between them has not been shown. RI is very sensitive to oxidative stress and is easily inactivated, while losing its ability to inhibit exogenous RNases (Dickson et al., 2005). RI inactivation can work as mechanism to switch on the RNase-mediated degradation of cellular RNA under stress conditions (Lomax et al., 2014). On the other hand, RI protects cells from oxidative stress, and its deprivation leads to decrease of GSH level as well as increase of oxidant-induced DNA damage (Monti et al., 2007). Decrease of RI activity in tumor cells induces their proliferation, increases migration and invasion, and reduces their adhesion (Chen et al., 2011). Upregulation of RI activity, on the contrary, causes the death of tumor cells (Tang et al., 2019). One can suggest that cytotoxic RNase influence on RI status by several ways. RNase reduction of ROS level, increased in tumor cells, protects RI from inactivation; exogenous RNase in cells can upregulate RI, which will reduce oxidative stress; interaction of RI with RNase may not lead to inactivation of the latter; however, it may change the properties of the RI, including protection against oxidation. These points should be clarified in future studies.

### RNASES AND HYPOXIA

Uncontrolled division of tumor cells changes their normal microenvironment. In particular, due to the lack of blood supply, cells at a certain stage of tumor development are in a state of hypoxia. Hypoxia serves as an additional risk factor that accelerates cell malignancy. Under hypoxia and/or altered redox status of cancer cells, the transcriptional factor hypoxia inducible factor 1 (HIF1) stabilizes in them, which makes the cells resistant to oxygen deficiency. Hypoxia is a characteristic marker for the development of tumors and has a significant impact on the course of chemotherapy, often negative (Carreau et al., 2011). It can also affect the expression level of oncogenes. In particular, during hypoxia, miR29b synthesis is suppressed, which normally suppresses the synthesis of KIT oncoprotein (Liu et al., 2010) and, consequently, the level of KIT in cells increases. Activation of KIT leads to an increase in the level of HIF1 (Zhang et al., 2011). It was shown that co-expression of AML1-ETO and HIF1 oncogenes in myelogenous leukemia cells leads to a higher rate of cell proliferation *in vitro* and to a more severe course of leukemia in mice (Gao et al., 2015). By itself, stabilization of HIF1 under conditions of chronic hypoxia is a risk factor for the spread of a tumor.

The work of endogenous RNases designed to suppress the development of neoplasia is impaired under hypoxic conditions. Nature has provided a mechanism for the destruction of cells with stabilized HIF1 by increasing the level of endogenous RNase T2 (Uccella et al., 2018). It inhibits angiogenesis and induces apoptosis of malignant cells. However, in aggressive cancers, the inhibition of cell growth by RNase T2 stops working. Reduced expression of DICER, the enzyme involved in microRNA processing, is frequently observed in cancer and is associated with poor clinical outcome in various malignancies. DICER expression is suppressed by hypoxia through an epigenetic mechanism that involves inhibition of oxygen-dependent H3K27me3 demethylases KDM6A/B and results in silencing of the DICER promoter. Subsequently, reduced miRNA processing leads to de-repression of the miR-200 target ZEB1, stimulates the epithelial to mesenchymal transition, and ultimately results in the acquisition of stem cell phenotypes in human mammary epithelial cells (Van den Beucken et al., 2014). It was found that miR-630, which is upregulated under hypoxic conditions, targets and downregulates DICER expression. In an orthotopic mouse model of ovarian cancer, delivery of miR-630 using 1,2-dioleoyl-sn-glycero-3-phosphocholine nanoliposomes resulted in increased tumor growth and metastasis and decreased DICER expression (Rupaimoole et al., 2016). Exogenous RNase, resistant to inactivation under conditions of altered redox status during hypoxia, can compensate for the loss of endogenous RNase function. Binase does not contain cysteine and methionine residues and, accordingly, is insensitive to changes of redox conditions. We tested the effect of binase on Kasumi-1 cells, and cervical cancer SiHa cells growing at different oxygen contents. A decrease in [O2] from 21 to 5 and 1% resulted in an almost two-fold increase in the proportion of apoptotic cells in the Kasumi-1 and SiHa cells treated by binase. The increased sensitivity of cancer cells to the effect of binase under decreased oxygen level is associated with a change in the expression of oncogenes and the activation of processes mediated by oncogenic proteins. This suggests that the response of malignant cells to RNases during tumor development may be enhanced by disrupting their adaptation to low oxygen conditions.

### RNASES AND PH

Disruption of oxygen supply leads to aerobic glycolysis in cancer cells (Warburg effect). The excess of protons produced during glycolysis, by the Na+/H+ exchanger is transferred to the extracellular environment. Change in pH is one of the markers of cancer cells (Webb et al., 2011). Malignant cells have a "reversed" pH gradient with a constantly elevated intracellular pH that is higher than the extracellular pH (Webb et al., 2011). The increase in intracellular pH leads to the induction of cell proliferation, increases their resistance to apoptosis, and metabolically adapts the cells to oxygen deficiency. It has been suggested that a decrease in the intracellular pH of malignant cells will lead to antitumor effects (Slingerland et al., 2013). We obtained preliminary data showing that the effect of binase (0.8 µM) on Kasumi-1 cells under 20%, 5%, 1%, and 0.2% [O2] leads to a decrease in the intracellular pH value by 0.2–0.5 units. Since the normal functioning of tumor cells is associated with high values of intracellular pH, the decrease of this parameter under the action of RNase should disrupt the intracellular regulation and reduce their adaptive potential.

### RNASES AND ADHESION

To adapt to the extracellular environment, cancer cells re-arrange their plasma membranes to sustain proliferation, avoid apoptosis, and resist anticancer drugs. This leads to changes in the cell deformability, which is important for invasiveness, membrane stiffness, and receptor function causing disruption of adhesion and intercellular signaling (Bernardes and Fialho, 2018). Membrane of normal cells is characterized by a lipid asymmetry between the inner and outer leaflets. A de-regulation of this asymmetry is often encountered in cancerous cells where phosphatidylserine is often exposed in the outer membrane resulting in a negative surface charge, leading to disruption of cell adhesion and promotion of metastasis and tumor invasion (Bernardes and Fialho, 2018). It has been shown that an increase in intracellular pH is necessary for the directed migration of malignant cells (Webb et al., 2011). During oncogenic transformation of cells, there is also a change in the structure and expression of the glycoproteins and glycolipids in the cell membrane, which form the so-called adhesive molecules (Hart et al., 1991). As a result, the membrane of tumor cells contains more acid glycoproteins and phospholipids than the membrane of normal cells. This adaptation marker is one of the reasons for the selective elimination of tumor cells by RNases. Indeed, the majority of known cytotoxic RNases are basic proteins, and cationization of RNases is considered to be an effective strategy for strengthening their antitumor properties (Mitkevich et al., 2014). The proliferation of tumor cells may be associated with disruption of endogenous RNases. It has been demonstrated that endogenous RNase L inhibits cell attachment and cell spreading (Dayal et al., 2017). Prostate cancer cells depleted of RNase L show greater migration in wound healing and trans-well migration assays in response to fibronectin and serum. Over-expression of RNase L suppressed cell migration compared to both endogenous levels and knockdown cells while activation of RNase L, which requires RNase L dimerization, which inhibited cell migration. This was attributed to the destabilization of the miR-regulated transcriptome by RNase L (Rath et al., 2015). However, it was later shown that the effect of RNase L on cell migration is mediated, in part, by protein–protein interactions and does not require enzymatic activity (Dayal et al., 2017). We have previously established that binase reduces metastasis in animals (Mironova et al., 2013). This allows suggesting that binase can reduce invasion and enhances the adhesion of tumor cells. One can expect that exogenous RNases are able to influence the adhesion and motility of tumor cells, preventing the spread of the tumor. It should be borne in mind that such effects on tumor cells may have upregulation of RI, so it is absolutely necessary to verify the role of cytotoxic RNases in regulating RI activity.

### RNASES AND ION TRANSPORT

Change in the functioning of ion transport systems plays a major role in the adaptability of malignant cells. The increase in intracellular pH in tumor cells is associated with changes in the activity of ionic transporters. Neoplasms induced by oncogenes or carcinogens have an abnormally high Na+ content, caused by membrane depolarization and the opening of potential-dependent Na+ channels (Lobikin et al., 2012). Membrane depolarization serves as a marker for carcinogenesis and suggests that the development of neoplastic changes can be delayed by increasing the expression of ion channels, which leads to hyperpolarization of membranes for example, K+ and Cl channels. Some pharmacological antitumor agents act precisely by changing the activity of ion channels (Lobikin et al., 2012; Peigneur et al., 2012; Schweikart et al., 2013). We have found that cytotoxic bacterial RNases, binase, and 5K RNase Sa (highly cationic mutant of RNase Sa from *Streptomyces aureofaciens*) can change the functional activity of calcium-activated potassium channels artificially introduced into embryonic kidney cells (Ilinskaya et al., 2004; Ilinskaya et al., 2008). Binase leads to an increase in the level of Ca2+ (Mitkevich et al., 2013), which may be due to inhibition of the Na,K-ATPase, leading to the activation of the Na+/Ca2+ exchanger. Na,K-ATPase is the main ion transport system in mammalian cells. It creates a transmembrane gradient of K+ and Na+ ions in cells. This pump plays an important role in the regulation of cell volume, transmembrane potential, intracellular pH, and Ca2+ level through Na+/H+ and Na+/Ca2+ exchangers (Slingerland et al., 2013). In tumor cells, the activity of Na,K-ATPase is increased, which may be due to the need to remove Na+, which accumulates as a result of intensive activity of the Na+/H+ exchanger. In addition, maintaining the level of intracellular K+ with Na,K-ATPase is also necessary to suppress apoptosis. Our preliminary experiments showed that, in Kasumi-1 cells, under the action of binase (0.8 µM, 24 h), the hydrolytic activity of Na,K-ATPase was reduced by 80% (**Figure 1A**). This is due to both a 25% decrease in the level of the enzyme in cells (**Figure 1B**) and an increase in its glutathionylation level by 50% (**Figure 1C**, **D** ). It was shown that glutathionylation led to inactivation of Na,K-ATPase and associated with a change in the cell redox status (Petrushanko et al., 2012). The decrease in the activity of Na,K-ATPase leads to a violation of ionic homeostasis and precedes the manifestation of the apoptogenic effects of binase. Obviously, this is one of the components of the mechanism of the cytotoxic effect of binase on onco-transformed cells. Thus, the effect of exogenous RNases on ion transport systems may be another reason for the selective cytotoxicity of RNases on cancer cells.

### RNASES AND VESICULAR TRANSPORT

Exosomal vesicles secreted by tumor cell play an active role in oncogenesis and metastasis (Kim et al., 2017). They communicate between tumor cells and their microenvironment. Exosomes transporting miRNAs and proteins can cause neoplastic transformation and aid tumor development (Zhang et al., 2015). miRNAs absorbed by the recipient cell from exosomes can act as a regulator of gene expression; there is growing evidence that they are involved in the onco-transformation of cells (Kim et al., 2017). RNases bound to negatively charged molecules on the surface of cells penetrate them by endocytosis. It has been established that RNase A and BS-RNase bind actin *in vitro* and induce its polymerization (Simm et al., 1987). Since the process of polymerization/depolymerization of actin governs its dynamic properties during endocytosis, it can be assumed that RNases can influence the endocytosis of tumor cells, changing their interaction with the microenvironment.

We did not find information on how RNases influence production and composition of the exosomes of malignant cells. However, it can be assumed that since RNases change the properties of tumor cells, affecting ionic homeostasis and redox status, the number and composition of the exosomes produced by the cells will also change. An indirect confirmation of our assumptions may be the data from (Mironova et al., 2013). The authors found that RNase A reduces the level of miRNA in blood stream of mice with Luis lung carcinoma. This decrease is not related to the level of miRNA in tumor cells. MiRNA degradation by RNase A did not observed *in vitro* conditions. Given that part of extracellular miRNAs in the blood is enclosed in exosomes, it can be assumed that the decrease of miRNA level is associated with a decrease in the production of exosomes by cancer cells after treatment by RNase A. Probably, RNases affect the number and properties of RNA molecules entering into exosomes. In this case, the cancer cell will lose "communication channels" with the microenvironment, which will inevitably reduce its adaptive potential and reduce the risk of tumor spread.

### IMMUNE-MODULATORY EFFECTS OF RNASES

One of the main factors in the survival of tumor cells in the organism is their ability to circumvent immune surveillance. Tumor cells may escape from immune control and proliferate in an unrestricted manner (Mittal et al., 2014). This escape can be mediated through various mechanisms, such as reduced immune recognition, increased resistance to attack by immune cells, or the development of an immunosuppressive tumor microenvironment (Mittal et al., 2014). Human

S-glutathionylated (GSS-α1/α1) form of the protein normalized to its total amount. The levels of S-glutathionylated Na,K-ATPase α1-subunit and total α1-subunit were estimated using immunoblotting as described in (Mitkevich et al., 2016). Mouse monoclonal anti-glutathione antibody MAB5310 (Millipore) and mouse monoclonal anti-Na,K-ATPase α1 antibody clone C464-6 (Millipore) were applied to detect glutathionylated proteins and total amount of α1-subunit correspondingly. Mouse monoclonal antibody AM4302 (Ambion) was used for β-actin detection. After staining with horseradish peroxidase–conjugated secondary antibodies, membrane was stained using a commercial kit SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific), and chemiluminescence was detected using Bio-Rad ChemiDoc MP instrument. Densitometric analysis was performed by Image Lab (Bio-Rad) program, and the results were represented as ratio of glutathionylated α1- subunit to total α1-subunit band intensity ([GSS- α1]/total α1) or as ratio of α1-subunit to β-actin band intensity (α1/β-actin). (D) The original immunoblotting readouts. Data are mean values for three independent experiments with triplicates ± standard diviation. The differences among the groups were analyzed by Student's t-test, and p < 0.05 was considered statistically significant. Statistica 7 software was used for analysis. \*p < 0.005.

extracellular RNases are mainly expressed in innate cells and display a variety of immune modulation activities (Lu et al., 2018). They can participate in host immune responses, working as alarmins and safeguard molecules against infection and inflammation. Secretion of RNase in the focus of inflammation сontributes to tissue repair and remodeling. RNases also function as cytokines and chemokines, displaying anti-inflammatory activities and inducing chemoattraction of innate cells (Lu et al., 2018). Activity of host RNases is often lowered in tumor tissue (Shlyakhovenko, 2009) and significantly reduced in serum of patients with different types of cancer (Huang et al., 2014). It turned out that exogenous RNases also have an immunomodulatory effect (Lu et al., 2018) and can be used to replenish the pool of extracellular RNases and modulate the immune response. For example, macrophage immune regulation by binase can trigger the host cell antitumor response (Makeeva et al., 2017). Binase also showed its effectiveness in restoring interferon sensitivity (IFN) of SiHa cervical cancer cells, in which the IFN response was initially suppressed by HPV (Mitkevich et al., 2017).

### DISRUPTION OF TUMOR CELL ADAPTABILITY BY RNASES AS A POTENTIAL CANCER THERAPEUTIC TARGET

From the presented data, it is evident that the influence of exogenous RNases on the adaptation of malignant cells to the external environment has not been properly explored. Such adaptation mechanisms give malignant cells competitive advantages, ensuring their survival and distribution in the body. Influencing these mechanisms with the help of exogenous RNases will allow the weakening of the adaptation potential of tumor cells, inducing their death and making them more susceptible to therapeutic agents (**Figure 2**). A comprehensive

FIGURE 2 | Disruption of the adaptation mechanisms of cancer cells by RNase. Upper panel: cancer cell adapts to the environment (decreased pO2 level) and changes it for itself, thus avoiding the immune system cells and drugs. The properties of malignant cell that ensure its survival, proliferation, and spread include altered redox status (increased reactive oxygen species [ROS]), resistance to hypoxia (hypoxia inducible factor [HIF1] increased and stabilized), changes in the functioning of ion transport systems (dysregulation of N+,K+ and H+,K+ pumps and exchangers), increased intracellular pH value, reduced adhesion, disruption of endogenous RNases (endoRNase), and production of specific exosomes (vesicles) containing specific cancer miRNA. Lower panel: exogenous RNases (exoRNase, green stars) act on a whole range of adaptive properties of tumor cells—decrease level of HIF1, ROS, and pH value; reduce activity of ion transporters; suppress exosomes production and degrade miRNA; and stabilize cell adhesion. These make tumor cells available for elimination by the immune system and reduce their drug resistance.

### REFERENCES


study of the adaptive properties of cells that are influenced by RNases will determine the mechanisms of their selective cytotoxicity to malignant cells and establish conditions affecting the effectiveness of their antitumor activity. This will make it possible to effectively use RNases in combination with other therapeutic agents to which the tumor cells are initially resistant. Drug resistance of tumor cells is largely related to their ability to adapt to the environment. Reducing the adaptive capacity of neoplastic cells with RNases can be an effective strategy for overcoming drug resistance. A similar approach was demonstrated by us using the example of SiHa cells, the oncotransformation of which is due to HPV. Viral oncoproteins suppress the IFN response in cells, as a result of which they become resistant to interferon therapy. Binase restores IFN signaling, enhances IFN sensitivity and apoptosis in SiHa cells (Mitkevich et al., 2017). The advantage of binase and, probably, other exogenous RNases is that under conditions of low oxygen content in the environment its antineoplastic properties are increased, which accentuates the therapeutic potential of RNase for chemo-resistant tumors. Since, RNases act on a whole range of adaptive properties of tumor cells and retain their activity even in a changing micro-environment (with altered redox status, pH, oxygen, etc.), they are promising molecules for the treatment of various types of tumors. The combination of RNases with other types of anticancer drugs can help solve the problem of drug resistance of tumor cells.

### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and the supplementary files.

### AUTHOR CONTRIBUTIONS

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

### FUNDING

The study was supported by Russian Foundation for Basic Research (Grant #17-00-00061).


based on the inhibition of KV channels. *Mol. Pharmacol.* 82, 90–96. doi: 10.1124/mol.112.078188


Shlyakhovenko, V. A. (2009). Ribonucleases in tumor growth. *Exp. Oncol.* 31, 127–133.


**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 Mitkevich, Petrushanko and Makarov. 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.*

# Characterization of the *Puumala orthohantavirus* Strains in the Northwestern Region of the Republic of Tatarstan in Relation to the Clinical Manifestations in Hemorrhagic Fever With Renal Syndrome Patients

*Yuriy N. Davidyuk1, Emmanuel Kabwe1, Venera G. Shakirova2, Ekaterina V. Martynova1, Ruzilya K. Ismagilova3, Ilsiyar M. Khaertynova2, Svetlana F. Khaiboullina1,4, Albert A. Rizvanov1 and Sergey P. Morzunov5\**

*1 OpenLab Gene and Cell Technologies, Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, 2 Department of Infectious Diseases, Kazan State Medical Academy, Kazan, Russia, 3 Research Laboratory "Omics technology", Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, 4 Department of Microbiology and Immunology, University of Nevada, Reno, NV, United States, 5 Department of Pathology, University of Nevada, Reno, NV, United States*

Over 1,000 cases of hemorrhagic fever with renal syndrome (HFRS) were recorded in the Republic of Tatarstan (RT) in 2015. HFRS is a zoonotic disease caused by several different Old World hantaviruses. In RT, *Puumala orthohantavirus* (PUUV) is a prevalent etiological agent of HFRS. We looked for the genetic link between the PUUV strains isolated from the bank voles and from the infected humans. In addition, possible correlation between the genetic makeup of the PUUV strain involved and different clinical picture of HFRS was investigated. Partial PUUV small (S) genome segment sequences were retrieved from 37 small animals captured in the northwestern region of RT in 2015. Phylogenetic analysis revealed that 34 PUUV sequences clustered with strains of the previously identified "Russia" (RUS) genetic lineage, while 3 remaining PUUV sequences clustered with the known lineage from Finland (FIN). Sequence comparisons showed that the majority of the S-segment sequences isolated in the current study displayed 98.2–100.0% sequence identity when compared with the strains isolated earlier from the HFRS patients hospitalized in Kazan city. HFRS patients infected with PUUV strains of either RUS or FIN genetic lineages were observed to have consistent differences in clinical presentation of the disease and laboratory findings. These findings indicated a strong genetic link between the infected bank voles and human HFRS cases from the same localities. Thus, S-segment sequences of the PUUV strains isolated from HFRS patients could serve as a molecular marker for determining the likely geographic area where infection occurred.

Keywords: HFRS, RNA, PUUV, *Puumala orthohantavirus*, Republic of Tatarstan

#### *Edited by:*

*Hector A. Cabrera-Fuentes, University of Giessen, Germany*

#### *Reviewed by:*

*Sudheer Kumar Ravuri, Steadman Philippon Research Institute, United States Jan Clement, KU Leuven, Belgium Alexei B. Shevelev, Russian Academy of Sciences, Russia*

#### *\*Correspondence:*

*Sergey P. Morzunov smorzunov@medicine.nevada.edu*

#### *Specialty section:*

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

*Received: 11 January 2019 Accepted: 29 July 2019 Published: 05 September 2019*

#### *Citation:*

*Davidyuk YN, Kabwe E, Shakirova VG, Martynova EV, Ismagilova RK, Khaertynova IM, Khaiboullina SF, Rizvanov AA and Morzunov SP (2019) Characterization of the Puumala orthohantavirus Strains in the Northwestern Region of the Republic of Tatarstan in Relation to the Clinical Manifestations in Hemorrhagic Fever With Renal Syndrome Patients. Front. Pharmacol. 10:970. doi: 10.3389/fphar.2019.00970*

## INTRODUCTION

The Republic of Tatarstan (RT) is among the regions of the Russian Federation (RF) with the highest annual load of the human cases of hemorrhagic fever with renal syndrome (HFRS). In 2015 alone, more than 1,000 cases of HFRS were diagnosed in the RT (Matrosov, 2016), thus placing this region into the fourth place with respect to the prevalence of this zoonosis infectious disease after Udmurtia, Bashkiria, and Mordovia in the Volga Federal District (VFD) of the Russian Federation (RF). *Puumala orthohantavirus* (PUUV) is one of the known causative agents of HFRS in humans, which causes a mild form of the disease with the fatality rate <0.43% (Lee, 1999; Khismatullina et al., 2016). PUUV belongs to the genus *Orthohantavirus*, family *Hantaviridae*. The main primary reservoir for PUUV is the bank vole, *Myodes glareolus* (Vapalahti et al., 2003; Tkachenko et al., 2012).

The PUUV tripartite genome consists of the S (small), M (medium), and L (large) single-stranded negative-polarity RNA segments: the S-segment codes for the nucleocapsid protein (NP), the most abundant protein produced in infected cells; the M-segment codes for the precursor of the envelope glycoproteins (Gn, Gc); and the L-segment codes for the RNA-dependent RNA polymerase (Plyusnin et al., 1996). Currently, eight genetic lineages of PUUV have been documented in various regions in Europe and western Siberia (Razzauti et al., 2013). Earlier, two distinct genetic lineages of PUUV were identified circulating within the bank vole populations in Russia. "Russia" (RUS) genetic lineage includes strains from the Samara region, Bashkiria, Udmurtia, and Tatarstan (Plyusnin et al., 1994; Lundkvist et al., 1997; Kariwa et al., 2009), and "Finland" (FIN) genetic lineage includes strains from Karelia and western Siberia (Asikainen et al., 2000; Dekonenko et al., 2003; Yashina et al., 2015). Between these two lineages, S-segment nucleotide sequences diversity reaches over 15% (Razzauti et al., 2012), while within a local rodent population, nucleotide sequence diversity does not exceed several percent within one lineage (Avsic-Zupanc et al., 2007; Kariwa et al., 2009; Razzauti et al., 2009). Although significant achievements have been made in the study of PUUV genetic diversity in Eurasia, only few investigations touched a question of the possible link between genetic composition of the PUUV strains in the bank voles and in the HFRS patients with different clinical manifestations of the disease (Horling et al., 1995; Plyusnin et al., 1997; Bahr et al., 2006; Garanina et al., 2009).

In our previous research (Davidyuk et al., 2017), we identified the PUUV strains in 25 patients diagnosed with HFRS in RT and 8 patients from the Republic of Mordovia. In the current study, genetic link between the PUUV strains isolated from the lungs of the bank voles and the RT PUUV strains identified earlier from infected humans was investigated. In addition, possible correlation was evaluated between genetic compositions of the PUUV strains infecting humans and differences in the clinical picture seen in the corresponding HFRS patients hospitalized in Kazan City.

### MATERIALS AND METHODS

*Patients.* Clinical manifestations of HFRS were studied in 74 patients who met the case definition and were receiving treatment in the Tatarstan Republican Clinical Infectious Diseases Hospital in Kazan City in 2015. According to the reports obtained from the patients, the infections occurred in the northwestern part of RT. Patients were hospitalized for 5–6 days of illness. The first sampling was carried out on the first day of admission to the clinic. Initial HFRS diagnosis based on the clinical and epidemiological criteria was confirmed by ELISA detection of antihantavirus IgM and IgG antibodies. The PUUV partial S-segment sequences were obtained from 25 patients in our previous work (Davidyuk et al., 2017). Three patients were excluded from the analysis because their clinical data were incomplete. The Institutional Review Board of the Kazan Federal University approved this study, and informed consent was obtained from each study subject according to the guidelines approved under the corresponding protocol.

*Tissue samples.* Frozen rodent lung tissue samples and information about trapping localities were obtained from Federal Healthcare Institution "Center for Hygiene and Epidemiology in the Republic of Tatarstan (Tatarstan)." Rodents were trapped in the northwestern regions of the Republic of Tatarstan in May and September 2015 (information about the geographic locations of the trapping sites is shown in **Figure 1** and **Table 1**). All PUUV sequences obtained from the serologically positive bank voles were named to include their corresponding virus name, trapping region, strain designation, and year (for example PUUV/ Vysokogorsky/MG\_054/2015).

#### RNA Extraction, cDNA Synthesis and PCR

Total RNA was extracted from lung tissues of bank voles with TRIzol Reagent kit (Invitrogen Life Technologies™, USA) following the manufacturer's recommendations. The concentration of RNA was determined with a NanoDrop 2000 UV–Vis spectrophotometer ("Thermo Fisher Scientific," USA). cDNA was synthesized using Thermo Scientific RevertAid Reverse Transcriptase ("Thermo Fisher Scientific," USA). PCR was performed using TaqPol polymerase kit ("Sileks," Russia) with primers PuuV-For 5-CTGCAAGCCAGGCAACAAACAGTGTCAGCA-3' and PuuV-Rev 5'-TCTGCCACATGATTTTTGTCAAGCACATC-3' (Kariwa et al., 2009). The resulting PCR products were purified with Isolate II PCR and Gel Kit ("Bioline," UK) and subsequently sequenced using ABI PRISM 310 big Dye Terminator 3.1 sequencing kit (ABI, USA). Obtained sequences were deposited in the GenBank database under accession nos. MG573266-573302.

### Phylogenetic Analysis

The nucleotide alignments and phylogenetic analysis of the PUUV strains based on the partial S-segment sequences 171 nucleotides in length (nucleotides 424–594) were performed with MegAlign program (Clustal W algorithm) located in the DNASTAR software package Lasergene (DNASTAR, USA) and MEGA v6.0 (Tamura et al., 2013). The parameters were adjusted manually.

PES, LAI-1, and LAI-2 (numbered 1–5, respectively).

TABLE 1 | Bank vole trapping site locations, number of trapped bank voles and nucleotide sequences obtained per location.


Phylogenetic trees were constructed using maximum parsimony method included in MEGA v6.0. For comparison, several partial S-segment sequences of the genetically distinct PUUV strains were downloaded from GenBank database (NCBI). These were included the following strains: Samara\_49/CG/2005, AB433843; Puu/Kazan, Z84204; PUUV Udmurtia/894Cg/91, Z21497; CG17/ Bashkiria-2001, AF442613; Sotkamo 2009, HE801633; PUUV/ Konnevesi/Mg\_O22B/2005, JQ319168; Kuchuk170/Mg/2007, KJ292966; and CRF366, AF367071. Tula virus strain Sennickerode Sen05/205, EU439951 was used as an outgroup.

#### RESULTS AND DISCUSSION

#### Identification and Sequence Comparisons of the PUUV Strains Found in the Bank Vole Populations in RT

Altogether, 129 bank voles were captured at 13 sites in May and September 2015. PUUV RNA was detected in 37 bank voles trapped at 11 sites, while viral RNA was not detected in 21 bank voles trapped at the sites S8 and S10 (**Table 1** and **Figure 1**). The proportion of infected bank voles in different sites vary. The high prevalence of infected bank voles was in sites S1–S4, S13, and S5–S7 located in two forests near the bank of the Volga river. In contrast, the lower proportion of infected animals was found in the isolated bank vole populations located in sites S9–S12 away from the Volga river. Plausible explanation is that bank voles are migration from sites S1–S4 to sites S5–S7. In addition, spatial distribution of hantavirus positive animals in different locations could be related to population densities and rates of virus transmission (Abbott et al., 1999; Mills et al., 1999). PUUV partial S sequences (564 base pairs, nt 242-805) were obtained from all reverse transcription PCR positive bank voles, analyzed and used for constructing phylogenetic trees. In addition, sequences obtained earlier from the strains circulating in the patients diagnosed with HFRS in 2015 (Davidyuk et al., 2017) and sequences of the selected PUUV strains from GenBank database were included in the analysis. The region of the S segment was selected based on the fact that it is shown to be one of the most variable region and often used for the analysis of genetic variability of hantaviruses (Asikainen et al., 2000; Sironen et al., 2001; Johansson et al., 2004; Klempa et al., 2006; Garanina et al., 2009; Ali et al., 2015). Sequence comparisons showed that PUUV sequences from each of the sites S1, S2, S4, S7, and S11– S13 displayed from 99.4 to 100.0% within-site nucleotide identity, although the sequences obtained from the site S3 were more variable (from 96.5 to 100% nucleotide identity). The sequences PUUV/Vysokogorsky/MG\_054/2015 and PUUV/Vysokogorsky/ MG\_058/2015 from the site S5 had 99.4% nucleotide identity while diverging significantly (83.6–84.8% nucleotide identity) from the rest of the sequences within the site S5 (see **Tables 1** and **2**). Nucleotide sequence identity between the sequences PUUV/Vysokogorsky/MG\_064/2015 and PUUV/Vysokogorsky/ MG\_066/2015 from the site S6 was as low as 83.0%.

Between-site nucleotide sequence comparisons revealed that 12 samples from the sites S1–S4 (sample pool from these four sites was designated as ZEL) shared 99.4–100.0% sequence identity. The only sequence that diverged slightly was PUUV/Zelenodolsky/ MG\_113/2015 (96.5–97.1% nucleotide identity with the rest of the ZEL pool). Likewise, the nucleotide sequences of 12 samples from the sites S5–S8 taken together (designed VYS-1 pool) displayed 98.8–100.0% nucleotide identity, and the nucleotide identity between the samples PUUV/Vysokogorsky/MG\_054/2015 and PUUV/Vysokogorsky/MG\_058/2015 from S5 and PUUV/ Vysokogorsky/MG\_064/2015 of S6 (designated VYS-2 pool) was 100.0%. The lowest nucleotide identity was observed between the sequences from VYS-1 and VYS-2 pools (83.0–84.2%).

The nucleotide sequences of the samples from the sites S11 and S12 (designated LAI-1 pool) were 100% identical. These samples also showed 95.9% nucleotide identity with the PUUV strains isolated from the site S13 (designated LAI-2 pool). The results of the comparison of the PUUV sequences from the different pools are shown in **Table 2**.

As expected, genetic variability of the PUUV S-segment nucleotide sequences obtained in the current study showed good correlation with geographic locations of the bank vole trapping sites. Sites S1–S4 are located in the western part of the sampled region, while sites S5–S8 are situated in the eastern part of the large forest (from Zelenodolsk to Vysokaya Gora), which borders the west and north of Kazan. There are no natural barriers to prevent migration of bank voles within the boundaries of this forest. Therefore, determining the boundaries of individual bank vole populations is difficult. All bank voles in this forest can be considered as one large population, which may include multiple local populations.

Likewise, sites S11 and S12 are located in the forest around Laishevo, and bank voles in this forest most likely form a distinct population; this conclusion is supported by the lower similarity observed between the PUUV sequences in the LAI-1 pool and in the ZEL/VYS pools. On the other hand, bank voles trapped in a relatively small isolated forest to the south of Pestretsy (S9) belong to a different population, and the PUUV strain identified in this forest differed significantly from the other groups.

The site S13 (LAI-2 pool) is located in the forest along the left bank of the Volga river west of Teteevo. The small mammals trapped there represent a group which is isolated from other populations included in the current study. As expected, corresponding PUUV sequences from LAI-2 pool are relatively distinct from ZEL/VYS pools and are most similar to those from LAI-1 pool.

Geographical locations of the trapping sites S9, S11, S12, and S13 and comparisons of the nucleotide identity values of the PUUV sequences from the corresponding PES, LAI-1, LAI-2 pools (**Figure 1** and **Table 2**) suggest that these three pools might be phylogenetically closer to each other than to ZEL and VYS pools.

Nucleotide sequence comparisons of the partial S-segment PUUV sequences with the corresponding sequences of the previously known PUUV strains found in GenBank showed the sequences from VYS-2 pool to be 100.0% identical to the corresponding nucleotide sequence of the PUUV strain Sotkamo and 93.0% identical to the sequence of the PUUV strain Konnevesi, both of the FIN lineage. The sequences of this pool displayed the lowest level of nucleotide identity (81.9–84.8%) with the strains of the RUS lineage previously known from the Volga region of Russia, such as Samara, Kazan, Udmurtia, and Bashkiria strains (**Table 3**). On the other hand, nucleotide sequences of ZEL, VIS-1, PES, LAI-1, and LAI-2 pools were more closely related to the RUS genetic lineage (91.2–97.7%


nucleotide identity) and more genetically distant from the sequences of the FIN genetic lineage (81.9–83.6% nucleotide identity) (**Table 3**). Hence, most of the PUUV strains identified in the current study belong to the RUS genetic lineage, while only three PUUV strains, PUUV/Vysokogorsky/MG\_054/2015, PUUV/Vysokogorsky/MG\_058/2015, and PUUV/Vysokogorsky/ MG\_064/2015, belong to the FIN genetic lineage. Thus, two PUUV genetic lineages were detected circulating in the bank vole populations in the forests north of Kazan between Yash-Ketch and Vysokaya Gora. The possibility of such co-circulation of two distinct PUUV genetic lineages in one territory was demonstrated earlier (Razzauti et al., 2009). Since the PUUV strains of the FIN lineage were found in the several widely distributed geographic locations including central and southern parts of Finland (Vapalahti et al., 1992; Plyusnina et al., 2012; Razzauti et al., 2013), Karelia (Asikainen et al., 2000), and western Siberia (Dekonenko et al., 2003), it could be suggested that these strains are distributed continuously from northern Europe (Finland) to at least western Siberia (Omsk region). In this case, forests in the Republic of Tatarstan, the Mari El Republic, and the Republic of Udmurtia could represent another zone of contact between the bank vole populations carrying PUUV strains of these two distinct genetic lineages. Consequently, the emergence of the reassortant and recombinant PUUV strains could happen in these areas. The verification of this assumption requires further investigation.

#### Phylogenetic Analysis of the Rodents and HFRS Patients' PUUV Strains

It is noteworthy that PUUV strains from PES, LAI-1, and LAI-2 pools showed slightly higher sequence identities with the corresponding sequences of Kazan and Udmurtia strains than the strains from ZEL and VYS-1 pools (see **Table 3**). Perhaps, this finding reflects the bank vole migration from the south to the north along the Volga River in the postglacial period (Dekonenko et al., 2003). Interestingly, only the PUUV sequences PUUV/ Vysokogorsky/MG\_054/2015, PUUV/Vysokogorsky/MG\_058/ 2015, and PUUV/Vysokogorsky/MG\_064/2015 demonstrated high identities (99.4–100.0%) to the strains, which were obtained earlier from HFRS patients infected with the strains belonging to FIN genetic lineage (Davidyuk et al., 2017). Nucleotide sequence identities between the PUUV strains of the RUS genetic lineage obtained from bank voles and the PUUV strains of the same lineage found in HFRS patients exceeded 90.0%. In particular, the comparisons revealed that most sequences of the PUUV strains from HFRS patients and the strains from bank vole pools (specifically, RT024, RT031, RT033, RT036, RT038, RT039, RT057 versus ZEL; RT048, RT058 versus LAI-1; RT055 versus LAI-2; RT002, RT005, RT006, RT008, RT010, RT011, RT013, RT014, RT043, RT050 versus VYS-2; and RT065 versus PES) share from 98.2 to 100.0% nucleotide identity. Based on these data, one could assume with high confidence that corresponding patients acquired hantavirus infection in the northwestern region of the Republic of Tatarstan and even determine the approximate location where infection occurred. In other words, the sequences of the PUUV strains isolated from the HFRS patients could be used as a molecular marker for determining the probable area of infection (**Figure 1**). Recently, similar data were published using HFRS cases diagnosed in some Europian countries, such as Germany, Finland, and Belgium (Plyusnin et al., 1997; Escutenaire et al., 2001; Ettinger et al., 2012). The data obtained may be used in the future for the development of the epidemiological measures aimed at preventing infection and reducing the incidence of HFRS in RT.

Surprisingly, one HFRS patient-originated PUUV strain RT012 did not show close genetic relationship to the strains identified in bank voles. This sequence displayed from 90.1 to 93.0% nucleotide identities with the sequences of the RUS genetic lineage isolated from bank voles and was 98.8% identical with "Bashkiria" strain. The patient might have contracted HFRS in the Volga region within the area where "Bashkiria" strain circulates including possibly unexplored regions of RT.

Phylogenetic tree of the PUUV strains obtained in the current study was reconstructed on the basis of the partial S-segment sequences (171 bp, nt 424-594) isolated from bank voles and HFRS patients (**Figure 2**). The PUUV nucleotide sequences obtained formed two distinct clades, which corresponded to the FIN and RUS genetic lineages. The RUS clade included the LAI-1, LAI-2, PES, and ZEL+VYS-1 subclades, containing isolates from RT and Bashkiria. Each subclade includes PUUV strains isolated from bank voles and HFRS patients. It should be noted that strains of the "ZEL+VYS-1" subclade are equidistant to all previously known RUS lineage.

The majority of the PUUV sequences obtained from bank voles (34 from 37) clustered with the RUS lineage, while the remaining three sequences from VYS-2 pool clustered with the FIN lineage. Interestingly, sequences obtained from the HFRS patients showed different pattern: 12 sequences clustered with the RUS lineage, while 10 sequences were placed in the FIN lineage. Relatively high proportion of the patients infected with

TABLE 3 | Nucleotide sequence identity of the FIN and RUS genetic lineages to the PUUV strains discovered in the northwest region of RT.


The compressed branches marked "hum + ZEL," "VYS-1," and "hum + VYS-2" includes the following: (i) strains RT024, RT031, RT033, RT036, RT038, RT039, and RT057 and nine strains from the bank voles trapped in sites S1–S4; (ii) 12 strains from VYS-1 pool; (iii) strains RT002, RT005, RT006, RT008, RT010, RT011,

the FIN lineage strains could possibly be explained by: i) such strains circulate in the bank vole populations in the northwestern territory of RT that have not been sampled yet; ii) the infections could have occured within the limited geographic area around Yash-Ketch and Vysokaya Gora. More genetic analysis of PUUV circulating in RT and neighboring regions is required to explain variations in the number of patients infected with FIN and RUS virus lineages.

RT013, and RT 014 and three strains from VYS-2 pool, and strain Sotkamo 2009, respectively.

### Different Clinical Manifestations in the Patients With HFRS Caused by the PUUV Strains of the FIN and the RUS Lineages

In order to compare clinical manifestations observed in the HFRS patients infected with the PUUV strains of the RUS versus FIN genetic lineages, clinical and laboratory data from 22 patients (17 male and 5 female) were analyzed. The patients were divided in two groups: group 1 included 12 patients from whom the PUUV strains of the RUS lineage were recovered (group "RUS"), and group 2 consisted of 10 patients infected with the PUUV strains of the FIN lineage (group "FIN"). The average age of the patients was 35.9 ± 12.06 years in "RUS" group and 49.8 ± 14.6 years in "FIN" group, respectively (*p* > 0.05). The average hospitalization period and the febrile and oliguric periods did not differ significantly in both groups (**Table 4**). Urinalysis revealed proteinuria in 10 and 6 cases (*р* > 0.05) for "RUS" and "FIN" groups, respectively. Hematuria was observed in six "RUS" group and four "FIN" group cases (*р* > 0.05). Bleeding in the form of scleral hemorrhage was revealed in only one patient from "RUS" group. The kidney edema was observed in nine patients of the "RUS" group and in five patients TABLE 4 | Clinical and laboratory characteristics of HFRS patients, infected with PUUV strains of "RUS" versus "FIN" genetic lineages.


*aUltrasound diagnosis.*

*bOliguria— < 500 ml of urine for 1–3 days.*

*cProteinuria— > 30 mg/dl of protein in urine.*

*dHematuria— > 60 red blood cells per high-power microscopic field.*

*eThe bolded p values are statistically significant to clinical manifestations in different groups.*

of the "FIN" group (*р* > 0.05). Increased serum levels of urea were detected in 5 out of 12 in "RUS" and 3 out of 10 cases in "FIN" groups (*р* > 0.05).

Furthermore, five of the patients from "RUS" group and four from "FIN" group experienced acute lung injury (*р* > 0.05). Gastrointestinal (GA) symptoms including stomach pain, nausea, vomiting, and diarrhea were documented in seven "RUS" group and three in "FIN" group (*р* > 0.05), while oliguria was found in nine "RUS" group and three "FIN" group of HFRS cases (*р* > 0.05). Impaired vision was observed only in the "RUS" PUUV lineage infected patients (4, *p* < 0.05), while pain in the lumbar region was experienced by 11 of the "RUS" patients and by 5 of the "FIN" patients (*p* < 0.05) (**Table 4**). Among the laboratory parameters investigated, patients of the "RUS" group showed an increased level of urea and creatinine as compared to the "FIN" group: 12.92 ± 3.20 mmol/l vs. 6.25 ± 0.82 mmol/l (*р* > 0.05) and 208.08 ± 56.14 μmol/l vs. 104.90 ± 3.52 μmol/l (*p* > 0.05) (**Table 4**). The "RUS" PUUV lineage-infected patients had a significantly lower platelet count (72.16 ± 12.16 × 109 /l) and higher aspartate aminotransferase (AST) level (50.83 ± 7.77 U/l) as compared to the "FIN" group, where thrombocyte counts were 131.0 ± 14.71 × 109 /l (*p* < 0.05) and AST level was 31.6 ± 3.76 U/l (*p* < 0.05).

Overall, the clinical manifestations seen in the HFRS patients of both groups mentioned above matched typical HFRS clinical manifestation caused by PUUV and were in agreement with the description given in the works of many authors (Trusov et al., 2004; Shakirova et al., 2011; Rasmuson et al., 2013). HFRS was characterized by acute onset with fever, malaise, headache, pain in the eyes, and abdominal pain. Acute kidney injury diagnosed in 40–100% of patients depending on the disease severity. Hemorrhages are detected in some HFRS cases, including petechia, gastrointestinal, and nasal bleeding. Acute lung injury is evident in some patients and often characterized by mild to moderate dyspnea or dry cough (Rasmuson et al., 2013). In severe cases, lung edema was found (Trusov et al., 2004; Rasmuson et al., 2013). However, the significant differences in the course of the disease and biochemical findings observed in the current study suggest that the PUUV strains of the RUS lineage cause a disease with more severe clinical symptoms, including more pronounced hemorrhages and renal manifestations, unlike the PUUV strains of the FIN lineage, which cause a milder form of HFRS.

We believe that the results of this investigation are of particular importance not only for the local health authorities in Tatarstan but also for the international scientific and public health community. As PUUV could be found throughout most of Europe and Asia, PUUV strains currently circulating in RT are integral part of the Old World HFRS epidemics. Our current data confirm that PUUV strains circulating in RT (called Finnish and Russian genetic lineages) are genetically related to the virus strains found in the Western and Central Europe. Thus, current data improve our understanding of PUUV distribution and genetic diversity. Ongoing climate change could influence the bank vole migration patterns and therefore affect future distribution and genetic composition of PUUV strains throughout its entire geographic distribution. PUUV strains could "migrate" together with their natural hosts into the new areas that were not endemic in the past. In addition, increased international trade could inadvertently promote introduction of PUUV rodent hosts into the new areas. Thus, comprehensive understanding of PUUV epidemiology in RT could assist in predicting the virus spread beyond its current geographic range and investigating future HFRS outbreaks.

### ETHICS STATEMENT

The Ethics Committee of the Kazan Federal University approved this study and informed written consent was obtained from each NE patient and controls according to the Guidelines approved under this Protocol (article 20, Federal Law "Protection of Health Right of Citizens of Russian Federation" N323- FZ, 11.21.2011).

## AUTHOR CONTRIBUTIONS

YD and SK performed experiments; EM, EK, and VS made the conceptualization and data curation; AR and IK analyzed data; SM, and RI analyzed data and wrote the manuscript.

### ACKNOWLEDGMENTS

The work was performed as a part of the Russian Government Program of Competitive Growth for the Kazan Federal University. Some of the experiments were conducted using the equipment of the Interdisciplinary Center of Shared Facilities and the Scientific and Educational Center of Pharmaceutics of Kazan (Volga Region) Federal University, Kazan, Russian Federation. Albert A. Rizvanov was supported by state assignments 20.5175.2017/6.7 and 17.9783.2017/8.9 of the Ministry of Science and Higher Education of Russian Federation.

### REFERENCES


and Research (Hantaviruses) Seoul, Korea: Asan Institute for Life Sciences. pp. 17–38.


S and M RNA segments: evidence for strain variation in hantaviruses and expression of the nucleocapsid protein. *J. Gen. Virol.* 73 (Pt 4), 829–838. doi: 10.1099/0022-1317-73-4-829


**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 Davidyuk, Kabwe, Shakirova, Martynova, Ismagilova, Khaertynova, Khaiboullina, Rizvanov and Morzunov. 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.*

# Surveillance of Tumour Development: The Relationship Between Tumour-Associated RNAs and Ribonucleases

*Nadezhda Mironova\* and Valentin Vlassov*

*Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of Russian Academy of Sciences, Novosibirsk, Russia*

Tumour progression is accompanied by rapid cell proliferation, loss of differentiation, the reprogramming of energy metabolism, loss of adhesion, escape of immune surveillance, induction of angiogenesis, and metastasis. Both coding and regulatory RNAs expressed by tumour cells and circulating in the blood are involved in all stages of tumour progression. Among the important tumour-associated RNAs are intracellular coding RNAs that determine the routes of metabolic pathways, cell cycle control, angiogenesis, adhesion, apoptosis and pathways responsible for transformation, and intracellular and extracellular non-coding RNAs involved in regulation of the expression of their proto-oncogenic and oncosuppressing mRNAs. Considering the diversity/variability of biological functions of RNAs, it becomes evident that extracellular RNAs represent important regulators of cell-to-cell communication and intracellular cascades that maintain cell proliferation and differentiation. In connection with the elucidation of such an important role for RNA, a surge in interest in RNA-degrading enzymes has increased. Natural ribonucleases (RNases) participate in various cellular processes including miRNA biogenesis, RNA decay and degradation that has determined their principal role in the sustention of RNA homeostasis in cells. Findings were obtained on the contribution of some endogenous ribonucleases in the maintenance of normal cell RNA homeostasis, which thus prevents cell transformation. These findings directed attention to exogenous ribonucleases as tools to compensate for the malfunction of endogenous ones. Recently a number of proteins with ribonuclease activity were discovered whose intracellular function remains unknown. Thus, the comprehensive investigation of physiological roles of RNases is still required. In this review we focused on the control mechanisms of cell transformation by endogenous ribonucleases, and the possibility of replacing malfunctioning enzymes with exogenous ones.

Keywords: tumour-associated RNA, extracellular miRNAs, tumour development, ribonucleases, RNA degradation

## TUMOUR-ASSOCIATED RNAS AND THEIR ROLE IN CARCINOGENESIS

Tumour development is accompanied by rapid cell proliferation, loss of differentiation, the reprogramming of energy metabolism, loss of adhesion between tumour cells and matrix, evasion of immune surveillance, angiogenesis induction, infiltration growth, and metastatic spreading (Hanahan and Weinberg, 2011). Tumour-associated RNAs play an important role at all stages

#### *Edited by:*

*Hector A. Cabrera-Fuentes, University of Giessen, Germany*

#### *Reviewed by:*

*Karla Mayolo-Deloisa, Monterrey Institute of Technology and Higher Education (ITESM), Mexico Ester Boix, Autonomous University of Barcelona, Spain*

> *\*Correspondence: Nadezhda Mironova mironova@niboch.nsc.ru*

#### *Specialty section:*

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

*Received: 19 April 2019 Accepted: 09 August 2019 Published: 13 September 2019*

#### *Citation:*

*Mironova N and Vlassov V (2019) Surveillance of Tumour Development: The Relationship Between Tumour-Associated RNAs and Ribonucleases. Front. Pharmacol. 10:1019. doi: 10.3389/fphar.2019.01019*

**144**

of tumour progression: intracellular coding RNAs determine the route of metabolic pathways (Zheng, 2012), cell cycle control, angiogenesis, adhesion, apoptosis (Suizu et al., 2000; Wadehra et al., 2005; Rocnik et al., 2006; Segura et al., 2007; Liu et al., 2013; Yan et al., 2016) and pathways responsible for transformation, such as PI3K/AKT (phosphatidylinositol-3 kinase/protein kinase B), TGF-β (tumor growth factor beta), JAK/STAT (Janus kinase/Signal transducer and activator of transcription) and MAPK (mitogen-activated protein kinase) (Ikushima and Miyazono, 2010; Ciuffreda et al., 2014; Thomas et al., 2015; Doi et al., 2017; Ouyang et al., 2017). Intracellular and extracellular non-coding RNAs participate in the regulation of the expression of target proto-oncogenic and oncosuppressive mRNAs as well (Dalmay and Edwards, 2006; Esquela-Kerscher and Slack, 2006).

Different groups unequivocally demonstrated the presence of various RNAs, including tumour-derived and tumour-associated RNAs, in the blood samples of patients with oncological diseases (Wieczorek et al., 1985; Kopreski et al., 1999; Vlassov et al., 2010). The range of RNAs discovered in blood plasma samples is rather wide and includes fragments of ribosomal RNA (rRNA), messenger RNA (mRNA), transport RNA (tRNA), mitochondrial RNA, small non-coding RNA particularly miRNA being detected in plasma or serum in RNA/protein complexes and within extracellular membrane vesicles (EV) (Valadi et al., 2007; Savelyeva et al., 2016; Savelyeva et al., 2017).

Data are accumulated that EV of cancer origin comprise pathogenic components, including mRNA and miRNA, that together with other components such as DNA proteins, transcriptional factors and lipids can take part in paracrine signalling in the tumour microenvironment (Fujita et al., 2016). These evidences also predict the role of EV-mediated transfer of cancer-associated biomolecules to distant organs contributing to the initiation of pre‐metastatic niche formation. In support of this, an increasing number of publications has demonstrated the contribution of EV RNA in events of cancer development, namely, cell proliferation (Hong et al., 2009; Kogure et al., 2011; Zhang et al., 2015a), drug resistance (Challagundla et al., 2015), angiogenesis (Kosaka et al., 2013), immune modulation (Fabbri et al., 2012), and pre‐metastatic niche formation (Fong et al., 2015).

### Extracellular Circulating mRNAS

Among the extracellular miRNA (ex-miRNA), mRNA fragments encoding tumour-associated antigens were detected, i.e. proteins that are expressed at low level in the cell, but overexpressed upon tumour progression. In the blood samples of patients with breast and thyroid cancer, malignant melanoma, and hepatocellular carcinoma, elevated levels of mRNA fragments encoding the telomerase components HTR (telomerase RNA component) and HTERT (telomerase reverse transcriptase) were detected (Chen et al., 2000; Novakovic et al., 2004). Overexpression of telomerase in cells can lead to epithelial-mesenchymal degeneration and tumour progression.

High levels of mRNA encoding mammaglobin, CK-19 (keratin 19) (Silva et al., 2001) and HER2/neu (Erb-B2 receptor tyrosine kinase 2) (Nicolini et al., 2017), which are specific markers of breast cancer (Lianidou et al., 2015), are also found in the blood plasma of patients with given disease. CK-19 is involved in maintaining the stability of epithelial cells and high expression levels are detected in tumour tissue (Lianidou et al., 2015). The HER2/neu oncogene is amplified at a high level in approximately 20% of all breast cancer cases and is related to rapid tumour proliferation and a poor prognosis.

In the case of lung cancer, the elevated levels of mRNA fragments encoding HER2/neu, hnRNP-B1 (heterogeneous nuclear ribonucleoproteins B1) and 5Т4 (oncofetal trophoblast glycoprotein) were detected in blood (Kopreski et al., 2001; Fleischhacker et al., 2001; Sueoka et al., 2005). hnRNP-B1 was found to play a significant role in the splicing of tumour suppressors and is considered an oncogene. It has also found to be crucial in the development of glioblastoma, hepatocellular carcinoma and lung cancer (Zech et al., 2006; Golan-Gerstl et al., 2011; Shilo et al., 2014). Oncogene 5T4 is involved in the modulation of cell adhesion and is over-expressed in many types of cancer cells, as along with such oncogenes as MYC (protooncogene c-Myc) and NRAS (NRAS proto-oncogene, GTPase).

For patients with malignant melanoma, the presence in blood of mRNA encoding for tyrosinase, MAGE-3 (melanomaassociated antigen 3), MCAM (melanoma cell adhesion molecule), p97 (transitional endoplasmic reticulum ATPase) and HMBS (hydroxymethylbilane synthase) is typical (Hoon et al., 2000; Hasselmann et al., 2001). Tyrosinase is involved in the synthesis of melanin, MAGE-3 is involved in malignant transformation and is the main tumour-specific melanoma antigen, and MCAM is involved in cell adhesion.

## Circulating Non-Coding RNAs

The pool of non-coding RNAs discovered in the bloodstream consists of long non-coding RNAs (lncRNA), short noncoding RNA including microRNAs (miRNA), and the recently discovered piwi-interacting RNAs. Long non-coding RNAs and piwi-interacting RNAs are described in detail in a recent review (Rapisuwon et al., 2016). Here we will focus on ex-miRNAs, which are important in the regulation of tumour development.

In 2008, the presence of extracellular miRNAs (ex-miRNA) in human blood was demonstrated by several research groups for the first time (Chim et al., 2008; Chen et al., 2008; Lawrie et al., 2008; Mitchell et al., 2008). In blood plasma, the main component of the ex-miRNAs are preserved enclosed in the ribonucleoprotein complexes comprising Ago2 protein, microvesicles and apoptotic bodies, and high and low-density lipoproteins (Valadi et al., 2007; Hunter et al., 2008; Zernecke et al., 2009; Vickers et al., 2011; Turchinovich et al., 2011; Arroyo et al., 2011; Bayraktar et al., 2017; Pucci et al., 2018). In this regard, miRNAs are highly stable and resistant to blood ribonucleases and can be transferred throughout the body, which makes their regulatory function virtually unlimited.

It is worth mentioning that miRNAs regulate fundamental cellular processes, such as proliferation, differentiation, metabolism, DNA repair, apoptosis, and transformation and can function as mediators in cell-to-cell communication, thereby acting like hormones (Ambros, 2004; Bartel, 2004; Croce and Calin, 2005; Calin and Croce, 2006; Cortez et al., 2011; Anfossi et al., 2018).

Circulating ex-miRNAs include both oncomirs and oncosuppressors. Attempts have been made to use elevated levels of particular miRNAs in the blood of patients with various tumours as biomarkers to diagnose the disease (Chen et al., 2008; Cortez et al., 2011; Cui et al., 2013; Lin et al., 2017; Pendlebury et al., 2017; Wang et al., 2017; Elghoroury et al., 2018). High levels of miR-21-5p were found in the blood of patients with colorectal cancer, gastric cancer, pancreatic ductal adenocarcinoma, and metastatic breast cancer (Wang et al., 2013; Muller et al., 2014; Khan et al., 2016; Emami et al., 2019). miR-155 is indicative of chronic lymphocytic leukaemia, breast cancer, and rectal cancer (Ferrajoli et al., 2013; Gao et al., 2017; Orosz et al., 2018). miR-125b-5p is found in blood of patients with metastatic breast cancer, non-small cell lung carcinoma and diffuse large B-cell lymphoma (Yuan et al., 2016; Cui et al., 2013).The miRNAs miR-125a-5p, miR-145, and mir-146a can be indicative of non-small cell lung cancer (Wang et al., 2015). Other miRNAs can also be indicative of other forms of cancer: miR-125a-3p – colon cancer (Wang et al., 2017), miR-200c and miR-141 – metastatic breast cancer (Zhang et al., 2017a); and the miR-200 family – prostate cancer and high-grade serous epithelial ovarian cancer (Lin et al., 2017; Pendlebury et al., 2017).

miR-21 represents one of the first microRNAs being defined as an oncomir, that regulate multiple tumour suppressors like the PTEN (phosphatase and tensin homolog), PDCD (programmed cell death), p53 (tumor suppressor p53) and TP63 (tumor protein p63) pathways (Meng et al., 2007; Asangani et al., 2008; Papagiannakopoulos et al., 2008). miR-155 act as an oncogene by inhibition of suppressor of cytokine signaling 1 (SOCS1) expression (Jiang et al., 2012). Taking into account the wellknown functional link between inflammation and cancer, and the fact that inflammation to some extent is mediated by miR-155, the oncogenic role of miR-155 becomes clear (Jiang et al., 2010). However, although there is a lot of evidence for the oncogenic role of miRNA-155, it can also act as a tumour suppressor (Higgs and Slack, 2013).

Besides oncomirs, increased levels of which are identified in the blood of patients with various oncological diseases, tumour suppressor miRNAs can also be found in the blood stream. In this regard two or more miRNAs are used as diagnostic (by the level of oncomirs) and prognostic (by the level of tumour suppressor miRNAs) markers (Yang et al., 2015; Elghoroury et al., 2018).

The levels of tumour suppressor miRNAs are usually decreased in the blood of cancer patients, such as miRNAs belonging to let-7 family. Specifically, let-7 miRNAs directly interact with mRNAs encoding proteins involved in the cell cycle and signal transduction pathways that lead to carcinogenesis (Büssing et al., 2008). The decreased expression of let-7b is usually observed in lymph node metastases of breast cancer cells (Thammaiah and Jayaram, 2016). In addition, down-regulation of let-7b/g is evidenced during gastric cancer development being associated with poor survival and lymph node metastasis (Kang et al., 2014). A decreased level of miR-152 has been also detected in various human cancer cell lines and tumour tissues, such as gastrointestinal (Chen et al., 2010), endometrial (Tsuruta et al., 2011) and ovarian cancer (Zhou et al., 2012), as well as hepatocellular carcinoma (Huang et al., 2010), indicating that miR-152 might act as a tumour suppressor in these tumours.

### Role of Tumour-Associated Extracellular RNA (ExRNA) in Transformation

The secretion of RNAs in ribonucleoprotein complexes (RNP) by cells, and the transfer of those RNP between mammalian cells, was for the first time established in the 1970s (Kolodny et al., 1972). RNAs in RNPs can be a specific product released from tumour cells that may mediate host-tumour interaction and regulation of gene expression (Rosenberg-Nicolson and Nicolson, 1994). The recent discovery of regulatory RNAs, particularly miRNAs, has led to a revolutionary hypothesis that ex-miRNAs can mediate cell-to-cell signalling by paracrine or even endocrine manner, especially playing a crucial role in the context of cancer and metabolism. This hypothesis arose because several research groups found a large amount of miRNA in the bloodstream and was largely supported by numerous subsequent publications demonstrating that miRNAs in RNPs or EV can enter the recipient cells, change gene expression, and cause functional effects (Valadi et al., 2007; Skog et al., 2008; Mittelbrunn et al., 2011; Montecalvo et al., 2012).

Cell-to-cell communication by ex-miRNA has also been proven for cells of the immune system. Exosomal-transfered pro-inflammatory miR-155 and immunosuppressive miR-146a from dendritic cells was demonstrated to reduce the level of their mRNA targets, and reprogrammed the response of cellsrecipients to endotoxin (Alexander et al., 2015). The mechanism of regulation of the activity and differentiation of mast cells was shown to be mediated by the transfer of mRNA and miRNA in exosomes between cells (Valadi et al., 2007). Recent studies have demonstrated that adipose-derived EV-circulating miRNAs participate in cell–cell crosstalk between adipose and liver tissues by altering mRNA expression and translation in target tissue (see review Thomou et al., 2017).

Communication between tumour and normal cells provides a route for tumours to manipulate their environment, making it more favourable for growth and invasion. Glioblastoma cells have been shown to secrete exosomes containing mRNA, miRNA, including oncogenic miR-21, and angiogenic proteins being uptaken by normal cells, including microvascular endothelial cells of the brain (Skog et al., 2008). One example is the destruction of tight contacts in the epithelial cells of blood vessels under the action of exosome-derived miR-105 secreted by neighbouring cancer cells, and the subsequent increase in metastasis (Zhou et al., 2014). EV-derived miRNA-21 secreted by hepatocellular carcinoma cells participates in tumour progression triggering the conversion of hepatocyte stellate cells to cancerassociated fibroblasts (Zhou et al., 2018). MiR-210 obtained from hepatocellular carcinoma cells as well, was found to promote endothelial cell migration along with angiogenesis, which was confirmed by the correlation between the elevated level of miR-210 in the blood of patients with hepatocellular carcinoma and high microvessel density (Lin et al., 2018). Thus, tumour-derived miRNAs are the tools for the transformation of the normal cells to malignant, and adjustment of their microenvironment for favorable tumor development.

### THE ROLE OF ENDOGENOUS MAMMALIAN RNASES IN THE CONTROL OF INTRACELLULAR EVENTS AND SIGNALLING PATHWAYS RESPONSIBLE FOR TRANSFORMATION

Benner hypothesized that a certain balance between RNAs, RNases, and ribonuclease inhibitors controls tissue development in higher organisms (Benner, 1988). Many factors participated in regulation of gene expression at the mRNA level. These factors are non-coding RNAs, RNA-binding proteins and RNases that maintain RNA-homeostasis in cells and discard aberrant RNAs through the degradation and turnover of transcripts, control of RNA decay, and biogenesis of miRNAs. Disruption of mentioned processes is mainly associated with distortion of expression or the improper functioning of these factors followed by cell transformation and tumour development. Among these factors, RNases play quite important role since they regulate the turnover of various transcripts at every stage of the cell cycle and participate in the processing of RNA involved in translation control.

In a number of studies up to 2009, it was shown that intracellular RNases are involved in both induction and suppression of tumour progression (see review Kim and Lee, 2009). Nowadays a lot of information has appeared expanding the supervisory function of exogenous ribonucleases in the RNA world (**Tables 1** and **3**) and their intracellular RNA-targets (**Figure 1**).

#### RNases of Conventional RNA Decay

Deadenylation is an essential way of regulation of mRNA stability and expression of genes responsible for the fundamental functions such as development and differentiation at cell level under normal or pathological conditions, including chronic inflammation and cancer (Zhang et al., 2010; Zhang et al., 2015b). The CCR4-NOT complex, a major deadenylase in mammals, plays dual roles in the control of tumour development. The mammalian CCR4-NOT complex was described to comprise eight subunits: CNOT1, CNOT2, CNOT3, CNOT6 or 6L, CNOT7 or 8, CNOT9, CNOT10, and CNOT11 with four of them, namely, CNOT6/6L/7/8 and CNOT3 exhibited deadenylase properties (Bartlam and Yamamoto, 2010; **Figure 1B**).

Knockdown of CNOT3, a subunit incorporated in the CCR4- NOT complex and responsible for deadenylase activity, was shown to induce tumour development in a sensitized drosophila eye cancer model (Vicente et al., 2018). Moreover, mutations of the CNOT3 gene were discovered in samples of T-cell acute lymphoblastic leukemia (T-ALL) patients (De Keersmaecker et al., 2013), suggesting the tumour suppressor role of CNOT3. However, contrary to the tumour suppressive function components of the CCR4-NOT complex, it can have an impact on cancer progression. For instance, the up-regulation of CNOT3 promotes the progression of non-small cell lung cancer (Shirai et al., 2019) and activity of CNOT7 may stimulate migration of mouse breast cancer cells (Faraji et al., 2016; **Table 3**).

It is known that deadenylases facilitate miRNA-induced mRNA decay resulted from their interaction with the miRNA-induced silencing complex (miRISC). A vast amount of publications give an evidence of participation of deadenylation complexes, such as CCR4-NOT and Pan2–Pan3, in miRNA-mediated deadenylation being necessary for regulation of gene expression and stability of mRNA (see review Zhang et al., 2010). Poly(A)-specific ribonuclease (PARN) is an important deadenylase: among its targets are migration and adhesion factors, as well as mRNAs of proteins involved in p53, FAK (fokal adhesion kinase), and MAPK signaling (Lee et al., 2012; Devany et al., 2013; **Figure 1A**). PARN may also participate in miRNA-mediated deadenylation due to association with Ago2 in the RNA-induced silencing complex (RISC), and promote degradation of the oncogenic miR-21 followed by restoration of tumor suppressor activity of corresponding protein targets such as PTEN and p53 (Boele et al., 2014; Zhang et al., 2015a). PARN inhibition was shown to induce p53 accumulation and decrease cancer cell viability (Shukla et al., 2019).

5′–3′ exonuclease XRN1 is an enzyme being involved in conventional RNA decay (Long and McNally, 2003), which is also implicated in cancer as a tumour suppressor (**Table 3**). The decreased expression and/or complete depletion of XRN1 mRNA were found in primary samples of osteogenic sarcoma (Zhang et al., 2002). XRN1 realized additional control over epithelial to mesenchymal transition (EMT) on the level of ex-miRNA decay. It was found that XRN1 degrades ex-miRNA-223 derived from extracellular vesicles of polymorphonuclear leukocyte neutrophils after penetration into tumour cells, thus promoting transient epithelial-mesenchymal transition (Zangari et al., 2017; **Figure 1C**). Recently obtained data show that XRN1 negatively regulates autophagy in mammalian cells that thus reduces cell survival, which reinforces the evidence that this is a suppressor (Delorme-Axford et al., 2018). Contrary to this, 5′–3′ exonuclease XRN2 promotes EMT and metastasis through regulation of the processing of pre-miR-10a to mature miR-10a, and is a candidate inducer of spontaneous lung cancer (Zhang et al., 2017b; **Figure 1C**).

### Stress Signal Induced RNases

A number of endogenous RNases (RNase L, IRE1α, and PMR1) are normally silent in the cell and are induced under specific stress signals to effect tumour-modulating functions. Human RNase L displaying endoribonuclease activity expressed in many types of normal and cancerous mammalian cells (Zhou et al., 2005). RNase L is single-stranded ribonuclease able to cleave viral RNA, rRNA, and some cellular RNA both in cells and cell-free systems, at phosphodiester bonds in UU and UA sequences (Wreschner et al., 1981; Li et al., 1998; **Figure 1D**). In human and mouse cells, RNase L controls the stability of mRNA encoded in mitochondria and destabilizes the mRNA of genes induced by the interferon response to a viral infection (Li et al., 1998; Le Roy et al., 2001). RNase L is normally involved in innate immunity and antiviral defence

(Malathi et al., 2007), however besides these functions it also plays a role as a tumour suppressor. Mutations in the RNase L gene were found to contribute to enhanced cell migration and invasion, and knockdown of RNase L in human prostate cancer cell line PC3 resulted in increase of tumour growth rate and metastases spreading *in vivo* (Banerjee et al., 2015; Dayal et al., 2017; **Table 3**). Cleavage of mRNAs encoding proteins involved in cell adhesion and migration appears a more likely mechanism for the inhibition of cell migration by RNase L (Banerjee et al., 2015). Interestingly, RNase L can discriminate and eliminate exogenous miRNA mimics (Nogimori et al., 2019; **Figure 1D**).

IRE1α is a serine/threonine kinase, an endoribonuclease, which is one of the major participants in endoplastic reticulum (ER) proteostasis and plays a dual role in cancer development (**Table 3**). It carries out both tumour-inducing and tumoursuppressing activity. Activation of IRE1α was observed in several types of tumors and was associated with overexpression of such oncogenes as BRAFV600E (mutant form V600E of B-Raf proto-oncogene, serine/threonine kinase gene), MYC, and HRAS (HRAS proto-oncogene, GTPase) (Croft et al., 2014). In turn, activation of IRE1α and its functioning as ribonuclease may lead to the process named RNA regulated IRE1‐dependent decay (RIDD) that represent degradation of mRNA and miRNA targets (Maurel et al., 2014). In mammalian cells, the substrates for IRE1α are its own mRNA, mRNA encoding XBP1 and CD59, and other mRNAs encoding proteins involved in the regulation of angiogenesis (see review Kim and Lee, 2009; **Figure 1E**).

Several studies demonstrated that inhibition of the expression or the RNase activity of IRE1 suppresses the development of several types of tumours, mostly because of the ablation of prosurvival effects of XBP1 on tumour growth (Chevet et al., 2016; Obacz et al., 2017). Recently inhibition of IRE1 ribonuclease activity was found to influence the tumour cell secretome and enhance its sensitivity to chemotherapy (Logue et al., 2018). The tumour suppressive function of IRE1 was also detected. In several studies on genome screening, it was found that IRE1α is often found in the mutant form in various types of malignancies (Parsons et al., 2008; Guichard et al., 2012). Overexpression of IRE1 leads to a decrease in the expression of CD59, being implicated in the progression of lung cancer (Oikawa et al., 2007). Thus, IRE1 is an important RNase that exhibits a dual role in cancer progression by directing cancer progression and cell death.

PMR1 exhibits the properties of a proto-oncogene and is an effector of the EFGR (epidermal growth factor receptor) signalling pathway. Recently obtained data shows that increased migration activity and invasiveness of MCF-7 breast cancer cells is associated with high PMR1 activity, the targets of which are miRNAs of the miR-200 family, which are responsible for controlling adhesion and invasion (Bracken et al., 2014; Gu et al., 2016; Perdigão-Henriques et al., 2016; **Figure 1F**).

#### Proteins Regulating mRNA Stability

RAS-GTPase-activating protein (SH3 domain)-binding proteins (G3BPs) represent a family of proteins capable of RNA binding and able to regulate mRNA stability and translation in response to environmental stresses (**Table 1**). The mammalian G3BP family consists of homologous proteins G3BP1, G3BP2a, and its splice variant G3BP2b with a similar molecular structure, which are located in the nucleus and cytoplasm. The different functions of G3BPs are summarized in a range of reviews (see revs Kim and Lee, 2009; Alam and Kennedy, 2019). From the point of view of its influence on the RNA world, it is important to note that GB3P1 participates in RNA metabolism including regulation of various cellular mRNAs and miRNAs. G3BP1 controls certain transcripts either due to its ability to stabilize mRNA like mRNA *tau* and *CDK7* (cyclin dependent kinase 7) (Atlas et al., 2004) or to cause mRNA degradation as in the case of mRNA MYC, BART (Epstein-Barr virus derived RNA encoding a set of miRNAs), *CTNNB1* (catenin beta 1), *PMP22* (peripheral myelin protein 22), *GAS5* (growth arrest specific 5), and *IGF2* (insulin like growth factor 2) (Gallouzi et al., 1998; Tourrière et al., 2001; Zekri et al., 2005; Taniuchi et al., 2011a; Taniuchi et al., 2011b; Winslow et al., 2013; **Table 3**). In earlier papers G3BPs were suggested to play a role in tumour development since their elevated levels were found in different types of proliferating cells and tumours (Guitard et al., 2001; Barnes et al., 2002; French et al., 2002). Moreover, G3BPs were found to be participants of key cell-growth associated molecular pathways important for tumorigenesis including RAS, the NF-κB, and MAPK pathways, and the ubiquitin proteasome system (Gallouzi et al., 1998; Prigent et al., 2000; Soncini et al., 2001; **Table 3**).

#### Proteins With Ribonuclease Activity Participating in Maintenance of DNA Integrity

In addition, it was found that enzymes involved in DNA replication and repair, such as APE1, and FEN1, also exhibit RNase activity. Under stress, or when a nuclear localization

TABLE 1 | Endogenous RNases and other proteins with ribonuclease activity participating in maintenance of normal RNA homeostasis of eukaryotic cells.


*CNOT3, CCR4-NOT transcription complex subunit 3; CNOT7, CCR4-NOT transcription complex subunit 7; PARN, poly(A)-specific ribonuclease; XRN1, 5'-3' exoribonuclease 1; XRN2, 5'-3' exoribonuclease 2; RNase L, ribonuclease L; IRE1a, serine/threonine-protein kinase/endoribonuclease IRE1; PMR1, ATPase secretory pathway Ca2+ transporting 1; ANG, angiogenin, ribonuclease, RNase A family, 5; G3BP1, GTPase activating protein (SH3 domain) binding protein 1; APE1, apurinic/apyrimidinic endodeoxyribonuclease 1; FEN1, flap structure-specific endonuclease 1; Drosha, double-stranded RNA-specific endoribonuclease, nuclear; Dicer, double-stranded RNA-specific endoribonuclease, cytoplasmic; Ago2, Argonaute RISC catalytic component 2.*

signal is lost, these enzymes redistribute between the nucleus and the cytoplasm, where they can affect the level of cellular RNA. Two enzymes, APE1 and FEN1, have recently attracted close attention because of their ability to cleave RNA and the fact that their expression is associated with oncogenesis.

Apurinic/apyrimidinic endodeoxynuclease 1 (APE1) is an enzyme that exhibits both deoxyribonuclease and ribonuclease activity (**Table 1**). APE1 is mainly associated with DNA repair and redox regulation of transcription factors. In base excision repair (BER), APE1 functions as an apurinic/apyrimidinic endodeoxyribonuclease and corrects DNA damage caused by oxidizing or alkylating agents. APE1 was also found to exhibit endoribonuclease activity targeting *MYC* mRNA (Barnes et al., 2009; **Figure 1H**), and cleaving several other RNAs at UA, UG, and CA sites in the single stranded regions *in vitro* (Bergstrom et al., 2006). APE1 participates in rRNA quality control processes during cell division (Vascotto et al., 2009). Thus, APE1 performs several functions in the cell and can encourage genetic integrity and modulate turnover of different mRNAs as a ribonuclease. Recently, it has been suggested that this protein can perform non-canonical, but, nevertheless, important functions in RNA metabolism, regulating post-transcriptional expression of genes (Tell et al., 2010; Antoniali et al., 2014; Antoniali et al., 2017a).

Increased expression of APE1 was detected in a number of tumours: osteosarcoma (Wang et al., 2004), multiple myeloma (Yang et al., 2007), hepatocellular carcinoma (Di Maso et al., 2007), gastric cancer (Qing et al., 2015) (**Table 3**). APE1 is a normally a nuclear protein, but when cells acquire a cancerous phenotype it is redistributed between the nucleus and the cytoplasm (Jackson et al., 2005). There is evidence that the level of endoribonuclease activity of APE1 in the cytoplasm correlates with the aggressiveness of tumour. Of great interest are accumulating evidences demonstrating that APE1 may be involved in the control of gene expression due to its unsuspected activities during RNA metabolism (Antoniali et al., 2014; Jobert and Nilsen, 2014; Vohhodina et al., 2016) including miRNA expression (Antoniali et al., 2017b), thus enhancing APE1's critical functions in tumour progression.

Human flap endonuclease 1 (FEN1), localized in the nucleus, exhibits endoribonuclease activity and is able to cleave *in vitro* both synthetic and natural RNA in double-stranded regions (Stevens, 1998). FEN1 functions include flap endonuclease activity resulting in the removal of RNA primers during DNA replication, 5'-3'-exonuclease activity, and gap-endonuclease and RNase H-like activities (Shen et al., 2005; **Figure 1H**). Similar to APE1, the FEN1 protein, in addition to its function of removing RNA primers during DNA replication, can also be involved in regulation of RNA level in a cell. In the development of tumours, FEN1 plays the role of an oncogene (**Table 3**). Overexpression of this protein is found in numerous aggressive fast-growing malignancies (Sato et al., 2003; Lam et al., 2006). There is a suggestion that the rate of RNA primer removal during DNA replication by FEN1 directly affects cell proliferation. So, in mouse models, it has been shown that FEN1 deficiency significantly contributes to the frequency and multiplicity of the occurrence of tumours (Kucherlapati et al., 2007).

### RNases Involved in miRNA Biogenesis

A wealth of data is accumulating that indicates a correlation between aberrant miRNA expression and tumorigenesis. Three RNases: Drosha, Dicer, and Ago are involved in miRNA biogenesis (Murchison and Hannon, 2004; **Figure 1I**), and, accordingly, disorders in their expression can influence cancer development. The increased levels of Drosha and Dicer, their intracellular redistribution, and malfunction, is observed in many types of cancer cells. Increased Drosha and Dicer levels also correlate with elevated levels of oncogenic miRNAs.

The biogenesis of miRNAs starts by RNA polymerase II (Pol II)-mediated transcription of the miRNA gene encoded in the genome. This process generates long primary (pri-miRNA) transcripts comprising a stem-loop hairpin structure (Kim et al., 2009). Drosha is an essential part of the microprocessor complex (with its cofactor DGCR8) that continues miRNA biogenesis *via* cleavage of pri‐miRNAs with the formation of precursor miRNA (pre-miRNAs) (Kim et al., 2009; **Figure 1I**). Mutations in the Drosha/DGCR8 microprocessor complex subunit miRNA microprocessor complex are associated with highrisk of development of blastemal type Wilms tumours (Wegert et al., 2015). Reduced expression level of Drosha was found in melanoma (Jafarnejad et al., 2013), ovarian cancer (Papachristou et al., 2012), neuroblastoma (Lin et al., 2010), endometrial cancer (Torres et al., 2011), nasopharyngeal carcinoma (Guo et al., 2012), and gallbladder adenocarcinoma (Shu et al., 2012; **Table 3**). Recurrent homozygous deletions of Drosha were found in pineoblastoma (Snuderl et al., 2018). Single nucleotide polymorphisms (SNPs) in the sequence of Drosha gene were also found to correlate with high risk of cancer development (Wen et al., 2018). However, elevated levels of Drosha were found for a number of neoplasias: basal cell carcinoma, squamous cell carcinoma, and smooth muscle neoplasms (Sand et al., 2010).

The second processing step in miRNA biogenesis is realized by the cleavage of pre-miRNA with the RNase III Dicer endonuclease and RISC-loading complex subunit TRBP, which generates an approximately 22-nt miRNA duplex (Kim et al., 2009; **Figure 1I**). Dicer, an important RNase III endonuclease involved in miRNA processing, is down-regulated in many tumours, such as neuroblastoma (Lin et al., 2010), endometrial cancer (Torres et al., 2011), nasopharyngeal carcinoma (Guo et al., 2012), gallbladder adenocarcinoma (Shu et al., 2012), transitional cell carcinoma of the urinary bladder (Wu et al., 2012), breast cancer (Khoshnaw et al., 2012), lung cancer (Karube et al., 2005), gastric cancer (Zheng et al., 2007), colorectal cancer (Sun et al., 2017), and ovarian cancer (Pampalakis et al., 2010; **Table 3**). Low expression of Dicer is linked to poor prognosis and recurrence of cervical cancer (He et al., 2014). In addition, some correlations were found between the single nucleotide polymorphisms of Dicer and development and prognosis of several froms of epithelial cancers and endometrial cancer (Guo et al., 2012; Oz et al., 2018). Deletion of Dicer1 in a mouse model enhanced tumorigenesis (Kumar et al., 2009). The phosphorylation status of Dicer correlates with endometrioid tumour invasion (Aryal et al., 2019). Downregulation of Dicer expression was observed in human cancers and has been identified in promoting cancer metastasis and tumorigenesis due to repression of global miRNA maturation (Kumar et al., 2007; Martello et al., 2010).

Analysis of data from The Cancer Genome Atlas evaluated a significant influence of alterations in miRNA machinery genes on the development of multiple forms of malignancies. In particular, incidence of Ago2 alterations is the highest among the other miRNA-machinery genes and its contribution varies from 12.3% in case of colon and rectum adenocarcinoma to 20.7–23.30% in case of breast invasive carcinoma, bladder urothelial carcinoma, and prostate adenocarcinoma (Huang et al., 2014). The explanation may be due to the decreased competition between different miRNA species in regulation of gene expression and facilitation of operation of oncogenic miRNAs following the overexpression of Ago2 (Vickers et al., 2007).

### EXOGENOUS RIBONUCLEASES WITH ANTITUMOR ACTIVITY AND MECHANISMS OF THEIR ACTION

The antitumor potential of exogenous RNases has been studied for more than 60 years, due to their main function—the degradation of nucleic acids. Up to date, the most well-studied RNases with established antitumor activity are: BS-RNase from bull testes (Pouckova et al., 2004), amphibian RNase onconase from oocytes of Rana pipiens (Lee et al., 2000), bovine pancreatic RNase A (Patutina et al., 2010; Patutina et al., 2011), modified variants of RNase 1 from humans (Rutkoski et al., 2013) that belong to the RNase A superfamily, microbial RNase barnase from *Bacillus amyloliquefaciens* (Prior et al., 1996) and binase from *Bacillus pumilis* (Mironova et al., 2013a; Makeeva et al., 2017) relative to RNase T1 superfamily (**Table 2**).

A lot of published data confirm that exogenous RNases target different RNAs in a tumour cell. As already discussed, the degradation of RNAs managed by endogenous RNases plays a significant role in controlling gene expression, maturation, and turnover of RNA, which may be associated with malignant cell transformation and tumour progression. It can be assumed that exogenous RNases may restore the expression and/or activity of endogenous RNases disturbed in the tumour cell and modulate functions of tumour-associated RNAs. The mechanism of the cytotoxic action of exogenous RNases, presumed and partially confirmed in various studies, consists of series of stages. Firstly, the exogenous RNase binds with a tumour cell, is then internalised, gains access to the cytosol, and finally degrades intracellular RNA. The binding mechanisms of RNases with

TABLE 2 | Exogenous RNases of different origin displaying antitumor activity.


tumour cells, and their following penetration, are described in detail in a number of reviews (Makarov and Ilinskaya, 2003; Chao and Raines, 2011; Mit'kevich et al., 2014b); here we will focus directly on RNA degradation by exogenous RNases. It should be noted that in addition to intracellular RNAs, RNases, when released into the bloodstream, can also cause the degradation of circulating exRNA (Simsekyilmaz et al., 2014; Zernecke and Preissner, 2016; Lu et al., 2018).

It is obvious that the central molecular targets of RNases are various RNAs: rRNA, mRNA, RNA in the RNP complexes, tRNA, and non-coding long and small RNA. *In vitro* RNases destroy rRNA and tRNA in equal amounts, however, a certain type of RNA is more preferable *in vivo* for each RNase: BS RNase destroys rRNA (Mastronicola et al., 1995; Liao et al., 1996), while onconase preferentially degrades tRNA (Iordanov et al., 2000; Saxena et al., 2002; **Table 4**). Other intracellular targets of onconase are also rRNA, mRNA, and miRNA (Goparaju et al., 2011; **Table 4**). Thus, the toxic effect of RNases on tumour cells is associated with their main function - the ability to cleave RNA. However some data indicate that the ribonuclease activity of RNases is not the only component that provides an impact on their antitumor activity, but it is realized through the destabilization of double-stranded RNA (Sorrentino et al., 2003) or its irreversible binding (Blaszczyk et al., 2004).

After penetration into the cell, RNases degrade cellular RNA, as a result of which protein synthesis is blocked and apoptosis is initiated. It was shown for binase that treatment of the cells by the enzyme results in a significant decrease in the total amount of RNA in the cells, which, nevertheless, does not correlate with the level of cytotoxic effect of binase (Mitkevich et al., 2010a). It has been suggested that changes in intracellular tumourassociated RNA levels observed after treatment with exogenous RNases may be the result of both direct degradation of mRNA and miRNAs that suppress the expression of certain genes, and/ or generation of new siRNA-like molecules that can participate in the regulation of intracellular processes by the mechanism of RNA interference (Zhao et al., 2008; Saxena et al., 2009). Thus, the catalytic activity of exogenous RNases is considered a key factor in determining the regulation of intracellular processes involving RNA.

### RIBONUCLEASES OF RNASE A SUPERFAMILY

#### Bovine pancreatic RNase A and Pancreatic RNase 1 of Human

Bovine pancreatic ribonuclease A (RNase A) represents a small protein consisted of 124 amino acids with molecular weight equal to 13.7 kDa that, nevertheless, has the highest catalytic activity among the proteins of its superfamily. RNases belonging to RNase A superfamily catalyse the cleavage of RNA at phosphodiester bonds after pyrimidine residues in single-stranded regions (Raines, 1998). RNase A is the first ribonuclease whose antitumor activity was studied *in vitro* (Ledoux and Revell, 1955; Ledoux, 1956) and *in vivo* (Ledoux, 1955a; Ledoux, 1955b; Aleksandrowicz, 1958a; Aleksandrowicz, 1958b; Telford et al., 1959).

However, the obtained results were contradictory. This enzyme, at the doses of 40–1,000 mg/kg, caused retardation in the growth of solid and ascitic tumours in mice and rats (Ledoux, 1955a; Ledoux, 1955b). In the other studies, it was shown that RNase A does not exhibit cytotoxic and antitumor effects even using high doses of the enzyme injected into solid tumours (De Lamirande, 1961; Raines, 1998; Leland and Raines, 2001). Attempts were made to investigate the pancreatic RNase 1 of humans that belongs to the RNase A superfamily, and displays high catalytic activity, as an antitumor drug but the enzyme showed a very weak cytotoxic effect in cell cultures. The absence of cytotoxic activity of RNase A and RNase 1 was explained by their inactivation with intracellular ribonuclease inhibitor (RI), which form an extremely strong complex with these RNases (Kd < 10-15 М) (Johnson et al., 2007).

In a number of studies, an increase in the cytotoxicity of RNase A and RNase 1 was achieved by conjugating these RNases with peptides, proteins and antibodies, which increased the efficiency of their capture by tumour cells (Rybak et al., 1991; Newton et al., 1992; Psarras et al., 1998; Futami et al., 1999). To obtain RI-resistant RNase A and RNase 1 variants, methods of protein engineering, chemical modification, or protein conversion with the formation of covalent dimers were used (Di Donato et al., 1994; Rutkoski et al., 2005; Rutkoski et al., 2011). By means of site-directed mutagenesis D'Alessio and colleagues developed artificial dimers of RNase A and RNase 1 that showed cytotoxicity towards cancer cells (Di Donato et al., 1994; Piccoli et al., 1999). This allowed the production of derivatives of RNase A and RNase 1 with high antitumor activity. High cytotoxic activity against cancer cells was achieved by the conjugation of transferrin with mutated variant of human pancreatic RNase hRNase (Gly89→Cys) and mutant eosinophil-derived neurotoxin (Suzuki et al., 1999). Recently obtained data revealed that RI-resistant variants of pancreatic RNase 1 of human displayed strong toxic effect toward lung cancer and melanoma cells and worked sinergically with protein kinases in the ERK pathway (Hoang et al., 2018). A number of immunoRNases on the base of variants of pancreatic human RNase were developed which being fused with antibodies against ErbB2 exhibit strong toxic effects to ErbB2-positive tumor cells (D'Avino et al., 2014).

Despite encouraging results, interest in the therapeutic potential of RNase A disappeared for a long time, but arose again several decades later, when RNase was able to exert cytotoxic effects on tumour cells at much lower doses than was used in the 1950s. In 2002, conjugation of RNase A with poly [N- (2-hydroxypropyl) methacrylamide] led to constructions that effectively suppressed the growth of melanoma in nude mice (Soucek et al., 2002). In 2004, the first information appeared on the cytotoxic effect of RNase variants, which were inactivated by RI (Naddeo et al., 2005). A cytotoxic variant of human pancreatic RNase PE5 containing a nuclear localization signal was developed, which, despite its sensitivity to RI, demonstrated high cytotoxicity on a panel of various tumour cells (Bosch et al., 2004). Further, additional modifications of the PE5 structure led to the appearance of variants of RNase 1 with high cytotoxicity (Vert et al., 2012).

From 2010 to the present, several clinical studies of pancreatic RNase A have been conducted to treat various types of tumours. The first study (Phase I, ClinicalTrials.gov Identifier: NCT01201018), conducted from September 2010 to June 2012, used RNase A in peroral form (O'Shadi R) for the treatment of patients with various cancers. Although the official report on this study has not yet been presented, four more clinical trials of RNase A have since started: for the treatment of metastatic non-small cell lung cancer (ClinicalTrials.gov Identifier: NCT02134990), mesothelioma (ClinicalTrials.gov Identifier: NCT01627795), basal cell carcinoma (ClinicalTrials.gov Identifier: NCT02007317), acute myeloid leukemia and lymphoid leukemia (ClinicalTrials.gov Identifier: NCT02462265). Variant of human pancreatic RNase 1 with 95% sequence identity named QBI-139 is studied in a phase I clinical trial for the treatment of advanced refractory solid tumors (ClinicalTrials.gov Identifier: NCT00818831).

Colleagues in our laboratory used several mouse tumour models to demonstrate that RNase A administered in very low doses exhibited antitumor and antimetastatic activity (Patutina et al., 2010; Patutina et al., 2011). Attempts to find molecular targets of RNase A in the tumour and blood of tumour-bearing mice (with the example of Lewis lung carcinoma) revealed that antitumor and antimetastatic action of RNase A is realized *via* degradation of extracellular circulating miRNAs and is accompanied by significant boost of miRNA synthesis in tumour tissue (Mironova et al., 2013b). The microRNA boost in the tumour was associated with the overexpression of genes involved in microRNA biogenesis such as *Drosha*, *Xpo5*, *Dicer*, and *Ago2*. Ribonuclease activity of RNase A was demonstrated to play crucial role both in antitumour/ antimetastatic activity and the influence on the expression of microRNA and the microRNA processing genes.

Moreover, it was found that RNase A affected the whole transcriptome of murine Lewis lung carcinoma (Mironova et al., 2017) providing the downregulation of 644 transcripts and upregulation of 322 transcripts. Major part of the genes are involved in signalling pathways that maintain energy metabolism, promote cell growth and transformation, modulate the cancer microenvironment and extracellular matrix components as well as stimulate cellular proliferation and differentiation. As a result of RNase A treatment, we also detected an upregulation in carbohydrate metabolism, the stimulation of inositol phosphate cascade and oxidative phosphorylation as well as re-arrangement of apoptosis, transcription, cell cycle control and adhesion processes. Taken together, these data suggest that reorganization of the intracellular network of tumour cells caused by RNase A led to enhancement of energy cascade activity, shift in cancerrelated cell growth and dissemination processes, and partial depletion of signalling pathways that have tumour-promoting activity (Mironova et al., 2017).

### Angiogenin

Angiogenin (ANG) was described as the first proto-oncogenes among ribonucleases, and for this reason, researchers have never tried to use it as an antitumor drug. Nevertheless, since ANG is one of the brightest representatives of the RNase A superfamily, although it belongs to endogenous RNase, we discuss it in this section. ANG participated in neovascularization events, and an increased level of its expression was noted in many types of cancer cells (see rev. Kim and Lee, 2009).

Originally, human ANG was isolated as an angiogenesis factor of tumour origin; however, the expression of ANG by cells of various tissues suggests that its functions are not limited to neovascularization. Similar to RNaseA, ANG cleaves RNA in single-stranded regions, displaying Pyr-X cleavage specificity (Russo et al., 1996), although the ribonuclease activity of ANG is 10−5−10−6-times lower than RNase A activity (Lee and Vallee, 1989). In spite of the very weak ribonuclease activity of ANG, this activity is important for its biological functions, allowing ANG to drive an orchestra of various RNAs in a cell (Yamasaki et al., 2009; Ivanov et al., 2011; Li et al., 2019).

However, ANG exerts its functions not only due to its own ribonuclease activity, but also due to ability to bind with certain promoter regions of DNA and histone proteins, thus acting as a chromatin remodelling activator. ANG induces the synthesis of rRNAs by binding with promoter region of ribosomal DNA (rDNA), thereby promoting transcription of the precursor of 47S rRNA (Sheng et al., 2014). ANG in the nucleus may be involved in regulation of mRNA transcription. Using a chromatin immunoprecipitation-chip assay, Sheng and colleagues identified 699 genes that may be regulated by ANG on the mRNA level, many of which are related to tumorigenesis, such as proteins of Wnt and TGF-β pathways (Sheng and Xu, 2016).

A number of data highlighted the contribution of cellular ANG to the metabolism of RNAs, both in the nucleus and cytoplasm. In experiments *in vitro* it was shown that the ability of nuclear ANG to cleave 28S and 18S rRNAs, which, together with data on its participation in cleavage of the first cleavage site (A0) of the 47S prerRNA, provides evidence that ANG may enhance rRNA processing (Shapiro et al., 1986; Monti et al., 2009; **Table 3**). In addition, ANG carry out an important function in the tRNA metabolism that takes place in the cytoplasm. Under conditions of oxidative stress, hypoxia, and starvation, ANG performs the cleavage of the conserved singlestranded 3′-CCA termini of tRNA or its anticodon loop inducing the formation of so named tiRNA (tRNA-derived, stress induced small RNA) (Yamasaki et al., 2009; **Figure 1G**). ANG-produced tiRNA plays a significant role in proliferation of breast and prostate cancer cells (Honda et al., 2015; **Table 3**). It was revealed that ANG enhances colorectal cancer growth and metastasis both in *in vitro* and *in vivo* systems, producing a higher level of a 5'-tiRNA from mature tRNA-Val (Li et al., 2019). The resulting tiRNAs reprogram the translation of proteins, promoting damage repair and cell survival (Ivanov et al., 2011). Among two tiRNAs generated by ANG, 5′-tiRNA (but not 3′-tiRNA) inhibits translation *in vitro*, however, not all 5′-tiRNA are active. Preliminary data from Sheng and colleagues shows that ANG can also participate in degradation of miRNAs (Sheng and Xu, 2016).

### Onconase/Ranpirnase (From Oocytes of Frog *Rana Pipiens*)

In the late 1980s, the Alfacell corporation conducted a study on an extract of oocytes or early embryos of the Northern Leopard

Frog (*Rana pipiens*), which has profound cytostatic and cytotoxic activity towards tumour cells (**Table 2**). The active component of the extract was a protein of small size (11.82 kDa), being found in unfertilized oocytes as well. The amino acid sequence of this protein, originally named P-30, and later onconase or ranpirnase, resembled the sequence of enzymes of the RNase A superfamily (Ardelt et al., 1991). Onconase is the smallest protein of the RNase A superfamily having only 104 amino acids. Onconase displays significant ribonuclease activity that is 102 -105 times lower in comparison with the activity of RNase A (Ardelt et al., 1991; Ardelt et al., 1994; Boix et al., 1996). Onconase exhibits the antiproliferative and cytotoxic activity through interference with cell-cycle regulation and induction of programmed cell death by a mechanism described in details previously (Lee and Raines, 2008; Porta et al., 2008).

Onconase is an extremely stable protein that is not inactivated by RI (Boix et al., 1996). In the beginning, the cytostatic and cytotoxic facilities of onconase were investigated on the cell lines of human HL-60 leukemia, carcinoma A-253, and Colo 320 CM colon adenocarcinoma (Darzynkiewicz et al., 1988). Onconase caused a retardation of cell proliferation by increasing the duration of the G1 phase of the cell cycle, accompanied by a reduction in the DNA replication frequency. Cytotoxicity of onconase was shown in tumour cell lines of different histogenesis: B cell lymphoproliferative disorders (Smolewski et al., 2014), chronic myeloid leukemia (Turcotte et al., 2009), lung carcinoma and pancreatic adenocarcinoma (Mikulski et al., 1990a), multiple myeloma, adenocarcinoma, and prostate cancer (Ita et al., 2008; **Table 4**).

Initial *in vivo* studies of onconase were performed on the Madison M109 carcinoma model of mice and it was demonstrated that the survival rate of tumour-bearing animals after treatment with onconase increases 12-fold compared with the control (Mikulski et al., 1990b). Recent studies on mesothelioma xenograft models have shown significant suppression of tumour growth by onconase (Nasu et al., 2011); studies on non-small cell lung cancer and mesothelioma xenograft models have shown suppression of tumour growth and angiogenesis when the combined action of onconase and dihydroartemisinin was used (Shen et al., 2016). A number of publications have demonstrated the antitumor activity of conjugates of onconase with antibodies specifically addressed the enzyme to tumour cells (Rybak, 2008; Newton et al., 2009; **Table 4**). In recent studies, a high antitumor activity of onconase conjugated with chlorotoxin has been shown on a mouse glioma xenograft model (Wang and Guo, 2015). The increase of cytotoxicity of onconase to tumour cells was reached *via*  obtaining dimers of the enzyme (Fagagnini et al., 2017).

Onconase was one of the first ribonucleases studied in preclinical and clinical trials (Costanzi et al., 2005). Onconase has been approved for clinical use as an orphan drug for treatment of unresectable malignant mesothelioma in the United States, Europe, and Australia (Mikulski et al., 2002; Altomare et al., 2010). Clinical trials have shown that onconase is well tolerated by patients, has low immunogenicity, but has high nephrotoxicity (Mikulski et al., 2002). However, recent clinical trials of onconase for the treatment of patients with non-small

#### TABLE 3 | Endogenous RNases and their role in cancer development.


smooth muscle neoplasms


*#references are done in accordance with RNA target.*

*\*references are done in accordance with in vitro/in vivo effects.*

*\*\*data are presented in details in review (Hata and Kashima, 2016).*

cell lung cancer (ClinicalTrials.gov Identifier: NCT01184287) and mesothelioma (ClinicalTrials.gov Identifier: NCT00003034) have been prematurely terminated.

Although the main targets for onconase are tRNAs (Iordanov et al., 2000), an ability to affect miRNAs has also been found (Qiao et al., 2012; **Table 4**). The model of onconase-mediated cytotoxicity predicates that onconase is internalized in cytosol of tumour cells and cleaves tRNAs followed by ubiquitous inhibition of protein translation and apoptosis induction (Lee and Raines, 2008). Biochemical studies performed by Qiao and colleagues evaluate miRNAs as the direct downstream RNA targets of onconase. Onconase was found to downregulate miRNAs by cleavage of its precursor forms, thus decreasing the amount of mature miRNAs arisen from Dicer activity (Qiao et al., 2012). In addition, onconase was demonstrated to exert miRNAmediated effects through downregulation of NF-kβ using specific miRNAs, particularly, upregulating miR-17 and downregulating miR-30c (Goparaju et al., 2011).

#### BS-RNase (Bovine Seminal)

BS-RNase was revealed independently by Hosokawa and Irie in 1971 (Hosokawa and Irie, 1971), D'Alessio with colleagues in 1972 (D′Alessio et al., 1972), and Dostal and Matousek in 1972 (Dostal and Matousek, 1972). It is singular among all ribonucleases in that it has a quaternary structure. BS-RNase is a natural dimer that comprises a couple of identical subunits connected by two disulfide bonds and non-covalent interactions (D′Alessio et al., 1991). The amino acid sequence of the BS-RNase subunit and its structure classify this enzyme as belonging to the pancreatic RNase A superfamily (Beintema et al., 1988).

The polypeptide chain of the BS-RNase subunit contains 124 amino acids and has 80% homology with RNase A. Two cysteine residues located at 31 and 32 positions of the BS-RNase represents the most important difference between BS-RNase and RNase A. These two cysteines are involved in the formation of an intermolecular disulfide bond between Cys31 of one subunit and Cys32 of the second subunit, followed by dimerization of enzyme (Di Donato and D′Alessio, 1973). The dimeric enzyme (27.218 kDa) represents is a composition of two different quaternary forms, denoted as M = M and M × M (Piccoli et al., 1992).

The enzyme displays cytotoxic activity towards tumour cells only in dimeric form, and the ribonuclease activity is absolutely crucial (Kim et al., 1995a). However, the groups of D'Alessio and Raines showed that a single subunit of BS-RNase, which has a higher catalytic activity than the dimer, does not exhibit a cytotoxic effect on tumour cells (Vestia et al., 1980; Kim et al., 1995b). The explanation was that a separate subunit, but not a dimeric form of the enzyme, is inactivated by cytosolic RI (Murthy and Sirdeshmukh, 1992). In addition, the dimeric form of BS-RNase, but not monomeric, was shown to destabilize the membranes of tumour cells, and this destabilization contributes to the observed antitumor effect of the enzyme (Mancheño et al., 1994).

The antitumor activity of BS-RNase has been studied mainly on tumour cell lines and, to a lesser extent, on *in vivo*  tumour models. BS-RNase exhibited a cytotoxic effect on various tumour cell lines: SVT2 and 3T3 fibroblast cells, ML-2 myeloid cells, neuroblastoma cells and thyroid carcinomas (Cinatl et al., 1999; Marinov and Soucek, 2000; Kotchetkov et al., 2001; **Table 4**).

Soucek and colleagues developed conjugates of BS-RNase with poly [N- (2-hydroxypropyl) methacrylamide], which protects the enzyme from degradation in the bloodstream, and demonstrated a significant inhibition of melanoma growth on nude mice, whereas intact BS-RNase was ineffective (Soucek et al., 2002). BS-RNase PHPMA conjugates also showed high efficiency in various human tumour models in CD-1 nude mice: melanoma, neuroblastoma, and ovarian cancer (Pouckova et al., 2004; **Table 4**).

#### TABLE 4 | Exogenous RNases displaying antitumor activity.



*#references are done in accordance with RNA target.*

*\*references are done in accordance with in vitro/in vivo effects.*

*n.d. – not detected.*

*PHPMA, Poly[N-(2-hydroxypropyl)methacrylamide]; ErbB2, Erb-B2 receptor tyrosine kinase 2; KIT, KIT proto-oncogene, receptor tyrosine kinase; AML1-ETO, fusion protein detectable in patients with acute myelogenous leukemia; RLS40, drug-resistant lymphosarcoma RLS40.*

### RNases of RNase T1 Superfamily (Bacterial and Fungal)

Recently, many RNases of bacterial and fungal origin have been discovered that exhibit cytotoxic activity (**Table 2**). In contrast to mammalian cytotoxic RNases, which belong to the RNase A superfamily, microbial RNases are related to the RNase T1 superfamily. The RNase T1 superfamily consists of 25 enzymes of fungal and bacterial origin that have a similar amino acid sequence and tertiary structure (Yoshida, 2001). RNases of the RNase T1 superfamily catalyse RNA cleavage at phosphodiester bonds after guanine residues (G↓X) in single-stranded regions.

Fungal RNases, being also denoted as ribotoxins (α-sarcin, mitogillin and restrictocin), perform the cleavage of the eukaryotic 28S rRNA of the large ribosome subunit at a single phosphodiester bond leading to the inactivation of protein synthesis, induction of apoptosis and cell death (Endo et al., 1983; Kao et al., 1998; Olmo et al., 2001; Lacadena et al., 2007). Although α-sarcin demonstrated high cytotoxic activity against a number of tumours, including sarcoma, it also displayed high hepatotoxicity and caused toxic heart damage in healthy animals.

The most well-known of the microbial ribonucleases that exhibit cytotoxic activity on tumour cells are RNase Sa (RNase from *Streptomyces aureofaciens*), barnase (RNase from *Bacillus amyloliquefaciens*), and binase (RNase from *Bacillus intermedius*; **Tables 2** and **4**). According to a recent genotypic identification, the strain known as *B. intermedius* belongs to the *B. pumilus* species, so it has been renamed accordingly (GenBank Accession No. HQ650161.1). Binase and barnase have no homology with mammalian RNases and are not recognized by RI (Rutkoski and Raines, 2008). Barnase and binase are small proteins that consist of 110 (12.382 kDa) and 109 (12.213 kDa) amino acids, respectively, with 85% structure homology (Hartley and Barker, 1972; Aphanasenko et al., 1979).

Cytotoxic effects of RNase Sa were demonstrated towards acute myeloid leukemia Kasumi-1 cells (Mitkevich et al., 2014a; **Table 4**). The Deyev group is engaged in a comprehensive study of the properties of barnase. The ability of barnase to eliminate malignant cells was shown in carcinoma cell lines and human leukemia (Edelweiss et al., 2008). For targeted delivery of enzyme into tumour cells, conjugates were obtained (Balandin et al., 2011; **Table 4**). These conjugates were consisted of two barnase molecules conjugated to a single-stranded variable fragment (scFv) of a humanized 4D5 antibody targeted to the extracellular domain of human epidermal growth factor receptor 2 HER2, which is overexpressed in many human carcinomas. On the basis of barstar, which is an inhibitor of barnase of bacterial origin, and barnase conjugated with fragments of various antibodies and nanoparticles, the development of multifunctional supramolecular structures for the elimination of malignant cells was proposed (Nikitin et al., 2010). Based on barnase fused with a MYC epitope, immunotoxins able to selectively downregulate MYC-specific B cells were developed. They were shown to have an important influence on the development of both systemic and organ-specific autoimmune diseases (Stepanov et al., 2011; **Table 4**).

The Makarov and Ilinskaya groups studied the cytotoxicity of binase using a number of cell lines, distinguished by expressed oncogenes: myeloid precursors FDC-P1; FDC-P1-N822K cells expressing the KIT oncogene; transduced FDC-P1 cells expressing the AML1-ETO oncogene; transduced FDC-P1- N822K cells expressing the AML1-ETO and KIT oncogenes; cells of acute myeloid leukemia Kasumi-1, also expressing both oncogenes, and human ovarian cancer cells (Ilinskaya et al., 2007; Mitkevich et al., 2011; Garipov et al., 2014; **Table 4**). The sensitivity of cells to binase was shown to be dependent on the level of oncogenes, and that the Kasumi-1 cell line was the most sensitive (Mitkevich et al., 2011). It was found that the high expression level of the oncogene KIT also increased the sensitivity of tumour cells to binase (Mitkevich et al., 2010b), suggesting that oncogenic mRNAs may be targets for binase. However, binase exhibits antitumor activity not only due to its own ribonuclease activity, but also due to its ability to bind to certain proteins. Direct interaction of binase and the oncogenic protein KRAS was demonstrated and resulted in the stabilisation of the inactive KRAS conformation and inhibition of MAPK signalling (Ilinskaya et al., 2016).

The Makarov and Zenkova groups first demonstrated the ability of binase to retard primary tumour growth and inhibit metastases formation in experimental murine tumour models: Lewis lung carcinoma, drug resistant lymphosarcoma RLS40, and melanoma B16 (Mironova et al., 2013a; **Table 4**). Treatment of animals bearing tumours of various histogenesis with binase leads to significant retardation of primary tumour growth and dissemination. The therapy was also found to have general systemic, immunomodulatory, and hepatoprotective effects and did not induce inflammatory response in the organism (Mironova et al., 2013a; Sen'kova et al., 2014).

#### CONCLUSION AND CRITICAL POINT OF VIEW

Cell transformation, uncontrolled proliferation, increased migration, and invasion are multistage processes, at all steps of which tumour-associated RNAs take part. The correct balance between intracellular RNAs encoding oncogenes and RNA encoding tumour suppressors, as well as the balance between regulatory oncogenic or oncosuppressive miRNAs, determines the normal cell phenotype, and the imbalance of this equilibrium leads to their transformation. Extracellular tumour-associated coding and regulatory RNAs make a significant contribution to tumour progression through distant transfection of normal cells and the formation of new tumour foci. So named "house-keeping" endogenous ribonucleases are involved in the control of RNA homeostasis and the discarding of aberrant RNAs thus providing the proper functioning of RNA orchestra in the cell. Therein, an alteration of the expression or activity of endogenous nucleases leads to their failure in RNA turnover as well as changes in the profile of regulatory RNAs, which leads to oncogenesis and progression of the tumour.

#### REFERENCES


Thus, it is obvious that exogenous RNases, whose cytotoxic activity was discovered at a time when the era of understanding the role of regulatory and coding RNA in oncogenesis was just beginning, are now recognized participants in control of cell transformation and events of tumor progression. Large amount of data is accumulated confirming that tumour-associated RNAs inside the tumour cell and in the pool of circulating exRNAs, whose levels are significantly increased in the process of tumour progression, can be the targets for exogenous RNases. Exogenous nucleases reducing the amount of both intracellular and circulating tumour-associated RNAs may maintain normal untransformed state of the cell and decrease the rate of tumour dissemination. Thus, exogenous RNase do not at all play the role of a "scavenger," equally likely cleaving all asseccible RNAs, but the role of a supervisor over the tumor development.

In connection with a certain selectivity of natural RNases to RNAs, they can be considered as useful tools for searching for tumor-associated RNA targets. This strategy let to identify a wide range of RNA targets for selective shutdown in a tumor cell. Numerous studies of the antitumor activity of natural RNases, including preclinical ones, in the long term may provide completely new anticancer drugs that work not only at the cell level, but also at the level of organism.

#### AUTHOR CONTRIBUTIONS

NM analyzed published data and prepared the manuscript. VV revised and corrected the manuscript.

#### FUNDING

This work was funded by the Russian State funded budget project of Institute of Chemical Biology and Fumdamental Medicine SB RAS # АААА-А17-117020210024-8 and grant RFBR (no. 17-00- 00059 and 17-00-00062).


apoptosis in human cancer cells. *PLoS ONE* 3, e2434. doi: 10.1371/journal. pone.0002434


and is an oncogenic driver in glioblastoma. *Cancer Res.* 71, 4464–4472. doi: 10.1158/0008-5472.CAN-10-4410


interferon sensitivity and apoptosis in SiHa cervical carcinoma cells by downregulating E6 and E7 human papilloma virus oncoproteins. *Oncotarget* 8, 72666–72675. doi: 10.18632/oncotarget.20199


nano-shuttles: dual role in tumor progression. *Target Oncol.* 13, 175–187. doi: 10.1007/s11523-018-0551-8


noncatalytic basic amino acid residues. *Biochemistry* 42, 10182–10190. doi: 10.1021/bi030040q


Zhou, Y., Ren, H., Dai, B., Li, J., Shang, L., Huang, J., et al. (2018). Hepatocellular carcinoma-derived exosomal miRNA-21 contributes to tumor progression by converting hepatocyte stellate cells to cancerassociated fibroblasts. *J. Exp. Clin. Cancer Res.* 37, 324 doi: 10.1186/ s13046-018-0965-2.

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

*Copyright © 2019 Mironova and Vlassov. 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.*

# Nucleic Acids Delivery Into the Cells Using Pro-Apoptotic Protein Lactaptin

*Olga Chinak1†, Ekaterina Golubitskaya2,3†, Inna Pyshnaya4, Grigory Stepanov2, Evgenii Zhuravlev2, Vladimir Richter1 and Olga Koval1,3\**

*1 Laboratory of Biotechnology, Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia, 2 Laboratory of Genome Editing, Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia, 3 Department of Natural Sciences, Novosibirsk State University, Novosibirsk, Russia, 4 Laboratory of Biomedical Chemistry, Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia*

#### *Edited by:*

*Hector A. Cabrera-Fuentes, University of Giessen, Germany*

#### *Reviewed by:*

*Carlo Gaetano, IRCCS Scientific Clinical Institutes Maugeri (ICS Maugeri), Italy Han Qiao, Shanghai Jiao-Tong University School of Medicine, China*

*\*Correspondence:*

*Olga Koval o\_koval@niboch.nsc.ru*

*†These authors have contributed equally to this work* 

#### *Specialty section:*

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

*Received: 10 May 2019 Accepted: 16 August 2019 Published: 18 September 2019*

#### *Citation:*

*Chinak O, Golubitskaya E, Pyshnaya I, Stepanov G, Zhuravlev E, Richter V and Koval O (2019) Nucleic Acids Delivery Into the Cells Using Pro-Apoptotic Protein Lactaptin. Front. Pharmacol. 10:1043. doi: 10.3389/fphar.2019.01043*

Cell penetrating peptides (CPP) are promising agents for transporting diverse cargo into the cells. The amino acid sequence and the mechanism of lactaptin entry into the cells allow it to be included into CPP group. Lactaptin, the fragment of human milk kappacasein, and recombinant lactaptin (RL2) were initially discovered as molecules that induced apoptosis of cultured cancer cells and did not affect non-malignant cells. Here, we analyzed the recombinant lactaptin potency to form complexes with nucleic acids and to act as a gene delivery system. To study RL2-dependent delivery, three type of nucleic acid were used as a models: plasmid DNA (pDNA), siRNA, and non-coding RNA which allow to detect intracellular localization through their functional activity. We have demonstrated that RL2 formed positively charged noncovalent 110-nm-sized complexes with enhanced green fluorescent protein (EGFP)-expressing plasmid DNA. Ca2+ ions stabilized these complexes, whereas polyanion heparin displaced DNA from the complexes. The functional activity of delivered nucleic acids were assessed by fluorescent microscopy using A549 lung adenocarcinoma cells and A431 epidermoid carcinoma cells. We observed that RL2:pDNA complexes provided EGFP expression in the treated cells and that strongly confirmed the entering pDNA into the cells. The efficiency of cell transformation by these complexes increased when RL2:pDNA ratio increased. Pretreatment of the cells with anti-RL2 antibodies partly inhibited the entry of pDNA into the cells. RL2-mediated delivery of siRNA against EGFP was analyzed when A549 cells were co-transfected with EGFP-pDNA and RL2:siRNA complexes. siRNA against EGFP efficiently inhibited the expression of EGFP being delivered as RL2:siRNA complexes. We have previously demonstrated that non-coding U25 small nucleolar RNA (snoRNA) can decrease cell viability. Cancer cell transfection with RL2-snoRNA U25 complexes lead to a substantial decrease of cell viability, confirming the efficiency of snoRNA U25 delivery. Collectively, these findings indicate that recombinant lactaptin is able to deliver noncovalently associated nucleic acids into cancer cells *in vitro*.

Keywords: nucleic acids, cell penetrating peptides, lactaptin, gene delivery, snoRNA

## INTRODUCTION

Successful delivery of exogenous DNA and RNA molecules into cells has major implications for gene-based technologies and therapeutic approaches. Since free DNA and RNA are not readily taken up by cells, various delivery systems were developed during the last twenty years (Bolhassani, 2011). Delivery molecules or vesicles help nucleic acids to cross biological barriers and prevent enzymatic degradation by endogenous nucleases. Besides the lipid-based delivery system, the polymer-based delivery system or inorganic nanoparticles, peptide carriers can be also used for nucleic acids transport into the cells (Zorko and Langel, 2005; Rathnayake et al., 2017). Among peptide carriers, the most promising are short amphipathic and cationic peptides, which translocate themselves across membranes and are collectively named as cell-penetrating peptides (CPP) (Hoyer and Neundorf, 2012; Morris and Labhasetwar, 2015). There is no unique classification for CPPs yet, and usually, they are classified according to their origin, specificity to cargo, and amino acid composition. In 1997, Morris et al. demonstrated that 27-residue CPP peptide vector being premixed with singleor double-stranded deoxyoligonucleotides efficiently delivered them into cultured mammalian cells (Morris et al., 1997). Later, two CPPs—transportan and penetratin—successfully delivered 21-mer peptide nucleic acid (PNA) complementary to the human galanin receptor type 1 mRNA resulted in the suppression of galanin receptor (Pooga et al., 1998). Thereby, CPPs deliver functionally active nucleic acids into the cells.

Lactaptin, an 8.6-kDa proteolytic fragment of human milk kappa-casein, selectively kills various tumor cells while it is used in micromole concentration (Semenov et al., 2010; Koval et al., 2012). We have earlier demonstrated that recombinant lactaptin (RL2) penetrates into cells partly through lipid raft-mediated dynaminindependent pinocytosis and partly through direct penetration across the plasma membrane (Chinak et al., 2016). The comparison of RL2 primary structure and the mechanism of its penetration into the cell suggests that it can be assigned to the CPP.

In the present work, we analyzed if lactaptin analog forms complexes with nucleic acids and if it acts as a delivery system with no cytotoxic effects.

### RESULTS

#### RL2 Forms Non-Covalent Complexes With Plasmid DNA

CPPs, like other cationic proteins or peptides, can condense DNA through their cationic residues by non-covalent electrostatic interactions (Tecle et al., 2003). Since RL2 has the properties of сell-penetrating peptides, its ability to deliver cargo molecules into the cells was studied using a plasmid DNA (pEGFP). This plasmid contains a coding sequence of the enhanced green fluorescent protein (EGFP) under the control of the cytomegalovirus (CMV) immediate early promoter, allowing expression of EGFP in eukaryotic cells.

The charge ratio N/P or peptide nitrogen per nucleic acid phosphate of RL2:pEGFP complexes was calculated as described in the Materials and Methods section for the analysis of the efficient ratio of RL2 and pDNA. Here, to optimize empirically the ratio of RL2 and pEGFP for the non-covalent complexes (RL2:pEGFP) formation, we have varied the RL2 amount, whereas pDNA amount was constant within N/P ratio of 0.5 to 5. UV-vis spectroscopy analysis of complexes (**Supplementary Figure 1**) demonstrates that RL2:pDNA complexes have maximum absorbance in the range of 260 to 280 nm, whereas pDNA has an absorption maximum at 260 nm and RL2 at 280 nm.

The efficiency of complex formation was analyzed by agarose gel retardation assay. Data obtained (**Figure 1A**) shows the conditions when pDNA movement retards in gel indicating complex formation. The increase of RL2 amount leads to the decrease of free plasmid DNA in a reaction mixture. Complete retardation was reached at a N/P charge ratio of five (**Figure 1A**). It indicates that all pDNA was totally involved into RL2:pEGFP complexes which had no mobility under the conditions of electrophoresis due to their larger size compared to the pores size, or the neutral charge.

To characterize RL2:pEGFP complexes, the hydrodynamic size and ζ-potential of the complexes were estimated using the dynamic light scattering (DLS) technique. **Figure 1B** shows that the complex size is near 110 nm with no significant differences for the charge ratio range from 5 to 9. The size of the RL2:pEGFP complex was smaller than the sizes of RL2 or pEGFP alone. Thus, when complexes form, both molecules condense.

Electrokinetic potential (ζ-potential) of pDNA and RL2 in a solution demonstrated a negative charge for DNA and a positive charge for RL2. For complexes, ζ-potential gradually increases with the N/P increasing from 5 to 9 (**Figure 1C**). We suppose that the increasing of ζ-potential is specified by the rising amount of surface-exposed cationic groups of RL2 molecules in RL2:pDNA complexes.

#### Ca2+ and Heparin Change the Stability of RL2:pEGFP Complexes

Since RL2 originates from milk k-casein, it can be sensitive to Ca2+ ions concentration. It is known that environmental Ca2+ can stimulate aggregation or dissociation of casein's micelles (Ye and Harte, 2013). To reveal the Ca2+ effects on the stability of RL2:pEGFP complex, complexes were formed under semiextreme N/P ratios. Under N/P = 7, there was no free pDNA and Ca2+ did not release pDNA from the RL2:pEGFP complex (**Figure 2A**). Under N/P = 1, only half of pDNA molecules were in the RL2:pEGFP complex, and these conditions allow us to analyze Ca2+ concentration when total pDNA is incorporated in complex with RL2. The increase of Ca2+ concentration rises the amount of pDNA incorporated in the complex with RL2 (**Figure 2A**).

Heparin is a stronger polyanion than DNA; therefore, it can competitively bound to RL2, releasing plasmid from RL2:pEGFP complexes. Heparin sodium salt (2.4 units per 1 µg DNA) was added to pre-formed RL2:pEGFP complexes, and the appearance of free pDNA was analyzed. We showed that heparin releases pEGFP from the complexes with N/P ≤ 11 (**Figure 2B**). Complexes with high N/P ratio were more resistant to heparin.

Thus, Ca2+ divalent cations stabilize complexes and polyanion heparin destabilizes RL2:pEGFP complexes.

from 0.5 to 5 in reaction mixtures; C- control (free pDNA). Particle size (B) and ζ potential (C) of RL2:pEGFP complexes were analyzed for N/P ratios 5, 7, and 9. Reactions were performed in MES buffer for 5 min at 37°C. Particle size was estimated by the DLS technique. The data are representative of three independent repeats and are shown as the mean ± SD.

#### Plasmid DNA Delivery

To examine whether RL2 delivers pEGFP into human cancer cells the functional-based method was used. A549 and A431 cells were incubated with RL2:pEGFP complexes with various N/P ratios. After that, a reporter EGFP protein was analyzed by fluorescence microscopy. EGFP fluorescence was observed in the cells treated with complexes characterized by the high N/P ratio (**Figure 3A**). No fluorescence signal was seen when cells were treated with naked pEGFP, indicating that plasmid DNA itself did not enter into the cells under these conditions. The increase of RL2 content in complexes led to the increase of the number of cells with EGFP, and the highest EGFP fluorescence was in cells incubated with a RL2:pEGFP complexes with N/P = 9.

The entering of RL2:pEGFP complexes into the A431 cells was also evaluated by RT-PCR analysis with EGFP-specific primers using whole cellular RNAs as a matrixes. EGFP-specific PCR products were detected in cells treated with RL2:pEGFP complexes within N/P ratio of 5 to 9 (**Figure 3C**). No PCR product was seen when cells were transfected with pEGFP alone or with control pDNA.

Thus, RL2 provides translocation of plasmid DNA into the cells and saves pEGFP functional activity upon internalization. Pre-treatment of the cells with anti-RL2 antibodies partly inhibited the entry of pEGFP into the cells (**Figures 3B**, **C**).

#### RL2:siRNA Complex Formation

To analyze RL2:siRNA complex formation, the well-described anti-EGFP siRNA, which targets the EGFP gene coding region, was used (Nagy et al., 2003). The efficiency of complex formation was analyzed with N/P ratios of RL2:siRNA from 1 to 15. Gel

antibodies (ab) for 3 h and next transected with a complex RL2:pEGFP. (C) Electrophoretic analysis of the products of RT-PCR reaction with EGFP-specific primers. Free pUC19 and pEGFP were used as matrixes in PCR reaction for the negative and positive controls, respectively. Representative pictures of three independent experiments.

retardation assay revealed that the free form of siRNA in the samples disappeared when the N/P ratio was 10 and higher (**Figure 4**). Thus, RL2 efficiently forms complexes with siRNA with N/P more than 5.

#### siRNA Delivery

To study RL2-based siRNA delivery, at first, A549 cells were transfected with pEGFP, and then cells were incubated with pre-forming RL2:siRNA complexes. For the stage of pEGFP transfection, the Lipofectamine 3000 was used to avoid excessive RL2 amount for the cell treatment (positive control). Cells transected with naked pEGFP (negative control) demonstrate no EGFP fluorescent signal (**Figure 5A**). EGFP was down-regulated in the RL2:siRNA-treated cells when N/P was 25 or higher. Flow cytometry analysis of transfected cells confirmed the data obtained and showed approximately a 50% decrease of EGFP for the complex with N/P = 35 (**Figures 5B, C**). These experiments demonstrate the specific effects of anti-EGFP siRNA, allowing us to conclude that siRNA efficiently penetrates into the cells treated with RL2:siRNA complexes.

#### Cytotoxic Activity of RL2:U25 snoRNA Complex

Non-coding RNAs have been shown to modulate various cellular responses as well as induce death of cancer cells *in vitro* (Van et al., 2012; Stepanov et al., 2016). Earlier, we demonstrated that non-coding artificial analogue of U25 box C/D snoRNA (snoRNA U25) decreased the viability of various cancer cells *in vitro* (Nushtaeva et al., 2018)*.* The cytotoxicity U25 snoRNA was in part due to the activation of inflammation-involved genes. U25 snoRNA is a single-stranded molecule in comparison with double-stranded pDNA or siRNA. The complex of RL2 with snoRNA was performed as described in the Materials

and Methods section. For the experiments, RL2 was used in low-cytotoxic concentration. A549 cells were treated with RL2:snoRNA complex, and 48 h after MTT analysis revealed the decrease of cell viability (**Figure 6A**). **Figure 6A** demonstrates that Lipofectamine-delivered U25 snoRNA decreased the viability of treated cells more substantially than RL2-delivered. Concentration-response curves show the rise of cytotoxic effects with the rising of RL2 concentration only for RL2:snoRNA complexes, but not for free RL2 (**Figure 6B**). The increase of RL2 quantity in the complex elevates the penetration potency of such complexes, this seems to be the same as the other investigated complex of nucleic acids with RL2. Analyzing the data obtained, we conclude that the decrease of viability of treated cells was induced by snoRNA activity, and this efficiency correlated with the penetration potency of RL2:snoRNA complex.

### DISCUSSION

Negative charge of DNA and RNA molecules hampers their movement across cellular membranes. To reach intracellular space, DNA or RNA molecules usually need to be packaged in vesicles or carriers. Usually, peptides are used as cell-targeting compounds or cell-specific ligands being conjugated with vesicles or carriers. Among various carriers, the CPP peptides are the potential delivery system for nucleic acids (Ignatovich et al., 2003; Järver et al., 2012; Shukla et al., 2014; Kizil et al., 2015). Forming protein-nucleic acid complex, CPPs can absorb on the surface of the double-stranded helix generally by electrostatic interaction, and such complexes can aggregate with size of up to 100 nm and have a positive charge (Rathnayake et al., 2017). CPPs were demonstrated as potential siRNA delivery systems for the suppression of gene expression. Low-cytotoxic arginine-rich CPPs were successfully used for the gene delivery to the brain (Morris and Labhasetwar, 2015).

Recombinant analog of lactaptin RL2 originates from human k-casein and demonstrates CPP-like properties (Chinak et al., 2016). Due to amphiphilic characteristics, the caseins themselves can selfassemble into micelles (Portnaya et al., 2008). In drug delivery on the organism level, casein spheres showed a superior biocompatibility being introduced orally and were successfully used as carriers of doxorubicin (Chen et al., 1987). Therefore, here we analyzed if lactaptin analog RL2 translocates nucleic acids into the cancer cells.

We observed that double stranded as well as single stranded nucleic acids were efficiently translocated into the cancer cells being in complex with RL2. The delivery of EGFP reporter genecontaining plasmid into the cells was tested for EGFP-expressing activity. Since the transcription of mRNAs in eukaryotic cells takes place strictly in the nucleus, we suggest that at least pEGFP enters the nucleus. Whether pEGFP releases from the complex in the cytoplasm and then free pEGFP is transported through the nuclear membrane or RL2-pEGFP complex enters the nucleus this requires more investigations.

The efficiency of delivery was dependent on the RL2 amount in the complexes:RL2-rich complexes were more efficient according to the cargo functional activity. The actual process of DNA or RNA condensation with CPPs has not yet been fully understood. Our

cytometry analysis. Representative images of EGFP-positive gate excludes EGFP-negative cells (relative to negative control). (C) Quantitative evaluation of EGFP

down-regulation. The data are representative of three independent repeats and are shown as the mean ± SD. \**P* < 0.05; \*\**P* < 0.03.

finding that pDNA : RL2 complex size is smaller than the sizes of RL2 or pEGFP alone indicates the condensation of DNA in the complex. Nowadays, the most used DNA-condensing cationic peptides in gene delivery are lysine-reach, like RPC or arginine-reach, like viral Tat peptides (Wender et al., 2000; Read et al., 2003). RL2 contains five lysines and eight arginines (**Supplementary Figure 2**), and this is likely enough to act as a DNA-condensing molecule. We suppose that the increasing of ζ-potential is specified by the rising amount of surface-exposed cationic groups of RL2 molecules in RL2:pDNA complexes.

Although stability of RL2:pDNA complexes has a few differences in the N/P range of 1.5 to 13 (**Figures 1A** and **2B**), the EGFP ectopic expression level increased with N/P increasing from 5 to 9. This could explain that RL2:pEGFP complexes with high amount of RL2 exert higher resistance to the nucleases. The same results were obtained for the siRNA delivery: despite of the findings that strong RL2:siRNA complexes were formed with N/P range from ten and higher, the efficiency of delivery increased with the increasing of RL2 in the complex. This could mean that a high RL2 concentration is needed for the successful delivery or alternatively, that complexes with high N/P value are fully shielded from extracellular enzymes or that the high RL2 concentration provides early endosome escape of siRNA also. Moreover, we suppose that RL2-mediated nucleic acids transport is non-dependent on the cargo size, but only on the RL2 amount that was confirmed by the effective delivery of big plasmid DNA and small siRNA.

We have also observed that Ca2+ stabilizes complexes of RL2 with nucleic acids. It is known that at low concentration divalent cations like Ca2+ and Mg2+ promote the condensation of CPP particles into agglomerates (McIntyre et al., 2017). Next, to study the plasmid protection outside the cells and intracellular release of cargo, we confirmed the pDNA release from the complexes by heparin displacement. Polyanion heparin is a suitable molecule for the modeling of nucleic acid displacement in the complexes because before the complexes reach the plasma membrane phospholipids, the delivering complexes interact with proteoglycans of the extracellular matrix, such as heparin. High heparin concentration induced total release of pDNA from the RL2:pDNA complexes.

Taken together, these findings indicate that recombinant lactaptin is able to deliver noncovalently associated nucleic acids into cancer cells *in vitro*.

#### MATERIALS AND METHODS

#### Materials

Heparin sodium salt (H4784-1G) and CaCl2 were from Sigma-Aldrich. DNA Gel Loading Dye (R0611) and Lipofectamine 3000 were from ThermoFisher, molecular weight markers Sky High (10 and 500 bp) were from BioLabMix, Russia. MTT (3-(4,5-dimethyl-2-thiazolyl) -2,5-diphenyl-2H-tetrazolium bromide was from Sigma-Aldrich. Plasmid pEGFP-C1 (GenBank Accession U55763; cat 6084-1) was purchased from BD Biosciences Clontec, monoclonal anti-RL2 mouse IgG (clone F14; Biosan, Russia).

The recombinant analogue of lactaptin RL2 was obtained from *E. coli* and purified as described previously (Semenov et al., 2010). The 98% purity of the isolated protein was confirmed by RP-HPLC chromatography on C5 reverse phase column (Discovery BIO Wide Pore C5; Sigma) in water (0.5% TFA)-acetonitil solvent system using HPLC Station (Bio-RAD Laboratories) as well as by RP-HPLC on C18 (ProntosSIL) using Milichrom A-02 station (EcoNova, Russia) (**Supplementary Figure 3A**) and by SDS-PAAG electrophoresis under reduction conditions (**Supplementary Figure 3B**).

#### Cell Culture

A549 human lung carcinoma cells (ATCC CCL-185) and A431 squamous carcinoma cells (ATCC CRL-1555) were grown in DMEM, supplemented with 10% fetal bovine serum (FBS), 2 mM l-glutamine, 100 U/mL penicillin, 100 mg/mL streptomycin, and 0.25 μg/mL amphotericin B, all from Gibco BRL Co. USA, at 37<sup>о</sup>C, 5% CO2.

#### N/P Calculation

N/P values were calculated by the formula:

$$\mathcal{N} \mid P = \frac{n\left(\begin{array}{c} \ast \mapsto \ast \text{ charged groups of } R\Lambda.2\right) \times \mathcal{C}\left(R\Lambda.2\right)}{m\left(\begin{array}{c} \ast \mapsto \ast \text{ charged groups of nuclei } \mathbf{acid}\right) \times \mathcal{C}\left(\text{nucleic acid}\right)}.$$

where C is the concentration (mole/L).

Amino-groups and imino-groups of amino acids in RL2 were estimated as fully protonated for lysine, arginine, and histidine under experimental conditions. Phosphate groups of nucleic acids were accounted as totally deprotonated with negative charge.

### Complexes Formation

RL2:pEGFP complexes were prepared by varying a charge ratios (N/P). RL2 (1.1 × 10−8 М to 9.7 × 10−8 М) and plasmid DNA pEGFP-C1 (2.9 × 10−8 М) were dissolved in 25 mM MES buffer, pH 5.5, and then they were mixed and incubated for 5 min at 37°C.

For the RL2:siRNA complex formation, at first siRNA duplexes were formed. Chemically synthesized oligoribonucleotides 5′-GAACGGCAUCAAGGUGAACTT-3′ (sense) and 5′-GUUC ACCUUGAUGCCGUUCTT-3′ (antisense) were from Institute of Chemical Biology and Fundamental Medicine, SB RAS, Russia. Oligoribonucleotides (1 × 10−5 М) were incubated in the buffer containing 50 мМ potassium acetate, 15 мМ HEPES-KOH, pH 7.4 and 1 мМ magnesium acetate for 3 min at 90°C, and then mixture was cooled to 37°C. RL2 was mixed with anti-EGFP siRNA at N/P ratios of 10 to 35 in 25 mM MES buffer, pH 5.5 for 5 min at 37°С.

For the RL2:snoRNA complex formation, artificial analogue of U25 box C/D snoRNA was synthesized as was described in the study of Stepanov et al. (2018). RL2 (0.05–0.2 mg/mL) and snoRNA (10 or 30 nM) were dissolved in 25 mM MES buffer, pH 5.5, and then were mixed and incubated for 5 min at 37°С.

### Gel Retardation Assays

Aliquots of noncovalent complexes (reaction mixture, 10 µL) were loaded onto 0.5% agarose gel (for RL2:pDNA) or onto 15% PAAG (RL2:siRNA) with DNA Gel Loading Dye. Retardation was analyzed by electrophoresis in Tris-Acetate-EDTA (TAE) buffer or Tris-Borate-EDTA (TBE) buffer. DNA was visualized with ethidium bromide by Gel Doc XR+ Image system (BioRad, USA), and siRNA was visualized with SYBR Green I Nucleic Acid Gel Stain by BioRad GelDoc XR+ Image system (BioRad, USA).

#### Dynamic Light Scattering (DLS)/ Determination of Particle Size, **ξ-**Potential, and UV-vis

The hydrodynamic diameter and zeta potential of RL2:pEGFP were determined by dynamic laser light scattering using a Zetasizer Nano ZS (Malvern Instruments Ltd, UK) at a wavelength of 623 nm and 25°C. All DLS results were calculated as the average of at least triplicate measurements and presented as mean ± SD.

UV-vis spectroscopy analysis of complexes was performed using NanoDrop 2000c (Thermo Scientific, USA).

### pDNA Delivery Microscopic Assays

A549 and A431 cells were seeded (2 x 104 cells/cm2 ) on 24-well plates 24 h prior the experiment. Cell media was replaced with fresh Opti-MEM (ThermoFisher, USA) containing 2 mM l-glutamine, and pre-formed complexes were added to the cells to the final concentration of pEGFP 1 ng/µL. After 3 h, medium was replaced with complete fresh DMEM media supplemented with 10% FBS, and cells were incubated for additional 24 h. EGFP fluorescence was analyzed using fluorescent microscope (ZOE Bio-Rad, USA).

### pDNA Delivery RT-PCR Assays

After microscopic analysis, cells were washed three times with PBS and total RNA was isolated by phenol-chloroform extraction using the Lira reagent (Biolabmix Ltd, Novosibirsk, Russia) according to the manufacturer's protocol. The quality of total RNA was assessed by agarose gel electrophoresis or capillary electrophoresis with an Agilent 2100 Bioanalyzer, using 28S/18S > 2 or RIN > 8.0 criterion. RT-PCR were performed in the one-tube reaction mixture BioMaster RT-PCR SYBR Blue (Biolabmix Ltd, Novosibirsk, Russia, www.biolabmix. ru) with gene-specific primers: EGFP 5′-GTAAACGGCC ACAAGTTCAG-3′ and 5′-GGTGCGCTCCTGGACGTAGC-3′; hypoxanthine-guanine phosphoribosyltransferase (HPRT): 5′-CATCAAAGCACTGAATAGAAAT-3′ and 5′-TATCTTCCA CAATCAAGACATT-3′. For RT-PCR analysis 1 µg of total RNA was used in the reaction under the following conditions: incubation at 25°C for 5 min followed by 42°C for 30 min and at 95°C for 4 min, next incubation at 59°C for 30 s, and at 72°C for 20 s followed by 30 cycles of denaturation at 95°C for 30 s, annealing at 59°C for 30 s, and elongation at 72°C for 20 s.

### Pre-Treatment With Anti-RL2 Antibodies

Cells were grown in 12-well plates up to 80% confluence before the replacement for the fresh DMEM:F12 medium supplemented with l-glutamine. Monoclonal anti-RL2 IgG (15 and 77 ng) were added to the cells and incubated for 15 min. Next, cells were treated with RL2:pEGFP complexes and incubated for 3 h. Subsequently, culture medium was replaced with fresh DMEM:F12 medium supplemented 10% FBS and l-glutamine and cells were incubated for 24 h at 37°C in CO2 incubator. Cells were analyzed by fluorescence microscopy (ZOE Bio-Rad, USA), and cellular lysates were used for subsequent RT-PCR analysis.

### siRNA Delivery Assay

A549 cells were seeded (2 x 104 cells/cm2) on 12-well plates. After 24 h, cell media was replaced with fresh Opti-MEM (ThermoFisher, USA) containing 2 mM l-glutamine. Plasmid pEGFP-C1 with Lipofectamine 3000 (1 µg DNA/well) and RL2:siRNA complex were added to the cells. Cells were incubated 3 h at 37° C, and after that, FBS and antibiotics were added to the cells. Cells were incubated for 48 h and analyzed by fluorescence microscopy (ZOE Bio-Rad, USA) or by flow cytometry (BD FACSCantoII). For flow cytometry analysis cells were fixed in 5% formaldehyde for 2 h. Fluorescence was detected in the FITC channel (emission 535 nm, excitation 488 nm). Data Analysis was performed using the software package BD FACSDiva software v6.1.3.

### snoRNA Delivery Assay (MTT)

A549 cells were seeded in 96-well plates at density of 2 x 103 cells per well. After 24 h cell media was replaced with 100 µL DMEM, containing 2 mM l-glutamine, and 10.5 µL RL2:U25 complexes per well. After 2 h, 50 µL DMEM supplemented with 30% FBS, 300 U/ mL penicillin, 300 mg/mL streptomycin, 0.75 μg/mL amphotericin B, and 2-mM l-glutamine was added, and cells were incubated for 48 h, and MTT analysis was performed as was described in the study of Koval et al. (2014). Cell viability was expressed as a means percentage of control ± SD for triplicate independent experiments.

#### Statistical Analysis

All experiments were repeated three times independently, and data are presented as mean ± SD. The Mann-Whitney U-test was used to define statistically significant differences between groups. All error bars represent standard error of the mean. P value that was less or equal 0.05 was considered as significant.

### DATA AVAILABILITY

The data that support the findings of this study are available from the authors upon reasonable request.

### AUTHOR CONTRIBUTIONS

OK designed the study and analyzed the data. OC and EG performed the experiments. IP performed determination of particle size and zeta potential. GS and EZ performed the experiments with snoRNA. The draft manuscript was prepared by OK with input from VR. All authors agreed the final version. All authors read and approved the final manuscript.

#### REFERENCES


#### FUNDING

This work funded by the Russian State funded budget project of ICBFM SB RAS АААА-А17-117020210023-1.

### ACKNOWLEDGMENTS

We acknowledge Russian State funded budget project of ICBFM SB RAS АААА-А17-117020210023-1 for supporting this work.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.01043/ full#supplementary-material

SUPPLEMENTARY FIGURE 1 | Analysis of formation of RL2:pEGFP complexes with N/P 5 – 11 by UV-vis spectroscopy.

SUPPLEMENTARY FIGURE 2 | Amino acid sequence of RL2.

SUPPLEMENTARY FIGURE 3 | Purity analysis of RL2. A – HPLC analysis of RL2 comparing with blanking chromatography on ProntoSIL 120-5-B – Electrophoretic analysis of RL2 in 13% SDS-PAGE in reducing conditions.


**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 Chinak, Golubitskaya, Pyshnaya, Stepanov, Zhuravlev, Riсhter and Koval. 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.*

# A New Antisense Phosphoryl Guanidine Oligo-2**′**-O-Methylribonucleotide Penetrates Into Intracellular Mycobacteria and Suppresses Target Gene Expression

#### *Edited by:*

*Olga N. Ilinskaya, Kazan Federal University, Russia*

#### *Reviewed by:*

*Avinash Padhi, Karolinska Institute (KI), Sweden Raghunand R. Tirumalai, Centre for Cellular & Molecular Biology (CCMB), India Xiyuan Bai, University of Colorado Denver, United States*

#### *\*Correspondence:*

*Dmitry A. Stetsenko dmitry.stetsenko@ntlworld.com Tatyana L. Azhikina tatazhik@ibch.ru*

*†These authors have contributed equally to this work*

#### *Specialty section:*

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

*Received: 20 April 2019 Accepted: 19 August 2019 Published: 19 September 2019*

#### *Citation:*

*Skvortsova YV, Salina EG, Burakova EA, Bychenko OS, Stetsenko DA and Azhikina TL (2019) A New Antisense Phosphoryl Guanidine Oligo-2*′*-O-Methylribonucleotide Penetrates Into Intracellular Mycobacteria and Suppresses Target Gene Expression. Front. Pharmacol. 10:1049. doi: 10.3389/fphar.2019.01049*

*Yulia V. Skvortsova 1†, Elena G. Salina 2†, Ekaterina A. Burakova 3,4†, Oksana S. Bychenko1, Dmitry A. Stetsenko3,4\* and Tatyana L. Azhikina1\**

*1 Shemyakin and Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Moscow, Russia, 2 Bach Institute of Biochemistry, Research Center of Biotechnology of the Russian Academy of Sciences, Moscow, Russia, 3 Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia, 4 Faculty of Physics, Novosibirsk State University, Novosibirsk, Russia*

The worldwide spread of multidrug-resistant *Mycobacterium tuberculosis* strains prompted the development of new strategies to combat tuberculosis, one of which is antisense therapy based on targeting bacterial mRNA by oligonucleotide derivatives. However, the main limitation of antisense antibacterials is poor cellular uptake because of electrostatic charge. Phosphoryl guanidine oligo-2′-*O*-methylribonucleotides (2′- OMe PGOs) are a novel type of uncharged RNA analogues with high RNA affinity, which penetrate through the bacterial cell wall more efficiently. In this study, we investigated the uptake and biological effects of 2′-OMe PGO in mycobacteria. The results indicated that 2′-OMe PGO specific for the alanine dehydrogenase-encoding *ald* gene inhibited the growth of *Mycobacterium smegmatis* and downregulated *ald* expression at both the transcriptional and translational levels through an RNase H-independent mechanism, showing higher biological activity than its phosphorothioate oligonucleotide counterpart. Confocal microscopy revealed that the anti-*ald* 2′-OMe PGO was taken up by intracellular mycobacteria residing in RAW 264.7 macrophages without exerting toxic effects on eukaryotic cells, indicating that 2′-OMe PGO was able to efficiently cross two cellular membranes. In addition, 2′-OMe PGO inhibited the transcription of the target *ald* gene in *M. smegmatis*-infected macrophages. Thus, we demonstrated, for the first time, a possibility of targeting gene expression and inhibiting growth of intracellular mycobacteria by antisense oligonucleotide derivatives. Strong antisense activity and efficient uptake of the new RNA analogue, 2′-OMe PGO, by intracellular microorganisms revealed here may promote the development of novel therapeutic strategies to treat TB and prevent the emergence of drug-resistant mycobacterial strains.

Keywords: multidrug resistance, tuberculosis, antibacterial agents, antisense oligonucleotides, cellular uptake, RNase H, macrophages

### INTRODUCTION

Tuberculosis (TB), a chronic infectious disease caused by *Mycobacterium tuberculosis*, is responsible for nearly 2 million fatalities annually, and in 2017, 10 million new TB cases were documented (WHO report, 2018). Only 5–10% of infected individuals develop progressive lung disease, whereas the rest have latent infection without symptoms, and to date, more than 25% of the total human population are believed to be asymptomatic carriers of the pathogen (Veatch and Kaushal, 2018), which, however, can be reactivated, indicating high risk of developing active TB in a large proportion of the population. Furthermore, the disease can be refractory to treatment because the widespread use of conventional antibiotics has led to the emergence of multidrug-resistant (MDR) and extremely drug resistant (XDR) *M*. *tuberculosis* strains, which accounts for a decreased success recovery rate: according to the latest WHO data, it is 52% for MDR-TB and 28% for XDR-TB vs. 83% for TB. Therefore, alternative therapeutic options overcoming limitations of traditional antimicrobial drugs are urgently required.

Antisense therapy is an approach to treat bacterial infections using short (15–25 nucleotides) single-stranded oligonucleotides, often chemically modified to gain stability in biological media (Sully and Geller, 2016). Antisense oligonucleotides (ASOs) act by forming a complementary duplex with their target mRNA through specific sites (Zamecnik and Stephenson, 1978; Uhlmann and Peyman, 1990) and thus can modulate gene expression in a sequence-selective manner by inhibiting translation through either steric block (i.e., physical arrest of ribosomal elongation) or RNase H-mediated degradation of the mRNA strand hybridized with the ASO (Kurreck, 2003). ASOs can be introduced into cells by gymnotic ("naked") uptake without the aid of lipophilic carriers, which is convenient and eliminates possible toxic effects of transfection agents (Dowdy, 2017; Shen and Corey, 2018).

Antisense antibacterials can be designed to target conserved mRNA regions critical for the life cycle or antibiotic resistance of the pathogen, while not leading to the emergence of drugresistant strains (Altman, 2014; Penchovsky and Traykovska, 2015). The approach is considered very promising in combating *M*. *tuberculosis* infection (AlMatar et al., 2017; Hegarty and Stewart, 2018), and several attempts have been made to use ASOs modified with phosphorothioate (PS) or peptide nucleic acids (PNAs) to target essential mycobacterial genes (reviewed in Hegarty and Stewart, 2018). However, the use of ASOs as antibacterial agents is limited, in particular, by their poor uptake into either extracellular or, especially, intracellular bacteria (Xue et al., 2018), and the majority of them have been designed to work in eukaryotic cell systems (Juliano, 2016), with the exception of peptide conjugates with phosphorodiamidate morpholino oligonucleotides (PMOs or morpholinos) (Daly et al., 2017) or, to a lesser extent, PNAs (Good et al., 2001). Relatively poor efficiency of cellular uptake for common ASO derivatives without specific transfection agents or delivery systems could be, at least partly, attributed to their large net negative charge (Fokina et al., 2017), which prevents their penetration through hydrophobic cell membranes.

Phosphoryl guanidine oligo-2′-*O*-methylribonucleotides (2′-OMe PGOs) are a novel type of charge-neutral nucleic acid analogues developed by the Stetsenko group (Kupryushkin et al., 2014). Previous studies indicate that phosphoryl guanidine oligonucleotides are structurally similar to natural oligodeoxynucleotides and demonstrate improved nuclease resistance (Lomzov et al., 2019; Su et al., 2019). Our preliminary experiments have revealed that 2′-OMe PGOs combine the benefits of several ASO types as they are chemically and biologically stable similar to deoxy PGOs, possess high RNA affinity comparable with that of oligo-2′-*O*-methylribonucleotides, and, as morpholinos, lack negative charge (Kupryushkin et al., 2014; Fokina et al., 2018). Therefore, we hypothesized that 2′-OMe PGOs may have improved cellular uptake and functional activity in inhibiting translation of mycobacterial genes.

Here, we report that 2′-OMe PGO reduced the growth of mycobacteria in culture and inhibited translation of the target mRNA more effectively than its PS counterpart. Furthermore, 2′-OMe PGO showed the ability to penetrate into intracellular macrophage-residing mycobacteria without the aid of transfection agents and suppress target gene expression, suggesting its potential as a novel and effective antimycobacterial agent.

#### MATERIALS AND METHODS

#### Oligonucleotides

Oligonucleotides used in this study are listed in **Supplementary Table 1**. Oligodeoxynucleotides were synthesized by Eurogen (Russia), and a PS oligodeoxynucleotide (**Figure 1A**) was purchased from DNA Synthesis (Russia). Mesyl phosphoramidate oligodeoxynucleotide μ-*ald* (**Figure 1B**) was synthesized as described previously (Chelobanov et al., 2017), and 2′-OMe PGO*ald* and FAM-PGO incorporating 1,3-dimethylimidazolidine-2 imino group at each internucleotidic position (**Figure 1C**) were obtained according to the published method (Stetsenko et al., 2014; Su et al., 2019). Synthesis was performed in an automated DNA/RNA synthesizer ASM-800 (Biosset, Russia) using standard 2-cyanoethyl 2′-*O*-methylribonucleoside phosphoramidites (Sigma-Aldrich, USA) and 2′-OMe-rU (Sigma-Aldrich) or 3′-(6-fluorescein) (Glen Research, USA) Controlled Pore Glass supports by substituting Staudinger reaction with 2-azido-1,3 dimethylimidazolinium hexafluorophosphate (TCI, Japan) for iodine oxidation. The reaction was carried out with 0.1 M of 2-azido-1, 3-dimethylimidazolinium hexafluorophosphate in acetonitrile for 15 min at ambient temperature. In case of FAM-PGO, incorporation of the first 2′-OMe-rU residue was followed by iodine oxidation to produce a single phosphodiester linkage, switching to the Staudinger reaction for the rest of the sequence. Oligonucleotides were cleaved from solid support and deprotected by treatment with AMA reagent (25% aq. ammonia and 40% aq. methylamine, 1:1 v/v) at 55°C for 15 min. After the removal of volatiles *in vacuo*, the oligonucleotides were dissolved in 50% acetonitrile and subjected to conventional reversed-phase high-performance liquid chromatography (HPLC) analysis and purification followed by mass spectrometry. Molecular masses of

oligonucleotides were determined by electrospray ionization (ESI) liquid chromatography–tandem mass spectrometry (LC-MS/MS) on an Agilent G6410A spectrometer (Agilent Тechnologies, USA) in a positive ion detection mode using standard device settings. The samples were dissolved in a buffer containing 20 mM of triethylammonium acetate (TEAA) and 60% acetonitrile to 0.1 mM. Sample volume was 10 µL, eluent was 80% aqueous acetonitrile, and flow rate was 0.1 mL/min. Molecular masses of the oligonucleotides were calculated using sets of experimental *m*/*z* values, which were evaluated for each sample.

### Bacterial Culture

*Mycobacterium smegmatis* mc2 155 was taken from frozen stocks (bacterial collection of Bach Institute of Biochemistry, Research Center of Biotechnology of the Russian Academy of Sciences, Moscow, Russia) and pre-cultured for 24 h at 37°C on an orbital shaker (200 rpm) in 40 mL of the rich medium Nutrient Broth (Himedia, India) supplemented with 0.05% of Tween-80. To test antibacterial activity of ASOs, the culture was re-grown to fresh medium (same composition) with concentration of approximately 3 × 103 CFU/mL.

### *In Vitro* Growth Inhibition Assay

Antibacterial activity of ASOs against *M. smegmatis* mc2 155 was examined using colony-forming unit (CFU) counting. Oligonucleotides in concentrations of 10 and 20 µM were added to *M. smegmatis* cells suspensions, and bacteria were cultivated at 37°C, 200 rpm; untreated *M. smegmatis* culture was used as control. Samples were collected at various time points, serially diluted 10-fold, plated on Nutrient Broth agar, and incubated at 37°C for 3 days before counting CFUs. Three independent experiments were performed in four replicates each.

### RNA Isolation

*M. smegmatis* mc2155 cultures (10 mL) were centrifuged at 4,000 × *g* for 15 min, and bacterial pellets washed twice with fresh

medium, rapidly cooled on ice, and centrifuged again. Total RNA was isolated by phenol-chloroform extraction and cell disruption with Bead Beater (BioSpec Products, USA) as described (Rustad et al., 2009) and treated with Turbo DNase (Life Technologies, USA) to remove traces of genomic DNA.

#### cDNA Synthesis and Quantitative (q) Reverse Transcriptase–Polymerase Chain Reaction (RT-PCR)

cDNA was synthesized from 1-mg total RNA using random hexanucleotides and SuperScript III reverse transcriptase (Life Technologies, USA) according to the manufacturer's protocol. qPCR was performed using qPCRmix-HS SYBR (Evrogen, Russia) in a LightCycler 480 Real-Time PCR system (Roche, Switzerland) at the following cycling conditions: 95°C for 20 s, 61°C for 20 s, and 72°C for 30 s repeated 40 times. Three biological and nine technical replicates were used to ensure reproducibility, and the results were analyzed by LinRegPCR v 2014.6. The results were normalized against 16S rRNA to correct the sample-to-sample variation. Calculations were performed according to Ganger et al. (2017) for the relative expression ratio. PCR primers are listed in **Supplementary Table 1**.

### RNase H Assay

The assaywas performed as previously described (Miroshnichenko et al., 2019) with some modifications. Briefly, ASOs (1 µM) were hybridized to complementary RNA (0.1 µM) in 20 mM of Tris–HCl (pH 7.5), 120 mM of KCl, 15 mM of MgCl2, and 0.65 mM of EDTA at 68°C to 25°C for 30 min. After annealing, 0.5 U of RNAse H (Thermo Fisher Scientific) was added, and samples were incubated at 37°C for 30 min to allow RNA digestion in RNA : DNA hybrids. The samples were heated with Gel Loading Buffer II (Thermo Fisher Scientific) at 95°C for 5 min and analyzed by denaturing polyacrylamide gel electrophoresis (PAGE) (15% acrylamide, 7 M of urea). RNA and DNA bands were visualized by SYBR Gold staining under UV light.

### Primer Extension Analysis of *Ald* Transcription

Primer 5′-TGACCGCTTCTTCGAGTTCG was labeled at the 5′ end by [γ-32P]-ATP using T4 polynucleotide kinase (Thermo Fisher Scientific), and 10 pmol was mixed with 1 μg of total RNA, denatured at 70°C for 10 min, and chilled on ice immediately. Reverse transcription was performed using Superscript III reverse transcriptase (Invitrogen, USA) at 55°C for 1 h and 70°C for 15 min according to the manufacturer's instructions. The reaction products were mixed with Gel Loading Buffer II, denatured at 90°C for 5 min, and loaded in a 6% acrylamide sequencing gel. After electrophoresis, the gels were subjected to autoradiography on X-ray films (Retina, USA).

### Western Blotting

Bacterial cells were collected and lysed using Bead Beater (BioSpec Products) and heated for 5 min at 95°C in 2× sodium dodecyl sulfate (SDS) sample buffer (100 mM of Tris–HCl, pH 6.8, 4% SDS, 0.2% Bromophenol Blue, 20% glycerol, and 200 mM of DTT). Protein was measured by the Bradford assay. Equal amounts of total protein (5 μg each) were resolved by SDS-PAGE in a 12% gel and transferred onto Hybond-P membranes (GE Healthcare, UK), which were blocked with 5% w/v nonfat dry milk (Bio-Rad, USA), and incubated with primary antibodies against l-alanine dehydrogenase (Ald) (LS-C184923/70850, 1:10,000; LSBio, USA) and then with secondary anti-rabbit horseradish peroxidase (HRP)-conjugated IgG (Cell Signaling Technology, USA). Specific signals were visualized on X-ray films (Retina, USA) using the Immun-Star HRP Chemiluminescent kit (Bio-Rad) and quantified by densitometry using the GelPro software. The specificity of the LS-C184923/70850 antibody to the Ald protein of *M. smegmatis* was additionally confirmed by reaction with recombinant Ald (**Suppl. Materials and Methods**; **Suppl. Figure 1**).

### Cell Viability Assay

The effect of 2′-OMe PGO and ald-PS on cell viability was determined by the MTS assay (CellTiter 96® AQueous One solution Cell Proliferation Assay; Promega, USA). Mouse macrophage RAW 264.7 cells (ATCC TIB-71, a kind gift from Dr. Elena Svirshchevskaya, Shemyakin-Ovchinnikov Institute of Bioorganic Chemistry, Moscow, Russia) were seeded onto 96-well plates at a density of 10,000 cells/well and incubated at 37°C in a humidified atmosphere of 5% CO2 overnight. After 24 h, fresh medium containing 10 and 20 μM of oligonucleotides was added for 24 or 48 h; untreated cells were used as controls. Cell survival was assessed by adding 20 μL of MTS reagent into each well for 2 h and measuring the absorbance at 490 nm using a microplate reader (Benchmark Plus Microplate spectrophotometer, Bio-Rad). The assay was done in three biological replicates.

## Confocal Microscopy

The infection procedure was performed as previously described (Bettencourt et al., 2010) with some modifications. RAW 264.7 cells were cultured in RPMI-1640 medium (Gibco Europe, UK) supplemented with 10% fetal bovine serum (FBS, Gibco). The day before infection, cells were seeded at the density of 5 × 104 per well in antibiotic-free medium on cover glasses (18 × 18 mm Menzel Gläsercoverslips, Thermo Fisher Scientific) placed in 6‐well culture plates (Costar, USA). *M. smegmatis* cells expressing fluorescent mCherryRed red protein were obtained by electroporation of bacteria with pSMT3-mCherry plasmid (Carroll et al., 2010) and used to infect 80% confluent RAW 264.7 cell monolayers at multiplicity of infection (MOI) 10:1. After 3 h, the medium was removed, cells were washed twice with phosphate-buffered saline (PBS), and fresh medium containing amikacin (100 mg/mL) was added for 3 h to eliminate extracellular mycobacteria. Then, cells were washed twice in PBS and incubated with 10 μM of fluorescein-labeled PGO (FAM-PGO) in RPMI-1640 supplemented with 10% FBS for 3 h. Hoechst 33342 solution (5 µg/mL) was added at the end of incubation for 5 min to stain cell nuclei. After being washed twice with PBS, infected cells were fixed in 1% paraformaldehyde for 10 min and then washed three times with PBS. Mycobacteria were visualized by mCherry fluorescence at 543 nm, FAM-PGO uptake was detected at 488 nm, and cell nuclei were visualized at 405 nm using an Eclipse TE2000 confocal microscope (Nikon, Japan).

## Infection of Macrophages

RAW 264.7 cells were cultured in RPMI-1640 medium (Gibco Europe) supplemented with 10% FBS (Gibco) in 25-cm2 culture flasks (Costar) until 70~80% confluency. *M. smegmatis* cells (OD 0.8, washed PBS) in 2 mL of RPMI-1640 with 10% fetal calf serum (FCS) were added to macrophages at MOI 10:1. After 3 h, the medium was removed, cells were washed twice with PBS, and fresh medium containing amikacin (100 mg/mL) was added for 3 h to eliminate extracellular mycobacteria. Then, cells were washed twice in PBS, and ASOs (*ald*-PGO, scr-PGO) were added to the final concentration of 20 µM. After 4 h, infected macrophage cells were collected, and total RNA was isolated by bacterial cell disruption and phenol-chloroform extraction as described in Section 2.4.

### Statistical Analysis

Statistical analysis was performed using Microsoft office Excel 2007 and GraphPad Prism 6.0 (GraphPad Software Inc., USA). The data were expressed as the mean ± standard deviation. For non-normally distributed data, the Mann–Whitney *U* test was used. Differences were considered statistically significant at \**p* < 0.05. At least three independent experiments were performed for each assay.

## RESULTS

2′-OMe PGOs are structural derivatives of phosphoryl guanidine oligodeoxynucleotides synthesized by replacing deoxyribose with 2-*O*-methylribose (**Figure 1C**) as described earlier (Kupryushkin et al., 2014). An anti-*ald* 2′-OMe PGO (*ald*-PGO) covering the −10- to +13-nt region of the *ald* mRNA was designed to overlap the Shine-Dalgarno sequence and include the start codon. To elucidate the effects of antisense 2′-OMe PGs as anti-mycobacterial agents, we used nonpathogenic fast-growing *M. smegmatis*, which shares about 79% nucleotide sequence homology with *M. tuberculosis*, is very similar to it in cell wall composition and metabolic processes (Chaturvedi et al., 2007), and is widely used as a laboratory model of pathogenic *M. tuberculosis*. The *ald* gene (MSMEG\_2659) encoding alanine dehydrogenase, which may play a role in cell wall synthesis as l-alanine is an important constituent of the peptidoglycan layer (Dave and Kadeppagari, 2019), was selected as a target because its downregulation is not fatal for mycobacteria.

#### Ald-PGO Slows *in Vitro* Growth of *M. smegmatis*

The bacteria were treated with two concentrations of oligonucleotides *ald*-PGO, *ald*-PS, and a scrambled 2′-OMe PGO (scr-PGO) used as control, and the CFUs were counted at 24 and 40 h after treatment. The results indicated that *ald*-PGO was more efficient in suppressing *M. smegmatis* growth than *ald*-PS. At both concentrations used, *ald*-PGO inhibited the growth of *M. smegmatis*, and the effect was statistically significant at 20 μM (54% inhibition compared with untreated control at 24 h, 62% at 40 h), whereas *ald*-PS and scr-PGO did not slow bacterial growth (**Figure 2**). After 48 h, the growth rates of all cultures were approaching those of the control (data not shown), probably because a single addition of *ald*-PGO to the culture was not sufficient to induce prolonged effects.

#### *Ald*-PGO Downregulates Ald Expression at the Translational Level

To determine whether the decrease of *M. smegmatis* growth by 2′-OMe PGO was related to specific inhibition of Ald synthesis, we assessed Ald protein expression in ASO-treated cultures by western blotting. The samples were equalized for total protein amount by the Bradford method, and then evaluated by densitometry (**Supplementary Figure 2**). The *ald*-PGO-treated bacteria showed a much more significant reduction (22% of the untreated control) of Ald expression than PS-treated cells (52%), whereas no apparent difference in Ald protein levels was observed for scr-PGO-treated cells compared with control (**Figure 3A**). These results suggest that the suppression of *M. smegmatis* growth by *ald*-PGO was associated with its specific antisense activity to inhibit Ald expression.

### 2**′**-OMe PGO Did Not Activate RNase H

ASOs can inhibit gene expression by steric blocking or through degradation of the RNA strand in a hybrid duplex by bacterial RNases. To determine the mechanism underlying the downregulation of gene expression by 2′-OMe PGO, we evaluated the sensitivity of ASO-complementary RNA hybrids to RNase H-mediated degradation. An unmodified oligodeoxyribonucleotide (*ald*), *ald*-PS, and mesyl phosphoramidate oligodeoxynucleotide (μ-*ald*) were used as positive controls, as these structure were shown to activate RNase H (Miroshnichenko et al., 2019), whereas oligo-2′-*O*-methylribonucleotide (*ald*-OMe), which forms an RNase H-resistant duplex, was used as a negative control. The

oligonucleotides were hybridized with *ald* RNA and treated with RNase H. As expected, there was a decrease in *ald* RNA band intensity for duplexes with *ald*, *ald*-PS, and µ-*ald*, indicating cleavage by RNase H, whereas no traces of digested *ald*-RNA were observed for the duplex with *ald*-PGO (**Figure 3B**). The PGO-RNA duplex did not penetrate the gel at pH 8.3 because of its neutral charge (Fokina et al., 2018); the visible traces of *ald*-RNA could be attributed to the possible presence of non-hybridized RNA. These results indicate that duplexes of 2′-OMe PGO with the target RNA are stable even under denaturing conditions and that 2′-OMe PGO does not activate RNAse H.

#### Target mRNA Remained Intact After 2**′**-OMe PGO Treatment of *M. smegmatis*

To validate the results of direct molecular interactions on the cellular level, we assessed the presence of intact *ald* mRNA in ASO-treated and control samples by the primer

electrophoresis; qRT-PCR, quantitative reverse transcriptase–polymerase chain reaction

extension method. RNA was isolated from mycobacterial cultures exposed to 20 μM of *ald*-PGO and *ald*-PS and reverse transcribed using a radioactively labeled *ald*-specific oligonucleotide. The identified transcriptional start site coincided with the predicted one at position −27-bp upstream of the ATG start codon (Feng et al., 2002). Furthermore, it was found that full-size *ald* mRNA was significantly decreased in mycobacteria incubated with *ald*-PS, but remained intact in those incubated with *ald*-PGO (**Figure 3C**). These results confirm that *ald*-PGO does not activate RNase H.

#### 2**′**-OMe PGO Treatment of *M. smegmatis* Decreases the Transcription of Its mRNA Target

To quantitatively evaluate *ald* transcription in mycobacteria exposed to ASOs, we performed qRT-PCR. Both *ald*-PS and *ald*-PGO at 20 μM decreased *ald* mRNA expression in *M. smegmatis* after 24-h treatment than did scr-PGO and control cultures (**Figure 3D**). Considering that *ald*-PGO did not activate RNAse H and prevented *ald* mRNA degradation in the duplex (**Figure 4**), these results suggest that 2′-OMe PGO may act as a steric blocker slowing transcription of the target mRNA.

## 2**′**-OMe PGO Is Not Toxic to Macrophages

As antisense 2′-OMe PGOs are supposed to be used against intracellular bacteria, we tested cytotoxicity of *ald*-PGO for macrophages, in which mycobacteria reside and manage to survive. The results indicated that *ald*-PGO in concentrations 10 and 20 μM did not affect the viability of RAW 264.7 macrophages after 24- or 48-h incubation (**Figure 4**). However, *ald*-PS at 20 μM caused a significant decrease in cell proliferation at 48 h than did control (*p* < 0.05).

#### 2**′**-OMe PGO Is Delivered by Gymnosis Into Mycobacteria Residing in Macrophages and Targets Intracellular Mycobacteria

To determine whether 2′-OMe PGO could penetrate the cell wall of intracellular mycobacteria, RAW 264.7 macrophages were infected by *M. smegmatis* expressing red fluorescent protein, treated by fluorescein-labeled *ald*-PGO (FAM-PGO), and analyzed for FAM-PGO uptake by laser confocal microscopy. The images of infected RAW 264.7 macrophages incubated with FAM-PGO (10 μM) for 6 h revealed that the FAM-PGO was abundantly present in the cytosol and, most importantly, was also taken up by mycobacteria, as evidenced by colocalization of FAM-PGO with fluorescent mCherry

protein labeling bacterial cells (**Figure 5A**). These results suggest that 2′-OMe PGO could be delivered to mycobacteria residing in macrophages in the absence of any transfection agent, that is, by gymnosis. To determine whether *ald*-PGO treatment influences the transcription of the target *ald* gene in *M. smegmatis*-infected macrophages, we performed quantitative RT-PCR analysis using total RNA isolated from infected macrophages. qPCR revealed a significant decrease of *ald* transcripts after *ald*-PGO treatment compared with untreated control, or scr-PGO treatment, indicating the intracellular effect of 2′-OMe PGO on mycobacteria (**Figure 5B**).

### DISCUSSION

In this study, we provided evidence that an antisense 2′-OMe PGO could inhibit growth of mycobacteria, downregulate the expression of bacterial genes, and penetrate into intracellular mycobacteria through gymnosis. The 2′-OMe PGO specific for the *ald* gene of *M. smegmatis* effectively reduced the expression of the Ald protein (**Figure 3A**), showing a stronger effect than the *ald*-PS ASO, despite its apparent inability to activate RNase H (**Figure 3B**). It is interesting that, while the target mRNA in the hybrid with PGO remains intact as evidenced by primer extension experiments, its expression is obviously decreased as shown by qRT-PCR analysis (**Figures 3C**, **D**). This effect needs further investigation, but it can be hypothesized that PGO strongly bound to its target RNA may somehow hamper its transcription through steric blocking. Most importantly, 2′-OMe PGO could enter *M. smegmatis* cells residing inside mouse macrophages without damaging the host cells and downregulate expression of a target mycobacterial gene (**Figures 4** and **5**). To the best of our knowledge, this is the first example of an ASO derivative passing successfully through two cellular barriers, the outer membrane of the eukaryotic cell and the bacterial cell wall. Cumulatively, these findings suggest that PGOs may become a potential platform for designing efficient and specific drugs against mycobacteria.

Nucleic acid derivatives have been long considered as promising candidate therapeutic agents because of their unique ability to selectively bind complementary sequences of target mRNA and downregulate gene expression through the antisense mechanism (Uhlmann and Peyman, 1990; Kurreck, 2003). Extensive synthetic chemistry research has produced a number of structural approaches to improve ASO target affinity, biological stability, pharmacokinetic properties, and safety, resulting in PS oligonucleotides, morpholinos, PNAs, and many others (Shen and Corey, 2018). Several oligonucleotide-based therapies have been approved for clinical application (Stein and Castanotto, 2017).

However, in general, oligonucleotides have poor cellular uptake, which is due to their physicochemical properties such as high hydrophilicity and large net negative charge, and mostly cross the cell membrane barrier through various mechanisms of receptor-mediated endocytosis, which have low efficiency (Juliano, 2016). To ensure a desired therapeutic effect, it is often necessary to use special delivery methods, which may be complicated and increase the cost of treatment. Therefore, ASOs that could be delivered by gymnosis, that is, without the aid of any transfection reagents, are currently under development.

To date, there are no examples of successful clinical use of ASOs as antibacterial agents. The emergence of MDR and XDR *M. tuberculosis* strains in recent years requires new more effective therapeutic strategies, which would not promote drug resistance of the pathogen (Parida et al., 2015). In their pioneering work, Harth et al. (2000) used synthetic PS oligonucleotide derivatives against *M. tuberculosis*, demonstrating a fundamental proof of principle by showing that PS ASOs inhibited synthesis of glutamine synthase I (GlnA1) and its activity and suppressed bacterial growth. A more pronounced effect was achieved by using a combination of three different PS ASOs targeting different parts of the *glnA1* gene or inhibiting three mycolyltransferase genes, *fbp A*, *B*, and *C* (Harth et al., 2002; Harth et al., 2007). However, this work was not continued, possibly because of insufficient affinity of PS ASOs to structured RNAs and ineffective penetration through the cell wall of mycobacteria.

More success was gained using PNAs, another class of ASOs conjugated with peptides. Kulyte et al. (2005) demonstrated that a PNA oligomer targeting the *inhA* gene of *M*. *smegmatis* caused a significant retardation of bacterial growth and induced changes in cellular morphology typical for InhA mutants. The potential

of antisense antimicrobials has been demonstrated by studies on peptide conjugates of PMOs, which showed significant inhibitory effects against MDR strains of *Burkholderia multivorans* and *Acinetobacter baumannii* both *in vitro* and *in vivo* (Sully and Geller, 2016; Daly et al., 2017) as well as against intracellular *Toxoplasma gondii* (Lai et al., 2012). However, their activity against mycobacteria is unknown.

A new type of RNA analogues 2′-OMe PGOs (**Figure 1С**) tested in this study have been shown to be electrostatically neutral, possess high resistance to cellular nucleases, and form stable duplexes with target RNA targets (Stetsenko et al., 2014). In the present study, investigation of the biological activity of 2′-OMe PGOs revealed that they exerted stronger inhibitory effects on mycobacteria, had higher affinity to the target mRNA than PS ASOs, and are taken up by intracellular mycobacteria.

Thus, our findings indicate that novel 2′-OMe PGOs have a potential as antisense antimicrobial drugs, which may find application in the future as promising therapeutic agents against TB, especially its drug-resistant forms.

### DATA AVAILABILITY

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

### AUTHOR CONTRIBUTIONS

DS designed the 2′-OMe PGO RNA analogues. TA and DS conceived and designed the experiments. EB carried out oligonucleotide syntheses and analyses. YS, ES, and OB performed biological experiments. TA and DS analyzed the data. TA, ES, YS, and DS prepared figures and graphs. TA and DS wrote the paper. All the authors have read and approved the final manuscript.

### FUNDING

This work was funded by RSF grants 15-15-00121 (to DS) and 18-15-00332 (to TA, red fluorescent *M. smegmatis* strain preparation, infection, and confocal microscopy experiments). EB and DS were partially supported by the Russian governmentfunded budget project VI.62.1.3, 0309-2016-0005 (2017-2020) "Therapeutic nucleic acids."

### ACKNOWLEDGMENTS

The authors thank Dr. B.P. Chelobanov (Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia) for chromatographic analyses and purifications of 2′-OMe PGOs, Dr. E.V. Svirshchevskaya (Shemyakin-Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Moscow, Russia) for RAW 264.7 cell line, and Dr. V.V. Evstifeev (Central Institute for Tuberculosis, Moscow, Russia) for providing pSMT3-mCherry plasmid.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.01049/ full#supplementary-material

### REFERENCES


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

*Copyright © 2019 Skvortsova, Salina, Burakova, Bychenko, Stetsenko and Azhikina. 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.*

# Evolutionary Trends in RNA Base Selectivity Within the RNase A Superfamily

*Guillem Prats-Ejarque, Lu Lu, Vivian A. Salazar, Mohammed Moussaoui and Ester Boix\**

*Department of Biochemistry and Molecular Biology, Faculty of Biosciences, Universitat Autònoma de Barcelona, Barcelona, Spain*

There is a growing interest in the pharmaceutical industry to design novel tailored drugs for RNA targeting. The vertebrate-specific RNase A superfamily is nowadays one of the best characterized family of enzymes and comprises proteins involved in host defense with specific cytotoxic and immune-modulatory properties. We observe within the family a structural variability at the substrate-binding site associated to a diversification of biological properties. In this work, we have analyzed the enzyme specificity at the secondary base binding site. Towards this end, we have performed a kinetic characterization of the canonical RNase types together with a molecular dynamic simulation of selected representative family members. The RNases' catalytic activity and binding interactions have been compared using UpA, UpG and UpI dinucleotides. Our results highlight an evolutionary trend from lower to higher order vertebrates towards an enhanced discrimination power of selectivity for adenine respect to guanine at the secondary base binding site (B2). Interestingly, the shift from guanine to adenine preference is achieved in all the studied family members by equivalent residues through distinct interaction modes. We can identify specific polar and charged side chains that selectively interact with donor or acceptor purine groups. Overall, we observe selective bidentate polar and electrostatic interactions: Asn to N1/N6 and N6/N7 adenine groups in mammals versus Glu/Asp and Arg to N1/N2, N1/O6 and O6/N7 guanine groups in non-mammals. In addition, kinetic and molecular dynamics comparative results on UpG versus UpI emphasize the main contribution of Glu/Asp interactions to N1/N2 group for guanine selectivity in lower order vertebrates. A close inspection at the B2 binding pocket also highlights the principal contribution of the protein β6 and L4 loop regions. Significant differences in the orientation and extension of the L4 loop could explain how the same residues can participate in alternative binding modes. The analysis suggests that within the RNase A superfamily an evolution pressure has taken place at the B2 secondary binding site to provide novel substrate-recognition patterns. We are confident that a better knowledge of the enzymes' nucleotide recognition pattern would contribute to identify their physiological substrate and eventually design applied therapies to modulate their biological functions.

#### *Edited by:*

*Olga N. Ilinskaya, Kazan Federal University, Russia*

#### *Reviewed by:*

*Arun Malhotra, University of Miami, United States Vladimir Alexandrovich Mitkevich, Engelhardt Institute of Molecular Biology (RAS), Russia*

> *\*Correspondence: Ester Boix Ester.Boix@uab.es*

#### *Specialty section:*

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

*Received: 16 April 2019 Accepted: 12 September 2019 Published: 09 October 2019*

#### *Citation:*

*Prats-Ejarque G, Lu L, Salazar VA, Moussaoui M and Boix E (2019) Evolutionary Trends in RNA Base Selectivity Within the RNase A Superfamily. Front. Pharmacol. 10:1170. doi: 10.3389/fphar.2019.01170*

Keywords: RNase, RNA, purine, catalysis, molecular dynamics, evolution, RNase A superfamily

## INTRODUCTION

The interest to solve a biological problem is frequently correlated to its inherent difficulty. When entering the RNA world we are faced with a wide diversity of secondary and tertiary structures. An even higher level of complexity is encountered when trying to identify the rules that guide the RNA binding protein recognition process. During the last decades, many efforts have been applied to unravel the structural determinants for protein RNA recognition (Draper, 1999; Allers and Shamoo, 2001; Draper, 2015; Terribilini et al., 2007). We are currently witnessing significant advances within the RNA field thanks to the novel RNA sequencing methodologies that have laid the path to an RNA-omics era. Nowadays, we have access to many protein-RNA binding predictors (Miao and Westhof, 2016) and the main basic rules that drive the protein–nucleotide interaction process have been identified (Luscombe, 2001; Denessiouk and Johnson, 2003; Morozova et al., 2006; Kondo and Westhof, 2011). The study of RNA cleaving enzymes poses additional complexity. Efficient RNases should first recognize a specific RNA target, and then provide a proper active site configuration to promote catalysis and ensure the proper cleavage of the substrate. A particular pharmacological interest relies on the design of tailored enzymes with specific RNA cleavage targets (Tamkovich et al., 2016). Recent work on RNases' action within a cellular environment is helping to unravel their natural *in vivo* substrates (Honda et al., 2015; Lyons et al., 2017; Mesitov et al., 2017). A proper knowledge of the RNases' active site architecture should lead to the design of specific inhibitors of their biological functions (Chatzileontiadou et al., 2015; Chatzileontiadou et al., 2018).

In this work, we have explored the nucleotide base preference within the vertebrate-specific RNase A superfamily. The bovine pancreatic enzyme RNase A was one of the earliest enzymes to be studied in the 20th century and is still one of the best characterized (Cuchillo et al., 2011). All the family members share a common three-dimensional fold, catalytic triad and mechanism of action on single-stranded RNA. During the last decades, the modular subsite arrangement of RNase A for the recognition of bases, ribose and phosphates has been characterized (Parés et al., 1991; Nogués et al., 1998). The enzyme cleaves the 3′5′ phosphodiester bonds with specificity for pyrimidines at the main anchoring site (B1) and preference for purines at the secondary site (B2) (Richards and Wyckoff, 1971; Raines, 1998). In a previous work, we analyzed the enzyme residues that were reported to participate in the specific binding of adenine (A) and guanine (G) bases at the B2 site among the RNase A superfamily members (Boix et al., 2013). A high evolutionary conservation was observed for B1, whereas a significant variability was visualized for the secondary base selectivity. Interestingly, the observed structural differences at the secondary base site correlate with their substrate specificity and catalytic efficiency (Tarragona-Fiol et al., 1993; Sorrentino, 1998; Boix et al., 2013). Likewise, the analysis of the protein conformational changes induced upon nucleotide binding by NMR and molecular dynamics highlighted an evolutionary trend in base interaction selectivity (Gagné and Doucet, 2013; Narayanan et al., 2017; Narayanan et al., 2018a). Conserved conformational rearrangements upon ligand binding within closely related members suggested a link between shared protein networks and their characteristic biological properties (Narayanan et al., 2018a). The RNase A superfamily includes a series of proteins with antimicrobial and immune-modulatory activities and is considered to have emerged with an ancestral host-defense role (Boix and Nogués, 2007; Rosenberg, 2008; Lu et al., 2018). Family members were classified according to their structural, enzymatic and biological properties into eight canonical types (Sorrentino and Libonati, 1997; Sorrentino, 2010). A better understanding of the structural determinants that govern the RNases' substrate specificity can help us to explain the divergent functionalities within the family.

Here, we have committed ourselves to undertake a comprehensive comparative analysis of representative family members and explore the structural drift that has taken place through evolution to shape the substrate specificity of the secondary base binding site. First, we have performed a kinetic characterization of the first seven human canonical RNases using dinucleotides. Secondly, we have selected representative RNase A superfamily members from lower to higher order vertebrates and have performed molecular dynamics simulations of the proteindinucleotide complexes.

### MATERIALS AND METHODS

#### Expression and Purification of the Recombinant Proteins

RNase A was purchased from Sigma Aldrich. The cDNA for RNase 1 was a gift from Prof. Maria Vilanova (University of Girona, Spain) and cDNA for RNase 5 was provided by Prof. Demetres Leonidas (University of Thessaly, Greece). RNase 4 synthetic gene was purchased from NZYtech (Lisboa, Portugal) and RNase 6 was obtained from DNA 2.0 (Menlo Park, CA, USA). RNase 2, RNase 3 and RNase 7 genes were obtained as previously described (Torrent et al., 2010). The recombinant proteins were expressed and purified as previously described (Boix, 2001; Prats-Ejarque et al., 2016). Briefly, the gene was cloned into the pET11c expression vector (Novagen), the protein was expressed in *Escherichia coli* BL21(DE3) cells (Novagen) and then purified from inclusion bodies. Finally, the protein was purified by cationic exchange FPLC on a Resource S column (GE Healthcare) and lyophilized. Protein purity was confirmed by SDS-PAGE and mass spectrometry.

#### Spectrophotometric Kinetic Analysis

UpA, UpG and UpI (Biomers, Söflinger, Germany) were used as substrates, and the kinetic parameters were determined by a spectrophotometric method as described (Boix et al., 1999b). Assays were carried out in 50 mM sodium acetate, 1 mM EDTA, pH 5.5, at 25°C, using 1 cm path length cuvette. Substrate concentration was determined spectrophotometrically using the following extinction coefficients: ε260 = 24,600 M−1 cm−1 for UpA, ε261 = 20,600 M−1 cm−1 for UpG and ε260 = 16,400 M−1 cm−1 for UpI. The activity was measured by following the initial reaction velocities using the difference molar absorbance coefficients, in relation to cleaved phosphodiester bonds during the transphosphorylation reaction: Δε286 = 570 M−1 cm−1 for UpA, Δε280 = 480 M−1 cm−1 for UpG (Imazawa et al., 1968), Δε280 = 316 M−1 cm−1 for UpI (experimentally determined). Final enzyme concentrations were adjusted depending on the RNase activity for each assayed substrate in a range between 0.005 and 10 μM. The reactions were performed in triplicate with 100 μM of substrate and the activity was normalized at an enzyme/ substrate ratio of 1:100.

#### Molecular Dynamics Simulations

All the molecular dynamics (MD) simulations were performed using GROMACS 2016.2 (Abraham et al., 2015). The force field used was a modification of AMBER99SB (Best and Hummer, 2009). Charges of inosine were derived by R.E.D server (Vanquelef et al., 2011). The modifications of the force field to include inosine parametrization are detailed in the Supplemental Materials (**Figure S1**). All the complexes were centred in a dodecahedral cell with a minimum distance boxsolute of 1.0 nm. The unit cell was filled with TIP3P (transferable intermolecular potential 3P) water (Jorgensen et al., 1983) in neutral pH conditions supplemented with 150 mM of NaCl.

Neighbor search was performed using a Verlet cut-off scheme (Páll and Hess, 2013) with a cut-off of 0.9 nm for both Van der Waals and coulombic interactions. For long range interactions, smooth particle mesh of Ewald (PME) (Darden et al., 1993; Essmann et al., 1995) was used with a fourth-order interpolation scheme and 0.1125 nm grid spacing for FFT. The bonds were constrained with the P-LINCS algorithm (Hess, 2008), with an integration time step of 2 fs.

The energy of the systems was minimized using the steepest descendant algorithm and equilibrated in two steps. First, an initial constant volume equilibration (NVT) of 1 ns was performed with a temperature of 300 K using a modified velocity rescaling thermostat (Bussi et al., 2007). Then, 1 ns of constant pressure equilibration (NPT) was run at 1 bar with a Berendsen barostat (Berendsen et al., 1984) at 300 K and the same thermostat. Finally, 100 ns production runs were performed under an NPT ensemble without applying restraints. Three independent simulations in periodic boundary conditions were conducted for each complex. Dinucleotides were generated by modifying the dCpA ligand of an RNase A–d(CpA) complex (Zegers et al., 1994), maintaining the same initial coordinates.

### RESULTS

#### Comparison of Canonical RNases' Catalytic Activity on Dinucleotide Substrates: A Trend From Guanine to Adenine Selectivity At the B2 Secondary Base Site

In an effort to deepen into our knowledge of the evolutionary pressure that has guided the nucleotide base preference within the RNase A superfamily, we have compared the catalytic activity of the human canonical members on dinucleotides (**Figure 1**).

FIGURE 1 | Sequence alignment of the eight canonical human RNases together with the RNase A superfamily members analyzed by molecular dynamics (sequences correspond to mature proteins, without the signal peptide). Protein regions identified to participate in B2 site are highlighted in yellow: L4, spanning from b2 to b3, end of β6 (residues 109 and 111) and one of the two catalytic histidines together with a close by residue at β7 (residues 119 and 121). TT indicates the presence of a β-turn. Dots label every 10 residues of the reference protein used (Bt-RNase 1, known as RNase A). The disulphide bonds are labeled with green numbers. Full species names are included in Table S1. Labels are as follows: red box with white character for strict identity; red character for similarity within a group; and character with blue frame for similarity across groups. The alignment was performed using *Clustal Omega* (Sievers and Higgins, 2018), and the picture was drawn using *ESPript* (Robert and Gouet, 2014).

First, each canonical RNase was expressed using a prokaryote recombinant expression system. We successfully expressed and purified with high yield the first seven human canonical RNases using the T7 promotor and the pET expression system. Unfortunately, using the same prokaryote expression system we were unable to obtain a properly folded and catalytically active human RNase 8. In fact, inspection of the RNase 8 coding transcript by Rosenberg and co-workers revealed an unusual gene organization and protein disulphide pairing, suggesting a significant functional divergence from the canonical characteristic structure of the family (Chan et al., 2012). The authors do not discard the possibility that RNase 8 is not expressed as a standard secretory RNase. Therefore, we decided to perform our kinetic study using the first seven human canonical RNases. This is the first simultaneous comparison of the catalytic activity of all seven proteins within a single laboratory.

The catalytic activity of the RNases was assayed using dinucleotide substrates, where the first pyrimidine was kept invariable as a uridine and the secondary base was substituted by the natural standard purines and the modified base inosine. Together with the two natural purines incorporated in RNA during transcription, we have also selected inosine, a modified base frequently present in cellular RNA, as one of the main post-transcriptional modifications. Kinetic activity on UpA, UpG and UpI was measured by a spectrophotometric assay and the relative preference for the secondary base was estimated for each protein. Bovine pancreatic RNase A was taken as a reference control.

Interestingly, the respective catalytic activities of the seven human canonical RNases indicate a shift of the secondary base specificity, from a poor A/G discrimination to a pronounced preference for A (**Table 1**). In particular, the human RNase 5, which is the canonical member more closely related to ancestral RNases (Sorrentino, 2010), shows only a mild preference for adenine over guanine. In turn, the pancreatictype RNase 1 shows a significant preference for adenine at B2 position. Last, the more evolved RNase subgroups (types 2/3 and 6/7) do not have any detectable activity using UpG as a substrate (**Table 1**, **Figure 2**).

On the other hand, when we analyze the kinetic characterization of other family members available in the literature, we can infer

a shift at the substrate secondary base predilection, from lower to higher order vertebrates, from guanine to adenine (Boix et al., 2013). Basically, the characterized fish, amphibian and reptile RNases show a marked preference for G at B2 site (Hsu et al., 2003; Ardelt et al., 2008), while mammalian prefer A (Richards and Wyckoff, 1971; Zhao et al., 1998; Prats-Ejarque et al., 2016). We can group the family members, according to their relative activity on dinucleotide substrates, within three main subcategories by their base preference at the B2 site: G > A, G ~ A and A > G (**Figure 2**). The results suggest that an evolutionary pressure has taken place to promote selectivity for the adenine base within the family's more recently evolved members, coming from an ancestral precursor with a marked preference for guanine.

Last, we have studied the RNases' activity on UpI dinucleotides. Inosine (I) was selected as an appropriate model to inspect the particular effect of the presence of a C=O group at the purine C6 atom and the influence of the NH2 group at the C2 position, in comparison to the other two purine base structures. Detectable activity for the inosine dinucleotide was mainly registered for the RNases 1, 2, 4 and 5 (**Table 1**). Overall, kinetic results indicate that no important differences are observed between the proteins' enzymatic activity on UpG and UpI, although a slight preference for I over G is shown. Interestingly, the family members that have a closer relationship to lower order vertebrates (RNases 1, 4 and 5) present a significant activity against dinucleotides with inosine at the B2 position, but no detectable activity in the presence of a guanine. The results suggest that A/G discrimination within the RNase A superfamily relies partly in the recognition of N1/N2 group.

#### B2 Base Selectivity Within the RNase A Superfamily by Molecular Dynamics

Following, to complement the enzymatic characterization of the canonical RNases, we performed a comparative analysis within the RNase A superfamily by molecular dynamics. To gain insight into the structural determinants of the protein recognition pattern at the B2 site, we have selected here representative members for each vertebrate family subgroup. Ten representative RNases were chosen and their binding mode to dinucleotides was predicted by MD simulations. **Figure 2** illustrates the selected proteins and their evolutionary relationships. When no

#### TABLE 1 | Kinetic activity of RNase A and the human RNases 1–7.


*The reactions were performed using 100 μM of substrate. Initial velocity (V0) of dinucleotide phosphodiester bond cleavage is indicated. The average of three replicates is shown. Standard error of the mean is shown. n.d.: not detected at the assayed conditions. Bt, Bos taurus; Hs, Homo sapiens.*

percentage of trees in which the associated taxa clustered together is shown next to the branches. Initial tree(s) for the heuristic search were obtained automatically by applying Neighbour-Join and BioNJ algorithms to a matrix of pairwise distances estimated using a JTT model, and then selecting the topology with superior log likelihood value. The tree is scaled, with branch lengths measured as the number of substitutions per site. Evolutionary analyses were conducted in MEGA X (Kumar et al., 2018). A more complete phylogenetic tree of RNase A family members is included in Figure S2. Blue star highlights the selected representative RNases to perform MD simulation runs.

solved 3D structure was available (RNase 4 of *Ornithorhynchus anatinus* and RNase 1 of *Iguana iguana*), a prediction model was generated using the *Modeller* software by the *ModWeb* server (Webb and Sali, 2016). From lower to higher order vertebrates, the following organisms were analyzed: *Danio rerio* (Dr), *Rana pipiens* (Rp), *Iguana iguana* (Ii), *Chelonia mydas* (Cm), *Gallus gallus* (Gg), *Ornithorhynchus anatinus* (Oa), *Bos taurus* (Bt) and *Homo sapiens* (Hs). Representative organisms were selected based on the available information on the evolutionary relationships and previous structure-functional characterization studies (Goo and Cho, 2013). We also indicate, when known, the canonical type of each selected RNase (Sorrentino and Libonati, 1997). Within the placental mammals, we have included two representative human members (RNases 2 and 6; UniProtKB P10153 and Q93091), which are expressed during infection and inflammation and are endowed with a high catalytic activity. We have also selected the bovine pancreatic RNase, or RNase A (UniProtKB P61823), which is the family reference member and one of the best characterized enzymes (Cuchillo et al., 2011). Bovine pancreatic RNase belongs to the RNase 1 type. On the other hand, for early mammals, we selected the platypus (*O. anatinus)*, an egg-laying animal and precursor to higher order vertebrates, before divergence of placental RNases. Accordingly, the platypus RNase belongs to type 4 (UniProtKB F6ZXU1), and was previously identified as the predecessor of higher order mammalian RNase types (Goo and Cho, 2013). Following,

representative members of avian, reptiles, amphibian and fishes were chosen, based on the availability of previously solved 3D structures. Chicken RNase 1 was taken (UniProtKB P27043) as the only member with a known 3D structure (Lomax et al., 2014). In turn, reptiles have been represented by turtle (UniProtKB P84844) and iguana (UniProtKB P80287) (Nitto et al., 2005). Next, we selected the northern leopard frog (*R. pipiens)* RNase (also named Onconase, UniProtKB P22069), which has been extensively characterized because of its antitumoral properties (Boix et al., 1996; Lee and Raines, 2003; Lee et al., 2008). Lastly, for fish representative sequences, we selected *D. rerio* RNases (Dr-RNase 1 and Dr-RNase; UniProtKB A5HAK0 and E7FH77), also named as zebrafish RNases 3 and 5 respectively. Both RNases were previously reported to display a high catalytic activity in comparison to other fish homologues (Cho and Zhang, 2007; Pizzo et al., 2011). In particular, the zebrafish 5 (Dr-RNase) was classified as one of the most ancestral family members, showing a high catalytic activity along with both antimicrobial and angiogenic properties (Pizzo et al., 2011). In all cases, previously reported 3D structures were taken as a reference, except for the platypus RNase, where a prediction model had to be generated.

To compare the RNases' selectivity at the B2 site, the three dinucleotides, UpA, UpG and UpI, were selected (see **Figure S3** for atom nomenclature). Molecular dynamics were performed using GROMACS software as detailed in the methodology. Triplicates for each protein complex were carried out at 100 ns. The RMSD between the dinucleotide positioning during the simulation is shown in **Figure S4**. The following common criteria were established to confirm at the end of each modelling run that the nucleotide is positioned in a productive orientation, favorable for catalysis: phosphate location at the RNase catalytic triad and pyrimidine proximity to B1 site. Equivalent residues to RNase A (H12/K41/H119 at the catalytic triad and T45 at B1 site) were taken as a reference for each protein.

The dinucleotides' RMSD fluctuations during each production run indicate a reduced substrate mobility, oscillating within a value range of 0.1–0.4 nm (**Figure S4**). The total hydrogen bond interactions per residue were calculated for each simulation and expressed as a fraction of occurrence. **Figure S5** illustrates the interacting residues with the purine moieties.

Overall, we observe at the end of each simulation run a similar productive positioning of the dinucleotides at the active site cleft for most of the studied proteins (**Figures 3** and **S6**). However, comparison between all different RNase–nucleotide complexes and among triplicates highlights that most variability is located at the purine moiety (**Figure 4**). Likewise, time course analysis for each dynamic run shows significantly much higher mobility for the purine nucleoside in comparison to the pyrimidine main nucleoside and phosphate portions. We can confirm that the protein phosphate p1 and base B1 sites are mostly conserved among all the family members and provide stronger and more specific interactions.

Following, we have analyzed the specific binding interactions at the B2 purine portion. Specific binding residues at the B2 site were identified. In the majority of complexes, the purine base is fixed by the L4 loop and β6 strand structures (**Figure 4**). Contribution of each interacting residue was monitored as a function of time. Each run was subdivided into initial, central and late periods. Although some mobility of the substrate positioning is observed during the 100 ns MD production runs (**Figure S4**), overall no major significant differences are identified as a function of time. The most representative interacting residues and atom types involved in each modelled complex are summarized in **Figure 5**.

**Figure 5** illustrates the main residues that contribute to B2 base recognition. The figure indicates the main residues that were found involved in interactions with the purine ligand for at least one third of the total 100 ns molecular dynamics run. We observe the contribution of polar and charged residues that act as acceptors/donors to purine representative groups. We can identify the protein residues that can provide a bidentate anchoring with the purine base and selectively interact with unique base groups. In particular, we find specific discriminators for adenine (N1/N6 and N6/N7 groups) versus guanine (N1/O6 and N1/N2 groups). Likewise, discrimination between guanine and inosine binding was identified by looking for the residues with specific interactions at the base N1/N2 group, unique to guanine.

Each studied family member was analyzed taking bovine pancreatic RNase A sequence numbering as a reference (see **Table S2** and **Figure 1**) (Raines, 1998; Boix et al., 2013). The adenine base is fixed in bovine RNase by residues Asn67, Gln69, Asn71, Glu111 and His119 (see **Figure 5**). The reliability of the dynamic simulation was first evaluated by comparing the obtained results for RNase A using UpA with the previous structural work by

X-ray crystallography on RNase A–dinucleotide complexes (Boix et al., 2013). In particular the adenine binding residues identified by molecular dynamics were compared with the RNase A–d(CpA) complex (Zegers et al., 1994), where the same residues for adenine binding had been identified (Asn67, Gln69, Asn71, Glu111 and His119). Specific bidentate interactions for adenine are provided by Asn71/Gln69 at N1/N6, Asn67 at N6/N7 and Glu111 at N6. In particular, our molecular dynamics results corroborate the key contribution of all Asn71 counterparts in mammalian members for adenine specificity. On the other hand, we observe the flexibility of residues such as Glu111, which can offer a bidentate anchoring at either the NH2 group at C6 position in adenine or at N2/N1 groups in guanine (**Figure 5**).

In addition, we observe the contribution of the His119 catalytic residue by π-π interactions with the purine 5-membered ring in all the predicted complexes for any of the three assayed dinucleotides (**Figure 5**). Previous structural studies have revealed that the catalytic His119 in the free protein can adopt two conformations (A and B), where only one rotamer (A) is compatible with catalysis and purine interaction (Berisio et al., 1999; Merlino et al., 2002). Favored stacking interactions of the His imidazole with the purine ring are suggested to participate in nucleotide discrimination (Gagné and Doucet, 2013). In our molecular dynamics study we cannot find any significant differences between the complexes obtained with any of the three dinucleotide types. On the other hand, significant differences are observed for some particular RNases, where the purine ring is also establishing cation-π interactions with other residues, in particular arginine (such as Arg68 in Hs-RNase 2, Arg66 in Gg-RNase, Arg117 in Cm-RNase and Arg8 in Dr-RNase 1 (see

interactions are detailed for each location. The box labeled with a star includes the shared common van der Waals interactions for the three base types.

are listed according to RNase A numbering reference (see Table S2 for structural residues overlapping between the analyzed homologues). Polar and electrostatic

**Figure 5**). Overall, we observe that most differences among the studied family members are located at the L4 Loop. The loop mobility is restricted by a disulphide bridge (Cys65–Cys72 pair in RNase A), that is conserved in most mammalian RNases (except in RNase type5/angiogenin-like), but absent in all the non-mammalian vertebrate groups (see **Figure 1**).

We can conclude from the analysis of predicted protein– dinucleotide complexes that the main key residues for purine interactions (Asn71 and Glu111) are mostly conserved among all the studied family members, although distinct binding modes are identified depending on the nature of the purine base. Asn71 in RNase A, and equivalent residues both in human and platypus proteins, specifically bind by a bidentate interaction at the N1/ N6 of the adenine ring. Likewise, the Asn side chain can establish equivalent interactions for guanine or inosine binding, by shifting their NH and C˭O amide groups and thereby interacting with the respective N1/O6 groups. However, these interactions are not so often observed for guanine/inosine interaction and frequently only the Asn binding to the O6 group is identified.

When we inspect the non-mammalian vertebrate members, we find a similar scenario: an Asn residue (Asn71 RNase A counterpart) can also interact with both N1/N6 groups for A and N1/O6 in G/I in turtle, frog and fish proteins. Significant differences are found in chicken RNase, where an Arg is located at the same position. On its turn, the nearby residue Gln69 would contribute to provide a specificity for adenine. A Gln at this position is only present in the pancreatic RNase 1 type. Substitutions of Gln by an Argin Hs-RNases 2 and 6 and platypus RNase favor the bidentate interaction with G/I at the N7/O6 group. The equivalent counterpart in fish is an Asn (), which shows a preference for guanine/inosine binding. No equivalent residues are found in any other lower order vertebrates, due to a deletion in the loop L4 region from residues 65 to 71 (see **Figures 1** and **S7–S11**). In addition, we find another Asn residue in mammalian RNases that is also favoring the adenine versus guanine binding: Asn67 (**Figure 5**). In this case, the Asn is providing a bidentate interaction to N6/N7 adenine groups. The presence of an additional Asn is also found in zebrafish RNase 5 (Dr-RNase) but is missing in all the other studied lower order vertebrates. Interestingly, the shorter L4 loop version in the fish protein still permits the appropriate Asn positioning.

The molecular dynamics results also highlight two other protein regions, which are also participating in the purine binding: residues 109–111 (β5) and 119–121 (β6). In particular, we observe the main contribution of Glu111 in Bt-RNase A and the respective Glu/Asp counterparts in the other studied family members (**Figure 5** and **Table S2**). Both the Glu/Asp bidentate anionic side chains are observed to bind at both the NH2/N6 adenine and the N2/N1 guanine specific groups. However, Glu substitution by an Asp residue (found in Hs-RNases 2 and 6) prevents, or reduces drastically, the base interactions. Similar interactions at the adenine N6 NH2 group and the guanine N2/ N1 group are established by Asp121 at Bt-RNase A and their counterparts in mammals and chicken RNases. Although a Glu/ Asp is present in all the studied proteins, frog RNases show significant differences. Interestingly, the zebrafish 3 counterpart (Glu122) interacts with guanine base but is not involved in adenine binding. Finally, another substitution that is observed to favor guanine binding in non-mammalian RNases is Ala122 to Arg. The Arg counterpart residues in fish and turtle RNases can interact by bidentate interactions with the O6/N7 group of the guanine/inosine bases (**Figure 5**).

Overall, although key residues for purine binding are mostly conserved in all the studied members, such as Asn71, His119 and Glu111, our molecular dynamics analysis indicates that distinct binding modes could promote a shift from G to A at the B2 site.

#### An Evolutionary Trend Shaping the B2 Selectivity Within the RNase A Superfamily Lineage

To validate the significance of the residues identified by MD to participate in purine recognition, we have supplemented our study with the comparative analysis of other family member close homologues. Accordingly, each representative member analyzed by MD simulations has been compared within its own vertebrate subgroup. By close inspection of sequence alignments, we have identified the counterpart to the key residues for binding of a purine at B2 location. **Figures S7–S11** include the respective sequence alignments within each vertebrate subgroup. The relationships between all the aligned sequences of family homologues are illustrated in the phylogenetic tree included in **Figure S2**.

First, we have analyzed the fish RNase sequences, taking as a reference *D. rerio* RNase 1 (Pizzo et al., 2011), also named zebrafish RNase 3 (ZF3). Acharya and co-workers solved the crystal structure of this RNase together with a polymorphism variant (Kazakou et al., 2008). The chosen protein structure corresponds to the variant identified as ZF3e. Overall, the researchers identified five protein variants, with substitutions at six sequence locations. Among them, we observe that one of the residues involved in the purine binding (Arg123) is only present in the ZF3e polymorphism and is substituted by a Lys in the other variant. On the other hand, comparison with the other fish RNase sequences (**Figure S7**) highlights the presence of one or two conserved Asn residues at L4 loop region. The loop is present in fishes in a short-reduced version in comparison to the extended version present in more evolved mammal RNase types: 2/3–6/7/8 (**Figure 1**). However, the Asn residue at position 72/74 (corresponding to positions 67 and 71 in RNase A) can also participate in the adenine interaction but would preferably interact with the N1-O6 group of a guanine. Noteworthy, several fish RNase sequences display an Asp at 74 position, which according to molecular dynamics results is a suitable binder for guanine. Two other anionic residues at the protein C-terminus are key for the studied RNase complexes; that is Glu114 and Glu122 (corresponding to Glu111 and Asp121 in RNase A counterparts). While most fish RNases show a Glu at 114 position, we also find in some cases the presence of an Asp. This is the case of zebrafish 5 (Pizzo et al., 2011), which was reported to have a relative much higher catalytic activity than the other characterized fish RNases (Pizzo et al., 2011). Likewise, residue Glu122 is either conserved or substituted by an Asp residue. Finally, molecular dynamics reveal the presence of an Arg residue at the zebrafish proteins' N-terminus that shows favored interactions to guanine and inosine. The Arg is only present in about 50% of the analyzed fish sequences.

Following, we inspected the residues potentially involved in purine binding in amphibians. In this vertebrate group we also observed a short version of the L4 loop. However, in comparison to fish RNases, the analyzed amphibian members show a less optimal loop conformation. The loop is orientated to the opposite direction respect to RNase A, and lacks one of the key Asn found in mammalian RNases. In particular, in northern leopard frog *R. pipiens* RNase (Onconase) we can identify Asn56 (counterpart of Asn71 in Bt-RNase A) but no other equivalent residues in the region (**Figure 1**). Comparative structural alignment only reveals the presence of a conserved Glu residue at position 91 (Glu111 counterpart in RNase A). Molecular dynamics results on Onconase interaction with dinucleotides have been compared with the previous reported solved crystal structure in complex with a tetranucleotide (Lee et al., 2008). Raines and co-workers studied in detail the enzyme binding to the d(AUGA) substrate analogue and observed that while an equivalent binding pocket is conserved for the pyrimidine base at B1, significant differences are found for the B2 site. In particular, specific bidentate interactions of Glu91 with the guanine base were identified at B2 position. Moreover, the authors confirmed by site-directed mutagenesis that this residue was responsible for the frog RNase preference of guanine over adenine. In addition, the authors also highlighted the importance of the nature of the nearby residue located at position 89 (Onconase counterpart of Ala109 in RNase A). Ala109 is conserved in all mammalian and most reptile sequences but shows a significant variability in fishes and amphibians. Interestingly, ZF3 presents an Ala at this position, as observed in mammals, whereas other fish RNases have a polar or cationic residue (Thr/Lys or Arg), as observed in frog RNases. Substitution of Thr89 in Onconase by an Asn residue reduced the enzyme's G > A preference. The authors suggested that long-range electrostatic interactions were key for the enzyme turnover activity on cellular RNA substrate in physiological conditions (Lee et al., 2008). The hypothesis was further backed up by recent NMR and molecular dynamics studies by Doucet and collaborators, that emphasized the key role of network interactions connecting distant protein residues (Narayanan et al., 2018a). Interestingly, site-directed mutagenesis in Onconase revealed also the contribution of the N-terminus in the B2 base discrimination (Lee et al., 2008). In particular, insertion of an Arg at position 5 is significantly enhancing the frog RNase catalytic activity. Likewise, in our molecular dynamics analysis we observe equivalent Arg residues at the protein N-terminus of turtle and fish RNases that contribute to purine binding (**Figure 5** and **Table S2**).

Molecular dynamics results of Onconase were also compared with the structural information reported for bullfrog (*R. catesbeiana*) RNase purified from oocytes (RC-RNase), the most catalytically active frog RNase (Chang et al., 1998; Lee and Raines, 2003). A structural study by NMR of bullfrog oocyte RNase analyzed the enzyme interaction with tetranucleotides (Hsu et al., 2015). The authors reported a much higher catalytic activity for oocyte RC-RNase in comparison to RC-RNase 2 and RC-RNase 4. The contribution of the L4 loop to guanine binding was also highlighted, although distinct conformations are observed among the bullfrog RNases that could account for the higher catalytic activity displayed by the oocyte RC-RNase. When we overlap the reported NMR structures with our modelled structures in complex with dinucleotides, we also observe that the oocyte RC-RNase is the only one that has an Asn residue at an equivalent position to Asn71 in RNase A, that can establish interactions with the N1-O6 group of the guanine. Therefore, the higher catalytic efficiency of bullfrog oocyte RNase respect to Onconase could be mostly attributed to residue Asn57, which is substituted by a Ser in the latter. When we compare the sequence identities of the distinct frog RNases, we observe a high variability at the loop L4, where Asn residues are mostly substituted by either a Ser or an Asp (**Figure S8**). Besides, presence of Pro and short amino acid insertions in other amphibian RNases might also modify significantly the interaction at this site. Interestingly, whereas most *R. catesbeiana* RNases show a particular four amino acid insertion, we found several *Xenopus* species that display an alternative loop version, with a slightly extended insert (**Figure S8**).

In turn, reptiles present a short version of the L4 loop (see **Figure S9**), with a similar length to the one observed in fishes, although encompassing a higher sequence divergence at the region. In particular, most species include only one Asn within the region. In our molecular dynamics study of iguana and turtle RNase-dinucleotide complexes we can identify one Asn (Asn68 in turtle RNase and Asn67 in iguana) equivalent to the Asn71 counterpart in RNase A (Table S2). Close inspection of sequence alignment identifies few reptile species with two Asn residues at 67/71 positions (such as the species of the *Micrurus* or *Boiga* genera), whereas other species show an Asn to Asp substitution at position 71. However, we observe an overall higher variability at L4 loop, which incorporates non-conserved substitutions (**Figure S8**). We have also analyzed within reptiles the other residues that were identified in turtle or iguana to potentially participate in binding at the B2 site (**Figure 5**). Rosenberg and co-workers characterized the RNase from iguana, which is mostly expressed in the pancreas and displays a significantly high catalytic activity (Nitto et al., 2005). In the present work, productive binding conformations obtained by molecular dynamics of turtle RNase with dinucleotides highlight the contribution at the protein C-terminus of Asp116 and Arg117 (Asp116, counterpart of Asp121 in RNase A, is only present in few reptile sequences). In turn, the presence of an Arg at position 117, not shared by all the family homologues, is rare.

Avian RNases present the shortest L4 loop version that incorporates the most significant deviation from the L4 loop consensus sequence (**Figure S10**). Most L4 sequences do not include any Asn residue. In our modelled complex of chicken RNase, Asn65 is equivalent to Asn67 in RNase A. However, we did not observe any direct participation of Asn65 in purine binding. In turn, the neighboring residue Arg66 is significantly participating in B2 binding and was observed to bind to any of the three purine bases. Arg66 position can be equated to Bt-RNase A Asn71 counterpart, although the loop conformation is very divergent at this region (Table S2). Arg66 is only found in few bird sequences but is located close to Arg66/70 in some other mammalian RNases (Hs-RNase 2, Hs-RNase 6 and Oa-RNase in our study; **Figure 5** and **Table S2**). Most strikingly, there is no equivalent Glu/Asp counterpart to Bt-RNase A Glu111. On the contrary, a Trp is present at that location in the studied chicken RNase (**Figure 1**). Trp is conserved in some chicken and snake (*Boiga*) RNase sequences (**Figures S9**, **S10**). In other avian sequences we find another bulky hydrophobic residue, followed by Asp, which might substitute the Glu111 function. Interestingly, in our molecular dynamics study we find the contribution of stacking interaction of Trp105 with the purine base. On the other hand, the presence of a residue equivalent to Asp121 is only observed in some of the sequences, whereas others show a substitution by an Ala. In any case, a higher proportion of non-productive dinucleotide binding is obtained by molecular dynamics (>75% of all run assays) in relation to the other studied members (<30% in fish, frog, turtle or platypus), which might be attributed to the chicken RNase's different conformations of the L4 loop and the presence of Arg66 and Trp105, that tend to establish stacking interactions with the purine base. Noteworthy, Rosenberg and co-workers performed a comparative study of available sequences for chicken RNases and concluded that the evolution within this group of proteins might not respond to functional constraints directly related to the enzyme catalytic activity (Nitto et al., 2006). Comparison of two chicken leukocyte RNases identified key regions for either antimicrobial or angiogenic activity. By construction of hybrid proteins, they concluded that following a duplication event, a selective evolutionary pressure unrelated to the protein enzymatic activity had taken place. In our study, we have selected the only available 3D structure of a chicken RNase. Unfortunately, this RNase corresponds to the angiogenin-type RNase instead of the other characterized chicken RNase (leukocyte RNase-A2 or RSFR-RNase), which displays a much higher catalytic activity (Nitto et al., 2006).

In contrast to lower order vertebrates, we observe that all mammalian RNases, except the RNase 5 type, share an extended L4 loop fixed by a disulphide bridge (Cys65 and Cys72 in RNase A). In contrast, all the non-mammalian vertebrate RNases have either a single Cys or none at this location (**Figure 1**). Mammalian RNases' extended L4 loop includes in all cases the RNase A Asn71 counterparts and, in the majority of cases, the RNase A Asn67 residue. On the other hand, more variability is observed at 69 position, where either a Gln, Ser or Arg is found (**Figure S11**). On its side, Glu in position 111 is mostly conserved but can also be substituted by an Asp or even a Lys residue. Our MD results indicate that the presence of the shorter Asp residue is associated in Hs-RNase 2 and Hs-RNase 6 with scarce or null interactions with the purine (**Figure 5**). Structural crystallographic data on Hs-RNases 2 and 3 complexes (Mohan et al., 2002) also highlighted the different interaction mode of Asp at this position. Hs-RNase 6 structural studies also indicated that the Glu to Asp substitution might significantly modify substrate specificity (Prats-Ejarque et al., 2016). On the other hand, it is also interesting to note the presence of an Arg at position 122 (Arg132 in RNase 2), which is shared with some fish and other lower order vertebrates (**Table S2**) and might provide a significantly differentiated specificity. Overall, we can conclude that counterpart residues to Asn71 and Glu111 in Bt-RNase A, shared by all the mammalian RNases, were already present in most ancestral RNases; but the observed purine selective specificity is modulated in each family member by complementary interactions of environment residues.

#### DISCUSSION

The RNase A superfamily is currently a reference model for evolutionary and enzymology studies. Although a wealth of information is available on ruminant evolution and the pancreatic-type RNases (Beintema and Kleineidam, 1998; Goo and Cho, 2013; Xu et al., 2013; Lomax et al., 2017), a comprehensive full understanding of the whole family is still missing. Following a pattern characteristic of host-defense proteins, the RNase A family has undergone frequent duplication and gene sorting events (Rosenberg et al., 1995; Zhang et al., 2000; Zhang et al., 2002; Liu et al., 2015). Many studies have tried to unveil the structural determinants for the distinct RNases' biological activities (Lu et al., 2018); however, we find much less information on the evolutionary trends that shaped the family's enzymatic diversity. Nonetheless, the understanding of the evolutionary processes that determined the enzymes' substrate selectivity is key to unravel their physiological roles. Distinct nucleotide specificities should respond to an adaptation to their respective biological functions (Narayanan et al., 2018b). Undoubtedly, mastering the structural basis for protein nucleotide recognition is essential to assist the design of novel anti-infective and immunomodulatory drugs.

Here, we have compared for the first time the catalytic activity of the human canonical RNases. The analysis of all the recombinant proteins, obtained by the same expression system and using the same kinetic characterization methodology, ensures a reliable comparative evaluation of their respective efficiencies. To unravel the enzyme specificity for the binding of the purine secondary base, we have tested the respective catalytic activity of the distinct canonical RNases using UpA, UpG and UpI dinucleotides as a substrate. Interestingly, when we compare the A/G ratio at B2 site for the studied seven canonical RNases (**Figure 2**), we observe a pronounced evolutionary tendency from guanine to adenine preference. Previous evolutionary studies identified within the RNase A superfamily the phylogenetic relationship between the eight canonical subtypes (Cho and Zhang, 2007). By comparative analysis of the family coding sequences we can order the different RNase types from ancestral to modern as follows: 5, 4, 1 and 6/7/8–2/3 group (Zhang, 2007; Sorrentino, 2010) (see the family phylogenetic tree in **Figure S2**). Kinetic results of human canonical RNases follow the same ordering when considering the UpA/UpG ratio (Table 1). The present result is in agreement with previously reported kinetic data, where lower order vertebrates, such as amphibians and reptiles, show a preference for G (Liao, 1992; Irie et al., 1998), while mammalian RNases have a clear preference for A (Boix et al., 2013). In addition, the recent kinetic characterization of human RNase 6 corroborated the previously reported preference for adenine at B2 for human RNases 2 and 3 (Boix et al., 1999a; Sikriwal et al., 2007; Prats-Ejarque et al., 2016). On its turn, human RNase type 5 shows a much less pronounced preference for adenine over guanine (Acharya et al., 1994; Shapiro, 1998). Early kinetic characterization of human RNase 5 (angiogenin) already reported its poor discriminating ability on the purine located at B2 position (Shapiro et al., 1986; Harper and Vallee, 1989). Vallee and co-workers engineered an RNase 5 hybrid protein by replacing the L4 loop residues 60-70 with the RNase A counterparts, successfully enhancing the enzyme catalytic activity on UpA dinucleotides (Harper and Vallee, 1989). Further work by site-directed mutagenesis in angiogenin suggested that an Asn at Gln69 position in RNase A would provide most of the purine selective binding. On the contrary, replacement of the Glu111 RNase A counterpart (Glu108 in RNase 5) did not significantly alter the enzyme B2 selectivity (Curran et al., 1993). Indeed, structural and molecular dynamics simulation studies indicate that Glu111 in RNase A could contribute to either adenine or guanine binding by alternative modes, by direct or water-mediated interactions. The nature of the nearby residue (109 in RNase A) could determine the potential participation of the corresponding Glu residue (Glu111 in RNase A) in the purine binding. The hypothesis was elegantly confirmed by sitedirected mutagenesis studies in Onconase (Lee et al., 2008). Raines and co-workers demonstrated that the B2 site specificity could be shifted from guanine to adenine preference by impeding the long distance network interactions that Glu establishes for the purine recognition (Lee et al., 2008). Likewise, substitution of Glu111 by the shorter Asp side chain in human RNases 2 and 6 could enhance the adenine versus guanine discriminating power in relation to ancestral RNases, such as RNase-type 4 and 5, as observed in our kinetic comparison studies (**Table 1**, **Figure 2**).

In an effort to unravel the structural determinants underlying the observed differentiated kinetic behaviors, we have carried out a molecular dynamics analysis of RNase-dinucleotide complexes. Representative family members were chosen from lower order vertebrates to placental mammals. Overall, molecular dynamics corroborate the observed shift from guanine to adenine preference by kinetic analysis (**Table 1**, **Figure 2**). Notwithstanding, results also highlight that conservative sequence identities are frequently not accompanied by equivalent substrate binding. A similar conclusion was reached by NMR analysis of RNases' nucleotide binding (Narayanan et al., 2018a). Therefore, no straightforward conclusions can be directly inferred from the identification of individual interactions to nucleotides by mere structural overlapping analysis. On the other hand, although molecular dynamics considers the protein–ligand complex as an entity in motion and provides the equivalent freedom and flexibility that could be found in experimental conditions, the methodology has also its own limitations when trying to simulate the enzyme behavior. Fortunately, Bt-RNase A, the family reference member, has been one of the best enzymes ever characterized (Raines, 1998; Cuchillo et al., 2011). RNase A was classified by earlier studies as an almost "perfect" enzyme, where the transphosphorylation state is not limited by the transition state (Albery and Knowles, 1976). Raines and coworkers analyzed the behaviour of RNase A on UpA substrate by experimental kinetics and concluded that the cleavage efficiency is mostly limited by the substrate desolvation (Thompson et al., 1995). Early crystallographic and NMR studies of RNase A in complex with mono-, di- and tetranucleotides identified the main residues that conformed the RNase A substrate binding subsites (Fontecilla-Camps et al., 1994; Nogués et al., 1998; Hsu et al., 2015).

Notwithstanding, despite the RNase protein small size and structure stability, that facilitated the pioneer biochemistry works during the first half of the 20th century, the polymeric nature and structural complexity of the substrate is still challenging the enzymologists. In this context, it is important to analyze the protein family members as a whole dynamic entity. The protein has a kidney-shaped structure conformed by two domains that delimitate the catalytic active site groove. The open and closed conformation of enzymes were compared in the presence of nucleotide ligands (Watt et al., 2011; Gagné and Doucet, 2013). Key residues involved in the RNase protein motion would have co-evolved to shape the enzyme catalytic efficiencies, as described for other enzyme families (Maguid et al., 2006; Ramanathan and Agarwal, 2011; Narayanan et al., 2018a). Within the RNase A superfamily we observed the conservation of key domains involved the protein motion (Merlino et al., 2003; Gagné and Doucet, 2013). Notwithstanding, comparative studies from lower to higher order family members infer an inverse relationship between the protein's structural rigidity and its catalytic efficiency (Merlino et al., 2005; Holloway et al., 2011).

Although our molecular dynamics runs using dinucleotides are overall in agreement with the reported crystal complex structures (Fontecilla-Camps et al., 1994; Zegers et al., 1994; Leonidas et al., 2001; Mohan et al., 2002; Lee et al., 2008), we do observe some significant differences. This might be due to the allowed protein flexibility during the molecular dynamics simulations, a fact that could enable a better accommodation of the nucleotide substrates. Besides, MD studies permitted us to work with the natural enzyme substrates, rather than the analogues, commonly used in crystallographic studies. On the other hand, NMR titration studies using mononucleotides could only mimic the enzyme interactions that were to take place with the enzyme reaction product (Narayanan et al., 2018a). Interestingly, when we analyze the results of our molecular simulation, we can observe significant differences among the residues that participate in the distinct periods of the reaction. Mostly, interactions with the purine base are frequently lost at the end of the production run. Interestingly, in our modelling studies we observe how the substitution of Glu111 by an Asp residue in human RNases 2 and 6 is only participating in the purine binding at the initial step of the reaction. In contrast, the positioning of the pyrimidine base, located at the main B1 site, and the phosphate are mostly retained during all the simulation run, as reported in previous molecular dynamics using RNase A or angiogenin (Madhusudhan and Vishveshwara, 2001). Indeed, Raines and co-workers' kinetic studies indicated that the RNase A catalytic mechanism relies mostly on the substrate association step (delCardayre and Raines, 1994). A high catalytic efficiency would mostly be associated to the enzyme facility to throw away the product from the catalytic site. In this context, previous studies emphasized the importance of the active site flexibility for substrate recognition, catalysis and product release (Sanjeev and Vishveshwara, 2005; Gagné et al., 2012; Gagné and Doucet, 2013). The authors identified two main clusters involved in the protein motion that participate in substrate recognition and product release. In particular, L4 loop plays a key role in the protein motion (Gagné et al., 2012). In addition, a distant residue, Ala109, was identified in RNase A to work as a hinge and promote the active cleft opening and product release (Gagné et al., 2015). Ala109 is conserved in almost all the studied vertebrate members, except in frog and chicken RNases. To note, chicken family members are characterized by a much lower catalytic efficiency. On the other hand, comparison of zebrafish proteins indicates that presence of an Ala or Gly at this position is associated to high catalytic efficiency (Kazakou et al., 2008). On the other hand, a network of sequential hydrogen bond interactions was found mostly dependent on His48 protonation state, where deprotonation is associated to product release (Doucet et al., 2009; Watt et al., 2011). Interestingly, His48 is close to the protein family signature CKXXNTF and is conserved in most members, except in fish and amphibian sequences (**Figures S7–S11**).

Early dynamic predictions could also clearly differentiate between the main B2 residue (Asn71), which directly interacted with the adenine base, and other contributing residues, such as residues Gln69 and Glu111, which participated through watermediated interactions (Seshadri et al., 1995; Madhusudhan and Vishveshwara, 2001). The results helped to interpret previous results obtained by site-directed mutagenesis and kinetic characterization (Tarragona-Fiol et al., 1993). Likewise, the NMR analysis of several frog RNases in complex with a deoxytetranucleotide also highlighted the key role of Asn71 counterparts for guanine binding, even if the respective L4 loops are significantly shortened in contrast to the bovine RNase A structure (Hsu et al., 2015). Moreover, the studies by Hsu and Chen corroborated the importance of Glu111 counterpart in specific guanine recognition at the N1/N2 group (Hsu et al., 2015).

On the other hand, significant divergence is evidenced at the guanine-binding mode between the present molecular dynamics analysis and previous structural characterization by X-ray crystallography. Mostly, although our data emphasizes the preference for adenine at B2 site in mammal RNases, we do not observe any impediment for guanine positioning at the enzyme base secondary site, nor any tendency of guanine to bind at the main B1 base site. Surprisingly, RNase A crystallographic studies using 2′5′-UpG and d(UpG), both in soaking and co-crystallization conditions, showed an unusual binding mode (Lisgarten et al., 1995). Specifically, the guanine was located at B1 instead of B2 site; this peculiar non-productive positioning was classified as a "retrobinding" mode (Aguilar et al., 1992). In addition, not only was "retrobinding" reported by independent researchers for RNase A for both d(CpG) and d(UpG) (Aguilar et al., 1992; Lisgarten et al., 1995; Vitagliano et al., 2000), but also for bullfrog RNase binding to d(CpA) (Chang et al., 1998). Noteworthy, the present kinetic results are also emphasizing a much more pronounced substrate selectivity at B2 site than the MD data reveal (**Table 1**, **Figure 2**).

Overall, our molecular dynamics study using UpA and UpG enabled us to outline the main residues involved in the RNases' distinct specificities for B2. **Figure 5** illustrates the main interactions that participate in the purine recognition. First, bidentate interactions can mainly discriminate between binding to either adenine or guanine at N1/N6 or N1/O6 groups respectively. In addition, we observe specific interactions at N7/ N6 for adenine versus N7/O6 for guanine; and eventually specific binding at guanine N1/N2 group. A summary of the most representative residues that provide selectivity for each base is shown in **Figure 6**. Although no universal rules can be written for protein-nucleotide base binding, the residues identified in our study for RNase A superfamily members match most of the previously reported in the literature (Luscombe, 2001; Kondo and Westhof, 2011). Our previous statistical analysis of protein– nucleotide complexes available at the Protein Data Bank also highlighted the main contribution of Asn/Gln, Arg and Glu/ Asp that provide bidentate interactions at N1/N6 and N1/O6 or N1/N2 groups, respectively (Boix et al., 2013). Other polar or charged secondary residues, such as Thr, Ser or Lys could also be identified (Luscombe, 2001; Boix et al., 2013). Complementarily, stacking interactions are also significantly influencing the protein binding mode (Luscombe, 2001; Boix et al., 2013). Interestingly, another structural feature reported by Westhof and co-workers as characteristic for adenine binding is the combined contribution of amino acid side chain and the peptide backbone atoms (Kondo and Westhof, 2011). Our molecular dynamics analysis highlights the conserved binding mode for adenine of Asn71 in RNase A, and counterparts, together with L4 loop main chain atoms. This emphasizes the importance of Asn and loop L4 conformation in RNase A superfamily to favor adenine binding in mammals (**Figure 4**). On their turn, lower order vertebrates tend to present an Arg that facilitates the interactions at N7/O6 for guanine recognition, as reported for other nucleotide-binding proteins (Luscombe, 2001).

Last, together with the two natural purine bases found in RNA we decided here to analyse the modified base inosine. Inosine molecular structure was used as a purine binding model that served to visualize unique interactions at N7/O6 and N1/O6, in relation to guanine. Comparison of kinetic and MD results on UpG and UpI highlights the importance of specific Glu/Asp residues

FIGURE 6 | Schematic depiction of the main interactions identified by molecular dynamics according to the two main family classes (mammalian versus non-mammalian RNases). In dark and light red, the main and secondary specific interactions characteristic for either mammalian or nonmammalian RNases and in grey, the other interactions observed in all the RNases analyzed. Representative icon labels for humans and fishes illustrate the predominant interaction mode for each RNase class. Asn1 corresponds to Asn71 and Asn2 to Asn 67 in RNase A.

in non-mammalian RNases involved in guanine N1/N2 group recognition. Interestingly, we find in the literature an inosinespecific RNase that can accommodate the base in its active site groove and provides specificity by discriminating the modified base against the two natural purines (Versées et al., 2002). To note, the contribution of Trp side chain in packing the inosine base by stacking interactions is observed. Inosine represents one of the main posttranscriptional modifications in cellular transcripts. RNA modifications not only contribute to regulate the translation pathway, they are also involved in the generation of regulatory tRNA fragments (Lyons et al., 2018). It is important to highlight that specific tRNA cleavage participates in the host response in stress conditions (Thompson et al., 2008) and RNA posttranscriptional modification can alter the target specificity for cellular endonucleases. For example, base methylation can protect tRNA from cleavage by human RNase 5 (angiogenin) (Lyons et al., 2017). Overall, RNA modifications not only alter their own processing rate but also influence their association to selective binding proteins, participating in the cellular metabolism and physiology (Boccaletto et al., 2017). Besides, the complexity of cellular RNA structure and its organization into supramolecular complexes within the cell further difficult our understanding of the cellular RNA metabolism (Van Treeck et al., 2018). Definitely, we are still facing important methodological limitations to interpret the RNases' behavior in physiological conditions.

On the other hand, a comprehensive analysis of the protein nucleotide recognition pattern cannot disregard the existence of an extended substrate binding site architecture as demonstrated by many structural and kinetic studies (Boix et al., 1994; Fontecilla-Camps et al., 1994; Irie et al., 1998; Nogués et al., 1998; Raines, 1998; Hsu et al., 2015; Prats-Ejarque et al., 2019). Interestingly, recent work on the protein motion and ligand binding energies using a pentanucleotide suggests that induced conformational changes take place upon RNA interaction with secondary binding sites and can eventually provide a synergistic addition effect (Narayanan et al., 2017). The cooperative participation of secondary substrate binding sites could explain the enzyme low binding affinity for mono- and dinucleotides and is also significantly limiting the potency of molecular dynamics predictions, when working with such probes. However, our present results, together with previously reported data, are definitely indicating an evolutionary trend in B2 base selectivity within the vertebrate-specific RNase A superfamily that should respond to changing environmental conditions and adaptation to novel physiological needs. There is still a long path to walk to unveil the RNases' substrate selectivity *in vivo*. We are confident that the identification of the structural patterns for nucleotide recognition in host defense RNases would provide valuable tools for structure-based drug design.

### CONCLUSIONS

In this work, we have analysed the base selectivity at B2 site within the RNase A superfamily by kinetic assays and molecular dynamics simulations using dinucleotide substrates. Our results indicate an evolutionary drift tendency from guanine to adenine preference. Interestingly, a close inspection of the residues potentially involved in the enzyme B2 site reveals that the main contributors (Asn71 and Glu111 in RNase A and equivalent counterparts) are present in all the family members. Notwithstanding, significant differences in L4 loop extension and contribution of complementary residues can facilitate a distinct binding mode that confers discrimination between both purine bases. Overall, Asn, Glu/Asp and Arg bidentate side chains provide selective binding to adenine N1/N6 and N6/N7 versus guanine N1/O6, O6/N7 and N1/N2 groups.

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the manuscript/**Supplementary Files**.

### AUTHOR CONTRIBUTIONS

EB and GP-E conceived and designed the experiments. GP-E, LL, VS and MM performed the experimental work. EB, GP-E and LL analysed the data. EB and GP-E drafted the manuscript. EB, GP-E, LL, VS and MM revised the final manuscript. All authors approved the final manuscript version.

## FUNDING

Research work was supported by the Ministerio de Economía y Competitividad (SAF2015-66007P) and by AGAUR, Generalitat de Catalunya (2016PROD00060; 2017SGR1010), co-financed by FEDER funds and by Fundació La Marató de TV3 (20180310). GP-E is a recipient of a PIF (UAB) predoctoral fellowship. LL is a recipient of a CSC predoctoral fellowship.

### ACKNOWLEDGMENTS

The authors wish to thank Helena Carbó for laboratory technical support, Clara Villalba for her careful revision of the manuscript and the *Laboratori d'Anàlisi i Fotodocumentació*, Universitat Autònoma de Barcelona for providing the necessary infrastructure. We also wish to thank Dr. Marc Torrent for his help in molecular dynamic studies.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.01170/ full#supplementary-material

SUPPLEMENTARY FIGURE 1 | Modifications of the force field to include inosine parametrization.

SUPPLEMENTARY FIGURE 2 | Phylogenetic tree of representative sequences of pancreatic ribonucleases. The evolutionary history was inferred by using the Maximum Likelihood method and JTT matrix-based model (Jones et al., 1992).

The tree with the highest log likelihood (−27534.43) is shown. The percentage of trees in which the associated taxa clustered together is shown next to the branches. Initial tree(s) for the heuristic search were obtained automatically by applying Neighbour-Join and BioNJ algorithms to a matrix of pairwise distances estimated using a JTT model, and then selecting the topology with superior log likelihood value. The tree is drawn to scale, with branch lengths measured in the number of substitutions per site. This analysis involved 160 amino acid sequences. There were a total of 212 positions in the final dataset. Evolutionary analyses were conducted in MEGA X (Kumar et al., 2018). RNases are labeled with the species abbreviation (see Table S1) and its UNIPROT code, or, in its absence, with its NCBI code.

SUPPLEMENTARY FIGURE 3 | Atom nomenclature of the three dinucleotides used in the molecular dynamics simulations.

SUPPLEMENTARY FIGURE 4 | Mobility of the dinucleotides, calculated in RMSD (nm), during each 100 ns simulation run. Each color represents a different replicate.

SUPPLEMENTARY FIGURE 5 | Fraction of hydrogen bond interaction occurrence of the key protein residues involved in the binding to the purine base during each MD simulation run.

SUPPLEMENTARY FIGURE 6 | Schematic illustration of RNase-UpG complexes obtained by molecular dynamics simulations using GROMACS. The picture was generated using PyMOL 1.7.2 (Schrödinger, Inc).

SUPPLEMENTARY FIGURE 7 |Sequence alignment of representative sequences of fish RNases. Protein regions identified to participate in B2 site are highlighted in yellow (L4, spanning from b2 to b3, end of β6 and one of the two catalytic histidines together with a close by residue at β7). Main conserved key residues are: Asn 72/74, Glu 114 and Glu 122/Arg123. TT indicates the presence of a β-turn. Dots label every 10 residues of the reference protein used (Dr-RNase 1). The disulphide bonds are labeled with green numbers. The alignment was performed using Clustal Omega (Sievers and Higgins, 2018), and the picture was drawn using ESPript (Robert and Gouet, 2014). Labels are as follows: red box, white character for strict identity; red character for similarity within a group; and character with blue frame for similarity across groups.

SUPPLEMENTARY FIGURE 8 | Sequence alignment of representative sequences of amphibian RNases. Protein regions identified to participate in B2 site are highlighted in yellow (L4, spanning from b2 to b3, end of β6 and one of the two catalytic histidines together with a close by residue at β7). Main conserved key residues are: Arg5, Asn56 and Thr89/Glu91. TT indicates the presence of a β-turn. Dots label every 10 residues of the reference protein used (Rp-RNase). The disulphide bonds are labelled with green numbers. The alignment was performed using Clustal Omega (Sievers and Higgins, 2018), and the picture was drawn using ESPript (Robert and Gouet, 2014). Labels are

#### REFERENCES


as follows: red box, white character for strict identity; red character for similarity within a group; and character with blue frame for similarity across groups.

SUPPLEMENTARY FIGURE 9 | Sequence alignment of representative sequences of reptilian RNases. Protein regions identified to participate in B2 site are highlighted in yellow (L4, spanning from b2 to b3, end of β6 and one of the two catalytic histidines together with a close by residue at β7). Main conserved key residues are: Asn68 and Asp116/Arg117. TT indicates the presence of a β-turn. Dots label every 10 residues of the reference protein used (Cm-RNase 1). The disulphide bonds are labelled with green numbers. The alignment was performed using Clustal Omega (Sievers and Higgins, 2018), and the picture was drawn using ESPript (Robert and Gouet, 2014). Labels are as follows: red box, white character for strict identity; red character for similarity within a group; and character with blue frame for similarity across groups.

SUPPLEMENTARY FIGURE 10 | Sequence alignment of representative sequences of bird RNases. Protein regions identified to participate in B2 site are highlighted in yellow (L4, spanning from b2 to b3, end of β6 and one of the two catalytic histidines together with a close by residue at β7). Main conserved key residues are: Asn65, Arg66 and Trp105. TT indicates the presence of a β-turn. Dots label every 10 residues of the reference protein used (Gg-RNase 1). The disulphide bonds are labelled with green numbers. The alignment was performed using Clustal Omega (Sievers and Higgins, 2018), and the picture was drawn using ESPript (Robert and Gouet, 2014). Labels are as follows: red box, white character for strict identity; red character for similarity within a group; and character with blue frame for similarity across groups.

SUPPLEMENTARY FIGURE 11 | Sequence alignment of representative sequences of mammalian RNases. Protein regions identified to participate in B2 site are highlighted in yellow (L4, spanning from b2 to b3, end of β6 and one of the two catalytic histidines together with a close by residue at β7). Main conserved key residues are: Asn67/Gln69/Asn71, Ala109, Glu111 and Arg122. TT indicates the presence of a β-turn. Dots label every 10 residues of the reference protein used (Bt-RNase 1). The disulphide bonds are labelled with green numbers. The alignment was performed using Clustal Omega (Sievers and Higgins, 2018), and the picture was drawn using ESPript (Robert and Gouet, 2014). Labels are as follows: red box, white character for strict identity; red character for similarity within a group; and character with blue frame for similarity across groups.

SUPPLEMENTARY TABLE 1 | Phylogenetic classification and abbreviations used for the analysed species.

SUPPLEMENTARY TABLE 2 | Main key residues identified by molecular dynamic simulations of the studied RNase complexes with UpA, UpG and UpI. Equivalent residues by structural superposition are indicated. Cationic residues are indicated in blue, anionic residues in red and polar residues in black. Histidine residues involved in stacking interactions with the purine base are coloured in orange. Only specific interactions with purine atoms are included. Specific interaction to purine atoms are illustrated in Figure 5.


of high resolution structures. *Bioinformatics* 22, 2746–2752. doi: 10.1093/ bioinformatics/btl470


ribonuclease-like activity. *Biochemistry* 25, 7255–7264. doi: 10.1021/ bi00371a002


and building force field libraries for new molecules and molecular fragments. *Nucleic Acids Res.* 39, 511–517. doi: 10.1093/nar/gkr288


**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 Prats-Ejarque, Lu, Salazar, Moussaoui and Boix. 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.*

# Immunotherapy Based on Dendritic Cell-Targeted/-Derived Extracellular Vesicles—A Novel Strategy for Enhancement of the Anti-tumor Immune Response

#### *Oleg Markov\*, Anastasiya Oshchepkova and Nadezhda Mironova*

*Laboratory of Nucleic Acids Biochemistry, Institute of Chemical Biology and Fundamental Medicine SB RAS, Novosibirsk, Russia*

#### *Edited by:*

*Hector A. Cabrera-Fuentes, University of Giessen, Germany*

#### *Reviewed by:*

*Flavio Andres Salazar Onfray, University of Chile, Chile Marion E. G. Brunck, Monterrey Institute of Technology and Higher Education (ITESM), Mexico Di Yu, Uppsala University, Sweden*

> *\*Correspondence: Oleg Markov markov\_oleg@list.ru*

#### *Specialty section:*

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

*Received: 19 April 2019 Accepted: 06 September 2019 Published: 11 October 2019*

#### *Citation:*

*Markov O, Oshchepkova A and Mironova N (2019) Immunotherapy Based on Dendritic Cell-Targeted/- Derived Extracellular Vesicles—A Novel Strategy for Enhancement of the Anti-tumor Immune Response. Front. Pharmacol. 10:1152. doi: 10.3389/fphar.2019.01152*

Dendritic cell (DC)-based anti-tumor vaccines have great potential for the treatment of cancer. To date, a large number of clinical trials involving DC-based vaccines have been conducted with a view to treating tumors of different histological origins. However, DC-based vaccines had several drawbacks, including problems with targeted delivery of tumor antigens to DCs and prolong storage of cellular vaccines. Therefore, the development of other immunotherapeutic approaches capable of enhancing the immunogenicity of existing DC-based vaccines or directly triggering anti-tumor immune responses is of great interest. Extracellular vesicles (EVs) are released by almost all types of eukaryotic cells for paracrine signaling. EVs can interact with target cells and change their functional activity by delivering different signaling molecules including mRNA, non-coding RNA, proteins, and lipids. EVs have potential benefits as natural vectors for the delivery of RNA and other therapeutic molecules targeted to DCs, T-lymphocytes, and tumor cells; therefore, EVs are a promising entity for the development of novel cell-free anti-tumor vaccines that may be a favourable alternative to DC-based vaccines. In the present review, we discuss the anti-tumor potential of EVs derived from DCs, tumors, and other cells. Methods of EV isolation are systematized, and key molecules carried by EVs that are necessary for the activation of a DC-mediated anti-tumor immune response are analyzed with a focus on the RNA component of EVs. Characteristics of anti-tumor immune responses induced by EVs *in vitro* and *in vivo* are reviewed. Finally, perspectives and challenges with the use of EVs for the development of anti-tumor cell-free vaccines are considered.

Keywords: dendritic cells, extracellular vesicles, exosomes, tumor, anti-tumor vaccines

### DENDRITIC CELLS: A BRIEF HISTORY OF ANTI-TUMOR DC-BASED VACCINE DEVELOPMENT

The history of dendritic cells (DCs) began during the second half of the nineteenth century, when Paul Langerhans described for the first time star-shaped cells localized in the skin and mistakenly assumed a neuronal origin (Langerhans, 1868). Eventually, these cells were named Langerhans cells and were found to be a special skin resident subpopulation of DCs (Merad et al., 2008). DCs were rediscovered by Ralph M. Steinman and Zanvil A. Cohn a century later in 1973 (Steinman and Cohn, 1973), and this discovery eventually launched a new era in immunology. It was found that DCs are professional antigenpresenting cells, the main function of which is to capture, process, and present antigen material to T lymphocytes in complex with MHC I and II molecules, activating an antigenspecific T-lymphocyte immune response (Nussenzweig et al., 1980). It was understood how the antigen-specific T-cell immune response is triggered, and DCs were shown to play a principal role in this process (Dhodapkar et al., 1999).

Later, the method of generating a large number of DCs was developed on the basis of the incubation of monocyte and bone marrow DC progenitors in the presence of IL-4 and GM-CSF (Romani et al., 1994; Sallusto et al., 1995), which significantly facilitated investigations into the biology and immunotherapeutic potential of DCs. As a result of these studies, a novel type of immunotherapeutic anti-tumor vaccine was developed, namely, the DC-based vaccine (Constantino et al., 2016). Classically, this vaccine is based on DCs loaded with tumor antigens using different approaches ranging from passive loading of DCs with tumor proteins or peptides to transfection/transduction of DCs with nucleic acids (NAs)/viral vectors encoding tumor antigens (Palucka and Banchereau, 2013).

Several decades of intensive investigations of anti-tumor DC-based vaccines have revealed their high efficiency in various murine tumor models (Lin et al., 2015; Markov et al., 2017) and human xenografts in immunodeficient mice (Liu et al., 2019).

Now in clinical trials a great number of DC-based vaccines are explored (see revs. Van Willigen et al., 2018; Mastelic-Gavillet et al., 2019). The first therapeutic anti-tumor DC-based vaccine, Sipuleucel-T, was approved by the U.S. Food and Drug Administration in 2010 for use in castration-resistant prostate tumors. Sipuleucel-T activates the immune response against the antigen, PAP/PA2024, and increases overall survival in patients (Kantoff et al., 2010); however, Sipuleucel-T is highly priced cellular product with technology challenging preparation process which required highly qualified personnel (Van Willigen et al., 2018).

The Indian government agency (CDSCO-Central Drugs Standard Control Organization) approved in 2017 vaccine based on an autologous monocyte-derived DC loaded with tumor lysate (APCEDEN®) for treatment of prostate, ovarian and colorectal cancers, and non–small cell lung carcinoma (Kumar et al., 2017). In phase II clinical trial this vaccine demonstrated well tolerance by patients with refractory solid malignancies (Bapsy et al., 2014) and a survival benefit over 100 days (Kumar et al., 2017).

Current ongoing clinical trials using personalized DC-based vaccines are conducted for treatment of ovarian cancer; brain tumors; advanced melanoma, colorectal cancer, lung cancer, and so on (**Table 1**). Types of vaccines studied in clinical trials are varied by the composition of DC-pulsed antigen that can be in the form of tumor lysate, tumor-derived peptides, and mRNA-encoding TAAs. Some types of DC-vaccines are prepared by fusion of DC with tumor cells. Introduction in the treatment regimen of additional components including chemotherapeutics (NCT01957956, 2013; NCT01946373, 2013; NCT02503150, 2015; **Table 1**), immune response modifiers (NCT01808820, 2013; NCT01204684, 2010; NCT00799110, 2008; **Table 1**), and immune checkpoint inhibitors (NCT03014804, 2017; NCT02529072, 2015; NCT02678741, 2016; NCT03092453, 2017; NCT03152565, 2017; NCT03406715, 2018; **Table 1**), as well as cytokines providing the maturation of DC lead to the novel treatment schemes with enhanced efficacy. DC-based vaccines demonstrate their high potential, and some of them have already reached phase 3 of clinical trials: DC-vaccine loaded with tumor-derived RNA for treatment of uveal melanoma (NCT01983748, 2013, **Table 1**) and tumor-lysate pulsed DC-vaccine in combination with FOLFOX6 regiment for treatment of metastatic colorectal cancer (NCT02503150, 2015; **Table 1**).

The era of investigation of DC-derived membrane vesicles for immunotherapy of cancer is just beginning, and ongoing clinical trials are extremely scarce. For instance, a clinical trial of DC-derived extracellular vesicles (EVs) for treatment of patients with non-small cell lung cancer (NSCLC) is now completed (NCT01159288, 2010, **Table 1**). It was shown that DC-derived EVs exerted natural killer (NK) cell effector functions in patients with NSCLC, thus boosting the NK cell arm of antitumor immunity (Besse et al., 2016). These findings indicate the efficiency of novel immunotherapeutic anti-tumor approaches based on membrane vesicles and their great therapeutic prospects.

### EVS AS ALTERNATIVE CELL-FREE ANTI-TUMOR VACCINES

One of anti-tumor immunotherapeutic approaches is the application of EVs of DCs and tumor cells. All types of eukaryotic cells produce nano-sized vesicles with the capacity to shuttle NAs, proteins, and lipids to other cells and participate in cellto-cell communication, thus realizing paracrine regulation. EVs are a heterogeneous population of membrane vesicles that are classified into three major groups according to their subcellular origin and size: apoptotic bodies, microvesicles (MVs), and exosomes (Gurunathan et al., 2019), the latter of which are the most studied. Exosomes are nano-sized vesicles originating from multivesicular bodies (MVBs) in the endosomal pathway, with sizes ranging from 50 to 150 nm in diameter and membranes characterized by a high content of cholesterol and glycosphingolipids (Colombo et al., 2014). MVs are described as 100- to 1000-nm vesicles enveloped from the cell surface membrane by direct budding (Morel et al., 2011). Since there exists no perfect method to isolate only exosomes (Chulpanova et al., 2018), studies on the functional activity of EVs have been performed on exosome- or MV-enriched populations or a mixture of both types of EVs. Therefore, in the present review, the common term EVs will be used to describe primarily exosomeenriched vesicles.

It has been demonstrated that EVs can participate in immune regulation, matrix remodeling, signaling pathways, intercellular exchange with oncoproteins and oncogenes, induction of angiogenesis, and preparation of a pre-metastatic niche (Lee et al., 2011). It is known that DC-derived EVs carry functionally active molecules on their surfaces that take part in immunological synapses—complexes of MHC class I and II with tumor antigens, as well as co-stimulatory and adhesion molecules (such as CD80, CD86, and CD40)—needed for the induction of anti-tumor T-cell

#### TABLE 1 | Ongoing clinical trials of dendritic cell-based vaccines for treatment of various types of tumors.



*\*chemotherapeutic; \*\*immune response modifier; #poly ICLC, interstitial Cajal-like cells, TLR3 agonist; \$monoclonal antibodies, immune checkpoint inhibitor; &anti CD25 denileukin diftitox; §FOLFOX6 - a specific chemotherapy regimen of Oxaliplatin, 5-Fluorouracil and Leucovorin; ©the study has been completed; @TLPLDC-vaccine - autologous tumor lysate, particle-loaded, dendritic cell vaccine;* 

*OS, overall survival; PFS, progression free survival; QoL, quality of life; CR, number of participants with complete response; PR, number of participants with partial response; SD, number of participants with stable disease; PD, number of participants with progressive disease; RSDR, response/stable disease rate; RFS, relapse-free survival; irDCR, immune related disease control rate; irTTP, immuno-related time to progression; irORR, immuno-related overall response rate; irDOR, immuno-related duration of response; irTTR, immuno-related time to response; irPFS, immuno-related progression free survival; DCR, disease control rate; ORR, overall response rate; IR, immune response; RT, radiation treatment.*

immune responses (Munich et al., 2012). In addition, tumor cellderived EVs have been shown to have an immunostimulatory effect on anti-tumor DCs (André et al., 2002; Liu et al., 2018); hence, the application of both DC- and tumor cell-derived EVs as novel immunotherapeutic cell-free anti-tumor vaccines has great potential. Together with their highly therapeutic antitumor potential, cell-free vaccines based on EVs have advantages over classical DCs involved: (1) EV-based vaccines can be stored for a prolonged time without loss of immunotherapeutic activity (Jeyaram and Jay, 2018); (2) the more efficient capture of EVs rather than the soluble molecules of antigen-presenting cells (Zeelenberg et al., 2008). Undoubtedly, the use of EVs as antitumor vaccines possesses great potential and relevance (Viaud et al., 2010; Pitt et al., 2016).

### METHODS OF EV ISOLATION

The most commonly described method in the literature for the isolation of EVs is sequential centrifugation of conditioned medium samples or biological fluids (blood serum, urine, milk, etc.) (Petersen et al., 2014). Typically, low speeds with increasing centrifugal force are used to remove cells, cell debris, and large particles, followed by ultracentrifugation at 100,000*g* to 120,000*g* for at least 60 to 120 min to precipitate EVs (see studies in **Table 2**). This method is relatively laborious, timeconsuming, and requires special expensive equipment (Soung et al., 2017). The high heterogeneity of EVs and overlapping size with protein aggregates, as well as the need for several rounds of ultracentrifugation during the wash steps of EVs, inevitably results in EV loss, contamination, and low yields (Li et al., 2017). The widely used ultracentrifugation method for EV isolation results in the lowest recovery of particles; nevertheless, it is the most popular approach to date (Tang et al., 2017).

To increase the enrichment and purity of isolated EVs, centrifugation using a sucrose density gradient or sucrose cushion, in addition to ultrafiltration, is used in combination with ultracentrifugation or alone (Dai et al., 2008; Bu et al., 2011; Besse et al., 2016; Diamond et al., 2018; Guo et al., 2018). Sucrosebased density gradient ultracentrifugation of EVs allows isolation of the pure fraction of EVs due to their specific buoyant density exosomes float in a sucrose gradient of 1.13 to 1.19 g/mL (Théry et al., 2002). The ultrafiltration method sequentially removes larger particles from samples, with a final ultrafiltration step using a 100-kDa MWCO filter, such as Centricon Plus-70, Centriplus, or Amicon Ultra (Millipore). This method is rapid and technically easy but allows isolation of EVs with high purity. Furthermore, a combination of filtration and sucrose-based density gradient centrifugation has made it possible to efficiently isolate high purity EVs in sufficient amounts for clinical trials (Escudier et al., 2005; Morse et al., 2005).

Easy-to-use commercial kits, such as ExoQuick (System biosciences) (Rekker et al., 2014) or the Total Exosome Isolation Kit (Invitrogen) (Wang et al., 2015), are rarely used to isolate DC-derived EVs. The principle of such kits is salting out EVs from samples by the addition of water-excluding polymers, such as polyethylene glycol (PEG), which occupy water molecules and force less soluble components, including EVs, out of the solution (Li et al., 2017). Following overnight incubation at 4°C, precipitated EVs are isolated by short centrifugation steps at low speed (up to 10,000*g*). The advantages of commercial kits are less time-consuming method, possibility to isolate EVs from small volumes of medium, no need for special expensive instruments, such as an ultracentrifuge, and no technical challenges (Helwa et al., 2017). However, the main drawback of this method is the isolation of non-vesicular particles together with EVs (Yamada et al., 2012), which is the reason why this method is less frequently used.

TABLE 2 | Efficiency of antitumour vaccines on the base of tumour cell-/DC-derived EVs in animal tumour models *in vivo*, human cells *ex vivo* and in clinical trials.


#### A. Vaccines on the base of tumour cell-derived DC-targeted EVs.


#### A. Vaccines on the base of tumour cell-derived DC-targeted EVs.


#### B. Vaccines on the base of DC-derived EVs.


#### B. Vaccines on the base of DC-derived EVs.


#### B. Vaccines on the base of DC-derived EVs.


#### B. Vaccines on the base of DC-derived EVs.


#### B. Vaccines on the base of DC-derived EVs.


nude mice.

#### C. EVs-based anti-tumour immunotherapy in clinical trials.


#### C. EVs-based anti-tumour immunotherapy in clinical trials.


*aNK, activated NK (splenocytes activated with IL-2 for 6 days); CRCL, chaperone-rich cell lysate; CTLs, cytotoxic T-lymphocytes; DCs, dendritic cells; EVs, extracellular vesicles; GM-CSF, granulocyte-macrophage colony-stimulating factor; HS, heat stressed; ID, intradermal; iDCs, immature DCs; iEVs, extracellular vesicles from immature DCs; IFN, interferon; IL, interleukin; IT, intratumoural; lEVs, large extracellular vesicles; LPS, lipopolysaccharide; mEVs, extracellular vesicles from mature DCs; MVs, microvesicles; NK, natural killer; OVA, ovalbumin; PBMC, peripheral blood mononuclear cells; PT, peritumoural; SC, subcutaneous; sEVs, small extracellular vesicles; TGF, transforming growth factor; Th, T-helper cells; tmTNF, transmembrane tumour necrosis factor; TNF, tumour necrosis factor; Tregs, T-regulatory cells.*

Other methods of EV isolation, such as immunoaffinity capturebased and microfluidics-based isolation techniques, are more expensive and technically complicated, and in practice are not used for the isolation of DC-derived or DC-targeted MVs. Therefore, these approaches are not considered in the present review. The principles of the different EV isolation techniques are described in detail in other reviews (Li et al., 2017; Pariset et al., 2017).

### METHODS OF EV LOADING WITH THERAPEUTIC MOLECULES

For proper activation of an anti-tumor immune response, the antigen-presenting cells (APC), in particular DCs, should convey three signals to T cells: (1) presentation of tumor antigens in complexes with MHC class I and II molecules on the surface of DCs to T-cell receptors (2) signal transmission via interaction of co-stimulatory and adhesion molecules expressed on the surface of DCs to their receptors on T cells; and (3) production of T-cell stimulatory cytokines by DCs (Kapsenberg, 2003).

It is known that DC-derived EVs carry all the molecules required to activate anti-tumor T cell-mediated immune responses (Seo et al., 2018). On their surface, DC-derived EVs contain functionally active complexes containing tumor antigens and MHC molecules class I and II, as well as co-stimulatory and adhesion molecules (Morelli et al., 2004; Chaput et al., 2006). Furthermore, it has been shown that EVs are able of carrying cytokines (Gulinelli et al., 2012). Therefore, DC-derived EVs are assumed to interact with T lymphocytes and directly trigger anti-tumor immune responses that has been proven by many experimental data (Wang et al., 2013; Lu et al., 2017; Chen et al., 2018 etc.) (see **Table 2B**).

With respect to tumor cell-derived DC-targeted EVs, these vesicles may transfer tumor antigenic peptides and immunostimulatory molecules to DCs, resulting in the induction of DCs exhibiting high immunogenicity (Pitt et al., 2014).

Generally, for proper activation of immune responses, in addition to tumor-associated antigens and immunostimulatory molecules, DC-derived and DC-targeted EVs should carry tumor-targeted molecules. EVs can also carry therapeutic NAs, such as small non-coding regulatory RNA (siRNA, miRNA) for the targeting of certain genes in recipient cells, thus enhancing immunogenicity or inhibiting negative signals (Jiang et al., 2017). The efficient loading of EVs with therapeutic molecules is the most important step in the preparation of EV-based antitumor vaccines. Therapeutic molecules can be delivered to EVs indirectly by the loading of EV-secreting cells followed by the isolation of EVs or directly by the loading of preliminarily isolated EVs (Batrakova and Kim, 2015). Summarizing the investigations performed over recent decades, in the next part of the present review we will discuss the molecules used for the loading of antitumor EVs and provide an overview of the methods used for the indirect or direct loading of EVs with therapeutic molecules.

#### Loading of EVs With NAs

NA-based therapeutics is a promising tool for the modification of immune cell properties and treatment of a variety of human diseases. Plasmid DNA and mRNA encoding tumor antigens, immunostimulatory molecules, EV-targeted proteins, and regulatory short non-coding RNA (miRNA, siRNA), can be successfully used to launch or enhance the anti-tumor potential of EV-based vaccines (Van den Boorn et al., 2013). The main problems with NA application are instability of the naked NA in the presence of nucleases as well as non-targeted delivery. As a result, various types of chemical modifications of NAs and nanocarriers have been developed to ensure the stability of NAs and to provide their targeted delivery to both cells and EVs.

Overall, only two strategies are used to load EVs with NAs: (1) indirect loading – preliminary transfection/transduction/ electroporation of EV-secreting cells with NA followed by EVs isolation or (2) direct loading of preliminarily isolated EVs with the NA (Johnsen et al., 2014).

#### Indirect Loading of EVs With NA

Indirect loading of EVs is achieved by preliminary delivery of NA to EV-secreting cells by transfection, transduction, or electroporation. Subsequently, NA may be sorted into MVBs and secreted by exosomes (in the case of regulatory small noncoding RNA, such as siRNA or miRNA) or NA can be translated to peptides or proteins (tumor peptides, chimeric proteins, targeting molecules) that can be carried by EV to immune or tumor cells.

This approach has been well adapted for miRNA and siRNA loading. Overexpression of miRNA in EV-secreting cells is typically realized by transfection of cells with plasmid DNA or viral vectors encoding miRNA (Katakowski et al., 2013; Wang et al., 2016), while siRNA, in addition to the above-mentioned methods, can be delivered to cells in the form of siRNA duplexes. Recent reports have revealed that the use of siRNA duplexes is preferred (Zhang et al., 2014b) and allows packaging of approximately 0.001 to 0.14 pmol siRNA into 1 μg EVs (Zhang et al., 2014b; Liu et al., 2015). It was revealed that indirect loading of murine leukemia-derived EVs with *TGF-β1* shRNA resulted in robust maturation of DCs, activation Th1 immune response and pronounced inhibition of tumor growth *in vivo* (Huang et al., 2017).

The loading of high-molecular weight NAs, such as plasmid DNA and mRNA, into EVs is associated with additional difficulties, such as low loading efficiency due to the size of NAs and the possibility of losing functional activity of NAs. Thus, in the case of high-molecular weight NAs, the common strategy for the modification of EVs is preliminary transfection/ transduction/electroporation of EV-secreted cells.

Using the indirect NA delivery technique, DC-derived or tumor-derived DC-targeted EVs can be modified with different proteins, such as tumor-associated antigens or DC-activating molecules, respectively, by preliminary transfection/transduction of EV-producing cells with plasmid DNA/RNA/viral vectors encoding these proteins. Concerning the modification of DC-derived EVs with tumor antigens, DCs can be transduced with viral vectors encoding full-length tumor-associated proteins (Wang et al., 2013; Lu et al., 2017) or transfected with tumor RNA encoding a pool of tumor-associated proteins and peptides (Gehrmann et al., 2013). Following transfection/transduction of cells, intracellular processing of proteins occurs, and complexes of tumor peptides with MHC molecules class I and II are formed and exposed on the surface of EVs. Such DC-derived EVs carrying tumor-associated peptides in complex with MHC molecules are able to directly activate highly efficient T-cell antitumor immune responses both *in vitro* and *in vivo* (Gehrmann et al., 2013; Wang et al., 2013; Lu et al., 2017).

To place a particular peptide or protein on the surface of EVs, indirect loading of EV-producing cells with NA encoding chimeric proteins consisted of EV surface protein (lactadherin, Lamp2b, etc.) fused with protein or another molecule of interest (tumor antigen, immunostimulatory molecules as well as DC- or tumor-targeted molecules) can be applied.

For example, lactadherin, exosome-specific anchor expressed on the surface of EVs, can be used to modify EVs. EV-secreting murine melanoma cells were transfected with plasmid DNA encoding lactadherin fused to streptavidin; the EVs produced by these cells were shown to express streptavidin on their surface and could be modified with either biotinylated CpG DNA (Morishita et al., 2016) or GALA peptide (Morishita et al., 2017) for activation of DC maturation or enhancement of EV cargo release into the cytosol of recipient cells, respectively. Moreover, lactadherin was also fused to the tumor-associated antigens, CEA and HER2, to modify EVs produced by HER2+ breast carcinoma cells. Adenoviral vectors encoding lactadherin-CEA or lactadherin-HER2 fusion proteins were used to transduce EV-secreting cells. Enhanced expression of tumor antigens on the surface of EVs resulted in significant activation of antigenspecific anti-tumor immune responses in animal models of breast tumors (Hartman et al., 2011).

A mutant form of the HIV-1 Net protein (Netmut), that has been shown to have extraordinarily high levels of accumulation in exosomes (Lattanzi and Federico, 2012), can be used as anchors to modify EVs. Plasmid DNA encoding Netmut fused to the tumor antigens, HER2 or MART-1, was employed to indirectly modify human muscle cell-derived EVs. Modified EVs were demonstrated to possess great anti-tumor potential against breast cancer and lymphoblastoma *in vitro* and *in vivo* (Anticoli et al., 2018).

EVs possess the natural property of transferring their contents to target cells and tissues; however, in their unmodified form, EVs are known to accumulate mainly in the liver, kidneys, intestine, lungs, and spleen of laboratory animals (Wiklander et al., 2015). Accordingly, many studies have been conducted with the aim of targeting EV/NA complexes to organs and tissues of interest (Alvarez-Erviti et al., 2011; Tian et al., 2014; Liu et al., 2015; Bellavia et al., 2017). Modification of EVs with DC- or tumor cell-targeted molecules can allow an increase in the delivery efficiency of miRNA and siRNA to DCs or tumor cells and enhance the biological effect of therapeutic NA. This approach can be performed by transfection or transduction of EV-secreting cells with plasmid DNA or viral vectors, respectively, encoding EV surface proteins fused to DC-/tumortargeted or immunostimulatory molecules.

For instance, it has been demonstrated that the Lamp2b protein expressed on the surface of EVs can be modified by fusion with the RVG-peptide (neuron-specific rabies viral glycoprotein that binds to acetylcholine receptors) to target DC-derived EVs to neuronal cells *in vitro* and to brain cells *in vivo* (Alvarez-Erviti et al., 2011; Liu et al., 2015). Application of α-bungarotoxin (inhibitor of acetylcholine receptors) has been shown to result in the loss of function of RVG-EVs with respect to the ability to transfer BACE1 siRNA to Neuro2A cells (Alvarez-Erviti et al., 2011). In another study, DC-derived EVs were modified with Lamp2b by fusion with the iRGD peptide targeted to αv-integrin-positive breast cancer cells. It was demonstrated that these iRGD-modified EVs more effectively penetrated into MDA-MB-231 cells than unmodified EVs. Moreover, intravenous injection of iRGD-EVs into tumorbearing mice has been reported to result in the rapid accumulation of reprogrammed EVs in tumors (Tian et al., 2014).

Modification of DC-targeted tumor-derived EVs involves preliminary transfection of EV-producing cells with plasmid DNA encoding DC-targeted/-activating molecules, for example CD40L (Wang et al., 2014). It was shown that EVs derived from CD40L-modified Lewis lung carcinoma cells were highly immunogenic towards DCs *in vitro* and extremely efficient in a protective and therapeutic scheme of treatment for murine lung carcinoma *in vivo* (Wang et al., 2014).

It appears that reprogrammed EVs have a similar transfection efficiency to commercial transfection reagents. It has been reported that RVG-EVs have a transfection efficiency comparable with commercial TransIT LT1 transfection reagent (Mirus Bio). In this work, DC-derived RVG-modified EVs transferred three types of α-Syn siRNA to a human SH-SY5Y neuroblastoma cell line expressing mouse α-Syn-HA. The level of α-Syn mRNA/ protein reduction was similar, with a slight advantage for RVG-EVs over TransIT LT1-mediated delivery (Cooper et al., 2014). Moreover, a similar result has been demonstrated previously with RVG-modified murine DC-derived EVs (Alvarez-Erviti et al., 2011). The delivery of *GAPDH* and *cyclophilin B* siRNA was performed by reprogrammed RVG-EVs, and the level of gene silencing was comparable with that obtained following transfection with Lipofectamine 2000 (Thermo Fisher Scientific) transfection reagent (Alvarez-Erviti et al., 2011).

#### Direct Loading of EVs With NA

In addition to the indirect modification of EVs by preliminary cell transfection, another approach is the direct modification of EVs with DC-/tumor-targeted molecules.

Electroporation is often used for direct loading of EVs with low-molecular weight NAs, with an efficiency of approximately 3% to 24% transfected EVs (Alvarez-Erviti et al., 2011; Wang et al., 2017; Usman et al., 2018). This technique has been applied to EVs of different origins including DCs (Alvarez-Erviti et al., 2011; Wang et al., 2017). Electroporation is best suited to the direct loading of EVs with small non-coding regulatory RNAs and antisense nucleotides (ASOs). For instance, miRNA-155 was efficiently directly loaded into EVs derived from murine colon carcinoma cells by using electroporation (Asadirad et al., 2019). miRNA-155-EVs was shown to significantly enhance maturation of DCs and stimulate their immunostimulatory functions (Asadirad et al., 2019). One of the main problems with electroporation is the aggregation of EVs, which result in distortion of the true efficiency of NA incorporation into EVs (Kooijmans et al., 2013; Usman et al., 2018).

In addition, small non-coding RNA can be directly delivered to EVs in the form of conjugates with hydrophobic molecules, such as cholesterol. This approach is prospective, since it allows loading of a large number of NA molecules into EVs in the absence of any transfection reagents and additional manipulations (Didiot et al., 2016). Nevertheless, the functional activity of cholesterolconjugated siRNA can be lost in recipient cells (Stremersch et al., 2016). Indeed, melanoma B16 or monocyte/DC-derived EVs loaded with cholesterol-conjugated siRNA (chol-siRNA) were used to silence *CD45* and e*GFP* genes in JAWSII monocytes and H1299 non-small cell lung carcinoma cells, respectively. Highly efficient loading of EVs with chol-siRNA was shown; specifically, 15 μg EVs (~6.6 × 1010 vesicles) were able to bind 80% of the 10 pmol chol-siRNA, corresponding to approximately 73 cholsiRNA molecules per vesicle (Stremersch et al., 2016). EVs loaded with chol-siRNA were demonstrated to efficiently penetrate the target cells; however, no downregulation of gene expression in either tested cell line was observed (Stremersch et al., 2016). In another study, it was reported that approximately 74% of cholsiRNA was loaded into Neuro2A- or DC-derived EVs by using electroporation, and subsequently recommended to use roughly 15 molecules of chol-siRNA per vesicle to achieve this efficiency. In this case, EVs loaded with chol-siRNA were able to transport functionally active siRNA to target cells and downregulate target genes in a concentration-dependent manner (O'Loughlin et al., 2017). The loss of functional activity of chol-siRNA delivered by EVs observed in the Stremersch's investigation was possibly associated with high affinity between EVs and chol-siRNA, and hence, inability to release siRNA into the cytosol for biological action to occur. Furthermore, the structure and length of the linker between siRNA and cholesterol molecules are of great importance to ensure the biological activity of chol-siRNA conjugates (Petrova et al., 2012).

A technique for direct targeting EVs to tumor cells is the labeling of EVs with aptamers. This technique can be conveniently used for artificial EV-mimics derived from cell membranes. For example, the tumor-targeted anti-nucleolin aptamer, AS1411, conjugated to cholesterol-poly(ethylene glycol) was used to modify the membrane of murine DCs, which were further processed by extrusion to generate artificial mimics of natural EVs that significantly eradicated tumors in MDA-MB-231 breast tumor xenograft model (Wan et al., 2018). In another study, murine DC-derived EVs were also modified by AS1411. The binding efficiency of AS1411-modified EVs to breast cancer cells was roughly fourfold greater in comparison with unmodified EVs (Wang et al., 2017). Intravenous injection of miR-let-7–loaded AS1411-modified EVs or control EVs to breast tumor-bearing mice resulted in much higher intratumoral accumulation of the AS1411-modified EVs in comparison with unmodified EVs that resulted in more significant retardation of MDA-MB-231 tumor growth (Wang et al., 2017).

At the end of this section, we briefly consider some other techniques for NA loading into EVs that were recently used to prepare anti-tumor EVs. In some studies, transfection reagents, for example Lipofectamine 2000 (Thermo Fisher Scientific), were used for the direct loading of EVs with NA (Wahlgren et al., 2012; Shtam et al., 2013; George et al., 2018). The main problems of using chemical transfectants to directly load EVs are formation of surface conglomerates of NA/transfectant complexes with EVs that can drastically alter the original vesicle composition (Wahlgren et al., 2012). In other studies, CaCl2 associated transfection of isolated EVs combined with heat shock at 42°C has been reported (Zhang et al., 2017; Zhang et al., 2018). This approach allows the loading of approximately 200 copies of siRNA per EV. Sonication has also been used for NA loading into EVs, showing a significant decrease in siRNA aggregation in comparison with electroporation (Lamichhane et al., 2016). Moreover, the promising approach of using freezing/thawing and extrusion methods, in addition to a permeabilization technique with saponin, has been demonstrated for the efficient loading of EVs (Haney et al., 2015).

#### Direct and Indirect Modification of EVs With Proteins and Peptides

To use EVs as immunotherapeutic cell-free vaccines, DC-derived EVs can be modified not only with NAs but also with tumorassociated proteins or peptides. Similar to NA delivery into EVs, there are two techniques for loading EVs with proteins/peptides: indirect loading of EV-secreting cells and direct loading of EVs.

This technique allows to prepare EVs that are able to activate tumor antigen-specific immune responses. Both indirect and direct loading of EVs with proteins/peptides leads to the presentation of tumor-associated peptides in complexes with MHC class I and II molecules on DC-derived EVs and promotes the interaction of these EVs with T-lymphocytes or NK cells and subsequent activation of anti-tumor immunity.

Indirect loading of DC-derived EVs was demonstrated with the tumor-specific proteins or peptides MAGE3 (Viaud et al., 2009) and HPV early antigen 7 (Chen et al., 2018), model tumor antigens, such as OVA and OVA-derived peptide SIINFEKL (Gehrmann et al., 2013; Naslund et al., 2013; Yao et al., 2013; Damo et al., 2015; Wahlund et al., 2017), tumor lysates (Guan et al., 2014; Bu et al., 2015), and even HOCl-oxidized B16 melanoma cells containing both proteins and NAs (Damo et al., 2015). Immature DCs possess the intrinsic ability to capture proteins and peptides from surrounding fluids and tissues; therefore, there is no need to use special reagents or techniques for loading (Savina and Amigorena, 2007). Captured proteins and peptides are processed and tumor antigens in complex with MHC class I and II molecules are subsequently exposed on the surface of DCs, and as a consequence, on DC-derived EVs (Pitt et al., 2014; Markov et al., 2016). The presence of tumor peptides on the surface of EVs is proven in most cases by pentamer staining, ELISA, or western blotting assays. It should be mentioned that the maturation status of DCs loaded with proteins/peptides is essential for expression of costimulatory and adhesion molecules on the surface of DC-derived EVs, which are needed for the induction of an efficient anti-tumor immune response (Lutz and Schuler, 2002). Hence, the choice of suitable maturation stimuli for the treatment of DCs is of great importance. Different compounds have been used to stimulate the maturation of DCs, such as LPS, poly(I:C), CpG-oligonucleotide, and TNF-α. It has been shown that mature DC-derived EVs indirectly loaded with tumor antigens activate efficient anti-tumor immune responses *in vitro* and *in vivo* (Naslund et al., 2013; Bu et al., 2015; Damo et al., 2015; Wahlund et al., 2017; Chen et al., 2018) (see DC-Derived EVs and **Table 2B**).

The technique of direct loading of EVs with tumor antigenic peptides was developed by Hsu and co-authors (Hsu et al., 2003). Direct binding of peptides to MHC I and II molecules on the surface of DC-derived EVs was performed under mildly acidic loading conditions. It was demonstrated that MHC class I molecules on EVs can be directly loaded with peptides, such as HLA-A2–restricted MART1 tumor peptide, at much greater levels than indirect loading, requiring a 100- to 1000-fold lower peptide concentration. Moreover, it was demonstrated that MHC I molecules efficiently bind tumor peptides at pH 4.2 in the presence of β2-microglobulin (β2m); in the absence of β2m, MHC I molecules also efficiently bind tumor peptides under less acidic conditions (pH 5.2). Although the efficiency of peptide binding at pH 5.2 in the absence of β2m is approximately 50% of that observed at pH 4.2 in the presence of β2m, the biological activity of DC-derived EVs prepared in accordance with both methods is comparable (Hsu et al., 2003). MHC II molecules that are expressed in high density on the surface of DC-derived EVs also efficiently bind tumor peptides under the same conditions. Obtained complexes of MHC I or MHC II with tumor peptides maintain complete functional activity; thus, the technique of direct loading of DC-derived EVs with peptides consists not of peptide delivery inside EVs but the binding of peptides with MHC class I and II molecules on the surface of EVs. In this form, EVs can stimulate both CD8+ cytotoxic T-lymphocytes and CD4+ T-helper cells, which are crucial for triggering an efficient antitumor immune response (Hsu et al., 2003). More recently, this technique was used to load DC-derived EVs with the tumorassociated MAGE3 peptide to treat advanced melanoma in clinical trial (Viaud et al., 2009). Vaccination of patients with obtained EVs resulted in activation of NK cells (Viaud et al., 2009).

### ANTI-TUMOR POTENTIAL OF DC-DERIVED AND DC-TARGETED EVS *IN VITRO* AND *IN VIVO*

#### Tumor Cell-Derived DC-Targeted EVs

The anti-tumor potential of DC vaccines primarily depends on the efficiency of DC loading with tumor antigens. Tumor-derived EVs are known to be natural transport vectors that carry a large variety of tumor antigens, and thus can serve as vesicles for the direct delivery of tumor antigens to DCs. Current success in application of tumor-derived DC-targeted EVs to treat tumors in different murine models *in vivo* is summarized in **Table 2A**.

The ability of tumor cell-derived EVs to induce an anti-tumor immune response depending on their size has been investigated, and large-sized tumor cell-derived EVs (MVs, diameter 100−1000 nm) were shown to be more immunogenic than small-sized EVs (exosomes, diameter 60−100 nm). Treatment of tumor-bearing mice with large- and small-sized EVs resulted in the inhibition of tumor growth, with 50% and 12.5% of mice, respectively, remaining tumor-free. Furthermore, treatment of mice with tumor cell-derived large-sized EVs resulted in the activation of anti-tumor cytotoxic T-lymphocytes (CTLs) that were 1.5-fold to 2-fold more efficient in comparison with CTLs activated by treatment with small-sized EVs (Zhang et al., 2014a). It should be mentioned that despite the successful application of cell-free vaccines based on DC-targeted tumorderived EVs, DCs pre-loaded with tumor-derived EVs inhibited tumor growth significantly more efficiently in comparison with EV-based vaccines in all tested tumor models (Zhang et al., 2014a). This can be explained by the fact that tumor-derived EVs are not specifically targeted to DCs, and following subcutaneous inoculation in mice, interact with cells of different origins, not only with DCs, thus losing their antitumor potential.

Indeed, it was revealed that DCs loaded with EVs produced by murine L1210 leukemia cells activate a more efficient anti-tumor immune response in both prophylactic and therapeutic regimens in comparison with cell-free EV-based vaccines. In a model of L1210 murine leukemia, it was shown that subcutaneous injection of DC/L1210-EVs at a dose of 4×106 cells per mouse results in complete eradication of tumors in a therapeutic setting and protection of all mice in a prophylactic setting, whereas vaccines based on tumor-derived EVs are not as efficient (Yao et al., 2014).

The next question is whether DCs loaded with tumor-derived EVs are more immunogenic than classic DC-based vaccines loaded with tumor proteins or NA encoding tumor antigens. On the basis of the facts that tumor vesicles contain large amounts of different tumor antigens (Wolfers et al., 2001) and have a natural ability to efficiently deliver cargo to the cells (Ha et al., 2016), it appears that the use of tumor-derived EVs as a source of tumor antigens to load DCs is superior to commonly used sources of tumor antigens (tumor proteins, peptides, or NAs). Indeed, it has been shown that human DCs loaded with glioma cell-derived EVs activate anti-tumor CTLs *ex vivo* that kill glioma cells twofold more efficiently in comparison with CTLs primed with tumor lysate-pulsed DCs (Bu et al., 2011). Using a mouse AB1 malignant mesothelioma model, it was revealed that treatment of tumor-bearing mice with DCs loaded with either tumor cellderived EVs or tumor lysate results in infiltration of CD4+ T cells, CD8+ T cells, and DCs into tumor tissues. However, in the case of EV-loaded DCs, a higher increase in overall and median survival of experimental animals was observed as compared with lysate-pulsed DCs: the overall survival (on the 52nd day of tumor development) was 0%, 16.7%, and 33.3%; and the median survival was 13, 18.5, and 29.5 days for mice treated with PBS, lysate-pulsed DCs, and EV-loaded DCs, respectively (Mahaweni et al., 2013). Greater immunogenicity of tumor cell-derived EV-loaded DCs has also been demonstrated. Treatment of WEHI3B myeloid leukemia-bearing mice with such EV-loaded DCs resulted in a more significant retardation of tumor growth and survival in animals as compared with treatment with lysatepulsed DCs (Gu et al., 2015). The efficiency of DC-based vaccines loaded with tumor cell-derived EVs was shown to be almost equal to that of protein-pulsed DCs. Thus, DCs loaded with EVs derived from B16-F10-OVA mouse melanoma cells were shown to activate anti-tumor and anti-metastatic immune responses *in vivo* with slightly less intensity than that evoked by OVA-loaded DCs: tumor occurrence was observed in one of eight mice and zero of eight mice, and the mean number of metastases was 5 and 0, respectively (Yao et al., 2013). Such high efficiency of proteinpulsed DCs is suggested to be associated with the model tumor antigen OVA, which is abundantly expressed in tumor cells and does not reflect the actual expression of tumor-associated antigens. When tumor-associated antigens were used to load DCs or DC-derived EVs, higher immunogenicity of EV-loaded DCs over oncoprotein-loaded DCs could be explained by prolonged

presentation time and enhanced recovery of EVs in comparison with tumor proteins (Gu et al., 2015)

To enhance the immunogenicity of tumor-derived EVs, EV-secreting tumor cells or EVs themselves undergo modifications with different molecules or exposure to radiation. Tumor cell-derived EVs were shown to be indirectly (see Loading of EVs with NA) modified with the DC-targeted/-activating molecule CD40L (Wang et al., 2014), the Rab27a molecule that takes part in exosome secretion (Li et al., 2013), the shRNA silencing immunosuppressive TGF-β1 (Huang et al., 2017), or degraded cytosolic DNA (Diamond et al., 2018), as well as the pH-sensitive GALA-peptide that contributes to the endosomal release of EVs into the cytosol of DCs (Morishita et al., 2017) or the immunostimulatory CpG DNA (Morishita et al., 2016). All modified tumor cell-derived EVs mentioned above showed high immunostimulatory activity that exceeded the activity of unmodified EVs. Thus, it was demonstrated that DC-targeted tumor cell-derived EVs can be successfully used for the loading of DCs, and as such, tumor EV-loaded DC vaccines had higher anti-tumor efficiency in comparison with classic DC vaccines.

Nevertheless, it should be mentioned that it is necessary to carefully test tumor-derived EVs during preparation of anti-tumor vaccines, since tumor-derived EVs despite immunotherapeutic potential possess immunosuppressive properties that can lead to inhibition of immune cell functions and as a result to escape of tumor from immunosurveillance and formation of metastatic niche. A short review of immunosuppressive action of tumor-derived EVs on immune cells is presented below.

#### Immunosuppressive Action of Tumor Cell-Derived DC-Targeted EVs

As indicated above, tumor-derived EVs were successfully applied as efficient source of tumor antigens to load DCs that eventually activated robust anti-tumor immune response in different murine tumor models (see Tumor Cell-Derived DC-Targeted EVs and **Table 2**). Nevertheless, immunotherapeutic potential of tumorderived EVs is confused and controversial. It was demonstrated that tumor-derived EVs along with immune stimulatory action could also deliver tolerogenic signals to immune cells (Whiteside, 2016b).

The immunoinhibitory effects of tumor-derived EVs on immune cells can be direct (when signals or cargo delivered by EVs inhibit the targeted cell functions) or indirect (when EVs reprogram the differentiation program of targeted cells, which then suppress functions of other cells) (Whiteside, 2016b). Indeed, on the one hand, tumor-derived EVs carry FasL and TRAIL molecules and can directly induce apoptosis of DCs (Peng et al., 2011; Ning et al., 2018) or effector CD4+ and CD8+ T-lymphocytes (Whiteside, 2016a). On the other hand, tumorderived EVs can block differentiation of myeloid progenitor cells into CD11+ DCs and direct them to differentiate into myeloidderived suppressor cells (Xiang et al., 2009; Ning et al., 2018) or inhibit maturation and migration of DCs (Yang et al., 2011; Ning et al., 2018) that results in immunosuppression and favors tumor immune escape.

It was demonstrated that HSP72 and HSP105 expressed on the surface of tumor-derived EVs promoted DCs to produce high levels of pro-inflammatory cytokines (IL-6, PGE2, IL-1β, TNF-α) in a TLR2- and TLR-4–dependent manner (Shen et al., 2017). High levels of IL6 dramatically promoted tumor invasion and metastasis of murine melanoma by upregulation of transcription activity of STAT3 and production of STAT3-dependent MMP9 in tumor cells. It should be mentioned that depletion of IL-6 converted tumor-derived EVs from tumor promoters to inhibitors of tumor metastasis *in vivo* (Shen et al., 2017).

Human pancreatic cancer cell-derived EVs were shown to carry miR203 that inhibited expression of TLR4 and blocked secretion of IL-12 and TNF-α immunostimulatory cytokines in human DCs (Zhou et al., 2014).

Furthermore, tumor-derived EVs could drive differentiation and expansion of Treg cells (Huang et al., 2013; Whiteside, 2016a; Ning et al., 2018). Tumor cell-derived EVs induced the conversion of human conventional CD4+CD25– T cells to CD4+CD25hiFOXP3+ Treg cells in a TGF-β1–dependent manner (Wieckowski et al., 2009). Treg cells coincubated with tumor EVs were shown to upregulate the expression of FasL, TGF-β, IL-10, CTLA4, granzyme B, and perforin and exhibited enhanced suppressor functions (Szajnik et al., 2010).

NK cells being highly important in anti-tumor immunity due to induction an antigen-independent immune response against malignant cells (Fang et al., 2017) could be also inhibited by tumor cell-derived EVs (Whiteside, 2016b). It was demonstrated that NKG2D ligands (MICA, MICB, and ULBP) carried by tumor-derived EVs bound to the inhibitory receptor, NKG2D, on the surface of NK cells, with delivering of inhibitory signals and blocking of anti-tumor cytotoxicity of NK cells (Clayton et al., 2008; Lundholm et al., 2014). Another mechanism of tumor EV-mediated inhibition of NK functions was attributed to the presence of TGF-β1 in EVs cargo, a cytokine suppressing cytotoxicity of NK cells (Szczepanski et al., 2011).

It can be assumed that the choice of the immunostimulatory or immunosuppressive effects of the tumor cell-derived EVs depends on many factors: isolation methods of EVs from tumor cells, DC loading conditions, used immunostimulating molecules, the stage of the tumor process, and so on. In any case, preparation of antitumor vaccines on the basis of tumor cell-derived EVs should be associated with comprehensive and thorough investigation of their all biologic effects on immune system. On the other hand, the immunosuppressive properties of the tumor cell-derived EVs can be successfully applied to treat autoimmune diseases or to reduce graft-tohost immune reactions.

#### DC-Derived EVs

DC-derived EVs carry all the molecules needed for activation of a T-cell immune response and can act alone as cell-free anti-tumor vaccines. To efficiently activate anti-tumor immune responses by DC-derived EVs, the proper choices of tumor antigens for the loading of EV-producing DCs and factors stimulating the maturation of DCs, are of great importance. Significant success in the treatment of tumors by DC-derived EVs has been achieved in murine tumor models *in vivo* and human cells *ex vivo* (see **Table 2B**).

The anti-tumor potential of cell-free vaccines based on murine DC-derived EVs indirectly loaded with α-fetoprotein (AFP) has been evaluated (Lu et al., 2017). It was demonstrated that AFP-EVs activate an efficient antigen-specific immune response that causes significant retardation of tumor growth and an increase in the overall survival of mice with hepatocellular carcinoma Hepa1–6. Together with these effects, AFP-EVs reshaped the tumor microenvironment by attracting CD8+ T-lymphocytes to tumor sites and causing an increase in the levels of the immunostimulatory cytokines IFN-γ and IL-2 combined with reductions in immunosuppressive CD25+FoxP3+ T-regulatory cells and the levels of the immunosuppressive cytokines IL-10 and TGF-β (Lu et al., 2017).

Another tumor-associated antigen, HPV early antigen 7 peptide (E749–57), which is the main target antigen in cervical cancer, was used to load murine EV-producing DCs (Chen et al., 2018). The treatment of TC-1 cervical cancer-bearing mice with E749–57-loaded EVs induced an anti-tumor CTL response and activated potent protective and therapeutic immune responses *in vivo* (Chen et al., 2018).

Sources of multiple tumor antigens, such as tumor lysates or tumor total RNA, have advantages over single tumor antigens, since they contain a full set of tumor antigens and have the ability to activate a broad spectrum of polyclonal anti-tumor immune responses (Rizzo et al., 2014). Murine DC-derived EVs indirectly loaded with the lysate or chaperone-enriched lysate of GL261 mouse glioma cells have been obtained (Bu et al., 2015). DCs treated with EVs loaded with chaperone-enriched glioma lysate promoted the proliferation of CD4+ and CD8+ T cells and the activation of anti-tumor CTLs *ex vivo*, as well as the significant inhibition of tumor growth and prolonged survival of tumor-bearing mice. It was found that the chaperone-rich lysate contained at least four chaperone proteins including heat shock protein (Hsp) 70, Hsp 90, calreticulin, and glucose-regulated protein 94 (GRP94). Thus, chaperone-rich lysate is a superior source of tumor antigens for the loading of DCs and EVs in comparison with the lysate of tumor cells prepared using the conventional freezing–thawing method (Bu et al., 2015).

Human DC-derived EVs indirectly loaded with total tumor RNA or tumor lysate have been used to activate CTLs against human gastric adenocarcinoma BGC823 *ex vivo* (Guan et al., 2014). It was shown that DCs loaded with tumor RNA and EVs derived from these cells were more potent with respect to stimulating the proliferation of T cells and activated more efficient anti-tumor CTLs *ex vivo* in comparison with lysateloaded DCs and their EV derivatives. However, all tested DCs or EVs inoculated in combination with T cells into tumorbearing mice inhibited tumor growth with similar efficiency in a xenograft BGC823 tumor model (Guan et al., 2014)*.*

The anti-tumor potential of T-cell vaccines stimulated with DC-derived EVs has been investigated (Wang et al., 2013). CD4+ T lymphocytes were primed with EVs isolated from DCs expressing HER2/Neu or HER2 tumor antigens *ex vivo* and used as anti-tumor immunotherapeutic vaccines. It was demonstrated that this immunotherapeutic approach provides significant results: efficient tumor-specific CTLs are activated and protective immunity is triggered in mice with HER2/Neu+ Tg1-1 breast cancer or HLA-A2+HER2+ B16-F10 melanoma.

Immunotherapeutic anti-tumor approaches have been widely investigated in tumor cells stably expressing ovalbumin (OVA), which are used as a useful model of tumor-specific antigens. OVA and the OVA-derived MHC I-restricted peptide, SIINFEKL, were used for the indirect loading of DC-derived EVs, and it was demonstrated that only OVAloaded EVs, but not EVs loaded with the OVA-derived MHC I-restricted peptide, caused the induction of OVA-specific cytotoxic CD8+ T cells *in vivo*. The activation of CTLs was dependent on OVA-specific CD4+ T cells and B lymphocytes, since the full-length OVA protein contains both Th- and B-cell epitopes, whereas the SIINFEKL peptide is CD8+ T cell-specific. OVA-loaded EVs were superior in protecting the mice against B16-OVA tumor growth in comparison with peptide-loaded EVs (Naslund et al., 2013). A subsequent study by the same scientific group was devoted to comparing the ability of small- and large-sized EVs produced by OVA-loaded DCs to stimulate OVA-specific immune responses (Wahlund et al., 2017). Small-sized EVs (exosomes precipitated by ultracentrifugation at 100,000*g*) carried high levels of intact OVA inside and OVA peptides on their surface, whereas large-sized EVs (MVs pelleted by centrifugation at 10,000*g*) contained barely detectable levels of OVA. OVA-loaded smallsized EVs were shown to induce more potent OVA-specific Tand B-lymphocyte immune responses *in vivo* in comparison with large-sized EVs (Wahlund et al., 2017). It should be mentioned that small-sized EVs derived from immature DCs were shown to interact with T cells via CD80 signaling, inducing the secretion of Th1 cytokines (IFN-γ), whereas large-sized EVs induced Th2 cytokine synthesis (IL-4, IL-5, and IL-13) via CD40 signaling. Maturation of EV-producing DCs with LPS abolished the functional differences between small- and large-sized EVs: both types of EVs induced Th1 specific IFN-γ synthesis (Tkach et al., 2017).

DC-derived EVs have been indirectly loaded with a combination of model antigens, OVA and α-galactosylceramide (αGC), the latter of which is a ligand of invariant NKT cells (iNKT cells). Such EVs induced potent innate (NK and γδT lymphocytes) and OVA-specific T- and B-cell adaptive immune responses *in vitro* and *in vivo*. Using a mouse B16-OVA melanoma model it was demonstrated that DC-derived EVs loaded with OVA/ αGC caused significant retardation of tumor growth, promoted infiltration of tumor tissues with antigen-specific CD8+ T cells, and increased the median survival time of tumor-bearing mice in comparison with those treated with a combination of soluble OVA and αGC (Gehrmann et al., 2013).

Undoubtedly, NK cells play a significant role in tumor immunosuppression; therefore, the potent anti-tumor potential of DC-derived EVs to activate NK cells and induce adaptive immunity is of great importance. EVs produced by nonloaded mouse DCs and EVs secreted by human DCs loaded with MAGE3 peptides have been used to treat mouse YAC-1 lymphoma and human advanced melanoma (stages IIIb and V) in phase I clinical trials. It was demonstrated that both mouse and human DC-derived EVs carried functional IL-15Rα receptors that could transpresent IL-15 to NK cells to stimulate their activation, proliferation, and IFN-γ synthesis *in vitro* and *in vivo*. Additionally, activation of NK cells was shown to occur via the interaction of NKG2D ligands (ULBP-1, MICA/B), carried on the surface of DC-derived EVs, with NKG2D receptors on NK cells (Viaud et al., 2009). NK cells can also be activated through the interaction of TNF receptors on NK cells with TNF ligands expressed on the surface of DCs and their EVs. It has been clearly demonstrated that even intact EVs produced by mature DCs carrying transmembrane TNF are able to activate NK cells and stimulate IFN-γ synthesis (Munich et al., 2012). Furthermore, these EVs were shown to express other TNF superfamily ligands, such as FasL and TRAIL, and cause the direct induction of apoptosis of B16 melanoma, KLN205 lung squamous cell carcinoma, and MC38 colon adenocarcinoma cells *in vitro* (Munich et al., 2012). It has also been demonstrated that DC-derived EVs can bind LPS and Pam3CSK4, ligands of toll-like receptors (TLR) 4 and TLR1/2, respectively. DCs loaded with such EVs were shown to upregulate the expression of transmembrane TNF, activate NK cells, and stimulate NK cells to secrete of IFN-γ (Sobo-Vujanovic et al., 2014).

The level of anti-tumor immune response triggered by DC-derived EVs directly depends on the degree of maturity of DCs and the type of maturation stimuli. The anti-tumor potential of EVs derived from murine DCs loaded with OVA or HOCloxidized B16-OVA cells that were matured with poly(I:C) (TLR3 ligand), LPS (TLR4 ligand), and CpG-B oligonucleotide (TLR9 ligand) have been compared in a model of B16-OVA melanoma *in vivo* (Damo et al., 2015). It was demonstrated that all types of EVs were able to significantly retard tumor growth. However, EVs derived from necrotic B16 cell-loaded DCs treated with poly(I:C) were the most efficient and induced robust activation of melanoma-specific CD8+ T cells in tumor-draining lymph nodes, spleen, and tumor tissues and recruited NK and NK-T cells to the tumor site, resulting in drastic inhibition of tumor growth and an increase in survival in tumor-bearing animals (Damo et al., 2015). Moreover, EVs produced by poly(I:C)-matured DCs loaded with HPV early antigen 7 markedly inhibited murine TC-1 cervical tumor growth and improved the survival rate of tumor-bearing mice (Chen et al., 2018). Thus, it has been revealed that the TLR3 ligand poly(I:C) is a favourable TLR agonist for DC maturation during antigen loading, which significantly increased the potential for anti-tumor immunity induced by antigen-loaded EVs, and could be suggested as a promising maturation stimulus for DC-derived EVs.

EVs produced by heat shock-exposed DCs and tumor cells have proven themselves as efficient immunotherapeutic vaccines. Tumor cells are characterized by the unusual overexpression of heat-shock proteins, such as HSP70, HSP90, and HSP72, which protect tumor cells from apoptosis induction (Ferrarini et al., 1992; Jäättelä et al., 1992). Thus, these heat-shock proteins can serve as tumor-specific antigens to activate anti-tumor immune responses. Heat-stressed EVs (HS-EVs) are commonly isolated from DCs or tumor cells that are exposed to hyperthermia at 42°C to 43°C for 1 to 4 h *in vitro* (Chen et al., 2011; Zhong et al., 2011; Wang et al., 2015) or from human tumor cells isolated following induction of hyperthermia in patients at 39°C for 1 h (Guo et al., 2018). HS-EVs contained high levels of HSP70 and HSP60, indicating significant immunotherapeutic properties (Zhong et al., 2011). Such HS-EVs promoted maturation of DCs, activated proliferation of T-lymphocytes, and induced tumorspecific immune responses *in vivo* (Zhong et al., 2011; Guo et al., 2018). HS-EVs were also shown to contain the chemokines CCL2, CCL3, CCL4, CCL5, and CCL20, which chemoattracted CD11c+ DCs and CD4+/CD8+ T cells into tumor tissues *in vivo* (Chen et al., 2011). DCs treated with HS-EVs were able to convert immunosuppressive T-regulatory cells to Th17 cells, contributing to the rejection of established prostate cancer in mice (Guo et al., 2018).

As it has been reviewed above, accumulated material on the potential of DC-derived EVs confirmed that these EVs are able to directly initiate strong anti-tumor innate and adaptive immune responses *in vivo*, protect experimental animals against tumors in prophylactic settings, and significantly reduce tumor growth and metastasis in therapeutic regimen.

Furthermore, the immunogenic potential of DC-derived EVs has been clearly demonstrated in phase I and II clinical trials (Escudier et al., 2005; Morse et al., 2005; Dai et al., 2008; Viaud et al., 2009; Besse et al., 2016) (see **Table 2C**). DC-derived EVs were shown to be safe for patients and to stimulate anti-tumor CTLs and NK cells. However, the actual clinical potential of DC-derived EVs to trigger anti-tumor immune responses, reduce tumor size and metastasis, and increase the survival time in patients remains undefined. It is hoped that DC-derived EVs will be more successful anti-tumor vaccines in comparison with DC-derived vaccines.

### CONCLUDING REMARKS—EVS PROBLEMS AND OPENED QUESTIONS

As mentioned above, great success is achieved on application of DC-targeted/-derived EVs to treat tumors in different murine tumor models and promising results are obtained in phase I and II clinical trials. Experimental data obtained to date points to similar or even superior ability of DC-targeted or DC-derived EVs to activate anti-tumor immune responses in comparison with classic DC-based vaccines. To enhance the anti-tumor and immunogenic potential of EVs, they are modified with a variety of molecules, such as tumor-associated antigens, tumoror DC-targeted molecules, small non-coding regulatory RNA, and immunostimulatory molecules. Almost all methods of EV modification to possess anti-tumor properties are considered in the present review.

Nevertheless, some problems and opened questions on antitumor application of EV-based vaccines are still unsolved.

First, physiological activity of EVs is still poorly understood. In general, it is not yet entirely clear how EV-mediated paracrine regulation between cells occurs. The mechanism of selection of biologically active molecules (such as miRNA, cytokines, peptides, etc.) to EVs is not completely elucidated. Data on targeted delivery of natural EVs to specific cells and tissues are insufficient. Therefore, further comprehensive and in-depth investigations of biological properties of EVs are of immediate interest.

One of the main problems of EVs to investigate their biological properties and use them in clinical trials is inability of modern methods to isolate pure fractions of EVs without any mixture of residual subpopulations of other vesicles or even non-vesicular particles (Chulpanova et al., 2018). Hence, it is challenging to establish Clinical Good Manufacturing Practice (cGMP)-grade EVs preparations. Novel isolation methods are required to enrichment of the specific EVs subtypes. Therefore, better knowledge of specific markers of EVs subtypes is required.

Application of EV-based anti-tumor vaccines is associated with additional questions. How should dosing of inoculated EVs be determined? It is known that EV-mediated signaling is dose-dependent (Yu et al., 2007), so variation of EVs dose is able to impact on the balance between deleterious and therapeutic potential of administered EVs. In addition, does the route of administration (subcutaneous, intradermal, intravenous, etc.) impact the efficiency of EV-based antitumor therapy?

Additionally, to prepare sufficient amounts of therapeutic EVs, a big number of EV-producing cells and a huge volume of EV-containing condition media should be processed, that significantly complicates technological process and increases the price of potential antitumor EV-based vaccine. Highly scalable methods for mass production of EVs are required at all stages of the manufacturing process.

Finally, targeting of EVs to specific cells and tissues requires further optimization, as well as more efficient techniques for loading EVs with NAs, proteins, lipids should be developed.

In the near future these problems and questions will need to solve by enthusiastic multidisciplinary collaboration of molecular biologists, immunologists, biochemists, together with physicians to develop highly efficient next generation of EV-based antitumor vaccines.

## AUTHOR CONTRIBUTIONS

OM analyzed published data and prepared the manuscript. AO prepared the part of the manuscript devoted to delivery of nucleic acids using extracellular vesicles. NM revised and corrected the manuscript.

## FUNDING

The present work was funded by the Russian Science Foundation (grant 17-74-10144) (OM), the Russian Foundation for Basic Research [grants 17-04-01136 (AO), 17-04-00999 (NM)], and the Russian State funded budget project of ICBFM SB RAS АААА-А17-117020210024-8 (NM).

### REFERENCES


leukemia suppress natural killer cell function *via* membrane-associated transforming growth factor-β1. *Haematologica* 96, 1302–1309. doi: 10.3324/ haematol.2010.039743


T cell expansion and induce apoptosis in tumor-reactive activated CD8 + T lymphocytes. *J. Immunol.* 183, 3720–3730. doi: 10.4049/jimmunol.0900970


**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 Markov, Oshchepkova and Mironova. 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.*

# Are Small Nucleolar RNAs "CRISPRable"? A Report on Box C/D Small Nucleolar RNA Editing in Human Cells

*Julia A. Filippova1†, Anastasiya M. Matveeva1,2†, Evgenii S. Zhuravlev1, Evgenia A. Balakhonova1, Daria V. Prokhorova1,2, Sergey J. Malanin3, Raihan Shah Mahmud3, Tatiana V. Grigoryeva3, Ksenia S. Anufrieva4,5, Dmitry V. Semenov1, Valentin V. Vlassov1 and Grigory A. Stepanov1,2\**

#### *Edited by:*

*Hector A. Cabrera-Fuentes, University of Giessen, Germany*

#### *Reviewed by:*

*Stephanie Kehr, University Hospital Leipzig, Germany Christopher Holley, Duke University, United States*

#### *\*Correspondence:*

*Grigory A. Stepanov stepanovga@niboch.nsc.ru*

*†These authors have contributed equally to this work*

#### *Specialty section:*

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

*Received: 07 June 2019 Accepted: 27 September 2019 Published: 04 November 2019*

#### *Citation:*

*Filippova JA, Matveeva AM, Zhuravlev ES, Balakhonova EA, Prokhorova DV, Malanin SJ, Shah Mahmud R, Grigoryeva TV, Anufrieva KS, Semenov DV, Vlassov VV and Stepanov GA (2019) Are Small Nucleolar RNAs "CRISPRable"? A Report on Box C/D Small Nucleolar RNA Editing in Human Cells. Front. Pharmacol. 10:1246. doi: 10.3389/fphar.2019.01246*

*1 Institute of Chemical Biology and Fundamental Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia, 2 Department of Natural Sciences, Novosibirsk State University, Novosibirsk, Russia, 3 Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, 4 Department of Biological and Medical Physics, Moscow Institute of Physics and Technology (State University), Moscow, Russia, 5 Laboratory of Cell Biology, Federal Research and Clinical Center of Physical-Chemical Medicine, Federal Medical Biological Agency, Moscow, Russia*

CRISPR technologies are nowadays widely used for targeted knockout of numerous protein-coding genes and for the study of various processes and metabolic pathways in human cells. Most attention in the genome editing field is now focused on the cleavage of protein-coding genes or genes encoding long non-coding RNAs (lncRNAs), while the studies on targeted knockout of intron-encoded regulatory RNAs are sparse. Small nucleolar RNAs (snoRNAs) present a class of non-coding RNAs encoded within the introns of various host genes and involved in post-transcriptional maturation of ribosomal RNAs (rRNAs) in eukaryotic cells. Box C/D snoRNAs direct 2'-O-methylation of rRNA nucleotides. These short RNAs have specific elements in their structure, namely, boxes C and D, and a target-recognizing region. Here, we present the study devoted to CRISPR/ Cas9-mediated editing of box C/D snoRNA genes in *Gas5*. We obtained monoclonal cell lines carrying mutations in snoRNA genes and analyzed the levels of the mutant box C/D snoRNA as well as the 2'-O-methylation status of the target rRNA nucleotide in the obtained cells. Mutations in *SNORD75* in the obtained monoclonal cell line were shown to result in aberrant splicing of *Gas5* with exclusion of exons 3 to 5, which was confirmed by RT-PCR and RNA-Seq. The obtained results suggest that *SNORD75* contains an element for binding of some factors regulating maturation of Gas5 pre-lncRNA. We suggest that METTL3/METTL14 is among such factors, and m6A-methylation pathways are involved in regulation of *Gas5* splicing. Our results shell light on the role of *SNORDs* in regulating splicing of the host gene.

Keywords: snoRNA, Gas5, genome editing, CRISPR/Cas9, box C/D snoRNA, RNA modification, alternative splicing, m6A

## INTRODUCTION

Box C/D snoRNAs present one of the two subclasses of small nucleolar RNAs (snoRNAs) responsible for post-transcriptional maturation of ribosomal RNAs (rRNAs) in eukaryotic cells: they guide 2'-O-methylation (2'-O-Me) of rRNA nucleotides (Kiss-László et al., 1996). Ribose methylation (2'-O-Me) is one of the most frequent types of nucleotide modification (alongside with pseudouridylation) in eukaryotic rRNA, with each of the 2'-O-Me sites being modified by a specific box C/D snoRNA (Cavaillé et al., 1996; Kiss, 2001; Piekna-Przybylska et al., 2007). Box C/D snoRNAs, in their turn, can have one or two targets: there are one or two 10–21 nucleotide guide sequences in the structure of snoRNA, which exhibit complementarity to a specific region within rRNA. There are also conserved elements, so-called boxes C and D, in the structure of these regulatory RNAs; these elements are required for the recognition of snoRNA-associated proteins followed by formation of the functionally active small nucleolar ribonucleoprotein (snoRNP) complexes (**Figure 1A**) (Terns and Terns, 2002; Reichow et al., 2007; Massenet et al., 2017). Terminal regions of a box C/D snoRNA molecule are complementary to each other; they, altogether with boxes C and D, form a stem-bulge-stem structure named "Kink-turn" (K-turn) (**Figure 1B**) (Watkins et al., 2000; Szewczak, 2005). The K-turn is recognized by core box C/D snoRNA proteins and required for proper processing, functioning, and localization of a mature snoRNP (Cavaillé and Bachellerie, 1996; Watkins et al., 1996; Škovapačková et al., 2010).

Apart from the canonical role of box C/D snoRNAs, several members of the class are known to perform other functions in the cell. According to the snoRNABase (www-snorna. biotoul.fr) (Lestrade, 2006), over a half of all human box C/D snoRNAs are orphan snoRNAs, since they have no identified 2'-O-methylation targets, while the real function in the cell remains unknown for most of them (Deschamps-Francoeur et al., 2014). However, for some of these snoRNAs, noncanonical functions have been elucidated. For instance, U3, U8, and U13 snoRNAs perform endonucleolytic cleavage of pre-rRNA (precursor of rRNA) and ensure correct folding of the resulting rRNA (Kass, 1990; Peculis and Steitz, 1993; Cavaille et al., 1996). SNORD115 (M/HBII-52) is involved in the regulation of the serotonin 2C receptor (5-HT2CR) mRNA level through alternative splicing and control of the target mRNA editing (Vitali et al., 2005; Kishore, 2006). SNORD115 and SNORD116 (M/HBII-85), both of which are encoded within the same locus (SNURF-SNRPN), are processed into smaller RNA forms, which, in their turn, are associated with splicing of various mRNA precursors (Kishore et al., 2010). A series of box C/D snoRNAs were shown to be further processed into miRNA-like derivatives, namely, snoRNA-derived RNAs (sdRNAs). Some of these snoRNA derivatives not only undergo Dicer-dependent processing pathway and associate with Ago family proteins (Ender et al., 2008; Burroughs et al., 2011) but also demonstrate miRNA activity (Brameier et al., 2011; Li et al., 2011; Patterson et al., 2017). Long RNA forms containing snoRNA in their structure (sno-lncRNAs) have been also detected in human cells (Yin et al., 2012). A novel function has been found for two orphan box C/D snoRNAs in yeasts: guiding of acetylation of two cytosine residues in 18S rRNA (Sharma et al., 2017). Recent papers also demonstrate evidence that individual snoRNAs guide 2'-O-methylation of tRNA (Vitali and Kiss, 2019) and mRNA (Elliott et al., 2019). A series of studies have also demonstrated the involvement of snoRNAs in such cellular processes as PKR activation (Youssef et al., 2015; Stepanov et al., 2018), cellular response to lipotoxicity (Michel et al., 2011; Holley et al., 2015), cholesterol trafficking (Brandis et al., 2013), and glucose metabolism (Lee et al., 2016).

One of the main protein components of a box C/D snoRNP is fibrillarin, which presents a 2'-O-methyltransferase (Tollervey et al., 1993; Kiss-László et al., 1996). Small nucleolar RNA species that do not perform ribose methylation are associated with different proteins than 2'-O-methylating ones. Their ribonucleoprotein complexes lack fibrillarin and other canonical snoRNP proteins (Falaleeva et al., 2016). Instead of that, these snoRNAs are a part of a spliceosome or associated with various RNA-binding proteins such as hnRNPs, ELAVL1, and RNA helicases (Tycowski et al., 1996; Soeno et al., 2010). Thus, the recent results indicate a vast variety of snoRNA roles as well as their structural forms that are found in the cells and required for implementation of their non-canonical functions.

Taking into account the specificity of the structure of box C/D snoRNAs and their target recognition ability, this class of regulatory RNAs presents a promising model for obtaining novel regulators of various processes, including post-transcriptional maturation. In addition, due to the fact that snoRNAs are encoded within the introns of various host genes, their knockout or mutation will not result in a frameshift and not necessarily lead to any drastic changes in the expression of the host gene, which is an another crucial aspect for selecting snoRNAs as a model in such studies. The aim of the study was to assess the possibility of selective editing of snoRNA genes in human cells using CRISPR/Cas9 tools.

#### MATERIALS AND METHODS

#### Plasmids and Oligonucleotides

A number of protospacer sequences were selected for specific cleavage of snoRNA genes encoded within *Gas5* (growth arrest-specific 5) introns. The protospacers were tested for possible off-target effects using Benchling tool (Benchling, RRID : SCR\_013955). Plasmid pSpCas9(BB)-2A-GFP (pX458) (Addgene, #48138) was used as the expression vector (Ran et al., 2013). The corresponding oligonucleotides ("top and bottom strands" in **Figure 2C**) were annealed and cloned into the pX458 vector using BstV2I restriction endonuclease (SibEnzyme, Russia) and T4 DNA ligase (Thermo Fisher Scientific) according to (Ran et al., 2013). Competent TOP10 *Escherichia coli* cells were transformed with the obtained constructs, spread onto LB agar plates supplemented with ampicillin and incubated overnight at 37°C. Colonies containing pX458 plasmid with single guide

strands of sgRNAs. Overhangs for ligation into the pair of BstV2I sites in pX458 are shown in blue italics.

RNA (sgRNA) insertion were selected by colony PCR and Sanger sequencing; CRISPR/Cas9 expression vectors were isolated using "EndoFree Plasmid Maxi Kit" (Qiagen).

### Cell Culture and Transfection

Human 293FT cell line (Thermo Fisher Scientific) was used in the study. Cells were cultured in DMEM/F12 medium containing 10% FBS and supplemented with 1x solutions of MEM NEAA, sodium pyruvate, GlutaMax, and Anti-Anti with addition of MycoZap Prophylactic (200 µl per 100 ml of medium) at 37°C and 5% CO2. All medium components, except for MycoZap (Lonza), were purchased from Gibco. Cells were seeded in six-well plates at a density of ~0.3 x 106 cells per well 24 h prior to transfection. Transfection of the cells with the expression vector was performed in RPMI medium using Lipofectamine 3000 Reagent (Thermo Fisher Scientific) according to the manufacturer's instructions. Cells transfected with pX458 plasmid without sgRNA were used as the control.

#### Single Clone Selection and Identification of Mutations

Cells were seeded at the amount of 1 cell per well in a 96-well plate by FACS (S3e Cell Sorter, Bio-Rad) 48 h after transfection. After reaching a ~80–90% confluency, cells were divided into two equal portions and seeded in two 96-well plates. One of the plates was used for mutation screening by T7 endonuclease I (T7EI) cleavage assay (Kim et al., 2009). Genomic DNA was isolated using genomic DNA Isolation Kit (BIOLABMIX Ltd., Novosibirsk, Russia), PCR was performed using specific primers (5'-AGCCTTTGTCTGCTAAGGTCA-3' and 5'-GTTGCCAT TAACCGATGTCGA-3' for *SNORD74*, 5'-TGGTATGTTACC TGCATCATTGG-3' and 5'-TAGGTGTACTCTCTATGTT CCC-3' for *SNORD75*, 5'-GAGTGCTAGAATGATGAGG-3' and 5'-TCCAGCTTTCTGTCTAATGCC-3' for *SNORD77*, 5'-ATTACAGGCATGTGACACC-3' and 5'-CACTCCCA TCTACAGATTAAGG-3' for *SNORD80*), and the amplification products were annealed and subjected to T7EI (NEB) according to the manufacturer's protocol. Mutations were identified by TA-cloning of the PCR products using CloneJET Kit (Thermo Fisher Scientific) and *E. coli* strain XL1-Blue followed by Sanger sequencing with further analysis of the obtained data by Tracking of Indels by Decomposition (TIDE) assay (Brinkman et al., 2014).

### Isolation of Total Cell RNA

Total RNA was isolated from control cells and clones using LIRA reagent (BIOLABMIX Ltd., Novosibirsk, Russia) according the manufacturer's protocol and analyzed on a 1.5% agarose gel or using an Agilent 2100 Bioanalyzer.

### Differential Gene Expression Analysis

PolyA RNA fraction analysis with sequencing was performed using an Illumina NextSeq platform. Sequencing data (FASTQ formatted reads) were applied to the RNA-Seq workflow, which includes removing of adaptor-sequences with Trimmomatic V 0.38 (Bolger et al., 2014), mapping reads with HiSAT2 (Kim et al., 2015) on hg19, and transcriptome assembly with Cufflinks (NCBI RefSeq). The comparison of the expression levels of genes and transcripts in RNA-Seq experiments was carried out using CuffDiff (Trapnell et al., 2012). The list of differentially expressed genes (CuffDiff FDR adjusted after Benjamini–Hochberg correction of *p*-value for multipletesting *q* < 0.05) was applied to the gene enrichment analysis powered by the Enrichr platform (Kuleshov et al., 2016). The RNA-Seq data have been deposited in ArrayExpress database under accession number E-MTAB-8269. Differential splicing analysis was performed using rMATS splicing tool as described (Anufrieva et al., 2018).

### Real-Time RT-PCR

Prior to RT-PCR, total RNA was isolated from the cells and treated with DNase I (Thermo Fisher Scientific). Quantitative RT-PCR was performed using BioMaster RT-PCR SYBR Blue reaction mix (BIOLABMIX Ltd., Novosibirsk, Russia) on a LightCycler 96 (Roche, Switzerland) with the following primers: U74: 5'-CTGCCTCTGATGAAGCCTGTG-3' (U74-f) and 5'-CCACCATCAGAGCGGTTG-3' (U74-r) or 5'-GAGCGG TTGGCATTCATC-3' (U74-all-r); U75: 5'-GTCGTATCCAGT GCAGGGTCCGAGGTATTCGCACTGGATACGACAG CCTC-3' (U75-SL-sl-r), 5'-GTATACAGCCTGTGATGCTTT-3' (U75-SL-f), 5'-GTGCAGGGTCCGAGGT-3' (U75-SL-r), and 5'-FAM-TGGATACGACAGCCTCAG-BHQ1-3' (U75- SL-probe); U77: 5'-AGATACTATGATGGTTGC-3' (U77-f) or 5'-ATGATGGTTGCATAGTTCAG-3' (U77-all-f) and 5'-GA TACATCAGACAGATAG-3' (U77-r); U80: 5'-ACAATGATGA TAACATAG-3' (U80-f) and 5'-GATAGGAGCGAAAGACT-3' (U80-all-r) or 5'-CATCAGATAGGAGCGAA-3' (U80-r); Gas5: 5'-GAGGTAGGAGTCGACTCCTGTGA-3' (exon 1 forward), 5'-GTGGAGTCCAACTTGCCTGGAC-3' (exon 6 forward), 5'-CTGCATTTCTTCAATCATGAAT-3' (exon 9 reverse); U1: 5'-CAGGGGAGATACCATGATCACGAAG-3' and 5'-CGC AGTCCCCCACTACCACAAAT-3' U6: 5'-TCGCTTCGGCAG CACATATACTAAAAT-3' and 5'-GAATTTGCGTGTCATCCT TGCG-3' U8: 5'- AATCAGACAGGAGCAATCA-3' and 5'-ATC GTCAGGTGGGATAATCCT-3' HPRT: 5'-CATCAAAGCACTG AATAGAAAT-3' and 5'-TATCTTCCACAATCAAGACATT-3' B2M: 5'-CGCTCCGTGGCCTTAGCTGT-3' \_and 5'-AAAGA CAAGTCTGAATGCTC-3' 18S rRNA: 5'-GATGGTAGTC GCCGTGCC-3' and 5'-GCCTGCTGCCTTCCTTGG-3' U47: TaqMan MicroRNA Assay #001223 (Thermo Fisher Scientific).

For assessment of the level of wild-type snoRNAs, the following primers were used: U74-f and U74-r (U74 RNA); U75-SL-sl-r, U75-SL-f, U75-SL-r and U75-SL-probe (U75 RNA); U77-f and U77-r (U77 RNA); and U80-f and U80-r (U80 RNA). For evaluation of the total level of all mutant RNA forms of the target snoRNA in the corresponding clone, the following primers were used: U74-f and U74-all-r (U74 RNA forms); U75- SL-sl-r, U75-SL-f and U75-SL-r (U75 RNA forms); U77-all-f and U77-r (U77 RNA forms); and U80-all-r and U80-f (U80 RNA forms).

The expression of target genes is presented as values normalized to the endogenous level of 18S rRNA, *HPRT, B2M* mRNA, U1, U6, U8, and U47 RNA. The mean values [ ± standard deviation (SD)] from three independent experiments were represented.

#### Analysis of the Relative Level of 2'-O-Me of the Target rRNA Nucleotide Partial Alkaline Hydrolysis

Partial alkaline hydrolysis of RNAs was performed as described in (Kiss-László et al., 1996). Reverse transcription was performed using primers containing a 5'-terminal [32P] label: 5'-CGTTCCCTTGGCTGTGGT-3' (C3820 28S rRNA, U74 RNA), 5'-GCCTCACCGGGTCAGTGA-3' (C4032 28S rRNA, U75 RNA), and 5'-GTCAGGACCGCTACGGACCTC-3' (A1521 28S rRNA, U77/U80 RNA). Sequencing of the region of 28S rRNA was performed as described in (Filippova et al., 2015).

#### RT-PCR-Based Method

Reverse transcription followed by PCR with modificationspecific primers was performed using total RNA samples isolated from the control and monoclonal cells. The following primer sets were used for analysis of the methylation status of C3820 28S rRNA, C4032 28S rRNA, and A1521 28S rRNA, respectively: forward 5'-GAACGAGATTCCCACTG-3,' reverse 5'-CCGTTC CCTTGGTGTG-3,' inside primer 5'-GATTCCCACTGTC CCTACC-3'; forward 5'-CCGCCGGTGAAATACCA-3,' reverse 5'-AACTCCCCACCTGGCACT-3,' inside primer 5'-GAA ATACCACTACTCTGATCG-3'; and forward 5'-AGGACCCGA AAGATGGTGA-3,' reverse 5'-GTCAGGACCGCTACGGAC CTC-3,' inside primer 5'-AAGATGGTGAACTATGCCTG-3.' For each of the samples, reactions with 1.0 mM (or 1.5 mM) and 3.0 mM (or 0.01 mM) dNTP concentrations were performed in parallel. Relative change in the modification level of the target nucleotide was evaluated based on the difference between the amplification level for the study and control samples at suboptimal dNTP concentration. The approach is based on the method of identification of 2'-O-Me groups in rRNA by RT-PCR first presented by Belin et al. (2009).

#### RNase H- and HPLC-MS/MS-Based Method

For analysis of the 2'-O-methylation status by HPLC-MS/MS, rRNA was separated from short RNA forms using miRNA isolation kit LRU-100-50 (BIOLABMIX Ltd., Novosibirsk, Russia). A total of 3 µg of rRNA were incubated with 1µM oligonucleotides in a buffer containing 20 mM Tris-HCl, 40 mM KCl, 8 mM MgCl2, and 1 mM DTT (pH 7.8) at 37 °C for 30min. The following pair of oligonucleotides was used: 5'-CACTTATTCTACACACCTC-3' and 5'-CTCCCCCCACGGCACTGTC-3'

Next, RNase H (Thermo Scientific, USA) was added to the reaction mixture to a final concentration of 80 U/ml and incubated at 37°C for 2 h. The RNase H cleavage products were precipitated in 2% LiClO4/acetone and then separated on a denaturing Page gel (**Supplementary Table 1**). The fragment of interest was subjected to enzymatic hydrolysis to nucleosides (Chan et al., 2010). High performance liquid chromatography coupled with tandem mass spectrometry (HPLC-MS/MS) was performed for quantitative assessment of the 2'-O-methylation status of the target nucleotide as described earlier (Stepanov et al., 2018).

### Real-Time Cell Proliferation Analysis

The viability and the number of cells were evaluated on the automated cell counter LUNA-II (Logos Biosystems, South Korea) using Trypan Blue Dye (Bio-Rad Laboratories). Cell proliferation was assessed by real-time cell analysis using electrical impedance assay—xCELLigence System (ACEA Bioscience, San Diego, CA, USA). RTCA software was used to determine CI values through the measured impedance recordings. Briefly, cells were plated in 16-well E-plates (ACEA Bioscience) at a density of 20,000 cells per well in a total volume of 200 µl of complete medium and were monitored in real-time mode. The data were recorded every hour for 62 h; cell indexes were calculated using RTCA Software 1.2 (ACEA Bioscience). Cell index is a parameter reflecting the impedance of electron flow caused by adherent cells.

#### Statistical Analysis

Unpaired Student's t-test was used to confirm the statistical significance of the differences (data are presented as means ± SD). The differences were considered statistically significant at *p*-value < 0.05 (\*), *p*-value < 0.01 (\*\*), and *p*-value < 0.001 (\*\*\*).

### RESULTS

### Design of sgRNAs Targeted at snoRNAs

We selected *Gas5* (*growth arrest-specific 5*) as the target gene, since it presents a well-studied multi-small-nucleolar-RNA host gene. A total of 10 box C/D snoRNAs are encoded within the introns of *Gas5*: *SNORD74*, *SNORD75*, *SNORD76*, *SNORD77*, *SNORD44*, *SNORD78*, *SNORD79*, *SNORD80*, *SNORD47*, and *SNORD81* (**Figure 2A**) (Smith and Steitz, 1998). Analysis of the genomic sequences of *Gas5* snoRNAs demonstrated the presence of protospacer adjacent motifs (PAMs, 5'-NGG-3') in all of these box C/D snoRNAs. However, not all of the snoRNAs contain PAM sequences in the vicinity of the key functional elements, namely, boxes C, C,' D, and D' (**Figure 2**). The desired PAM positions were those located adjacent or within the conserved elements, since these regions are responsible for the recruitment of snoRNA-specific proteins and formation of the functionally active snoRNP complex. Even point mutations at specific positions, especially within the structure of functional elements, are known to abrogate snoRNA function (Cavaillé et al., 1996).

*SNORD74* and *SNORD75* contain more than one PAM in their structure (**Supplementary Table 3**). Of these, we have selected the cleavage sites located within the regions of boxes D and D' in *SNORD74* and *SNORD75*, respectively (**Figure 2B**). *SNORD77* and *SNORD80*, on the contrary, contain only one PAM site, which is located within the regions of boxes C and D, respectively (**Figure 2B**, **Supplementary Table 3**). Four sgRNAs were constructed, with each of them targeting cleavage of either *SNORD74*, *SNORD75, SNORD77*, or *SNORD80*: 74-4, 75-2, 77-1, and 80-1, respectively (**Figure 2C**). Plasmids expressing the designed sgRNAs were obtained using the oligonucleotides presented in **Figure 2C**.

### CRISPR-Mediated Mutations in Conserved Elements Resulted in Downregulation of the Target snoRNAs

After transfection of 293FT cells with the obtained plasmids, they were sorted for GFP-positive cells, and viable monoclonal lines carrying mutations in the target snoRNA-encoding genes were selected. T7 endonuclease I assay demonstrated the presence of mutations in *SNORD74* (293FT-74-4 line), *SNORD75* (293FT-75-2 line)*,* and *SNORD77* (293FT-77-1 line) (**Figure 3A**).

Further Sanger sequencing followed by TIDE assay revealed the presence of a 5-nt and a 11-nt deletions (74-4\_5del and 74-4\_11del, respectively) in *SNORD74* for clone 293FT-74-4 (**Figure 3B**). TA-cloning with sequencing of individual alleles demonstrated that both mutations partially covered the D box region as well as the 3'-terminal region involved in the Kinkturn formation (**Figure 3C**) (Watkins et al., 2000; Szewczak, 2005). A 9-nt and a 10-nt deletions (75-2\_9del and 75-2\_10del, respectively) were identified for *SNORD75* in the clone 293FT-75-2 (Figures 3B, C). These deletions covered partially (75- 2\_10del) or entirely (75-2\_9del) the D' box region. Clone 293FT-77-1 was shown to carry a 1-nt insertion (77-1\_1ins) and a deletion of 11 nucleotides (77-1\_11del) in the region adjacent to the C box sequence in *SNORD77* (**Figures 3B**, **C**).

Initially, no endonuclease I digestion products were obtained for the clone 293FT-80-1 (**Figure 3A**). We suggested that this is due to the presence of identical mutations in both alleles of *SNORD80*, since downregulation of the target snoRNA was observed in the obtained monoclonal cells. Indeed, TA-cloning with Sanger sequencing and analysis using TIDE software revealed a 9-nt deletion in the D box terminal region of the two alleles of *SNORD80* (**Figure 3B**). The mutation partially covered the D box and the 3'-terminal regions (**Figure 3C**).

Thus, four monoclonal cell lines were obtained, with each of them carrying mutations in both alleles of one of the target snoRNA genes: *SNORD74*, *SNORD75*, *SNORD77*, and *SNORD80*.

First, we confirmed the absence of wild-type forms of the four target snoRNAs in the corresponding monoclonal cells by qRT-PCR (**Figure 4A**). Next, in order to evaluate the level of the target snoRNA in the corresponding clone, we used primers that allow detection of the mutant snoRNA forms. Real-time RT-PCR analysis demonstrated a decrease in the total level of each of the four snoRNAs in the corresponding monoclones. For instance, mutant *SNORD77* was expressed at 25% of the wildtype level, while mutant U74 RNA was not detected by RT-PCR in 293FT-74-4 (**Figure 4B**). The level of other *Gas5 SNORDs* also changed but not as dramatically as that for the target *SNORDs* (**Figure 4B**). The most significant changes were observed for *SNORD74* in 293FT-77-1 (1.6-fold increase compared the level of intact *SNORD74* in the control cells), *SNORD77* RNA in 293FT-75-2 (1.5-fold decrease), and *SNORD80* RNA in 293FT-75-2 (1.5-fold increase) and 293FT-77-1 (two-fold increase) cells.

We analyzed the nature of mutations in order to determine whether they can provide functional snoRNA forms that can further form snoRNP complexes and guide 2'-O-methylation of their target nucleotides. Since mutations in most of the obtained monoclones (except for *SNORD75* in 293FT-75-2) cover terminal conserved elements (**Figure 3C**), they might prevent formation of the proper K-turn structure. It is known that association of the snoRNA with the core snoRNP proteins, especially 15.5 kDa, is impossible in case if an aberrant C/D-motif is formed (Watkins et al., 2002).

We concluded that, of all of the target *SNORDs*, only *SNORD77* in 293FT-77-1 might provide functional snoRNA forms. Mutations in the alleles of *SNORD77* differ significantly: there is a 11-nt deletion of the 5'-terminal area in one of the alleles (77-1\_11del) and a 1-nt insertion in the C-box-adjacent area (**Figure 3C**). Deletion of such a vast terminal region of 11 nt in 77-1\_11del seems deleterious for formation of a proper canonical stem structure. We tried to estimate whether the region could be substituted with the adjacent intronic sequence in the mature RNA form (**Figure 5**). However, this sequence does not provide enough complementarity to form a canonical stem of the K-turn (Chagot et al., 2019). On the opposite, addition of one nucleotide (adenosine) beyond the functionally important region might not have such a tremendous effect, and U77\_1ins might still provide a functional snoRNA (**Figure 5**).

However, the overall downregulation of *SNORD77* in the monoclone 293FT-77-1 indicates that the addition of one nucleotide in the K-turn area changes the spatial structure of the snoRNA and affects its interaction with snoRNA-associated proteins thus resulting in its dysfunction. The mutation carried in 77-1\_1ins allele might affect the spatial parameters such as the angle between the two stems (canonical and non-canonical stem) within the K-turn structure and impair the snoRNA stability.

Mutations in *SNORD75* cover such essential elements as the D' box area (75-2\_9del) and the guide region (75-2\_10del) (**Figure 3C**). Despite of the partial or complete deletion of the D' box in 75-2\_10del and 75-2\_9del, respectively, the element can be substituted with a D/D'-box-resembling sequence (CUGA), which is located in the structure between the boxes D and D' in wild-type U75. However, both mutations result in a significant shortening of U75 sequence sequence, as well as the distance between the boxes C and D' and boxes D and D,' and these parameters are known to be crucial for snoRNA functioning (Qu et al., 2011). Thus, it is unlikely that, if the mutant forms are somehow processed, the resulting snoRNA is not functional.

CRISPR/Cas9 cleavage of *SNORD74* and *SNORD80* resulted in impairment of the D box region (**Figure 3C**). Analysis of the structure of the mutant forms for these snoRNAs demonstrated that these variants cannot form a proper K-turn structure (**Supplementary Table 1 Figure 1**).

No significant differences in the growth rate were observed for all of the obtained clones compared to the control 293FT cells (**Figure 6**). 293FT-74-4 and 293FT-77-1 clones were characterized by insignificantly divergent growth rates compared to 293FT cells (**Figure 6B**). Functional analysis of RNA-Seq profiling of gene expression in obtained monoclonal and control cells did not reveal any significant activation of cell death pathways (**Supplementary Table 4**). The obtained results demonstrate that CRISPR/Cas9-mediated cleavage of snoRNA genes does not affect essential cellular processes and therefore can be used for obtaining stable cells expressing mutant snoRNAs.

FIGURE 3 | (A–C) CRISPR/Cas9-mediated mutations in the *Gas5 SNORD*s in the obtained monoclonal cell lines. (A) T7 Endonuclease I assay with specific primers. Lengths of the Endonuclease restriction products correspond to the position of the CRISPR/Cas9 cleavage site in the target *SNORD*. (B) Sequencing of the genomic region of the target *SNORD* in the obtained cell lines followed by TIDE analysis. Position on the X axis indicates the number of deleted (negative value) or inserted (positive value) nucleotides, while Y value shows percentage distribution of the detected mutations. (C) Mutations in the alleles of the target *SNORD* in the obtained clones. Genomic sequences of the target snoRNAs are shown in capital letters. Conserved elements are framed in green and purple. The nucleotide denoted in ochre yellow in the guide region is complementary to the 2'-O-methylation site in the target rRNA. Insertion of 1 nucleotide in *SNORD77* is denoted by red font. Deletion is indicated by dash line. WT, wild type allele.

#### CRISPR-Mediated Mutations in snoRNAs Affect 2'-O-Methylation of rRNA

Since mutations in the structure of box C/D snoRNAs can lead to the loss of their functional ability to guide 2'-O-methylation, the target rRNA nucleotide might not contain the modification. In order to test this hypothesis, we analyzed the level of ribose methylation of the target sites in 28S rRNA for each of the snoRNA in the corresponding clones using the approach proposed by Belin et al. (2009).

The method is based on termination of reverse transcriptase enzyme at 2'-O-methylated sites in the template RNA at low dNTP concentrations (Maden et al., 1995; Maden, 2001). In some cases, the use of high dNTP concentrations instead of decreased concentrations might yield higher specificity and more accurate results (Filippova et al., 2015). Two reactions of reverse transcription of the total RNA sample are performed in parallel: at optimal (1.0 or 1.5 mM) and suboptimal (3.0 or 0.01 mM) dNTP concentrations (Filippova et al., 2017). The length of the

FIGURE 5 | Kink-turn motif structure for wild-type U77 and potential structures for U77 mutants in 293FT-77-1. Insertion of 1 nucleotide in U77\_1ins is denoted by red "A." CS, canonical stem; NS, non-canonical stem.

reverse transcription product corresponds to the position of the 2'-O-Me in the template RNA. Therefore, the full-length cDNA product is amplified when using primers flanking the site of interest in case if the site is not modified while no amplification product is observed in case of truncated cDNA (in the presence of 2'-O-Me). Thus, the 2'-O-methylation level of the target site reversely correlates with the level of the PCR product.

The method of RT termination followed by PCR (Aschenbrenner and Marx, 2016) revealed a decrease in the 2'-O-methylation level of C3820 (~34% decrease), C4032 (~42% decrease), and A1521 (~60% decrease) 28S rRNA in 293FT-74-4, 293FT-75-2, and 293FT-80-1 monoclones, respectively, compared to the control cells (**Figure 7**). The only monoclone that demonstrated no changes in the 2'-O-Me level of the target rRNA site (A1521) was 293FT-77-1.

To confirm incomplete abrogation of the target site modification, we used independent approaches. The method of RNase H treatment followed by HPLC-MS/MS was used to verify the data on the 2'-O-methylation status of C4032 28S rRNA in the clone 293FT-75-2: a decrease in the 2'-O-methylation level was shown (**Supplementary Table 1 Figures 2A, B**). However, the method of partial alkaline hydrolysis demonstrated that the modification was not abrogated completely (**Supplementary Table 1 Figures 2C**).

Of special interest was to analyze the 2'-O-methylation status of the target rRNA nucleotide for U77 and U80 RNAs in the corresponding clones, since both snoRNAs share the same target, namely, A1521 28S rRNA. While RT-PCR-based method showed a decreased 2'-O-methylation level of the target nucleotide in 293FT-80-1, no changes were observed for 293FT-77-1 cells (**Figure 6**). The absence of changes for A1521 28S rRNA in 293FT-77-1 was confirmed by a modified RT-based approach, which has been developed by us earlier (Filippova et al., 2015) (**Supplementary Table 1 Figure 3A, Supplementary Table 2**). Partial alkaline hydrolysis demonstrated that mutations in U77 and U80 RNAs do not abrogate 2'-O-methylation of the target nucleotide completely (**Supplementary Table 1 Figure 3B**).

#### CRISPR/Cas9-Mediated Cleavage of *Gas5* snoRNA Results in Downregulation of the Host Gene and Formation of an Alternative Splicing Variant

In order to study the effects caused by CRISPR/Cas9-mediated mutations in the genes encoding snoRNAs on the level and maturation of the host gene lncRNA Gas5, we performed qRT-PCR analysis with the sets of primers complementary to various exons of *Gas5*. As a result, a decrease in the level of Gas5 RNA was shown for all of the obtained clones (**Figure 8**). It should be noted that the level of each of the studied *Gas5* snoRNAs altered independently of the others in the obtained monoclones (**Figure 4B**), which indicates the existence of an individual mechanism for regulation of the level of each of the *Gas5* snoRNAs.

We also observed the presence of a truncated splicing variant lacking some of the exons in 293FT-75-2 (**Figure 9A**). Two transcript variants of Gas5 lncRNA are detected in each of the obtained monoclones, as well as in control 293FT cells (**Figure 9A**). Both of these forms present naturally occurring transcript variants (NR\_152525.1, 660 nt; NR\_152530.1, 621 nt), which differ in the length of exon 7: the shorter transcript contains a truncated exon 7 region (**Figure 9B**). RNA-Seq data and Sanger sequencing of the alternative variant product revealed the presence of a mature Gas5 variant lacking exons 3 to 5 in the clone 293FT-75-2 (Figures 9C–E). The effect might be due to the presence of a region within *SNORD75* involved in regulation of splicing of the host gene transcript. Furthermore, the nature of mutations in *SNORD75* in the clone 293FT-75-2 indicates that such regulatory element is located within the chr1:173866903-173866920 region, which corresponds to the deleted sequence in 75-2\_9del and 75-2\_10del forms. Analysis of *Gas5* splicing events using rMATS (**Supplementary Data Sheet 3**) and JunctionSeq (**Supplementary** 

**Data Sheet 2, Supplementary Table 2**) revealed numerous changes in the splicing pattern of *Gas5* in 293FT-75-2, while few changes were observed for the other monoclones.

Further, we analyzed the data on RNA-binding factors interacting precisely with the above-mentioned region within Gas5 lncRNA precursor transcript according to the POSTAR database (Zhu et al., 2019). The following factors were found to be associated with the region in *SNORD75*: METTL3, YTHDF2, YTHDС2, CPSF4, CSTF2T, ELAVL1, FIP1L1, FMR1, HNRNPC, and SSB. Annotation of these proteins revealed that almost all of them were splicing regulatory factors. Therefore, an excision of the regions within Gas5 pre-lncRNA binding some of these proteins might result in formation of the detected products of alternative splicing (**Figure 9**). The presence of binding sites for METTL3/METT14 and a group of m6 A recognition factors in this region (**Figure 10**) suggests that regulation of Gas5 lncRNA processing is m6 A-dependent, while formation of alternative splicing products in 293FT-75-2 is due to deletion of one of the m6 A methylation sites. Moreover, analysis of the *SNORD75*

representing comparison of the relative number of reads (expressed as % of the maximum number of reads) for exons 3, 4, and 5 in each of the obtained monoclones and control cells. (E) Sanger sequencing of the alternative transcript lacking exons 3 to 5 in 293FT-75-2 monoclone.

structure demonstrated the presence of the required consensus element DRACH (GGACA in *SNORD75*) and the typical stemloop structure recognized by m6 A-methylating complexes, which is absent in 293FT-75-2 due to deletion of the METTL3 recognition region in *SNORD75* (**Figure 11**). Furthermore, bioinformatics analysis of the dataset presented in NCBI GEO (GSE56010) (Liu et al., 2015b) confirmed that the maturation pattern for Gas5 lncRNA alters in similar manner upon knockdown of METTL3, METTL14, and HNRNPC (**Supplementary Tables 2 and 5, Supplementary Data Sheet 1**). We found the most numerous alterations in exon

7\*—Alternative splicing variant of Gas5 RNA containing a truncated exon 7 region. (C) RNA-Seq data representing the number of reads for exons 2 to 9 of *Gas5* in the clones and control cells. Exons 3, 4, and 5 that are absent in the novel alternative transcript in the clone 293FT-75-2 are encircled by red rectangle. (D) Graphs

and junction coverage after HNRNPC knockdown in the region of 3–8 and 11 exons.

#### DISCUSSION

The presented data indicates that snoRNAs can be edited using CRISPR/Cas9 tools with generation of viable cell lines expressing mutant snoRNAs. Our experiments demonstrated that snoRNA genes can be edited accurately, selectively, and efficiently without affecting other snoRNAs encoded within the introns of the same host gene. Mutations can be targeted at the regions of boxes C, D, C,' and D' for generation of monoclonal cell lines expressing mutant snoRNA forms. In addition, the region of the Kink-turn area beyond the conserved elements can be also used as the target in snoRNA cleavage experiments. In our experiments, we achieved almost complete downregulation of the wild-type snoRNA forms and a minimum four-fold decrease in the level of mutant snoRNA forms in the obtained monoclonal cell lines compared to the wild-type snoRNAs.

Interestingly, a decrease, but no abrogation of the target rRNA site modification, was observed for all of the obtained monoclones (except for 293FT-77-1) (**Figure 7**, **Supplementary Table 1 Figures 2 and 3**). Such data indicate that all of the four *Gas5* snoRNAs have back up partners, which guide modification of the same site in case if their "partners" are downregulated for some reason. For instance, U80 backs up the modification of A1521 28S rRNA in 293FT-77-1 in the absence of a functional U77 form and vice versa. Indeed, analysis of the expression of U80 RNA in 293FT-77-1 demonstrated an almost two-fold increase in its level (**Figure 4B**). However, the overall level of mutant U77 RNA forms is decreased significantly in 293FT-80-1 (**Figure 4B**). The existence of more than one snoRNA that guide methylation of the same position is known for several rRNA sites (Watkins et al., 2002; Kehr et al., 2014). In addition, high-throughput RNA sequencing experiments indicate that numerous sites share complementarity with more than one snoRNA (Gumienny et al., 2016; Jorjani et al., 2016). Of course, most of these targets have been only predicted based on sequence complementarity but not yet verified. It is considered that, in case of changed cell conditions or altered expression of a specific snoRNA, another snoRNA backs up the modification for it (Kehr et al., 2014).

The obtained cell lines encoding snoRNAs with modified structure present a convenient and useful model for the study of metabolic pathways involving the target snoRNA. Thus, the presented cell lines, as well as similarly obtained cells, can be used for the study of the role of individual snoRNAs in the regulation of gene expression in human cells. Numerous studies demonstrate that snoRNAs are also involved in regulation of mRNA expression and alternative mRNA splicing (Falaleeva et al., 2016). In addition, some snoRNAs are processed into short miRNA-like derivatives, which perform fine-tuning of some pathways (Ender et al., 2008; Taft et al., 2009; Brameier et al., 2011; (Martens-Uzunova et al., 2015). Hence, the developed strategy allows one to reveal novel non-canonical RNA targets for small nucleolar RNAs, map functionally significant sites of modification within ribosomal RNAs, and create models for elucidation of ribosome heterogeneity phenomenon (Byrgazov et al., 2013; Krogh et al., 2016; Genuth and Barna, 2018). The absence of significant activation/inactivation of key cellular metabolic pathways indicates that snoRNAs can be cleaved selectively without deterioration of essential cellular processes.

Some snoRNAs are known to be splicing-dependent, while others mature independently of the host-gene transcript splicing (Hirose et al., 2003). Apparently, snoRNA genes can also contain splicing regulatory elements and elements important for binding of various splicing regulatory factors. Our study indicates that *SNORD75* contains such element in the following region chr1: 173866903-173866920. Using the POSTAR database, we have found a series of RNA-binding factors interacting precisely with the above-mentioned region within lncRNA Gas5 precursor transcript. One of the identified factors is N6-adenosinemethyltransferase METTL3/METTL14. Furthermore, there are factors, including YTHDF2, YTHDF3, YTHDС2, and HNRNPC, which are known to bind to an m6A-modified RNA only, that are also associated with this region, indicating that the region is subjected to m6A modification. Recruitment of m6A-recognizing-factors in *Gas5* introns suggests possible m6A modification of these regions. In the past decade, it has been established that m6A is a dynamic regulator of the processes of maturation, export, and degradation of pre-mRNA and lncRNA precursors (Wang et al., 2014; Liu et al., 2015b; Wang et al., 2015; Cao et al., 2016; Zhu et al., 2018; Coker et al., 2019). In addition, there are evidences indicating that snoRNA function can be also regulated through N6-methylation. Such modification of adenosine residue in the D box region, which forms a *trans* sugar/Hoogsteen base pair with guanine residue of the C box, prevents formation of the proper k-turn structure and further binding of the 15.5kDa protein; as a result, no snoRNPs are formed (Huang et al., 2017). Therefore, it is reasonable to suppose that splicing of Gas5 pre-lncRNA is m6A-dependent and regulated by methylation of one of the nucleotides in the chr1:173866903-173866920 region located in *SNORD75*.

Indeed, a decrease in the level of the host transcript, Gas5 lncRNA, has been noted for all of the obtained monoclones (**Figure 7**). We suggest that this is due to abrogated processing of Gas5 transcripts, which is due to mutations at the sites recognized by splicing regulatory factors, since changes in the intronic regions more likely affect maturation than transcription and stability of the lncRNA. It is intriguing that the binding sites for METTL3/METTL14 complex and m6 A recognition proteins were found within (or at least overlap with) the *Gas5 SNORD*s and other multi-snoRNA host genes encoding lncRNAs (**Figure 10**) (Liu et al., 2015a; Liu et al., 2018; Zhu et al., 2019).

The N6-methylation of adenosine residues at the sites located within introns by the METTL3/METTL14 complex is also known to impede splicing and result in slowly processed introns and alternative splicing (Louloupi et al., 2018). We believe that control of Gas5 lncRNA maturation is m6 A-dependent. This hypothesis is in accordance with our results of analyzing public dataset (GSE56010) on *METTL3*, *METTL14,* and HNRNPC knockdown (**Supplementary Table 5, Supplementary Data Sheet 1**) (Liu et al., 2015b). Using JunctionSeq (Hartley and Mullikin, 2016), we found numerous alterations in exon and junction coverage after METTL3 and HNRNPC knockdown (**Supplementary Data Sheet 1**). It is important to note that changes in the representation of *Gas5* exons and junctions are similar for the cells with knockdown of m6A-recognizing factor HNRNPC and for the cell line carrying mutations in **Supplementary Data Sheet 2** and **3**. Further, analysis of the METTL3-binding sites using POSTAR2 database revealed a binding site (173,866,922…173,866,902) for METTL3 within the deleted region in both mutants (75-2\_9del and 75-2\_10del) of *SNORD75* in 293FT-75-2 cells. Furthermore, a typical consensus "DRACH"motif (where D stands for G, A, or U; R stands for purine; and H is either U or A) is found in the deleted region of *SNORD75* (GGACA) (**Figure 11**). Thus, we suggest that recruitment of METTL3/METTL14 complex itself in this region plays a crucial role in determining the splicing pattern of Gas5 lncRNA transcript. It is still unknown, whether it is the N6Amethylation or the binding that regulates splicing of *Gas5*. We analyzed all known m6 A sites in the *Gas5* region, including *Gas5 SNORDs*, presented in MeT-DB V2.0 database, and have not found any methylation sites in *SNORD75*. However, its absence may be due to the fact that the modification changes the stability of this intron in the cells. A peculiar phenomenon was recently observed for another enzyme catalyzing N6A-methylation, METTL16: the dwell-time of the protein at the 3' UTR region of MAT2A mRNA was shown to have an impact on the target gene splicing (Pendleton et al., 2017). Interestingly, that, according to the authors, it is not the methylation itself that contributes to the splicing of the target MAT2A RNA but the occupancy time of the METTL16 at one of the hairpins in the 3' UTR of the target transcript (Pendleton et al., 2017). Thus, one can suggest that m6A-modifying factors regulate maturation of pre-mRNA and pre-lncRNA gene by binding to a specific intronic region even without causing N6A-methylation.

In the present study, changes in the sequence of the METTL3/ METTL14-binding site resulting in the deletion of the consensus motif in *SNORD75* resulted in formation of an alternative splicing product (**Figure 9A**). Taken together, our data suggests that sites responsible for METTL3/METTL14-dependent regulation of Gas5 lncRNA splicing are located within *SNORDs*.

#### CONCLUSIONS

Box C/D small nucleolar RNAs can be edited *via* CRISPR/Cas9 mediated cleavage at the regions near the conserved elements boxes C, C,' D, and D,' and specific downregulation of a target box C/D snoRNA can be achieved. The 2'-O-methylation level of the target

### REFERENCES


rRNA nucleotide can be modulated through CRISPR/Cas9-mediated knockout of the corresponding snoRNA. Deletion of the terminal region with disruption in the K-turn area even in preservation of the C and D box structures was shown to affect proper snoRNA processing and result in its downregulation. *SNORD75* contains an element for binding of splicing regulatory factors, the deletion of which causes the alterations of Gas5 pre-lncRNA maturation. In the current work, we show that METTL3/METTL14 might be among the factors regulating lncRNA maturation, and that *Gas5* splicing might be m6 A-dependent due to intronic *SNORDs*.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in ArrayExpress (www.ebi.ac.uk/arrayexpress), Acession E-MTAB-8269.

### AUTHOR CONTRIBUTIONS

JF and AM designed and mainly did the study under the supervision of GS and VV. EZ, RM, SM, and TG executed RNA-Seq protocol. DS, EZ, and KA performed analysis of RNA-Seq data. EB and DP provided assistance with RT-PCR experiments. DP performed HPLC-MS/MS analysis. JF, AM, and GS wrote the manuscript. All authors have read and approved the content of the manuscript.

### FUNDING

The study was supported by the RFBR grant No 18-29-07073 and partially (in method development) by State Budget Program (0245-2019-0001).

### ACKNOWLEDGMENTS

The research was performed using the equipment of Interdisciplinary centre for shared use of Kazan Federal University (Kasan, Russia) and SB RAS Genomics Core Facility (ICBFM SB RAS, Novosibirsk, Russia).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.01246/ full#supplementary-material

Is Associated with Tumor Progression of Human Breast Cancer Cells. *PLoS One* 4, e7147. doi: 10.1371/journal.pone.0007147


Cholesterol Trafficking. *J. Biol. Chem.* 288, 35703–35713. doi: 10.1074/jbc. M113.488577


of m6A in Splicing Efficiency. *Cell Rep.* 23, 3429–3437. doi: 10.1016/j. celrep.2018.05.077


**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 Filippova, Matveeva, Zhuravlev, Balakhonova, Prokhorova, Malanin, Shah Mahmud, Grigoryeva, Anufrieva, Semenov, Vlassov and Stepanov. 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.*

# Differential Expression of HERV-W in Peripheral Blood in Multiple Sclerosis and Healthy Patients in Two Different Ethnic Groups

Rachael Tarlinton1\*, Belinda Wang1 , Elena Morandi <sup>2</sup> , Bruno Gran<sup>3</sup> , Timur Khaiboullin<sup>4</sup> , Ekatarina Martynova<sup>5</sup> , Albert Rizvanov <sup>5</sup> and Svetlana Khaiboullina5,6

<sup>1</sup> School of Veterinary Medicine and Science, University of Nottingham, Loughborough, United Kingdom, <sup>2</sup> Centre de Physiopathologie de Toulouse Purpan (CPTP), Université de Toulouse, UPS, INSERM, CNRS, Toulouse, France, <sup>3</sup> Clinical Neurology Research Group, Division of Clinical Neuroscience, University of Nottingham School of Medicine, Nottingham, United Kingdom, <sup>4</sup> Republican Research and Clinical Center of Neurology and Neurosurgery, Kazan, Russia, <sup>5</sup> Insitute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, Russia, <sup>6</sup> Department of Microbiology and Immunology, University of Nevada, Reno, NV, United States

#### Edited by:

Marina A. Zenkova, Institute of Chemical Biology and Fundamental Medicine, Russia

#### Reviewed by:

Sudheer Kumar Ravuri, Steadman Philippon Research Institute, United States Martin Sebastian Staege, Martin Luther University of Halle-Wittenberg, Germany

\*Correspondence: Rachael Tarlinton

Rachael.tarlinton@nottingham.ac.uk

#### Specialty section:

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

Received: 03 December 2018 Accepted: 16 December 2019 Published: 30 January 2020

#### Citation:

Tarlinton R, Wang B, Morandi E, Gran B, Khaiboullin T, Martynova E, Rizvanov A and Khaiboullina S (2020) Differential Expression of HERV-W in Peripheral Blood in Multiple Sclerosis and Healthy Patients in Two Different Ethnic Groups. Front. Pharmacol. 10:1645. doi: 10.3389/fphar.2019.01645 Overexpression of the Human endogenous retrovirus W (HERV-W) group of inherited retroviruses has been consistently linked with Multiple Sclerosis (MS). However most of the studies on this link have focused on European genetic groups with a very high risk of MS and it is not clear that this relationship holds for all ethnic groups. This study examined via qPCR the RNA expression in peripheral blood of HERV-W (the multiple sclerosis associated retrovirus variant MSRV) of MS patients and healthy controls from two ethnic groups with very different risk rates of MS. Population one was derived from the UK with a Northern European genetic background and an MS risk rate of 108/100,000, population two was derived from the republic of Tatarstan, Russian Federation, with a mixed Russian (Eastern European) and Tartar (Turkic or Volga/Urals) population with an MS risk rate of 21-31/100,000. The Russian population displayed a significantly higher basal level of expression of MSRV in both healthy and MS individuals when compared to the British control population with a trend in the Russian population towards higher expression levels in MS patients than healthy patients.

Keywords: human endogenous retrovirus W, multiple sclerosis, multiple sclerosis associated retrovirus, human endogenous retrovirus, ethnicity

## INTRODUCTION

Multiple sclerosis (MS) is a chronic, progressive autoimmune neurological disease. The disease is often debilitating with few effective treatment options currently available. The underlying triggers for the disease are complex and be best summed up as: MS occurring in genetically susceptible individuals who are exposed to the "right" environmental triggers. A familial association is evident in some cases and some distinct restricted genetic groups (such as Sardinians) (Pugliatti et al., 2001) have a high risk. Extensive GWAS studies have demonstrated a compelling support for the HLA allele DRB1\*15:01 as the most significant genetic risk factor for MS, with smaller effects attributed to another 11 HLA alleles and 200 non HLA genes (Didonna and Oksenberg, 2017). Other risk factors for MS include gender (women are at higher risk than men), prior Epstein Barr Virus infection, Vitamin D levels (those with lower levels at higher risk).

The prevalence of MS also shows distinct regional variations with the highest rate in Northern European and North American populations. Also, prevalence increases with latitude in countries and this is thought to be linked to UVB exposure and subsequent vitamin D levels. With notable exceptions for certain populations such as the Sardinians and artic peoples with diets very high in fish oils (and vitamin D). Relapses of clinical disease are also associated with season (with a peak in spring) and low vitamin D levels (Harding et al., 2017). Immunoregulatory effects of vitamin D are believed to play role in MS pathogenesis. This assumption is based on data demonstrating distinct seasonal patterns in cytokine profile (and therefore basal inflammatory states) in a variety of disorders including MS (Blatt et al., 2017; Goodin, 2016; Watad et al., 2016).

Endogenous retroviruses (ERVs) are vestiges of retroviral infections from evolutionary history that remain in the host genomes. They make up to 8% of the human genome and are present in all vertebrates (Hayward et al., 2013). While not replicating as infectious viruses in humans they can be expressed in some individuals and are linked to the disease pathogenesis. There is a strong epidemiological link between the expression of the HERV-W retroviral family and Multiple Sclerosis (Morandi et al., 2017a). HERV-W is an attractive candidate for the link between the genetic and environmental factors in MS for a number of reasons. The envelope protein or mRNA of certain HERV-W loci are expressed constitutively in normal human tissues, particularly in the placenta (the syncytin locus), with certain HERV-W RNA sequences isolated from MS patient samples known as multiple sclerosis associated retrovirus (MSRV) (Nellaker et al., 2009; Bhat et al., 2011; Soygur and Moore, 2016). Interestingly HERV-W is upregulated in the brain of MS patients (Morandi et al., 2017a) when compared with healthy controls. Upregulation of HERV-W can be reliably induced by Epstein Barr and other viral infections (Nellaker et al., 2006; Mameli et al., 2012). The HERV-W Env protein also induces a pro-inflammatory responses and neuronal pathology similar to that found in MS (Perron et al., 2013; Morandi et al., 2017b). Intriguingly HIV patients receiving antiretroviral medication are seemingly protected from MS and there have been clinical trials of antiretroviral drugs targeted at HERV-W as MS therapeutics (Morandi et al., 2017b).

Despite this epidemiological data few studies have looked at whether there are differences in HERV-W expression between populations with high and low MS prevalence, or populations of distinctly different genetics, despite this being highlighted as a potential confounding factor in the use of HERV-W expression as a marker of MS disease (Morandi et al., 2017a). This study reports HERV-W expression in MS patients and healthy controls in two distinct populations with very different ethnic makeups and MS risk. The two study locations, the East Midlands in the UK and the Republic of Tatarstan in Russia are at approximately the same latitude but have a different climate, the main differences being in the temperatures of the coldest months (below 0 Celsius in Tatarstan) (Peel et al., 2007). The prevalence of MS is very different between the two sites with a rate of 97-108/100,000 in the UK site (Kingwell et al., 2013) and 21-31/100,000 for the Tatarstan site (Simpson et al., 2011). The ethnic mix of the populations also very different with the East Midlands 85% White British (Northern European), with a sizeable (7%) South Asian minority (Office for National Statistics, 2011) and Tatarstan, 40% Russians (Eastern Europeans), 53% Tatars and large minorities of other groups such as the Chuvashes (3%) (Volga, Ural/Turkic) (Federal State Statistics Service, 2010). Within the Tatarstan population there is a documented lower incidence of MS in the Tatar population when compared with the Russian (Bakhtiiarova and Magzhanov, 2006). Within the East Midlands population there is also a lower incidence of MS in the South Asian compared with the White British population (Albor et al., 2017).

### MATERIALS AND METHODS

### Human Blood Samples

Blood samples from patients attending Nottingham University Hospitals NHS Trust or the Republican Clinical Neurological Center, Republic of Tatarstan, Russian Federation were collected in PAXgene Blood RNA tubes (Qiagen). The Russian samples were shipped to Nottingham as PAXgene tubes. MS diagnosis was established according to the McDonald criteria (Polman et al., 2011). All patients and health controls (HC) signed informed consent for which ethical approval was obtained. In total 43 UK patients (21 HC and 22 MS) and 25 Russian patients (7 HC and 18 MS) gave samples. Patients and HC age, gender, and clinical status are presented in Supplementary Table 1. Ethnicity data where available is presented.

Ethics statement: Informed consent was obtained from each subject according to the clinical and experimental research protocol, approved by the Nottingham Research Ethics Committee 2 (Ref 08/H0408/167) and the Biomedicine Ethic Expert Committee of Republican Clinical Neurological Center, Republic of Tatarstan, Russian Federation (N: 218, 11.15.2012).

### RNA Extraction and cDNA Synthesis

RNA processing for all samples was performed in Nottingham. Total RNA was purified using the PAXgene Blood RNA kit (Qiagen) following the manufacturer's instructions. The RNA was treated with DNase to remove trace amounts of bound DNA. After the wash steps, RNA was extracted in the elution buffer provided with the kit and stored at -80°C. RNA concentration was determined by measuring the absorbance at 260 nm using NanoDrop ND-100 (Thermo Scientific). For making cDNA, 0.5 mg RNA samples, 2 ml Random hexamers and 1 ml dNTPs mix (stock solution 10 mM) (all Promega) were mixed, followed by a 5 min incubation at 65°C for first strand cDNA synthesis. A master mix containing RNase inhibitor (Promega, UK), DTT (0.1M) and 5X First-Strand Buffer was added, along with Superscript III RT (220 U/ml) (all from Invitrogen, UK). Negative controls replacing the RNA template or the RT with DNase/RNase free H2O were included. Samples were incubated as followed: 5 min at 95°C, 60 min at 50°C and 25 min at 70°C. cDNA was stored at -80°C.

#### qPCR

Relative quantification was performed using Ubiquitin C (UBC) and Ubiquitin conjugating enzyme E2D2 (UBE2D2) as housekeeping genes as per published literature on stable reference genes for use in PBMC in Multiple Sclerosis (Oturai et al., 2016). For detection of HERV-W nine hundred nM of probe/forward and reverse primer mix "MSRV" (Mameli et al., 2009) forward primer CTTCCAGAATTGAAGCTGTAAAGC, reverse primer GGGTTGTGCAGTTGAGATTTCC, Probe FAM-5′-TTCTTCAAATGGAGCCCCAGATGCAG-3′-TAMRA (Invitrogen custom Taqman assay) UBC and UBE2D2 (TaqMan gene expression assays, Invitrogen, catalogue numbers Hs05002522\_g1 and Hs00366152\_m1 both FAM-MGB probes) were used. All samples were run in duplicates. Agarose gel electrophoresis of the MSRV primers used in endpoint PCR produced amplicons of the expected size (166 bp) (Supplementary Figure 1) QPCR was performed using a BioRad CFX Connect (BioRad) and Faststart Universal probe master (Roche) in 96-well plates with the following cycling conditions: 10 min at 95°C and 40 cycles of 10 s at 95°C followed by 30 s at 60°C. A control sample not subjected to reverse transcription (a "no RT control") was included in each batch of samples and did not amplify in any instance. Primer efficiencies were calculated from the slope of a standard curve and were 87.2% for HERV-W, 96.8% for UBC and 93.8% for UBE202. Calculation of the relative amounts of HERV-W was performed using the two reference genes with one HC used as a calibrator. Any change in gene expression between patient cohorts and the calibrator HC patient was expressed using the following formula

$$\text{Relative gene expression} = \frac{\left(E\_{GOI}\right)^{\text{acroov}}}{\text{GeoMean}\left[\left(E\_{rcf}\right)^{\text{acrof}}\right]}$$

Where GOI = HERV-W, ref = UBC, UBE2D2, E = amplification factor (10[–1/slope]), DCT = Calibrator CT-Sample CT.

#### STATISTICAL ANALYSIS

Statistical analysis was performed using Graph Pad Prism 5, Kruskal Wallis test with Dunn's multiple comparison test post hoc testing. Data are presented as medians and interquartile ranges.

#### RESULTS

There was a significant difference in HERV-W relative expression between the populations (Russian HC, Russian MS, UK HC, and UK MS) (P < 0.0001 Kruskal Wallis) (Figure 1). On post-hoc testing (Dunn's multiple comparison test), there were significant differences (P < 0.05) between the UK and Russian populations but not between the MS and HC within the two populations, though there is a clear trend in the Russian group towards a higher HERV-W expression in the MS patients (Figure 1). The Russian patients were divided into those sampled at first presentation (who had not yet had disease modifying therapy) and those seen at follow up visits. There was a significant difference in these cohorts when compared to the Russian healthy controls (P = 0.0095 Kruskal Wallis) (Figure 2). On post-hoc testing (Dunn's multiple comparison test), there were significant differences between the Russian patients on primary presentation and Russian healthy controls but not the other two groups Though a trend, albeit with quite a bit of variability, towards highest HERV-W expression levels in those on primary presentation, an intermediate level of HERV-W expression in those on follow up visits and the lowest levels in healthy controls can be seen (Figure 2).

#### CONCLUSIONS

This study quite clearly shows differences in the expression levels of HERV-W between the UK and Russian populations that are statistically significant. While not statistically significant there is a clear trend in the Russian cohort (but not the UK cohort) towards a higher expression of HERV-W in MS patients than the healthy controls. Other factors such as gender did not display clear differences. The differences between the MS and healthy patients are in line with the reported increase in detection of HERV-W in

FIGURE 1 | Relative expression of HERV-W against the reference genes UBC and UBE2D2, calibrated against a healthy control (UK) sample. Medians and interquartile ranges are indicated by bars. HC, healthy control, MS, multiple sclerosis, UK, United Kingdom, RUS, Russian (N = 21 UK HC, 22 UK MS, 7 RUS HC, 18 RUS MS), (Medians = 0.98 UK HC, 0.58 UK MS, 11.51 RUS HC, 17.40 RUS MS), Kruskal Wallis P < 0.001, Dunn's Multiple comparison test P < 0.05 for differences between the Russian and UK cohorts but not between MS and HC. HERV-W, Human endogenous retrovirus W; UBC, Ubiquitin C; UBE2D2, Ubiquitin conjugating enzyme E2D2.

FIGURE 2 | Relative expression of HERV-W against the reference genes UBC and UBE2D2, calibrated against a healthy control (UK) sample. Medians and interquartile ranges are indicated by bars. RUS MS1 = Russian patients on primary presentation (before any disease modifying therapy) RUS MS2 = Russian patients on follow up visits (on a variety of disease modifying therapies) (N = 25, RUS MS1 = 7, RUS MS2 = 11 and RUS HC = 7) (Medians = 17.63 RUS MS1, 16.67 RUS MS2 and 11.51 RUS HC) Kruskal Wallis P = 0.0047, Dunn's Multiple comparison test P < 0.05 for differences between the RUS MS1 and RUS HC cohorts only. HERV-W, Human endogenous retrovirus W; UBC, Ubiquitin C; UBE2D2, Ubiquitin conjugating enzyme E2D2; HC, healthy control; MS, multiple sclerosis; UK, United Kingdom; RUS, Russian.

MS patients in other studies, summarised in the systematic review and meta-analysis in (Morandi et al., 2017a). One caveat is that this study examined HERV-W RNA from whole blood in PaxGene tubes whereas previous studies in the (Morandi et al., 2017a) metaanalysis used a variety of blood derivatives including PBMC, and plasma so the results are not directly comparable. However to our knowledge no previous studies have examined ethnic or population differences in HERV-W expression.

There are a number of potential explanations or confounding factors that could potentially explain these differences. The choice of an unstable housekeeping gene is a common cause of false conclusions from this type of qPCR study, however we performed stability experiments with 5 patients from each cohort (UK MS, UK HC, RUS MS, RUS HC) using a selection of genes from the published work on selection of stable reference genes in PBMC in MS and healthy patients (Oturai et al., 2016) with concordant results to the published work and are confident that this difference is not a technical error. The primer/probe combinations used here not span exon junctions so genomic DNA contamination is a possibility, though at no stage did any of the controls not subjected to reverse transcriptase amplify, indicating that DNA contamination was not present. Even if there were DNA contamination in these samples we do not really expect the copy number of the reference genes or HERV-W loci to vary substantially between individuals so the use of normalisation against the reference genes still should take this into account as a potential confounder. The two populations live at approximately the same latitude removing that as a source of variation in MS risk factors, largely leaving background genetics/ ethnicity as the potential explanation of this large difference.

It is also possible that the handling conditions during transport of the Russian samples to the UK, or RNA extractions performed at different time points could have resulted in systematic bias in the HERV-W RNA stability. These loci do not produce virions (which might protect viral RNA while cellular RNA is degraded) and it is hard to image what other processes would have selectively affected the HERV transcripts and not the reference genes. The use of multiple reference genes in relative qPCR studies such as this is specifically designed to balance variations in sample quality and as far as is possible with this study design is the best method currently available to account for this potential confounding factor.

Sample size is also another potential confounding factor and as always these findings will require verification by other groups and with other cohorts of patients. Increasing the sample size is unlikely to change the statistical significance of the differences between the two populations as these are really quite large (there is indeed no overlap in the confidence intervals) and power calculations based on the results in this study give quite trivial numbers (single digits) for confirming the differences between the populations. The differences in the MS patients versus the healthy controls in each population are more subtle and in the case of the British population would require several thousand participants to confirm with confidence, a clearly very different scale of study to the current one.

This study was not designed to stratify differences between stage of disease and HERV-W expression with all the British patients being in the remission phase of the disease and the Russian patients at varying stages with too few patients in each group to adequately compare stage of disease and HERV expression therefore the differences between the Russian patients at first presentation and at later stages of disease need to be taken with some caution. It is possible that stage of disease has also contributed to the differences between the UK and Russian populations, as there is some indication in the Russian cohort that those at first presentation and not currently being treated with disease modifying therapies had higher HERV-W expression, though without longitudinal studies comparing the same patient through relapse and remission this is hard to quantify.

There are several potential explanations for this difference in HERV-W expression; the first is that there may be polymorphisms in the number (presence or absence) or sequence of the HERV-W loci between the two populations. This may affect either the expression profile of the loci or their ability to be detected with the qPCR used in this study. Such polymorphism has recently been reported for HERV-W (Thomas et al., 2018). With the advent of long range deep sequencing technologies capable of distinguishing repetitive loci such as HERVs it may now be possible to determine if such polymorphisms exist between the UK and Russian populations and this is an obvious follow on study to the work presented here. Furthermore, epigenetic dysregulation or cell specific promotor activation of genes other than HERV-W loci (but that may interact with HERV-W expression) are other factors that require exploration in the population differences in HERV expression in the current study.

It is possible that expression of particular defective alleles may even be protective against the expression of deleterious alleles (or exogenous virus) as has been described in other species such as JSRV in sheep (Armezzani et al., 2014). Another possibility is variation in cellular factors that normally suppress HERV expression such as the Trim 5a and APOBEC 3G systems (Fadel and Poeschla, 2011).

Why the expression of a known risk factor for MS is higher in a population with a lower risk of the disease is puzzling, though it seems likely that other genetic or epigenetic factors must be in play here. There has been some work indicating that epitopes of the HERV-W envelope protein could be recognised by the HLA alleles DRB1\*1501 and DRB5\*0101 (Ramasamy et al., 2017). Therefore, it could be suggested that the failure of HLA allele recognition of the higher levels of HERV-W expression in the RUS MS cohort may account for some of the difference in MS risk. It is however clear that there are population differences in MSRV expression and that this needs to be taken into account when comparing patient cohorts for this risk factor for MS.

#### ETHICS STATEMENT

Informed consent was obtained from each subject according to the clinical and experimental research protocol, approved by the Nottingham Research Ethics Committee 2 (Ref 08/H0408/167) and the Biomedicine Ethic Expert Committee of Republican

#### REFERENCES


Clinical Neurological Center, Republic of Tatarstan, Russian Federation (N: 218, 11.15.2012).

#### AUTHOR CONTRIBUTIONS

RT and AR held the grant funding. RT prepared the manuscript. BW and EMo performed the laboratory work in BG's and RT's laboratory. BG, TK, and EMa collected the samples and patient information, SK provided the initial concept and facilitated the Russian sample collection. All authors edited and approved the manuscript.

#### FUNDING

This project was funded by a Royal Society - International Exchanges grant (number IEC\R2\170037). This study was also partially supported by the Russian Government Program of Competitive Growth of Kazan Federal University. AR was supported by state assignments 20.5175.2017/6.7 and 17.9783.2017/8.9 of the Ministry of Science and Higher Education of Russian Federation.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019. 01645/full#supplementary-material


retroviruses and multiple sclerosis: a systematic review and meta-analysis. PloS One 12, e0172415. doi: 10.1371/journal.pone.0172415


increasing risk. Acta Neurol. Scand. 103, 20–26. doi: 10.1034/j.1600- 0404.2001.00207.x


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 Tarlinton, Wang, Morandi, Gran, Khaiboullin, Martynova, Rizvanov and Khaiboullina. 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.